{"origin": "authored", "source_key": "authored:73faecdbe0aabacdee7c5ad9", "occurrence_id": "a60c919aa38e312e63fa8a8c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Joint Household Consumption Survey of 2021, assembled by WIPO, entered the archive in three\ntransfers rather than one, and each transfer was checked against the manifest that accompanied it.\nWhere a field arrived empty it was left empty rather than imputed, and the empty count was recorded\nalongside the record so that later users could see the gap without reopening the file.\n\n**4.2 Handling and retention.**\n\nThe working copy is held in the tabular form in which it arrived; no variable was renamed, and no\ncolumn was reordered to suit the reading environment. Two checksums accompany the material, one taken\nbefore the transfer and one taken after it settled, and the archive treats a mismatch as an\ninstruction to re-request rather than to repair. Requests for derived extracts are answered from the\nworking copy, never from the reading environment, so that a downstream edit cannot quietly become the\narchived version. Retention follows the schedule agreed when the collection was accepted, and the\nschedule is reviewed whenever the producing office changes its own policy. Access is logged, and the\nlog is reviewed once a year to confirm that the material has not left the environment it was promised\nto stay in.\n\nMention: \"Joint Household Consumption Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The paragraph describes how the 2021 survey assembled by WIPO was transferred, checksummed and retained in the archive.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:9c6d1bf7c1e63566690101dd", "occurrence_id": "f798865be37187937d82b99b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Employment shares by sector are reported for the National Working Conditions Registry (NWCR) without\nbeing reproduced from it, which matters because the registry records hours rather than sectors and the\ngrouping had to be built before any share could exist. The grouping follows the classification the\nregistry itself prints in its notes, applied to the raw activity descriptions.\n\n\nWhere a description matched two headings the record was assigned to the one carrying the greater share\nof stated hours, and roughly one in forty records fell to the residual heading on that rule. The\nresulting shares are therefore the author's construction and not a field of the registry, and they are\npresented to one decimal place to avoid implying a precision the assignment step did not have.\nComparisons across periods use the same assignment rule throughout, so a change in a share reflects a\nchange in the underlying records rather than a change in how they were sorted.\n\nMention: \"National Working Conditions Registry (NWCR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"Sector shares were computed from National Working Conditions Registry (NWCR) activity descriptions rather than taken from the registry.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:8a214eff7948ed7a326ea44e", "occurrence_id": "925174df411dcc087b6c2a51", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Longitudinal Host Community Survey (LHCS), compiled by the Universal Postal Union, is the source\nof both figures in the preceding table, and neither figure was adjusted before it was printed. The\nsurvey followed the same set of localities across its rounds, which is why the two shares can be read\nagainst each other at all; a fresh sample in the second round would have made the movement\nindistinguishable from a change in who was interviewed.\n\n\nValues are printed to the release's own single decimal place, and the rounding that\nfollows from that choice is left uncorrected, so a reader who adds the components may arrive at a\ntotal that differs from the printed one by a tenth of a point. No reweighting was applied on our\nbehalf, and the published weights were not re-derived, since doing so would have produced shares that\ncould no longer be traced back to the release a reader could obtain.\n\nMention: \"Longitudinal Host Community Survey (LHCS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Both table figures come verbatim from the Longitudinal Host Community Survey (LHCS) released by the Universal Postal Union.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:48b2689294954a893e789608", "occurrence_id": "a779fb2dcb04df16543c96ba", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Using the Joint Living Standards Microdata of 2015, the household was taken as the unit of analysis\nand the individual records were collapsed into it, because the consumption module repeats at the\nindividual level while the housing module does not. Collapsing required a rule for the few households\nwhose members reported inconsistent tenure, and the rule adopted was to take the value reported by\nthe household head, which is the convention the microdata documentation already recommends.\n\n\nEstimates are unweighted unless stated otherwise. The survey carries sampling weights, but applying\nthem changes the composition of the sample rather than only its representation, and the comparison\nhere concerns composition. Where a household reported no earned income the record was retained with a\nzero rather than dropped, since dropping it would have raised the mean by construction and left the\nreader with no way to see that the increase came from the treatment of a missing value.\n\nMention: \"Joint Living Standards Microdata\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"Household-level estimates were computed from the 2015 Joint Living Standards Microdata, collapsing individual records by a stated rule.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:e4f100bde85221d3fa19a3ae", "occurrence_id": "26033ec9a6abb628a5a19180", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Firm size distribution for Mauritania in 2015 was derived from the Coordinated Enterprise and Jobs Survey, which enumerates establishments rather than firms and therefore required a parent identifier before any size class could be assigned. Establishments without a usable identifier were treated as standalone firms, a choice that moves a small number of large employers into the smallest class and biases the distribution in a direction we can name but not measure.\n\nThe distribution is reported in bands rather than as a continuous measure, for the ordinary reason that the underlying counts are thin at the top, and also because the survey's own top-coded records would have made any finer cut an artefact of the coding rather than of the economy. Where the survey suppressed a cell for confidentiality the band was left blank rather than estimated, and the blank is counted in the denominator of the shares that follow it, so those shares read low by roughly the share of suppressed cells.\n\nNo seasonal adjustment was attempted, and the reference period is a calendar year. Two regions were fielded in a second window because of flooding, and since the second window falls partly outside the reference year we keep the survey's own assignment rather than re-assigning the records ourselves, which would have changed the composition of the affected bands without any way to test the change afterwards.\n\nMention: \"Coordinated Enterprise and Jobs Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"Firm size bands for Mauritania in 2015 were derived from the Coordinated Enterprise and Jobs Survey, carrying a known top-coding bias.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:f2c53391368fb9b2a3f33304", "occurrence_id": "84828fac80925dd4c5613ee0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The coverage gradient described in the section was produced from the National Immunisation Coverage Index for 2020, released by the World Food Programme, by grouping administrative units according to their reported distance from the nearest fixed post and then averaging within each group. The index supplies the coverage estimate and the distance enters only through the grouping, so the gradient is a property of how the units were bunched rather than of any model fitted to them.\n\nAveraging was unweighted, which gives a sparsely populated unit the same influence as a large one; weighting by population flattens the gradient noticeably, and that version is reported in the annex rather than in the text, because the flatter result is harder to describe in a sentence and the difference does not alter the direction of the finding.\n\nUnits whose estimate carried a flag for insufficient denominator were excluded rather than carried at the flagged value, and the exclusion removes about a tenth of the units, concentrated in the largest distance group. A reader who reinstates them should expect the outermost group to move toward the centre, which would weaken the gradient without eliminating it, and we would rather say that plainly than leave the exclusion to be discovered in a footnote nobody reads.\n\nMention: \"National Immunisation Coverage Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The coverage gradient was calculated from the 2020 National Immunisation Coverage Index issued by the World Food Programme, with flagged units excluded.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:4b6f79cf2f6d4b08b9b6391a", "occurrence_id": "39406b48fc6629e92c7e5dd3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The registration rules described here are those in force when the Nationwide Enterprise and Jobs System, Edition 4, was compiled for 2023, and the section does not otherwise engage with the material. Two of the rules have since been amended, and the amendments are noted only because they explain why a reader who consults the current register will not find the categories in the same arrangement, and might otherwise conclude that the counts reported here were drawn carelessly.\n\nMention of the system here is to establish that comparable coverage exists for the period, since the argument of the section rests on administrative records of a different kind and of narrower scope. No value, threshold or definition from the material has been carried into the discussion, and the edition number is given solely so that a reader can identify which rules were in force at the time of compilation rather than at the time of reading.\n\nWhere the current register and Edition 4 disagree about the boundary of a category, this section follows Edition 4, since that is the version the figures were produced under. The choice is stated here rather than in the notes because it changes which records fall inside the smallest category, and a reader comparing our counts with a recent publication will otherwise see a discrepancy that has nothing to do with either source being wrong.\n\nMention: \"Nationwide Enterprise and Jobs System\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Nationwide Enterprise and Jobs System, Edition 4, for 2023 is cited only as context; no value or definition is taken from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:5c8effebb84498fb849cd31e", "occurrence_id": "0eeb8592abeb410a6098ff07", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Both training-placement figures in this section are taken directly from the Integrated Skills and Employment Report of 2022, published by the Commerce Department, and neither was recomputed. The report presents them as shares of a cohort, and the cohort is the same in both rows, which is why they can be subtracted without further qualification.\n\nWe keep the report's own rounding and its own cohort definition, including its practice of counting a placement that ended within a month as a placement. That convention inflates the first figure relative to the retention literature, and a reader comparing the two should know the direction of the difference rather than discover it later. Where the report prints a dash the cell is reproduced as a dash; we did not read a dash as a zero, and the distinction matters in the two rows where a dash replaces an otherwise expected value.\n\nThe report's regional breakdown was not used, because its regional categories were redrawn between editions and the redraw is not documented anywhere we could obtain. Combining the current regions with figures compiled under the earlier boundaries would have produced a table in which a change of boundary and a change of outcome were indistinguishable, so the national figures are reported alone and the regional comparison is left out rather than approximated.\n\nMention: \"Integrated Skills and Employment Report\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Both training-placement shares are quoted verbatim from the 2022 Integrated Skills and Employment Report published by the Commerce Department.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:c3f6fc3502705b0c85cc9c51", "occurrence_id": "66e07ffe83ef045bddeee003", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The seasonal outlook quoted above is taken from the Integrated Seasonal Forecast Index (ISFI), Fifth Phase, and is reproduced without modification. The index is published as a probability rather than as a rainfall amount, and the paragraph keeps that form deliberately: converting it to millimetres would have required a distributional assumption the release does not state, and the resulting number would have looked like a measurement rather than a forecast.\n\nTwo caveats travel with the figure and are repeated here because they are easy to lose in transcription from a table into running text. The phase matters, since categories are recalibrated between phases and a category label is therefore not comparable across them, which is the most common error we have seen in secondary use of this series. The grid is the one the release chooses, and coarser neighbourhoods are averaged before the category is assigned, so a reader working from a finer map will not always reproduce the quoted value for a single point, and should treat the nearest cell rather than the nearest pixel as the unit being described.\n\nMention: \"Integrated Seasonal Forecast Index (ISFI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The quoted seasonal outlook is reproduced from the Integrated Seasonal Forecast Index (ISFI), Fifth Phase, and kept as a probability.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:14df4820f162b52c899796a9", "occurrence_id": "ff272a5b5bda644d8f1ca17c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Coverage of the antenatal module was cross-checked against the Joint Maternal and Newborn Report of 2020, published by ZamStats, which reports a comparable share for an overlapping period. The two figures are not identical, and the difference sits within the range one would expect from a narrower reference window on our side rather than from a disagreement about who belongs in the population.\n\nThe check was performed once, on the published table, and was not repeated against the underlying records, which we do not hold and did not request. Its purpose was limited and should be read as limited: to confirm that the module was not reporting coverage an order of magnitude away from the figure a reader could find elsewhere, since an error of that size would have indicated a mislabeled variable rather than a definitional difference.\n\nNo adjustment was made to either figure to bring them closer, and no attempt was made to reconcile the definitions, which differ in the treatment of facilities that opened partway through the year and in whether a woman counted once or twice when she attended two sites. Both choices are defensible and neither is ours to settle, so the two figures are reported side by side with their own definitions attached.\n\nMention: \"Joint Maternal and Newborn Report\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"validate\", \"summary\": \"The 2020 Joint Maternal and Newborn Report from ZamStats was used only to cross-check antenatal coverage, not as a source of figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:800634681d05f168ec408726", "occurrence_id": "0e39422aead223faaecd1118", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Remittance dependence by region was constructed from the Annual Migration and Remittance Index, Round 2, for Burkina-Faso in 2019, by taking the share of households reporting at least one inflow and dividing it by the share reporting a member abroad, which yields a crude intensity measure rather than a share of income. The construction is ours; nothing resembling it appears in the index.\n\nThe division is not innocuous. Where migration is rare but strongly selected toward households with steady transfers, the ratio exceeds one and reads as though remittances arrived more often than households had migrants, which is an artefact of the denominator rather than a finding. We therefore cap the reported measure at one and mark the capped cells, rather than allowing the ratio to carry a meaning it does not have.\n\nThe round matters because the sampling frame was rebuilt between rounds, and a comparison with the first round would conflate a change in who was reachable with a change in what households received. We make no such comparison here. Regional boundaries follow the index as published, including its treatment of the two eastern regions that were enumerated together for security reasons, which means those two regions should be read as a single observation.\n\nMention: \"Annual Migration and Remittance Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"Remittance intensity by region was computed from the Annual Migration and Remittance Index, Round 2, for Burkina-Faso in 2019, not taken from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:f63b2e1ecba3c3ca667354be", "occurrence_id": "35d5b5fa1ef0319781a249ad", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: One component of the settlement estimate uses the Combined Population Grid Survey (CPGS) for Colombia in 2023, Version 9, which supplies enumerated counts on a fixed grid rather than administrative totals. Its contribution is confined to the rural strata, where the administrative registers are thinnest, and the urban component comes from elsewhere.\n\nGrid cells are joined to settlements by centroid, which misassigns cells that straddle a boundary; the affected cells are a small minority but they concentrate along the edges of larger towns, so the error is not random with respect to settlement size. We did not attempt to redistribute them, since doing so would have required a population surface of our own and would have replaced a known bias with an unverifiable correction.\n\nVersion 9 is stated because the grid geometry changed from the previous version, and counts compiled on the earlier geometry cannot be aligned cell by cell with counts compiled on this one. Where the survey reports a cell as unavailable rather than zero we preserve the distinction, and unavailable cells are excluded from the denominators of any share reported from the rural strata.\n\nMention: \"Combined Population Grid Survey (CPGS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The rural component of the settlement estimate was derived from the Combined Population Grid Survey (CPGS), Version 9, for Colombia in 2023.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:d22095dfc994fe81862d6991", "occurrence_id": "0573a00f4379bcc147c3a564", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The registration counts were compared with the Multi-Topic Refugee and Migrant Registry, maintained by IEG, which reports a comparable total for the same facilities over a slightly longer window. The two totals agree to within the margin we would expect from that difference in window alone, and the comparison stops there.\n\nWe did not reconcile the two because we hold neither underlying record, and a comparison of published tables can establish only that neither figure is wildly wrong. Its value is limited accordingly, and we state it in a single sentence rather than a paragraph so that the reader is not left with the impression that a definitional question had been settled. The registry's own treatment of duplicated registrations, which we cannot inspect, is the largest remaining source of disagreement between the two counts and is left entirely to the producing office.\n\nNo figure from the registry is reproduced in this section, and none of its category definitions were adopted for our own tabulations, which follow the definitions used elsewhere in the report so that the tables remain internally consistent.\n\nMention: \"Multi-Topic Refugee and Migrant Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Multi-Topic Refugee and Migrant Registry maintained by IEG was used only to sanity-check registration totals; no figure is taken from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:d41a682dbfa449191fe0c2f7", "occurrence_id": "df99c93bd05da4ca13d54067", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The methodology section cites the Regional Public Works Dataset (RPWD), compiled in 2021 by the IMPACT Initiative, solely to indicate that geocoded records of this kind exist for the period under discussion. The argument proceeds from municipal contract files, and no quantity, coordinate or coverage claim in it derives from the dataset.\n\nThe citation is retained because a reader unfamiliar with the literature may reasonably ask whether location information of this sort has ever been collected for these districts, and the dataset answers that question without our having to argue the point. Its release notes describe a collection effort with a wider remit than ours, and the two should not be treated as alternative measurements of the same thing.\n\nWe did not verify the coordinates it publishes, did not adopt its classification of works, and did not compare its coverage with ours, since the comparison would have required a shared definition of what counts as a public work and no such definition exists between the two efforts.\n\nMention: \"Regional Public Works Dataset (RPWD)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Regional Public Works Dataset (RPWD), 2021, from the IMPACT Initiative is cited only to note that geocoded records exist; no value is used.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:e7252171af74e9c8b52b54fe", "occurrence_id": "86fe1d35fb7d6c15e83ac9cb", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Enrolment by programme type is taken from the Combined Vocational Training Dataset for 2020 and is reproduced without adjustment. The dataset reports enrolment as recorded at the reporting date rather than as sustained through the year, and the figures in the table should be read with that in mind, since a programme that closed partway through the year still contributes its opening count.\n\nWe did not attempt to net out early leavers, because the dataset does not record exit dates and any correction would have been an assumption presented as a measurement. The consequence is that the shorter programmes are over-represented relative to any survival-weighted count, and the direction of that bias is constant across programme types, so the ordering in the table is unaffected even if the levels are not.\n\nWhere a provider reported to two regional offices the dataset counts the enrolment once, and we rely on that de-duplication rather than reconstructing it, since the provider identifiers needed to do so are not released outside the producing office.\n\nMention: \"Combined Vocational Training Dataset\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Enrolment by programme type is quoted directly from the Combined Vocational Training Dataset for 2020, keeping its reporting-date convention.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:cfcc4aea7e847c734e5818f6", "occurrence_id": "c7f88257b008b985448bc98b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Private participation in early schooling was estimated from the Nationwide Non-State Education Report by classifying each provider according to the fee schedule it reports and then summing enrolment within the classes. The report does not publish a participation share, and the share in the table is therefore our construction rather than a published figure.\n\nThe classification is where the estimate is made. A provider charging a nominal fee and receiving public subsidy falls between two of our classes, and we assigned such providers to the subsidised class rather than the independent one; moving them to the independent class raises the headline share by roughly two points, which is a larger effect than the change we set out to measure and is reported here rather than buried.\n\nEnrolment figures are taken as printed. Where the report gives a range for a region we took the midpoint, and where it gives a figure qualified as provisional we retained the figure and the qualification together, so that the reader can see which part of the total rests on a number the publishers themselves expected to revise.\n\nMention: \"Nationwide Non-State Education Report\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"Private participation in early schooling was computed from the Nationwide Non-State Education Report by classifying providers; the share is not published in it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:541b6624528663dd43df2d60", "occurrence_id": "27a84d8b6fce5edd3d2c0709", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Coverage of the transfer programme was tested against the National Cash Transfer Registry Collection, maintained by the Government of India, which reports recipient counts for an overlapping period at a coarser geography. The test was coarse as well: we asked only whether our count sat on the same order of magnitude as theirs once the populations implied by the two geographies were taken into account.\n\nIt does, and that is the whole of the finding. Neither figure was adjusted toward the other, and no attempt was made to align the reference periods, which differ by several months and would have required an assumption about the pace of enrolment that neither source supports. A reader who wishes to make the comparison precisely should use the registry's own period definitions rather than ours.\n\nNo value from the collection appears in any table in this report, and its category definitions were not borrowed for our tabulations, which use the definitions in force for the programme being evaluated.\n\nMention: \"National Cash Transfer Registry Collection\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The National Cash Transfer Registry Collection from the Government of India was used only to confirm the order of magnitude of coverage.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:dde585bc256605d41a56f656", "occurrence_id": "d3bc84598ebfc0014482085f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Income concentration in the metropolitan areas was derived from the Urban Household Income and Expenditure Index, First Version, published by IEA for 2019, by grouping households into quintiles and computing the share held by the top two. The index publishes the quintile boundaries but not the shares, so the shares are ours.\n\nBoundaries are used as published rather than recomputed, which means our quintiles are those the index defines and not those implied by the population weights we apply later. The inconsistency is deliberate and one-directional: recomputing boundaries under our weights would have shifted the top share by an amount comparable to the change we report, and we preferred to keep one quantity fixed rather than two.\n\nFirst Version is stated because the index's later versions revise the treatment of in-kind transfers, and a reader comparing our share with a figure drawn from a later version will find a gap that has nothing to do with the underlying distribution. Where the index reports a grouped top category we treat it as a single observation at its lower bound, which understates concentration.\n\nMention: \"Urban Household Income and Expenditure Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"Top-quintile income shares were computed from the Urban Household Income and Expenditure Index, First Version, by IEA for 2019.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:75b8cfda639f39236c44845a", "occurrence_id": "d6788d11bbc05a6b01d8e068", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Annual Household Consumption Survey of 2023, released by Latinobarómetro Corporation, is mentioned here only because the consumption aggregate it produces is the usual point of comparison for the figures discussed in this section, and a reader may wish to consult it. Our estimates are built from administrative records, and nothing in the table is derived from that survey or from any of its published tables.\n\nWe note the difference in reference period because it is the first thing that confuses a comparison: the survey's collection window and our administrative window are offset, and the offset is long enough that a price change across it accounts for a visible part of any gap between the two series. That is a caution about comparing them, not a claim that ours is better.\n\nThe survey's treatment of in-kind receipts is also worth the reader's attention, since it differs from the treatment adopted throughout this report, and the difference is larger than the difference between the two estimates for the most recent period shown.\n\nMention: \"Annual Household Consumption Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Annual Household Consumption Survey of 2023 from Latinobarómetro Corporation is cited for context; no value in the table comes from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:a434563c8fe41d3cc69d3da6", "occurrence_id": "e31ba6a63876b2bda767e2e3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Regional Household Consumption Survey (RHCS) was received as nine separate extracts over four months, and the register describes what arrived rather than what was requested. Each extract was inventoried on receipt, and the inventory records the fields present, the number of records, and the date the extract stopped changing, which is the only date the archive can honestly assert.\n\nNo extract was merged into a common schema at the point of deposit. Merging is what a downstream analyst does, and performing it at the archive would have made the merged file the object of record while silently discarding the differences between extracts that make such a merge questionable in the first place. The nine files therefore remain nine files, and the register points to each of them separately.\n\nRequests for a consolidated view are answered with the individual extracts and the inventory rather than with a derived file, and the requester is told which extracts have not been superseded and which were replaced by a later delivery covering the same scope, since that distinction is invisible once the files are concatenated.\n\nMention: \"Regional Household Consumption Survey (RHCS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The paragraph describes how the Regional Household Consumption Survey (RHCS) was deposited as nine extracts and inventoried rather than merged.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:895ec183bddc170c10bfbff2", "occurrence_id": "426072d734e0e4b6dae1a687", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Both figures in the discussion of documentation gaps are taken from the Coordinated Statelessness Index (CSI), and neither was recomputed before publication. The index reports them as shares of an adult population, and the two rows share that denominator, which permits the subtraction shown in the third column without any further adjustment on our part.\n\nWe preserve the index's own treatment of the categories, including its practice of placing a person holding documentation of uncertain validity in the undetermined category rather than the documented one. That choice raises the second figure relative to a stricter reading, and the effect is large enough that a reader comparing our numbers with those of a country statistical office should know which convention produced them.\n\nThe index's regional breakdown was not used, since its regions are constructed for comparability across countries and do not correspond to the administrative units on which the rest of this report is organised. Mixing the two would have produced totals that disagree by construction rather than by chance.\n\nMention: \"Coordinated Statelessness Index (CSI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Both documentation-gap figures are reproduced from the Coordinated Statelessness Index (CSI), keeping its uncertain-documentation convention.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:6fb181609b2535a3eadb1b7d", "occurrence_id": "15db793501fdcaae83f22e09", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The establishment-level analysis draws on the Regional Enterprise and Jobs Microdata released in 2020 by the U.S. Census Bureau, which supplies the employment band and the sector code for each record. It is used here as a source of observed values rather than as a frame, and the population is defined by the survey design rather than by what the extract happens to contain.\n\nEmployment bands are used as published. Because the bands are open at the top, any mean employment figure computed from them depends on the value assigned to the top band, and we assigned the lower bound, which biases every mean downward while leaving the ordering of sectors intact. A reader who prefers a different assignment will obtain larger means but, working from the same records, will obtain the same ranking.\n\nRecords without a sector code were retained in the totals and excluded from the sector tabulations, so the sector rows do not sum to the total row, and the difference is reported as a line in the table rather than distributed across the sectors.\n\nMention: \"Regional Enterprise and Jobs Microdata\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The establishment analysis uses observed values from the 2020 Regional Enterprise and Jobs Microdata released by the U.S. Census Bureau.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:b83c0469bfd61bd0a2b68bae", "occurrence_id": "d73670bc049ba4968246c600", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Regional Working Conditions Collection of 2019, held under CONARE, was accessioned after two earlier deliveries were rejected for truncation, and the accepted delivery is the third. The accession record notes the rejections because they explain a gap in the holdings that would otherwise read as a loss.\n\nRecords were counted on arrival against the count stated in the delivery note, and where the two disagreed the delivery was returned rather than repaired. This is slower than fixing a truncated file, and it is adopted because a repaired file cannot later be distinguished from a file that arrived intact, whereas a returned delivery leaves a durable record of what was wrong with it.\n\nThe collection is held in the delivered encoding, which is not the encoding the reading environment prefers, and no conversion was applied at deposit. Conversion is available on request and is logged as a derived use, so that a reader who encounters the material in another encoding can trace when it changed and who asked for the change.\n\nMention: \"Regional Working Conditions Collection\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The paragraph describes how the 2019 Regional Working Conditions Collection under CONARE was accessioned after two deliveries were rejected.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:2c2b83ab139834fad1315013", "occurrence_id": "78bf36c080b892c6ff0f2972", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The distributional comparison for Azerbaijan in 2016 rests on the Continuous Household Consumption Index, Seventh Version, from which we took the published deciles and computed the ratio between the highest and the lowest. The index does not publish that ratio, so the value quoted in the text is ours.\n\nRatios of this kind are sensitive to the treatment of the lowest decile, which the index defines against a population that includes households reporting no consumption at all. Excluding those households raises the ratio by a visible margin, and we did not exclude them, in order to keep the denominator identical to the one the index uses for its own published shares.\n\nThe seventh version is specified because the index revised its imputation procedure for missing income items between versions, and a reader who reproduces our ratio from a later release will obtain a different number for reasons entirely internal to the index. We made no attempt to reproduce the earlier imputation ourselves, lacking the item-level data it would have required.\n\nMention: \"Continuous Household Consumption Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The top-to-bottom decile ratio for Azerbaijan in 2016 was computed from the Continuous Household Consumption Index, Seventh Version.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:07ba8e5244014f992021ce51", "occurrence_id": "d8793095cbbd60e3661965d0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: That a panel of this kind exists is worth noting, and the Longitudinal Poverty and Vulnerability Survey carried out in 2020 under Landsat is the nearest instance of it. This section does not use its estimates, its weights or its definitions, and the citation is offered only so that the reader can see that repeated observation of these households has been attempted elsewhere and by other means.\n\nWe mention the panel rather than cite its results because our argument concerns the difficulty of attrition in this setting, and the design of the panel bears on that difficulty in a way its published estimates do not. No figure from it appears in any table here, and none of the comparisons drawn in this section should be attributed to it.\n\nA reader wishing to place the two efforts side by side should note that the panel's rounds are separated by a longer interval than the administrative cycles described above, which is a difference in design rather than in quality, and one that limits how directly the resulting series can be compared.\n\nMention: \"Longitudinal Poverty and Vulnerability Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The 2020 Longitudinal Poverty and Vulnerability Survey under Landsat is cited only as evidence that a panel design exists; no values are used.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:571ecc0f2ac7b5af57174b89", "occurrence_id": "decf0606a842a85b4608e466", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The household size figures quoted for Romania in 2018 come from the Decennial Socioeconomic Survey (DSS) and are reproduced as published, including the survey's practice of counting a student in residential accommodation at the parental address. That convention produces larger households than a register-based count of the same population would produce, and the difference is concentrated in the larger towns.\n\nWe did not adjust the figures toward a register-based definition, because the adjustment would have required assumptions about which students live away from home that the survey does not record and that we could not have tested. It seemed better to report a figure with a known convention attached than to report a figure whose construction no one could check.\n\nThe tabulation we built uses the household as enumerated, so a change in the convention between surveys would appear in our series as a change in household size rather than as a change in the survey, and any reader extending the series backwards should test for that before drawing a conclusion from the first difference.\n\nMention: \"Decennial Socioeconomic Survey (DSS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"Household size figures for Romania in 2018 are reproduced from the Decennial Socioeconomic Survey (DSS), with its student-residence convention intact.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:dcfbbf872e11e616ce265d39", "occurrence_id": "4ac89d0a82a81124347ab901", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The facility counts used in the annex are those released by the Continuous Facility Assessment Survey, taken as published and without any reclassification on our part. The survey enumerates facilities that were operating at the time of the visit, which is a different thing from facilities that were licensed, and the distinction matters wherever the annex is read against an administrative count.\n\nWe resisted the temptation to reconcile the two, since a reconciliation would have required access to licence records we do not hold and a rule for facilities that are licensed but closed, on which the two sources take opposite positions silently. The annex therefore reports the survey's own counts and notes the difference in a single line rather than resolving it.\n\nWhere the survey reports a facility as temporarily closed we retain it in the count, following the survey's own convention, and a reader who excludes such facilities will obtain smaller totals in the regions where closure was most common during the collection window.\n\nMention: \"Continuous Facility Assessment Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"Facility counts in the annex are reproduced from the Continuous Facility Assessment Survey, keeping its operating-at-visit definition.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:386e9af6feca951d64b47daa", "occurrence_id": "687dbcb60ac89318648f58a9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The shoreline change reported for the exposed districts was produced from the Household Coastal Erosion Dataset (HCED) by assigning each surveyed household to the nearest transect and averaging the reported change within that assignment. The dataset supplies the transect values; the household-to-transect assignment is ours and is where the estimate is made.\n\nNearest-transect assignment is crude where transects are widely spaced, and in two districts the spacing exceeds the distance over which change varies along the coast, so the household-level variation reported for those districts should be read as transect-level variation presented at a finer resolution than it possesses. We report the spacing alongside the result for this reason.\n\nHouseholds more than a fixed distance from any transect were dropped rather than assigned to the nearest one, and the dropped households are not a random subset: they sit further from the coast, which is precisely where the reported change is smallest, so the exclusion biases the average toward the larger values.\n\nMention: \"Household Coastal Erosion Dataset (HCED)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"Shoreline change by district was derived from the Household Coastal Erosion Dataset (HCED) using a nearest-transect assignment we constructed.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:cbabed33fdcff3205848da3f", "occurrence_id": "74ef2d32c215dd7f9cc7ddba", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The two well-being figures set out in the preceding table are taken from the Urban Socioeconomic Index (USI) and are reproduced without recomputation. The index reports them as normalised scores on a scale of its own construction, and we have not translated them into percentages, units or categories, since no conversion is published and any conversion we invented would have implied a comparability the scale does not have.\n\nWe keep the index's own treatment of missing components, which are imputed by the publishers before the score is formed, and we do not attempt to undo that imputation. A reader who prefers complete-case scores will obtain different values, particularly for the districts where component reporting was least complete, and the direction of the difference is not the same in every district.\n\nRankings drawn from the table should be read with the index's own statement of measurement uncertainty in view, since several districts are separated by less than that uncertainty and the ordering between them is not a finding worth defending.\n\nMention: \"Urban Socioeconomic Index (USI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Both well-being figures are quoted unchanged from the Urban Socioeconomic Index (USI), retaining its normalised scale and imputation.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:a82943eda737895da0654588", "occurrence_id": "accd8006e091f11f820974bd", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Completion rates by cohort were derived from the Longitudinal Vocational Training System for 2017, operated by UNESCO-IBE, by counting enrolments that reached a recorded completion event and dividing by the enrolments open at the start of the cohort. The system records events rather than rates, so every rate in this section is ours.\n\nThe denominator is the point of difficulty. Enrolments transferred between providers remain open in the originating record until a closure event arrives, which can be more than a year later, and a cohort observed at a fixed date therefore contains enrolments that have already finished elsewhere. We did not attempt to close them early, since doing so would have required a transfer rule the system does not implement.\n\nRates are reported for cohorts that have had at least the nominal programme duration in which to complete, and the two most recent cohorts are omitted on that basis rather than reported at a value that would understate completion by construction.\n\nMention: \"Longitudinal Vocational Training System\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"Completion rates by cohort were computed from the 2017 Longitudinal Vocational Training System operated by UNESCO-IBE, which records events rather than rates.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:a57684b3e31b5b2acadabf5c", "occurrence_id": "162ed9e76fbdc30fab8a88a9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Annual Skills and Employment Report was catalogued on arrival in Bolivia, where the receiving office took custody of the files before any assessment of their contents had been made, and the catalogue entry records custody rather than completeness. Two accompanying annexes were missing at that point and were requested separately.\n\nThe catalogue describes the material in the form in which it was received: files are listed individually, no attempt is made to present them as a single object, and the relationship between the main report and its statistical appendices is expressed as a note rather than as a structure. This is deliberate, because the appendices were revised independently and a single object would have had to carry a version that none of its parts possesses.\n\nAccess is granted at the level of the individual file, and the log records which files were requested together, which has proved more useful than any statement of use could have been in telling us which parts of the material are actually read.\n\nMention: \"Annual Skills and Employment Report\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The paragraph describes how the Annual Skills and Employment Report was catalogued and released file by file in Bolivia.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:63303352091db9178fad6840", "occurrence_id": "8e21e8b3c2812fbc54bba3ce", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Elevation is not a variable in this study, and the Annual Settlement and Elevation Index of 2022 is named here only because a reader may wonder why settlement lists in this region appear incomplete, which is a question about how settlements were defined rather than about anything the index contains. No value, threshold or classification from it enters the analysis.\n\nThe index is cited because it uses a minimum-population rule that differs from ours, and a reader who checks our list against it will find entries missing that are not in fact missing. The two rules are not reconcilable without a concordance that does not exist, and the difference is left as a caution rather than resolved.\n\nWe did not adopt the index's naming conventions, so a settlement appearing under one name here may appear under another there, and the index remains the better guide to official spellings for anyone requiring them.\n\nMention: \"Annual Settlement and Elevation Index\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The Annual Settlement and Elevation Index of 2022 is mentioned in passing to explain a naming difference; no value is used.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:ac7ad963347262896f86e2e1", "occurrence_id": "f7cd080d546ef38a0161686e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Movement estimates in the second panel are taken from the Continuous Forced Population Registry for 2018 and are reproduced without adjustment, including the registry's practice of counting a household once at its point of departure rather than at each site it passes through. A reader comparing our totals with site-level counts will therefore find ours smaller, and the difference is a matter of where the count is taken rather than of who was counted.\n\nWe did not redistribute the counts across sites, because the registry does not record the order of movement and any allocation we made would have been a guess presented as a measurement. The consequence is that the second panel cannot be read as a picture of site pressure, and it is not offered as one.\n\nWhere the registry marks a record as reported by more than one office, we take the earliest report and ignore the later ones, following the registry's own guidance, and we make no attempt to detect a record that was reported twice under different household identifiers.\n\nMention: \"Continuous Forced Population Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"Movement estimates in the second panel come from the Continuous Forced Population Registry for 2018, keeping its point-of-departure counting rule.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:ee4897b3a00fcf24a6639749", "occurrence_id": "64d8a3e2d7550da921573471", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Postings and vacancies reported in the tables are those recorded in the Coordinated Teacher Workforce Registry (CTWR) for 2021, maintained by World Bank/GFDRR, and neither figure was recomputed from underlying records. The registry is the only source we used for these two quantities, and the tables should be read as a restatement of it rather than as an independent estimate.\n\nA vacancy in the registry means a post that is funded and unfilled at the reporting date, which excludes posts that are unfunded and excludes teachers assigned to a post they do not occupy. Both exclusions matter in the districts where the report draws attention to shortage, and neither can be undone from the published data.\n\nWe did not adjust for the reporting lag, which the registry acknowledges and does not quantify, and we did not interpolate between reporting dates. Where a district submitted no return for the period, the cell is left empty rather than carried forward from the previous period, so a reader should not read an empty cell as continuity.\n\nMention: \"Coordinated Teacher Workforce Registry (CTWR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Postings and vacancies are restated from the Coordinated Teacher Workforce Registry (CTWR) for 2021, maintained by World Bank/GFDRR.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:e455fa7d8ce1757d402e9196", "occurrence_id": "041a262b0e918773e9051582", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The extract held for Republic of Zambia was taken from the Multi-Topic Food Security System (MFSS), operated by the World Bank Group, and the holding record describes the extract rather than the system: it notes which tables were taken, in what order, and which were re-requested after a checksum mismatch on the first attempt.\n\nNothing in the holding record asserts that the extract is complete, and it would be wrong to read it that way. Completeness is a property of the system as a whole, which we do not hold and cannot inspect; the record exists so that a later user can see exactly what this office possessed on a given date and can distinguish that from what the system contains.\n\nThe extract was not modified after receipt. Two tables arrived with a column order that differs from the published layout, and rather than reorder them we recorded the difference, since reordering would have destroyed the only evidence that the two deliveries came from different vintages of the same table.\n\nMention: \"Multi-Topic Food Security System (MFSS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The paragraph describes the held extract for Republic of Zambia from the Multi-Topic Food Security System (MFSS) by the World Bank Group.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:b5ded423dc00b1c1285637fd", "occurrence_id": "0813dc5cb92f96041ca4f97c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The welfare figures reproduced in this section are those published in the Combined Living Standards Survey (CLSS) by the Bureau, taken without recomputation, and we have deliberately not reweighted them even though the survey's own weights are available, because reweighting would have produced figures that no longer match anything a reader can obtain and verify.\n\nThe survey reports a design effect for its principal estimates, and we repeat it here rather than in a footnote, because the figures in this section are frequently quoted as though they carried the precision of a census. They do not, and the design effect is the most economical way of saying so in a single sentence.\n\nWhere the survey revised an estimate between its preliminary and final releases we use the final release throughout, and where a figure appears only in the preliminary release we say so at the point of use rather than leaving the reader to discover that part of the series rests on a number the publishers themselves later replaced.\n\nMention: \"Combined Living Standards Survey (CLSS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"Welfare figures are reproduced from the Combined Living Standards Survey (CLSS) published by the Bureau, without reweighting.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:544b3e1942bfbd57e5b2917b", "occurrence_id": "c5db732e803db1a96b6a51d5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The plot-level counts set out below are drawn from the Urban Agricultural Household Registry (UAHR), which records a household's claim on a plot rather than the area actually cultivated, and the distinction is the largest single reason our counts exceed those of the cadastral office. We reproduce the registry's counts rather than the cadastral ones because the analysis concerns claims, and the claim is what the surrounding argument is about.\n\nWe did not attempt to reconcile the two counts, and we did not compute a conversion factor between them, since the relationship between a claim and a cultivated area varies by district in a way that no published source documents. A reader who requires cultivated area should use the cadastral series and expect a systematic difference.\n\nDuplicate claims arising from household splits are retained as separate records, following the registry's own convention, which inflates the count of small plots relative to any household-level enumeration of the same neighbourhoods.\n\nMention: \"Urban Agricultural Household Registry (UAHR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"Plot-level counts are drawn from the Urban Agricultural Household Registry (UAHR), which records claims rather than cultivated area.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:b574f27cc166fd50c8481a4f", "occurrence_id": "fef01c5c8289ba90d0f4bffa", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The administrative figures underlying the district comparison are those released through the Annual Socioeconomic Registry, published by TRMM, and they are used exactly as released. No district was reclassified, no series was rebased, and no missing value was filled from an adjacent period, since each of those operations would have made the resulting table impossible to check against the source a reader can obtain.\n\nTwo districts were merged by the publisher for confidentiality in the most recent period and appear as a single row. We kept them merged rather than apportioning the combined figure back to the two, which would have required a population assumption the publisher did not supply and would have produced two numbers that look like observations.\n\nWhere the registry reports a figure as revised we use the revised value and note the revision, and where a figure has been withdrawn we omit it entirely rather than reproducing a value the publisher no longer stands behind.\n\nMention: \"Annual Socioeconomic Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The district comparison rests on figures released through the Annual Socioeconomic Registry published by TRMM, used exactly as released.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:2a816e84df31a56f8397745c", "occurrence_id": "173b716b50db0146e93ad4bb", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The hazard categories used to group the sites are those assigned by the Combined Climate and Flood Risk Index, and we adopted them without alteration rather than deriving our own, partly to preserve comparability with the assessments that cite the index and partly because we had no independent basis for reassigning a site.\n\nAdopting the categories means adopting their revision history, which is not published. A site that appears in the highest category here may have been placed there by a rule that has since changed, and we have no way of knowing which sites are affected. The groupings should therefore be read as a description of how the sites were classified at the time we read them, and not as a statement about the hazards themselves.\n\nWe did not merge categories to obtain larger groups, even where a group contains few sites, because the resulting shares would have mixed categories whose definitions differ more than their sizes do.\n\nMention: \"Combined Climate and Flood Risk Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"Hazard categories used to group the sites are adopted unchanged from the Combined Climate and Flood Risk Index.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:b1f2876617b714720e24fd40", "occurrence_id": "cc4c91b3c4e29c9beb6f83ad", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Threshold exceedances for Tamil Nadu in 2024 are reported from Wave 7 of the Decennial Heat and Cold Extremes Registry, and the count is the registry's own rather than a recount from station records. The registry defines an exceedance as a day on which the recorded maximum passes the station-specific threshold, and the thresholds are set locally, which is why two stations at similar latitude can produce different counts for the same heat spell.\n\nWe did not standardise the thresholds, because doing so would have required the station-level series that the registry summarises but does not release, and a reader who prefers a common threshold should treat the counts here as upper bounds for the cooler stations and lower bounds for the warmer ones.\n\nWave 7 is specified because the registry revised the treatment of days with incomplete observations between waves, and a count drawn from an earlier wave is not comparable with one drawn from this wave even for the same station and the same year.\n\nMention: \"Decennial Heat and Cold Extremes Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Threshold exceedances for Tamil Nadu in 2024 are reported from Wave 7 of the Decennial Heat and Cold Extremes Registry.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:642f5f965f4f60d8cce668f8", "occurrence_id": "f380575e5cfadaf69238f757", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The stunting estimates by age band were derived from the Annual Nutrition and Mortality Report (ANMR) of 2023 by classifying measured children against the reference the report adopts and then computing the share falling below the second standard deviation within each band. The report publishes the mean z-score by band but not the shares, so the shares are ours.\n\nComputing shares from published means is not possible, and we did not attempt it; the shares were formed from the child-level records released with the report, which are subject to a different licence and are held separately. A reader working only from the published tables will not be able to reproduce them, and we say so here rather than implying that the calculation is a simple restatement.\n\nChildren with no measured height are excluded rather than imputed, and the excluded children are more often from the households with the weakest attendance, which biases the reported shares downward in the bands where attendance is lowest.\n\nMention: \"Annual Nutrition and Mortality Report (ANMR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"Stunting shares by age band were computed from child-level records accompanying the Annual Nutrition and Mortality Report (ANMR) of 2023.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:babf0fc594e8bfe6b0e79824", "occurrence_id": "d39164badd033d75d0134182", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Participation rates quoted for the two cohorts were produced from the Joint Vocational Training Index, Third Wave, published for 2016, by dividing recorded enrolments by the relevant age population and expressing the result as a share. The index publishes the enrolments; the population denominators come from elsewhere, so the rate is ours and inherits the uncertainty of both inputs.\n\nThe denominator choice is not neutral. Using the single-year age population raises the rate relative to using the five-year group, and we used the five-year group because the enrolments include students recorded at ages outside the nominal range, though not by much. A reader who prefers the narrower denominator will obtain consistently higher rates for every cohort shown.\n\nThe third wave is cited because enrolment counting moved from headcount to registration between waves, and the change is large enough that a series drawn across waves would show a decline that reflects the counting rule rather than the participation it is meant to measure.\n\nMention: \"Joint Vocational Training Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"Participation rates were computed from the Joint Vocational Training Index, Third Wave, for 2016, using population denominators from a separate source.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:66d0daf08f2cae748688739a", "occurrence_id": "581f4f208ea42aa6b096c3ff", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Rapid Household Income and Expenditure Registry (RHIER) of 2020 is referred to here because its sampling frame is the one most readers in this literature will recognise, and the note exists to explain why our sample is smaller than the figures they may have in mind. The analysis uses administrative records, and no estimate, weight or published table from the registry contributes to any number in this report.\n\nThe registry was collected over a shorter window than the administrative cycle, and that difference is enough to explain a visible part of any gap between the two series during a period of price movement. We mention it as a caution about comparison rather than as a defence of our own figures, which have weaknesses of their kind that are described in the notes.\n\nWhere the registry and our records disagree about the composition of the poorest group, we have preferred to present both descriptions rather than to choose between them, since the disagreement is itself of interest and nothing in either source settles it.\n\nMention: \"Rapid Household Income and Expenditure Registry (RHIER)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Rapid Household Income and Expenditure Registry (RHIER) of 2020 is cited to explain a sample-size difference; no value is used.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:408c81a3a40a4b0125a49707", "occurrence_id": "c6c584c3b961ed4d9bbe0a3f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Landings and farm output shown for Chile are those printed in the Longitudinal Fisheries and Aquaculture Report (LFAR), and we have not converted either series into a common unit, since the report keeps them in the units in which they were logged and the conversion factors it offers apply to the national total rather than to the individual entries.\n\nThe two series therefore appear side by side without being summed, and the reader who wishes to add them should do so from the report's own annex rather than from our table, because the annex carries the factors together with the caveats attached to them and our table would carry only the numbers.\n\nWhere the report revises a previous year we use the revised figure and note the revision, and where it reports a figure as an estimate pending log verification we retain the word estimate at the point of use. A reader comparing our table with an earlier publication of the same series will find differences in the older years that arise from this practice rather than from any error in either publication.\n\nMention: \"Longitudinal Fisheries and Aquaculture Report (LFAR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Landings and farm output for Chile are reproduced from the Longitudinal Fisheries and Aquaculture Report (LFAR) in their original units.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:9df1836e43890e41c169395e", "occurrence_id": "1ee783624ffd6ec831a62626", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Consumption aggregates by division were computed from the Continuous Household Consumption System (CHCS) by summing the item groups it defines within each division and expressing the result per adult equivalent. The item groups are the system's; the adult-equivalence scale is ours, taken from the scale used elsewhere in the report so that the tables remain internally consistent.\n\nThe scale matters more than it appears to. A scale that weights children more heavily lowers the aggregate for the divisions with larger households, which are also the divisions the section is about, and the ordering of the two lowest divisions reverses under the alternative scale. We report the reversal rather than suppressing it, since the choice of scale is a judgement and not a measurement.\n\nRecords flagged by the system as partially complete are retained, and their item coverage is lower, which biases the aggregate downward in the divisions where the flag is most frequent. We did not scale those records up to compensate, because the system does not indicate which items are missing.\n\nMention: \"Continuous Household Consumption System (CHCS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"Consumption aggregates by division were computed from the Continuous Household Consumption System (CHCS) using an adult-equivalence scale we chose.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:49e9180b26dbef450fa39a18", "occurrence_id": "18f2f6e13ac0c3d8eeff70c3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The station counts used in the analysis were checked against the Periodic Heat and Cold Extremes Index for 2016, which reports the number of stations contributing to its published series. The two counts differ by a small number of stations, and the difference is explained by stations that contributed to the index but were excluded from our analysis for gaps in the daily record.\n\nThe check was performed to confirm that we had not overlooked a substantial part of the network, which is the only conclusion it supports. We did not reconcile the two lists station by station, since the index does not publish the identifiers, and a reader who requires the exact overlap should request it from the publishers rather than infer it from our table.\n\nNo value from the index appears in this section, and its thresholds were not adopted in place of the ones described in the method, which follow the definitions used by the national service rather than those used by the index.\n\nMention: \"Periodic Heat and Cold Extremes Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The station count was cross-checked against the Periodic Heat and Cold Extremes Index for 2016; no value from the index is used.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:b7ea5bd7102f8c5615e02f14", "occurrence_id": "74356b1eb1e5a9f457e03fb6", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The expenditure shares reproduced for Lao PDR are those held in the Decennial Household Income and Expenditure Registry, published by IIASA, and they are shown as received rather than as we would have constructed them. In particular we did not reallocate the imputed value of home-produced consumption, which the registry places in the food group and which several writers place in a separate group.\n\nThat placement moves roughly a fifth of the total for the rural divisions, so the food share reported here will read high to a reader accustomed to the alternative convention. We kept the registry's placement because the same registry supplies the comparison figures in the annex, and mixing conventions between the two tables would have made the comparison meaningless.\n\nWhere the registry reports a share for a group we reproduce that share rather than deriving it from the components, so that rounding differences do not accumulate across the rows of the table, and the residual row is reported explicitly instead of being absorbed into the largest group.\n\nMention: \"Decennial Household Income and Expenditure Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Expenditure shares for Lao PDR are reproduced from the Decennial Household Income and Expenditure Registry published by IIASA, keeping its imputation placement.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:c0b393655671912d0476dd08", "occurrence_id": "62c033a9b217e451096ef3e1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Changes in cover between the two dates were computed from the Rapid Land Cover Survey (RLCS), Phase 5, carried out in 2017 by ENIA, by taking the difference in the class share assigned to each mapped unit and reporting the result where the classification confidence exceeded the survey's own threshold. Units below the threshold were excluded rather than assigned, and the exclusion is the largest source of uncertainty in the table.\n\nThe excluded units are not scattered evenly: they concentrate along the margins between two classes, where change is most likely to be occurring, so the reported change is a conservative estimate of the total. We state the direction of that conservatism rather than quantifying it, since quantifying it would have required a confidence assignment of our own.\n\nPhase 5 is specified because the class definitions were consolidated between phases, and a change computed across a phase boundary would partly record the consolidation rather than the land. We did not attempt to unconsolidate the classes, lacking the pre-consolidation labels for the later date.\n\nMention: \"Rapid Land Cover Survey (RLCS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"Cover change per mapped unit was computed from the Rapid Land Cover Survey (RLCS), Phase 5, by ENIA in 2017, excluding low-confidence units.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:a2437ba4e5bc813bdb28bb7d", "occurrence_id": "4e24663aba8d849f25aed14d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Average class size by phase was derived from the Nationwide Teacher Workforce Registry (NTWR), Phase 3, by dividing enrolled pupils by teaching groups within each school and averaging across schools. The registry supplies enrolment and group counts; the average is ours, and it is an unweighted average of school-level values rather than a ratio of totals.\n\nThe distinction is material. Averaging across schools gives a small school the same weight as a large one, and because enrolment is concentrated in the larger schools the unweighted average understates the class size a pupil actually experiences. The weighted figure, which we report in the annex, is higher in every phase, and the difference is largest where small schools are most numerous.\n\nPhase 3 is stated because the registry changed the way it records part-time teaching groups between phases, and a comparison with an earlier phase would attribute a change in recording to a change in staffing. We made no comparison across that boundary.\n\nMention: \"Nationwide Teacher Workforce Registry (NTWR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"Average class size by phase was computed from the Nationwide Teacher Workforce Registry (NTWR), Phase 3, as an unweighted average across schools.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:d32ce06c8e6c45bf2d85d841", "occurrence_id": "5122b6bd094d26f803f665a7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The denominators used for the coverage rates for China in 2018 were taken from the Rapid Maternal and Newborn Census (RMNC), and the rates themselves were computed by us from our own numerator and that denominator. The census therefore supplies the population base and nothing else, and no coverage figure published by the census is reproduced here.\n\nUsing one source for the denominator and another for the numerator produces a rate that belongs to neither, and we flag this at each point of use rather than in a general note, because the effect is not uniform across the table. Where the census undercounted the group in question, as it is generally held to have done in the denser districts, the resulting rate is too high by an amount we cannot estimate from published material.\n\nThe census's own definitions of the age groups were adopted unchanged so that the rates can be compared with the census's population figures directly, and we did not regroup the ages to match the categories used elsewhere in the report.\n\nMention: \"Rapid Maternal and Newborn Census (RMNC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"Coverage-rate denominators for China in 2018 come from the Rapid Maternal and Newborn Census (RMNC); the rates themselves were computed by us.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:da770de5400e684425705c3d", "occurrence_id": "5db43b262f0f0640aa7ac4ce", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Our reading scores were compared with the Periodic Early Grade Reading Index (PEGRI) for 2024, published by Latinobarómetro Corporation, which reports a comparable proficiency share for the same grades. The comparison was made once, on the published figures, and it is reported here in full: our share is higher, and the difference is of a size that a difference in assessment method could plausibly account for.\n\nWe did not attempt to harmonise the two assessments, since harmonisation would have required the item-level results of the index, which are not public, and a comparison made at the level of the published share is the most that can honestly be done. The reader should treat the comparison as a check that our figure is not absurd rather than as evidence that it is correct.\n\nNo value from the index is reproduced in any table, and its proficiency thresholds were not substituted for the ones described in our method, which follow the curriculum rather than the index.\n\nMention: \"Periodic Early Grade Reading Index (PEGRI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"Reading scores were cross-checked against the 2024 Periodic Early Grade Reading Index (PEGRI) from Latinobarómetro Corporation; no value is adopted.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:7735cbb665b910b2293ed8f6", "occurrence_id": "5cf0b404587120ddae362663", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Fertility and childhood illness indicators in this section are those released in the Fourth Release of the Multi-Topic Demographic and Health Registry for 2024, and they are reproduced without recomputation. The release supersedes an earlier one that circulated for several months, and where the two differ we use the later figure and say which table it comes from, because the earlier release is still cited in the literature this section engages with.\n\nWe did not recalculate the indicators from the underlying records, which we hold under a licence that does not permit redistribution of derived products, and the tables are therefore a restatement of the release rather than an independent estimate. That is a limitation of the tables and not of the underlying work.\n\nWhere the release reports an indicator for a population subgroup without publishing its denominator we leave the cell empty, since inferring the denominator from a related table would have produced a rate whose construction we could not defend to a reader who asked.\n\nMention: \"Multi-Topic Demographic and Health Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Fertility and childhood illness indicators are restated from the Fourth Release of the Multi-Topic Demographic and Health Registry for 2024.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:75d1fb134ce1d23af3e8aa66", "occurrence_id": "6cd325f3544deb9799261e5c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The method section refers to the Periodic Multidimensional Poverty Index for 2015 only to note that a deprival-based measure of the same population exists, since a reader may reasonably ask why we did not use one. The discussion of thresholds proceeds from monetary measures, and no component, weight or cutoff from the index is used anywhere in this report.\n\nThe index was not adopted because its deprivation cutoffs are set for a national population and the analysis here is regional, where the same cutoff can classify a majority or a minority of households depending on the district. That is an argument about fit rather than about merit, and it does not extend to the index's own national estimates, which remain the best available for their purpose.\n\nWe did not compute a comparable measure ourselves, since doing so would have required indicators the household records used here do not contain.\n\nMention: \"Periodic Multidimensional Poverty Index\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The Periodic Multidimensional Poverty Index for 2015 is cited as context in the method section; no component or cutoff from it is used.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:a891947db75b879d69db5392", "occurrence_id": "1f7dd2f6be79ec0efa4a20d7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Tenure composition by agro-ecological zone was derived from the Integrated Land Use and Tenure Survey of 2018 by classifying each holding according to the instrument the holder reported and summing within the zone. The classification is ours; the survey records the instrument verbatim and does not publish a typology.\n\nThe typology is where the result is made, and it is stated in full in the appendix rather than summarised, because several categories are adjacent and a reader who disagrees with a single boundary can recompute the affected rows without redoing the whole classification. Merging the two weakest categories would move the headline composition by more than the change the section reports.\n\nHoldings reported by a holder other than the person interviewed were classified on the reported instrument, which is less reliable than a document inspection, and the affected holdings are concentrated among the smallest, so the composition of that group carries more uncertainty than the table conveys.\n\nMention: \"Integrated Land Use and Tenure Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"Tenure composition by zone was derived from the Integrated Land Use and Tenure Survey of 2018 using a typology we built and publish in full.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:22591aa9ef8cf59253b7fc7a", "occurrence_id": "d3938369357542dfa190c280", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The facility counts underlying the coverage rates for Ukraine were checked against the Urban Maternal and Newborn Collection (UMNC), which reports the number of facilities providing the relevant service over a partly overlapping period. The counts agree closely enough that we proceeded, and the check is reported here because a reader who repeats it with a different period boundary may not obtain the same agreement.\n\nWe did not merge the two sources or average them, and no denominator from the collection enters any rate in this section. The purpose of the comparison was to detect an error of magnitude rather than to refine a value, and a comparison designed for one purpose should not be read as serving another.\n\nWhere the collection includes a facility type our own frame excludes, the difference is noted in the accompanying table rather than adjusted away, since the exclusion reflects the scope of the service being evaluated and redefining it would change what the rates mean.\n\nMention: \"Urban Maternal and Newborn Collection (UMNC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"Facility counts behind the Ukraine coverage rates were cross-checked against the Urban Maternal and Newborn Collection (UMNC); no value is adopted.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:7c7157157778cf5158620c73", "occurrence_id": "4598e2482c0a2662fb252282", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Periodic Social Protection Collection (PSPC) for 2019 is named in the notes because it is the usual reference for administrative data of this kind, and the note exists to prevent a reader from assuming that our figures were drawn from it. They were not: the analysis uses national administrative extracts, and no value, definition or coverage statement in this report originates with the collection.\n\nWe mention it specifically because its harmonisation rules are widely applied, and a reader who applies them to our extracts would obtain a different series from the one published here. The difference would arise from the rules rather than from the underlying records, and it is worth stating plainly that we did not apply them.\n\nThe collection's country notes were read during preparation and informed the way the gaps in our own extracts are described, which is a influence on the presentation rather than on the numbers, and we would rather say that than imply a derivation that did not occur.\n\nMention: \"Periodic Social Protection Collection (PSPC)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Periodic Social Protection Collection (PSPC) for 2019 is cited in the notes to rule out a derivation; no value is taken from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:7accb0520cf475c36c509033", "occurrence_id": "471d83a27ee052035f143867", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Lengths and surface classes reported for Johor in 2019 are those held in Version 8 of the Nationwide Transport Network Registry, reproduced without adjustment. The registry records length by segment rather than by route, and the figures in the table are the sum of segments within each district, which double-counts nothing but does count a shared segment once for each district it touches.\n\nWe did not attempt to apportion shared segments, because the registry does not indicate the share of a segment lying in each district and any apportionment we invented would have been invisible in the resulting table. Districts containing many boundary segments therefore carry slightly inflated totals, and the effect is largest for the two smallest districts shown.\n\nVersion 8 is specified because surface classes were consolidated in an earlier version and a small number of classes from that version remain in use in the published series, so a comparison across versions will attribute a change of classification to a change in the network.\n\nMention: \"Nationwide Transport Network Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Road lengths and surface classes for Johor in 2019 are reproduced from Version 8 of the Nationwide Transport Network Registry.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:f9f9f4c7a73098f64c5c762a", "occurrence_id": "a6a270fb22b069ece71db7f2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Margins reported by farm size class were computed from the Combined Farm Economy Index (CFEI) by taking the gross margin each holding reports and dividing it by the utilised area, then averaging within the class. The index publishes neither the margin nor the area at the class level, so every figure in the row is ours rather than the index's.\n\nAveraging the ratio differs from taking the ratio of averages, and we report the former because it describes a typical holding, whereas the latter describes a typical hectare. In the two largest classes the two measures diverge substantially, and the section discusses the divergence rather than choosing between the numbers, since the question the section asks is about holdings.\n\nHoldings reporting zero utilised area were excluded rather than treated as missing, since a zero area is a reporting error that would make the ratio undefined, and the excluded holdings are listed in the appendix so that a reader can see which records were dropped and check whether they matter.\n\nMention: \"Combined Farm Economy Index (CFEI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"Margins by farm size class were computed from the Combined Farm Economy Index (CFEI) as an average of holding-level ratios.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:3a5ae02aa04497f3e22fa89c", "occurrence_id": "84aeadd15d00b1b6a25cb5ea", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Qualification mix by school type was derived from the Decennial Teacher Workforce Registry (DTWR) for 2020, held by TCC, by classifying each teacher against the highest qualification recorded and computing the share within each school type. The registry records qualifications as reported by the employer, and the classification into our four levels is ours.\n\nThe classification is not one-to-one. Several qualifications recorded in the registry straddle two of our levels, and we assigned each to the lower level, which understates the share holding a degree and overstates the share holding a certificate. The direction is constant across school types, so the comparison between them survives the choice even if the levels do not.\n\nTeachers holding more than one qualification are counted once at their highest, following the registry's own guidance, and a teacher recorded without a qualification is placed in the unqualified category rather than excluded, since exclusion would raise the qualified share in the school types where records are least complete.\n\nMention: \"Decennial Teacher Workforce Registry (DTWR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"Qualification mix by school type was computed from the 2020 Decennial Teacher Workforce Registry (DTWR) held by TCC, using a four-level classification we defined.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:68fb7cba6b01ddf01b61cf5e", "occurrence_id": "5b9dc2f98d4a224dbaddd1e2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The enterprise counts used for the informal share were checked against the Periodic Informal Sector System (PISS) for 2023, and the check is described here in full because it is the only external comparison our central figure receives. The system reports a count of informal units for a period overlapping ours, and the two counts agree to within a margin we regard as acceptable given how differently the two efforts define an enterprise.\n\nWe did not reconcile the definitions and did not adjust either count toward the other. Reconciliation would have required access to the system's frame, which is not published, and an adjustment made without it would have produced a figure that could be defended only by pointing at the adjustment itself.\n\nNo value from PISS appears in our tables, and its classification of activity was not adopted in place of the one described in the method. The comparison establishes that our count is not wrong by an order of magnitude, and we claim nothing beyond that.\n\nMention: \"Periodic Informal Sector System (PISS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The informal enterprise count was cross-checked against the 2023 Periodic Informal Sector System (PISS); no value from it enters the tables.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:2c2f72e85a05a044607df019", "occurrence_id": "12d44ddeedffaf46e1168810", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Coverage for Houston in 2017 is reported as published in the Continuous Immunisation Coverage Survey, produced by Treasury, and we have not recomputed it from the underlying records. The survey presents coverage as a proportion of a administrative target rather than of the enumerated child population, and that choice raises the reported figure relative to a population-based denominator.\n\nWe reproduce the survey's denominator rather than substituting our own because the target is the quantity the programme manages against, and a reader consulting this section for programme purposes would find a population-based figure useless. A reader needing the other denominator should expect a lower number and should obtain it from the survey's own technical annex.\n\nWhere the survey reports a coverage figure above its own target we reproduce it without capping, since capping would silently convert a measurement into a statement about plausibility, and the notes attached to the release explain why such figures occur.\n\nMention: \"Continuous Immunisation Coverage Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Coverage for Houston in 2017 is reproduced from the Continuous Immunisation Coverage Survey produced by Treasury, keeping its target denominator.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:acc378ad8efb244b4ee4d112", "occurrence_id": "dfa2e60401c05ff3c8c4d4b7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The corridor figures quoted for the United Kingdom in 2020 are those released by the Joint Migration and Remittance System, taken without adjustment. The system defines a corridor by the pair of countries at its ends and pays no attention to the route between them, which means a figure reported here for a given pair may include transfers that passed through a third country entirely.\n\nWe did not attempt to refine the routes, since the system does not publish them and no other source available to us records them at the level of granularity the section requires. The corridor label should therefore be read as a description of the endpoints and not of a path.\n\nWhere the system suppresses a corridor for confidentiality we leave the row out rather than estimating it, and the suppressed rows are counted in the narrative totals only where the system itself reports a total from which they can be recovered by subtraction.\n\nMention: \"Joint Migration and Remittance System\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"Corridor figures for the United Kingdom in 2020 are taken from the Joint Migration and Remittance System without adjustment.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:196e845187d70febd831c421", "occurrence_id": "3ca421cc62cd7d376b27489e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: A panel of informal enterprises exists, and the Longitudinal Informal Sector System (LISS), run in 2020 by Teikoku Databank, is the one a reader is most likely to have encountered. It is named here because its existence bears on a question about feasibility that the section raises, and nothing in the analysis is drawn from it.\n\nWe did not use its estimates because its frame is drawn from a register of enterprises that have voluntarily registered, which is precisely the population our own discussion sets aside. The mismatch is a matter of scope rather than quality, and it would not be cured by any adjustment we could have made from published material.\n\nThe system's retention record is mentioned in the notes because it is the only published account of attrition in this population that we could find, and the reader who wishes to assess our own attrition should consult it for the order of magnitude rather than for a comparison.\n\nMention: \"Longitudinal Informal Sector System (LISS)\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"source\", \"summary\": \"The Longitudinal Informal Sector System (LISS), 2020, by Teikoku Databank is cited as background; no estimate is drawn from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:9fb216e57b5ba2ff514c8ac4", "occurrence_id": "cada81c772947cb88f70f943", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The distribution reported in the third column was computed from the National Household Consumption Microdata, which we obtained in its public-use form and did not modify before analysis. No weights were applied, since the public-use file does not carry the sampling weights, and the consequence is that the distribution describes the sample rather than the population.\n\nThat distinction is stated at each point where the distribution is used, because the sample over-represents households that responded to a postal instrument, which in this case means households with higher reported income. The effect is to shift the upper tail, and the median is affected less than the top decile, so the comparison drawn in the text rests on the median rather than on the mean.\n\nRecords with an implausible consumption value were retained rather than trimmed, and the text reports the distribution both with and without the extreme records so that the reader can see how much of the result the trimming choice carries.\n\nMention: \"National Household Consumption Microdata\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The third-column distribution was computed from unweighted public-use records of the National Household Consumption Microdata.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:81f1483ad23c26355ac260d7", "occurrence_id": "344e7bb22ce01d6e325c193e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Fare levels reported for Malaysia are those printed in the Integrated Transport Network Report, reproduced without conversion to a common currency or a common date. Converting would have made the rows comparable at the cost of importing an exchange rate and an inflation assumption, neither of which the report uses, and the section's argument does not require comparability of that kind.\n\nWe kept the report's own fare categories, which are defined by service type rather than by distance, and which therefore do not correspond to the distance-based categories used in the comparison table. Rather than mapping one onto the other we present the two side by side and let the reader see that the categories are not the same, since a mapping would have concealed exactly the difference the discussion is about.\n\nWhere the report gives a fare as a range we reproduce the range, and where it gives a single value we do not infer a range from neighbouring entries.\n\nMention: \"Integrated Transport Network Report\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Fare levels for Malaysia are reproduced from the Integrated Transport Network Report in their original currency and categories.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:9ac281c00f11a80b36d1ab6b", "occurrence_id": "d0771981ecf884d149937ba7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Asset counts for Croatia in 2022 are those recorded in the Decennial Public Works Survey (DPWS), used as delivered and without reclassification. The survey records an asset where a local authority has acknowledged responsibility for it, which means an asset built but never handed over is absent, and the reader should expect the counts to be smaller than an inspection-based inventory for exactly this reason.\n\nWe did not attempt to add the unaccepted assets back in, since the survey does not record them and no alternative source covers the same authorities at the same date. The comparison drawn in the section is therefore between two administrative views of the same incomplete object rather than between a measurement and a estimate.\n\nWhere an authority returned no schedule the row is left empty rather than imputed from a neighbouring authority, and the empty rows are listed in the appendix so that the reader can see how much of the national total rests on returns actually received.\n\nMention: \"Decennial Public Works Survey (DPWS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Asset counts for Croatia in 2022 are used as delivered from the Decennial Public Works Survey (DPWS), with its accepted-asset definition.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:92aad691b6f5ff8c00207e59", "occurrence_id": "b6d2509adcc8d9c1e4575cb5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The exposure classes assigned to the sampled structures follow the Regional Disaster Exposure Survey, published by Census, and were adopted rather than derived. We had no independent basis for reassigning a structure, and adopting a published classification has the further advantage that a reader can check any assignment against a source we do not control.\n\nAdoption carries the survey's revision history with it, and that history is not published, so a structure placed in a given class here may have been placed by a rule that no longer applies. The classes should be read as a description of how the structures were classified when we read them rather than as a statement about hazard.\n\nWe did not merge classes to obtain larger groups even where a group holds few structures, because the definitions differ more than the sizes do, and a merged group would have produced shares whose meaning changed between the rows of the same table.\n\nMention: \"Regional Disaster Exposure Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"Exposure classes for the sampled structures are adopted unchanged from the Regional Disaster Exposure Survey published by Census.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:65540d15b74ece56f9539451", "occurrence_id": "9da65c794f881cf44ce53838", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Benefit receipt by household composition was computed from the Continuous Social Protection Survey by cross-tabulating the receipt indicator against the composition variable and expressing each cell as a share of the row. The survey publishes neither cross-tabulation, so the table is ours and can be reproduced only from the microdata rather than from the published reports.\n\nThe cross-tabulation is thin in several cells, and cells holding fewer than a stated number of responding households are suppressed rather than reported, with the suppression applied to the row as a whole so that a suppressed cell cannot be recovered from the row total. This leaves visible gaps in the table and we prefer them to estimates that would carry a confidence interval wider than the difference the section discusses.\n\nHouseholds reporting no benefit in the reference period are placed in the none category rather than treated as missing, since the survey's own instrument distinguishes the two, and collapsing them would have raised the reported receipt rate for the compositions where non-response is most common.\n\nMention: \"Continuous Social Protection Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"Benefit receipt by household composition was cross-tabulated from the Continuous Social Protection Survey; the table is not published by the survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:150f06fb0d5f752f8b48e174", "occurrence_id": "b7d32b92da7e78eac6ddc004", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The consumption figures reproduced in the annex are those released through the Regional Household Consumption Registry for 2019, taken as published and not reweighted, rebased or redefined for consistency with the other sources in this report. Where the registry's definition of a household differs from the one used in the main tables we say so at the point of use rather than reconciling the two, because the reconciliation would have required a record-level link that does not exist.\n\nTwo regions appear in the registry with boundaries that were revised after the reference period, and the figures are those compiled under the earlier boundaries. We did not restate them, since the restatement would have required population data allocated to the new boundaries for a period in which they did not exist.\n\nWhere the registry reports a figure as provisional it is shown with that qualification attached, and the reader should treat the affected rows as expected to change rather than as final values that happen to carry a label.\n\nMention: \"Regional Household Consumption Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Annex consumption figures are reproduced as published from the Regional Household Consumption Registry for 2019, with definitions left unreconciled.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:15668051ef7c3d6f229bfb4a", "occurrence_id": "d6376215aefc57891e396727", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The number of units at each level was derived from the Continuous Administrative Boundary Census (CABC) by counting the records carrying a given level code and removing those marked as superseded. The census publishes a total for the country but not the level-by-level breakdown, so the counts in the table are ours and were produced by a rule stated in the appendix.\n\nThe removal of superseded records is the step that matters. Counting them produces a larger number for every level, and the difference is not uniform across levels, since the lowest level was reorganised more often than the higher ones. A reader who reproduces our counts without applying the rule will obtain a series in which the apparent growth of the lowest level is largely an artefact of record keeping.\n\nWe did not attempt to date the changes, since the census records the current status of a record without recording when it changed, and a reader wishing to construct a series over time would need the earlier releases rather than this one.\n\nMention: \"Continuous Administrative Boundary Census (CABC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"Unit counts by administrative level were derived from the Continuous Administrative Boundary Census (CABC) by a supersession rule we state in the appendix.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:a49da312e1192db7455d7d8c", "occurrence_id": "8a1922275b29b871d07f3716", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Utilisation figures for 2021 are taken from the Nationwide Health and Household Collection (NHHC), released by NRC-IDMC, and are reproduced without recomputation from the underlying records. The collection defines utilisation as an episode recorded at a participating facility, which excludes care obtained outside the facility network and excludes facilities that did not participate in the period.\n\nWe did not attempt to correct for non-participation, since the collection does not report which facilities were invited and declined, and any correction would have assumed the missing facilities resembled those that took part. The resulting figures are therefore lower bounds on utilisation in the areas with the lowest participation, and we describe them that way rather than presenting them as measurements.\n\nWhere the collection reports a figure for a subgroup without publishing the denominator, the cell is left empty rather than inferred from a related table, since the inference would have produced a rate whose construction we could not show a reader.\n\nMention: \"Nationwide Health and Household Collection (NHHC)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"Utilisation figures for 2021 are reproduced from the Nationwide Health and Household Collection (NHHC) released by NRC-IDMC.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:eab435c6789bb383335206db", "occurrence_id": "2cde5d50c5bb1b20ecb3724a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Service readiness scores by facility type were computed from the Joint Service Delivery and Readiness Microdata, released by Reinhart and Rogoff, by scoring each facility against a checklist of five inputs and reporting the share of facilities meeting all five. The microdata supplies the individual items; the conjunction is ours and is the reason our figures are lower than those quoted elsewhere.\n\nRequiring all five items rather than counting items met is a deliberately severe rule, chosen because a facility missing one of the inputs cannot deliver the service the section discusses. A reader who prefers an additive score will obtain much higher figures for the same facilities, and the appendix reports that alternative so the difference is visible rather than implied.\n\nFacilities that refused part of the checklist are scored as failing the refused items rather than as missing, which biases the score downward in the facility types where refusal was most common, and we were unable to separate refusal from absence with the data available.\n\nMention: \"Joint Service Delivery and Readiness Microdata\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"Readiness scores by facility type were computed from the Joint Service Delivery and Readiness Microdata released by Reinhart and Rogoff.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:afb14eb620fe0e007aaccea3", "occurrence_id": "f0cfbd53e027f25ebb914276", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Both inequality measures in the table are those published by the National Household Income and Expenditure System (NHIES), reproduced without recomputation, and we have not substituted measures of our own even where the alternatives are more common in this literature. The reason is comparability: the series the section extends is published in these two measures and in nothing else, and a change of measure partway along the series would have been read as a change in inequality.\n\nWe keep the system's treatment of the top of the distribution, which truncates the highest records before computing either measure, and which therefore places both measures below the values an untruncated computation would give. The truncation point is published, and a reader who wishes to see its effect can compare our figures with the system's own untruncated appendix.\n\nWhere the system reports a measure for a shorter period than the one discussed here, the figure is placed against the period the system states rather than against the period we would have preferred, and the mismatch is noted at the point of use.\n\nMention: \"National Household Income and Expenditure System (NHIES)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Both inequality measures are reproduced as published by the National Household Income and Expenditure System (NHIES), without substitution.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:6bbcc75dd86d0c2485eaf6d6", "occurrence_id": "cf8696571f45cb21768dc629", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Our grid totals were checked against the Longitudinal Population Grid Survey (LPGS), Second Version, which publishes a population surface for the same extent. The check was a sum over the same cells, and it produced a difference of a size we could not attribute to rounding, which is why the totals appear in the appendix rather than in the body of the report.\n\nWe did not scale our totals to match the surface, since scaling would have concealed a genuine disagreement between two constructions of the same quantity and would have made the resulting figures impossible to interpret. The reader who needs a single number should use the surface and should know that we did not.\n\nThe second version is specified because the first version's cells were defined on a different geometry, and a comparison cell by cell with that version would compare different areas under the same labels. No value from the survey is reproduced in this report.\n\nMention: \"Longitudinal Population Grid Survey (LPGS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"Grid totals were checked against the Longitudinal Population Grid Survey (LPGS), Second Version; the unexplained difference moved our table to the appendix.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:cffe03dcedd658a8f0617535", "occurrence_id": "5e97ea8a1f687e42b5611006", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The boundary definitions used to allocate facilities to districts are those carried in the Nationwide Administrative Boundary Survey for 2019, released by FAO, and were adopted without amendment. Where a facility lies near a boundary the allocation follows the geometry rather than the address, and in a small number of cases the two disagree, which we report rather than resolve.\n\nWe adopted the survey's geometry rather than the addresses because the survey is the source the district statistics themselves are compiled against, and an allocation made from addresses would produce counts that cannot be compared with published district totals without a concordance that no one publishes.\n\nWe did not attempt to update the boundaries for changes made after the survey, and a reader comparing our district counts with a current publication should expect disagreement in the districts whose boundaries were revised in the intervening years, which are named in the appendix.\n\nMention: \"Nationwide Administrative Boundary Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Facility-to-district allocation follows the boundary geometry of the Nationwide Administrative Boundary Survey for 2019, released by FAO.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:b61a595e3a627bf9104f7e94", "occurrence_id": "1531ec077ba56bd77de40d04", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Area by tenure category was computed from the Annual Land Use and Tenure Census (ALUTC), published by FAPRI, by summing the reported holding area within each category and expressing the result as a share of the total recorded area. The census publishes counts of holdings by category but not areas, so the area shares in the table are ours.\n\nReported areas are self-stated and are known to be rounded, with a pronounced concentration at round figures. We did not correct the rounding, since any correction would have redistributed area between categories according to an assumption about which direction the rounding ran, and the categories in question are the ones the section discusses.\n\nHoldings reporting no area were excluded from both numerator and denominator rather than treated as zero, which lowers the number of holdings in the smallest category without changing the area shares by much, and the excluded holdings are counted in a footnote so that the count and the area columns of the table are not read as being on the same base.\n\nMention: \"Annual Land Use and Tenure Census (ALUTC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"Area shares by tenure category were computed from the Annual Land Use and Tenure Census (ALUTC) published by FAPRI, which publishes counts but not areas.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:3e91ac780c6d8b583aa82c2f", "occurrence_id": "ebb9d01df992d0db60e5b918", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Joint Living Standards Dataset (JLSD) is named here only to alert the reader that a second source for the same variables exists and disagrees with the one we used, since a reader who encounters both will otherwise assume an error. Our analysis uses the administrative series throughout, and no value, weight or definition in this report derives from the dataset.\n\nThe disagreement is largest for the group the section concerns, where the dataset reports a materially higher rate, and we have no way of adjudicating between the two from published material. Presenting both would have required a comparison we could not construct honestly, so we note the disagreement and proceed with the series whose construction we can describe in full.\n\nWe did not adopt the dataset's definitions in the notes either, since borrowing a definition while rejecting the corresponding figure would have left the reader with a description of a measurement that appears nowhere.\n\nMention: \"Joint Living Standards Dataset (JLSD)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Joint Living Standards Dataset (JLSD) is cited to flag a disagreement with the source we used; no value is taken from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:a4f7dcd4a8704bb2745e0d4f", "occurrence_id": "6cb7180f3f6252997cbf4893", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Decision counts for Ukraine in 2017 are those printed in the Regional Asylum and Protection Report, issued by WITS, and are reproduced without adjustment to the categories. The report distinguishes decisions at first instance from those on appeal and reports them in separate columns, and we keep the two apart rather than reporting a single resolved figure, since combining them would have concealed the rate at which first-instance decisions were later changed.\n\nWe did not attempt to match individual cases between the two columns, since the report publishes no case identifiers and any matching we performed would have been a guess about which appeal corresponded to which decision. The reader who wishes to compute a rate of change should use the report's own commentary on that question rather than our columns.\n\nWhere the report withholds a count to prevent identification of a small group the cell is left empty, and the empty cell is not treated as a zero in any total we compute, which means our totals for the affected years are smaller than the totals a reader would obtain by reading the table literally.\n\nMention: \"Regional Asylum and Protection Report\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Decision counts for Ukraine in 2017 are reproduced from the Regional Asylum and Protection Report issued by WITS, keeping first instance and appeal separate.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:e38aabd1a0c5739b7fdd880f", "occurrence_id": "50d95131e08360dbcd6950e8", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Our round counts were compared with the Household Displacement Tracking Registry for 2020, maintained by UNICEF, which reports movement over a partly overlapping window. The comparison was performed to establish whether our counts were of a plausible order, and they are, which is the whole of what the comparison establishes and is stated here so the reader does not read more into it.\n\nWe did not adjust our counts toward the registry or adopt its definitions of a round and of a household, both of which differ from ours, and the differences are of a size that would matter for any finer comparison than the one we attempted. A reader attempting such a comparison should work entirely within the registry's definitions.\n\nNo value from the registry appears in our tables, and its coverage map was not used to identify the areas in which our own counts are weakest, though that would have been possible, because the resulting caveat would have implied a precision neither source has.\n\nMention: \"Household Displacement Tracking Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"Round counts were cross-checked against the 2020 Household Displacement Tracking Registry maintained by UNICEF for order of magnitude only.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:deac78ea7fdadc4f97765f83", "occurrence_id": "e3ece291663c22c840b39075", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: That the distribution of nurses by facility type has been measured before is worth stating, and the Rapid Health Workforce Collection of 2019, Phase 5, is the instance a reader is most likely to want. This section does not use its estimates, its definitions or its weights, and the citation is offered to locate our own effort within an existing literature rather than to support any figure.\n\nWe mention the collection rather than cite its results because our argument concerns the difficulty of obtaining facility-level records at all, and the collection's design bears on that difficulty in a way its published tables do not. No figure from it appears in any table, and no comparison in this section should be attributed to it.\n\nA reader who wishes to compare the two should note that the collection's facility types are defined by ownership while ours are defined by service level, so the categories overlap without corresponding, and we did not construct a mapping between them.\n\nMention: \"Rapid Health Workforce Collection\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Rapid Health Workforce Collection of 2019, Phase 5, is cited as context for a prior measurement; no estimates are used.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:14cce3a5cd5f496e4012a028", "occurrence_id": "19b165e0c0e9492bfe735bfc", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Coordinated Vocational Training Registry, operated by the Turkish Statistical Institute and reporting for 2019, is referenced here because it defines the programme categories used in the surrounding literature, and a reader will need that reference to situate the categories this section uses. The analysis proceeds from provider returns, and no figure in this report is drawn from the registry.\n\nOur categories are narrower than the registry's, since we report only the programmes for which provider returns were complete, and the difference is not a matter of choice: the remaining categories lack the returns that would be needed to report them. A reader comparing our totals with the registry should expect ours to be smaller and should not read the difference as coverage.\n\nWe did not map our categories onto the registry's, since the mapping would have been many-to-one in several places and would have implied an equivalence between programme types that the two definitions do not share.\n\nMention: \"Coordinated Vocational Training Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Coordinated Vocational Training Registry of the Turkish Statistical Institute for 2019 is cited to locate our categories; no figures are drawn from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:1dae39eae98c27a22acc1bfc", "occurrence_id": "7cee0dde21d37fc5840ddc25", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The intention figures quoted in the discussion are those released by the Multi-Topic Movement Intentions Survey (MMIS), reproduced without modification, and we have not converted stated intentions into an expected number of movers, which is the transformation a reader might most want and the one the survey itself cautions against.\n\nWe keep the survey's own wording of the categories, since the wording carries the question that produced the answer and a paraphrase of a stated intention is a different quantity from the intention as asked. Where the survey offers a category we did not use, it is omitted from our table rather than merged into a residual, so that the reader can see which categories were reported and which were not.\n\nWhere the survey reports a figure for a subgroup whose denominator is not published the cell is left empty, and we have not inferred the denominator from the marginal totals, which would have produced a rate whose base we could not identify to a reader who asked.\n\nMention: \"Multi-Topic Movement Intentions Survey (MMIS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"Intention figures are reproduced unchanged from the Multi-Topic Movement Intentions Survey (MMIS), without conversion to expected movers.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:e0d81229d297c0e931c63459", "occurrence_id": "7f7e80cbdde1b7e3000d4996", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The provider returns for 2015 were accessioned from the Joint Vocational Training Registry and registered in the condition in which they arrived, which for three of the returns meant a spreadsheet whose header row had been overwritten with a title. The register records that defect rather than correcting it, so that a later user knows the file was received in that state and does not attribute the ambiguity to whoever processed it next.\n\nNo normalisation was applied at deposit. Variable names, category codes and the treatment of blank cells differ between the returns, and presenting them as a single tidy object would have required decisions that belong to an analyst rather than to an archive. The register therefore lists the returns separately and notes the differences between them.\n\nRequests for a consolidated extract are answered with the individual returns and the register entry, and the requester is told which of the differences have consequences for a consolidated view and which are only cosmetic, since that advice is cheaper to give at the point of deposit than to reconstruct later.\n\nMention: \"Joint Vocational Training Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The paragraph describes how the 2015 provider returns from the Joint Vocational Training Registry were accessioned and registered unnormalised.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:f97a3b9497c6ee2624819022", "occurrence_id": "e89aa2b673fda4763e289f0c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The expenditure figures in the second table are those published by the Periodic Household Consumption Survey for 2021 and are reproduced exactly, including the survey's practice of reporting the smallest group as a residual rather than enumerating it. We did not expand the residual, since its contents are known only to the survey and an expansion we invented would have appeared in our table as a measured quantity.\n\nThe survey's reference period is shorter than a year, and we reproduce the figures without annualising them, because annualising a short window imports a seasonal assumption the survey neither states nor supports and the section's argument does not require an annual figure.\n\nWhere the survey reports a figure as based on a small number of responding households we retain the qualification at the point of use rather than collecting such qualifications into a general note, since the reader needs to know which particular rows are affected in order to decide how much weight to give them.\n\nMention: \"Periodic Household Consumption Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"Second-table expenditure figures are reproduced from the Periodic Household Consumption Survey for 2021, residual group and all.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:b8a83828c43c4fe51e26fb79", "occurrence_id": "8f8c80c72334746da075ab17", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Nationwide Health Workforce Survey of 2019 is mentioned because it is the source most readers reach for when they want a workforce figure of this kind, and the note exists to prevent the assumption that ours came from it. Our figures are computed from registration records, and no value or definition in this report originates with the survey.\n\nThe two differ in population rather than in quality: the survey covers those present at a facility on the night of enumeration, while our records cover those holding a current registration, and the gap between those populations is larger in the districts where staffing is most contested. We state the difference rather than reconciling it, since no adjustment available to us would have made the two comparable.\n\nWe did not use the survey's estimates to validate ours, because a comparison of two differently constructed counts can show only that they differ, and the reader would reasonably have expected more from a paragraph than that.\n\nMention: \"Nationwide Health Workforce Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Nationwide Health Workforce Survey of 2019 is cited to rule out a derivation; our figures come from registration records.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:82eb17049be0c95c1110410b", "occurrence_id": "db6ad980124f833f87868385", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Urban Poverty and Vulnerability Survey (UPVS), Round 2, released by Amadeus for 2016, is referred to here because its poverty line is the one most comparisons in this area are made against, and the note explains why our counts do not use it. The analysis applies a line derived from the cost of a local basket, and no figure in this report is taken from the survey or from its published headcounts.\n\nThe two lines differ by enough to change which households fall below them, and the change is not uniform across the districts discussed, which is the reason a single comparison would have misled rather than informed. We therefore report the difference in the lines rather than a comparison of the resulting counts, since only the former can be stated without an assumption we cannot test.\n\nWe did not adopt the survey's household definitions in the notes, since borrowing a definition while rejecting the corresponding count would have left the reader with a description of a measurement that appears nowhere in this report.\n\nMention: \"Urban Poverty and Vulnerability Survey (UPVS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Urban Poverty and Vulnerability Survey (UPVS), Round 2, by Amadeus for 2016 is cited to explain a differing poverty line; no values are used.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:65809364f3a666eb0c1d7a20", "occurrence_id": "55298f9012e71f98a9199695", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Birthweight distributions were computed from the Regional Maternal and Newborn Microdata (RMNM) for 2024 by classifying measured infants against the standard the microdata documentation names and reporting the share below the lower cut. The microdata publishes the mean and the standard deviation but not the share, so the share in the table is ours.\n\nA share computed from a mean and a deviation would be an assumption rather than a count, and we did not do it; the classification was applied to the infant-level records, which are released under a separate access condition and are held by the office that holds the microdata rather than by us. A reader working from the published tables cannot reproduce the figure, and we say so rather than implying otherwise.\n\nInfants with no recorded weight are excluded rather than imputed, and the unrecorded are more frequent among the earliest deliveries at each site, which biases the reported share downward to an unknown degree that we would rather leave stated than conceal behind a model.\n\nMention: \"Regional Maternal and Newborn Microdata (RMNM)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"Birthweight shares were computed from infant-level records of the Regional Maternal and Newborn Microdata (RMNM) for 2024.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:51da87ba24316610856e9883", "occurrence_id": "2c206d3d644977e02370538c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Stocks by land-cover class were derived from the Rapid Ecosystem Carbon Survey of 2017 by combining the measured biomass per plot with the mapped area of each class and reporting the product. The survey supplies the per-plot measurements; the mapped areas come from a separate exercise, and the stock figure is the product of two sources rather than a measurement by either.\n\nMultiplying them compounds their uncertainties in a way neither reports separately, and we therefore present the stocks as an order of magnitude rather than as a value with a decimal place, and we omit the confidence interval that a naive calculation would produce, since it would reflect only the sampling of plots and not the mapping.\n\nPlots falling in a class mapped at less than a stated area were excluded, which removes the classes with the fewest plots and leaves the smallest classes in the table supported by a handful of measurements each. Those rows are marked in the table rather than quietly reported alongside better-supported ones.\n\nMention: \"Rapid Ecosystem Carbon Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"Carbon stocks by land-cover class were derived from the Rapid Ecosystem Carbon Survey of 2017 combined with separately mapped areas.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:6fe59c46d2791508a0ea28bc", "occurrence_id": "9c1d44a7dc4695ce0b126534", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The nutrition indicators reported for Ghana are those carried in the Continuous Nutrition and Mortality Index, reproduced without recomputation, and we have not recalculated them from survey microdata even where microdata for a comparable period exists, since two constructions of the same indicator in one report would have invited a comparison neither source supports.\n\nThe index derives its indicators from a modelled series rather than from direct measurement in each period, and we reproduce that character honestly by describing the figures as modelled where they appear rather than relegating the point to a methodological note. A reader who requires measured values should use the survey series and expect a different level.\n\nWhere the index reports a value for a period whose input data the index describes as sparse we retain the value and the description together, and we do not extend the series across the sparse period by interpolation, since the resulting line would imply continuity that the index itself declines to assert.\n\nMention: \"Continuous Nutrition and Mortality Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Nutrition indicators for Ghana are reproduced from the Continuous Nutrition and Mortality Index and described as modelled, as the index presents them.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:df8beb8974c6e99f68e72ba0", "occurrence_id": "6be3f6f1f3826d5e8b638ba7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Income shares reproduced in the first table are those published in the Annual Household Income and Expenditure Report (AHIER) for 2015, taken as printed and not reweighted to the population figures used elsewhere in this report. Reweighting would have made the tables internally consistent at the cost of producing shares that appear in no publication a reader can check, and we preferred the checkable figure.\n\nThe consequence is that the shares and the population totals rest on different bases, and a reader who multiplies one by the other will obtain a quantity that neither source reports. The appendix states the bases side by side rather than reconciling them, since reconciliation would have required a common microdata file that does not exist.\n\nWhere the report gives a share only for a broader grouping than the one used here, the row is left empty rather than split pro rata, because a pro rata split would present an assumption as an observation in the same typeface as the figures around it.\n\nMention: \"Annual Household Income and Expenditure Report (AHIER)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"Income shares in the first table are taken as printed from the Annual Household Income and Expenditure Report (AHIER) for 2015.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:2a1a2443222b272f7ba615b2", "occurrence_id": "8e9b7f9c5c26b3c3f8f78553", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The index is named in the notes because a reader asking whether irrigation has been tracked over time in this setting should know that it has, and where to look. The section's own figures come from the annual returns of the irrigation authority, and nothing in this report is drawn from the Longitudinal Irrigation and Inputs Index, whose estimates, definitions and coverage statements are not used.\n\nThe two efforts are not directly comparable, since the index follows a panel of schemes while the authority's returns cover whatever was operating in the year, and the difference between following a panel and counting a population is larger than any change the section reports. We state that rather than offering a comparison the difference would not support.\n\nWe did not use the index to fill gaps in the authority's returns, since doing so would have joined a panel estimate to a population count without a basis for the join, and the resulting series would have been impossible to interpret at exactly the points where it was incomplete.\n\nMention: \"Longitudinal Irrigation and Inputs Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Longitudinal Irrigation and Inputs Index is cited in the notes as evidence of prior tracking; no estimates or definitions are used.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:fd738354ff9740f1a02b7c10", "occurrence_id": "4f1ea3fc96a4122a08e98c6a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Our movement figures were tested against the Periodic Displacement Tracking Dataset (PDTD) for 2019, maintained by NASA, which reports displacement for the same extent over a wider window. The test was coarse and answered the question it was designed to answer: our figures are of the same order as theirs, and we proceeded.\n\nWe did not reconcile the two, since the dataset measures displacement from changes in observed settlement extent while our figures count registrations, and the two quantities diverge wherever a displaced household remains with relatives rather than in a new structure. That divergence is a property of the two methods rather than an error to be corrected, and no adjustment available to us would remove it.\n\nNo value from PDTD appears in our tables, and its settlement definitions were not adopted, since adopting them would have required us to restate our registrations in terms of structures we did not observe.\n\nMention: \"Periodic Displacement Tracking Dataset (PDTD)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"Movement figures were tested against the 2019 Periodic Displacement Tracking Dataset (PDTD) maintained by NASA; the two methods diverge and were not reconciled.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:ebf8782b6c3b8ebb12223b9f", "occurrence_id": "27cd620723eaf2d153813611", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Employment by employment-size band was computed from the National Enterprise and Jobs Registry (NEJR) for 2019 by summing the reported employment of each enterprise within its band and expressing the result as a share of the total. The registry publishes counts of enterprises by band but not the employment totals, so every share in the row is ours.\n\nThe registry records employment as reported by the enterprise at registration and does not revisit it, so an enterprise whose workforce has contracted since registration contributes its original figure. We did not attempt to update the records, since the registry offers no field in which an update would be recorded and an adjustment we made would have been invisible in the resulting data.\n\nEnterprises reporting no employment are excluded from the numerator and retained in the denominator, which lowers each share slightly and by a larger amount in the smallest band, where the reporting of employment is least reliable.\n\nMention: \"National Enterprise and Jobs Registry (NEJR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"Employment shares by size band were computed from the 2019 National Enterprise and Jobs Registry (NEJR), which publishes counts but not employment totals.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:5dc6e739d1310f4acf66dc0c", "occurrence_id": "e98a98b5a7c56e6aaf85a57e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The consumption levels in the comparison are those carried in the Urban Household Consumption Registry (UHCR), reproduced as released and not adjusted for the difference in reference period between it and the other source in the table. Adjusting would have required a price index for the specific basket the registry measures, and the index that exists covers a different basket.\n\nWe did not convert the levels into a common unit of comparison, since the registry's units are local currency at the reference period and the conversion we would have used is not the one its own comparisons employ. The table therefore shows the two series side by side without a ratio between them, which is less convenient and more honest.\n\nWhere the registry reports a level for a grouping whose definition changed between releases we use the release matching the period and note the change, rather than restating the earlier figures onto the later definitions, which would have required a bridging study that has not been done.\n\nMention: \"Urban Household Consumption Registry (UHCR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Consumption levels in the comparison are reproduced as released by the Urban Household Consumption Registry (UHCR), without period adjustment.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:ba528fadca0925ae7cd974d0", "occurrence_id": "19e363919837e6b8b52054cc", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Participation rates in the annex are those published by the Longitudinal Labour Force Survey (LLFS) and are reproduced without recomputation, including the survey's treatment of students seeking work as participants, which raises the rate for the youngest age group relative to a definition that excludes them. We kept the treatment because the annex compares the survey with itself over time, and a change of definition partway through would have appeared as a change in participation.\n\nWe did not reconcile the survey's figures with the administrative counts discussed in the body, since the two measure different populations by different methods and the difference between them is a matter of design rather than of accuracy. A reader who wants a single number for the period should know that this report deliberately declines to supply one.\n\nWhere the survey suppresses a rate for a thin subgroup the cell is left empty, and the empty cells are not filled from adjacent periods, since a carried-forward rate would have been indistinguishable in the table from a measured one.\n\nMention: \"Longitudinal Labour Force Survey (LLFS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Annex participation rates are reproduced from the Longitudinal Labour Force Survey (LLFS), keeping its student treatment and its suppressions.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:495a27a0c1b7437f8ac75382", "occurrence_id": "8e315a5a9a49725284ddaa5d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Deprivation incidence for Maputo in 2018 was computed from the Periodic Multidimensional Poverty Microdata by applying the cutoff stated in the microdata documentation to the household's count of deprivations and reporting the share above it. The microdata supplies the household counts; the cutoff was chosen by us from the options the documentation discusses, and the choice changes the result materially.\n\nA stricter cutoff lowers incidence in every district, and a looser one raises it, but not in the same proportions, so the ordering of the two most affected districts reverses under the looser cutoff. We report the reversal rather than defending one cutoff, since the documentation presents the options as defensible alternatives rather than as a single standard.\n\nHouseholds with missing values on any indicator are treated as not deprived on that indicator, which lowers the count for households in the areas where the module was most often skipped, and we did not impute the missing values, since imputation would have moved the incidence in a direction determined by the imputation model rather than by the households.\n\nMention: \"Periodic Multidimensional Poverty Microdata\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"Deprivation incidence for Maputo in 2018 was computed from the Periodic Multidimensional Poverty Microdata using a cutoff we selected and report alongside the reversal it causes.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:2d869d05da3577a86b84a0c3", "occurrence_id": "95b5574bc865077d67ee9905", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Decennial Living Standards Survey (DLSS) files were received in Switzerland under a transfer agreement that restricts onward sharing, and the register records the restriction alongside the holdings, since a catalogue entry describing only the files would have invited requests the office cannot honour. Two of the files are held under a narrower condition than the remainder and are listed separately for that reason.\n\nNothing was de-identified at deposit. De-identification is a task for an analyst with a specific release in mind, and performing it in advance would have removed variables whose value cannot be judged before that purpose is known. The register states what was received and what may be shared, and the two statements are kept apart so that neither can be read as the other.\n\nAccess requests are answered by referring to the agreement rather than to the files, which is slower and has on occasion produced a second request naming a file the agreement does not cover. The office considers the slower course preferable to a disclosure decision made by whoever happens to be on duty that week, since such a decision cannot be reviewed later by anyone who was not present when it was made.\n\nMention: \"Decennial Living Standards Survey (DLSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The paragraph describes how the Decennial Living Standards Survey (DLSS) files were registered in Switzerland under a restrictive transfer agreement.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:a39b528fab97781843788af6", "occurrence_id": "fe7abef178fbb83e032f9673", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The National Social Protection Survey (NSPS) was received from LRMIS in 2015 as a household file and a community file with no linking key between them, and the register records that the two arrived unlinked rather than presenting the holding as a single survey. Any linkage would have been constructed by an analyst and would have carried that analyst's assumptions into the archive, where they would have been invisible to everyone who used the file afterwards.\n\nThe register notes which variables appear in both files with identical names and which share a name but not a definition, since that list is the practical limit of what can be said about linkage without performing it. Two variables share a name and differ in wording, and the difference is recorded in the entry so that a future user does not treat them as the same field and build a result that quietly depends on the distinction.\n\nNo derived file was created at deposit, and requests for a linked extract are answered with the two files and the register entry, together with a plain statement of what the office does not know about the relationship between them.\n\nMention: \"National Social Protection Survey (NSPS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The paragraph describes how the 2015 National Social Protection Survey (NSPS) was registered as two unlinked files with no linking key.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:298485f74fc357354f576cae", "occurrence_id": "dc8ce31096a8aebf089e66a4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Data management arrangements.** The records gathered for this review were compiled into the Quarterly School Infrastructure Registry, which the Latinobarómetro Corporation maintains on a rolling quarterly cycle, and the paragraph describes how that collection was assembled rather than what it reports.\n\nEnumerators recorded water, electricity and roof cover at each school site on one shared instrument, and a school running two shifts was entered once against the site rather than once per session. Where a classroom was unusable, the reason was coded from a fixed list so that a later round would not invent its own wording.\n\nEach deposited copy travels with the checksum taken at the moment of archiving, and a correction is appended as a further entry instead of replacing the record it amends, so the archive keeps the sequence in which the facts arrived. A site that could not be contacted in one round is carried forward with its earlier entry intact and re-approached in the next.\n\nMention: \"Quarterly School Infrastructure Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The section describes how the Quarterly School Infrastructure Registry, maintained by the Latinobarómetro Corporation, is collected and archived rather than what it reports.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:7d1abcdfe728c4579f893a3a", "occurrence_id": "c06a88b06a81c2b7ac13040d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Basis for the estimates.** Two figures are reproduced here directly from the Verified Minimum Needs Registry (VMNR), the 2018 compilation held by UNFCCC, and neither was recalculated here.\n\nThe first counts households whose recorded transfers fell short of the minimum basket the registry prices, and the second counts households whose housing costs alone exceeded that basket once the transfer was received. both keep the one decimal place the release prints, and the gap between them is left unexplained here because the registry does not publish the components.\n\nSelection followed the register's own eligibility flag, so a household whose final record fell inside 2018 stays in the count even where it had left the programme afterwards. The instrument behind those records asked the dwelling's reference person about the previous month, and in-kind support was valued at the amount the administering body had recorded rather than at local prices.\n\nMention: \"Verified Minimum Needs Registry (VMNR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Both shortfall figures were quoted directly from the 2018 release of the Verified Minimum Needs Registry (VMNR) held by UNFCCC.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:48238409c228d4cdee5fd8b8", "occurrence_id": "23aeeee8b4ecf2b2448d4660", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The travel-time figures in the table were placed against the Midline Road Segment Registry for 2023 as a consistency check, and the agreement between the two is reported here as a test of the estimates rather than as a measurement of its own.\n\nSegments were matched on the identifiers the enumeration carried, and a stretch that appeared in one source as two entries was compared against the summed length of those entries rather than against either one. Where the two sources disagreed by more than a kilometre, the figure carried forward was the one the enumeration had used and the disagreement was flagged in the annex.\n\nBoth figures were restated on a common reference definition before they were compared, since the registry records surfaced length while the travel-time estimate is made along the driven route. No uncertainty bounds accompany either number, and the comparison is therefore read as descriptive.\n\nMention: \"Midline Road Segment Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The travel-time estimates were cross-checked against the 2023 Midline Road Segment Registry, which served as a supporting source rather than as a new estimate.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:99abbf84ef863d5abf4fe61d", "occurrence_id": "dffe73f2773e41ced71db33f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Documentation practice in this period is described at a general level in the Pilot Access and Documentation System, an administrative collection the UK Cabinet Office ran in 2017, and none of its values are used in the calculations that follow.\n\nThe system recorded whether an office published its decision times and its appeal routes, and it did so as a yes or no entry rather than as a measured duration, which is why it appears here only as background. Its own notes on sampling are not reproduced, and the definition it uses for an application is left with the system itself.\n\nWhat the surrounding discussion takes from that record is the direction of change rather than any count: the shift toward published decision times began before the period covered here, and the pilot is cited because it sits close to the turning point, not because it measures it.\n\nCoverage was uneven by office type, since the pilot enrolled offices that had volunteered rather than a drawn sample, and a return counted as complete only where both questions were answered. That is a second reason the record is used here for orientation: the missing returns are not scattered across the offices in any way the release describes.\n\nMention: \"Pilot Access and Documentation System\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The 2017 Pilot Access and Documentation System run by the UK Cabinet Office is cited for context, with none of its values entering the calculations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:2fa15bd005e9449265edb51f", "occurrence_id": "cb66243d2f2315f7e51d0f8c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Coverage and limitations.** The durations reported below were computed from the Rural Job Search Microdata for Australia, 2023, which is the section's principal evidence rather than a quotation from it.\n\nSpells were constructed from the monthly activity entries, and a spell was closed only where two consecutive months showed no search activity, so a short gap inside a single month does not end it. Respondents who reported a job start inside the reference period were censored at that month rather than carried to the end of the window.\n\nThe regional indicator was merged from the area frame rather than from the respondent's own description, and the resulting shares are weighted with the design weights supplied with the extract. Intervals were not carried through from the source, so the comparison between the two cohorts is presented as descriptive, and medians are reported instead of means because the upper tail is long.\n\nMention: \"Rural Job Search Microdata\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"Search-duration medians were computed from the Rural Job Search Microdata for Australia in 2023 rather than quoted from any release.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:7f43d4096c8981c8cd9f95bc", "occurrence_id": "f782202109772775ad18fe64", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The surfaced length quoted in the second column is taken from the Triennial Road Segment Registry for 2020, and it is reproduced here without adjustment because the purpose is to show what that collection records rather than to restate it.\n\nEntries cover maintained segments only, so an unclassified track linking two of the same settlements appears in neither the length nor the density figure. The registry counts a segment once at its maintained surface, which is why the density in the table falls short of the figure the network operator publishes for the same alignment.\n\nNeither share was recomputed; neither number was rounded again here. Where a segment was reclassified during the round, the entry reflects the classification in force when the enumeration visited, and the later change is not carried back into the table.\n\nThe round was run in the dry season, which is when the registry expects a maintained surface to be passable, and a segment destroyed by flooding after the visit kept its earlier entry. The table is therefore the network as recorded, not the network as it stood at the time of writing.\n\nMention: \"Triennial Road Segment Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The surfaced length and density shares were quoted as published from the 2020 Triennial Road Segment Registry.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:76c055bcfb3a4a202802791a", "occurrence_id": "74f72ccc202a408f2af6de18", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Data sources.** The two retreat figures reproduced below are quoted directly from the High-Frequency Coastal Erosion Registry, a collection maintained by DLA Piper, and neither was recomputed for this section.\n\nThe first gives the mean annual retreat along the monitored frontage and the second the share of surveyed fronts that lost material in the most recent observation window; the precision of both is left exactly as published. Observation windows differ in length between fronts, so the second figure counts fronts rather than months.\n\nWhere a front had no observation inside the window it was left out of the denominator rather than treated as stable, and a front whose extent was disputed was carried at the status the field check recorded. The definitions behind those figures are left with the source itself, and no interval accompanies either number.\n\nFrontages were resurveyed on a fixed repeat, against the same datum and the same marks where the marks survived, so a change in the figures describes the ground rather than a change in how the ground was read. A repeat taken just after a storm surge was flagged and is reported apart, because those observations carry the event rather than the trend.\n\nMention: \"High-Frequency Coastal Erosion Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Both retreat figures were reproduced directly from the High-Frequency Coastal Erosion Registry maintained by DLA Piper.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:e99eac3882a2d853d8590a28", "occurrence_id": "1fac6cd78f6563e388f1c969", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The counts in the table are taken directly from the Clustered Transport Network Registry, and they are reproduced without adjustment because this section reports what the registry holds rather than a restatement of it.\n\nSegments were drawn as clusters of contiguous surface type, so a route whose surface changes twice appears as three entries and the totals below count entries rather than routes. A segment recorded as under construction was retained where vehicles could already pass, and excluded where the works had closed the alignment.\n\nboth appear at the original's single decimal exactly as published, and neither was recomputed from the underlying geometry. The registry keys on the segment rather than on the settlement it joins, which is why the table cannot be read as a measure of settlement access.\n\nCluster boundaries were drawn where the surface type changed or where an administrative line crossed the alignment, and a conflict between the two was resolved in favour of the surface type. The convention matters for the totals above, since the alternative would have produced a larger number of shorter entries.\n\nMention: \"Clustered Transport Network Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The segment counts and shares were quoted directly, to one decimal place as published, from the Clustered Transport Network Registry.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:c16ac66d90957fe9c3ff8453", "occurrence_id": "83fc57dfa91ac6f7fb0b8524", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The participation shares in this section were derived from the Metropolitan Disability and Support Survey, an instrument fielded by the Banque de France, and they serve as a supporting comparison rather than as the section's main evidence.\n\nSupport was counted where the respondent named a service used in the previous month, and an unpaid family arrangement was recorded separately rather than folded into the same total, because the two move differently with age. Responses were weighted to the household frame before the shares were taken, and no interval accompanies the derived shares, so they are read as descriptive only.\n\nWhere a respondent reported no difficulty with the listed activities the record was retained with a zero entry rather than dropped, since dropping it would have lifted every share in the table. No definition from the survey is restated here.\n\nThe shares were taken for the age bands the instrument itself reports rather than for bands chosen afterwards, and a respondent below the working threshold was excluded from both the numerator and the denominator of the participation measure.\n\nMention: \"Metropolitan Disability and Support Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"Participation shares were derived from the Metropolitan Disability and Support Survey fielded by the Banque de France, as a supporting comparison.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:92b9b5c7fde004815163870c", "occurrence_id": "872a7ae08c604b7716523f86", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Completion rates quoted in the discussion come from the Ministerial Apprenticeship Registry (MAR) for 2021, and they are reproduced as published rather than recalculated from the underlying records.\n\nA programme was counted as completed where the registry held a final assessment record, and a withdrawal inside the first quarter was carried in the denominator, which is the convention that lowers the rates shown here relative to those circulated by the training providers. Providers were not asked to reconcile those withdrawals before the release.\n\nNeither share was recomputed; no re-rounding was applied to either. Trade codes follow the scheme the registry applied to the 2021 records, so a comparison with an earlier table drawn on the older codes should be made at the aggregate level only.\n\nThe registry holds one record per apprentice rather than per enrolment, so a re-started programme appears once and carries its most recent outcome. The rates quoted in the discussion are therefore for people, while the provider figures circulated beside this release are for enrolments and sit slightly higher.\n\nMention: \"Ministerial Apprenticeship Registry (MAR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The completion rates were quoted from the 2021 Ministerial Apprenticeship Registry (MAR) and left exactly as published.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:99e98e5806bdfaae27cc27a2", "occurrence_id": "9eba80965bbe103a7c2fe4a2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The facility delivery estimate carried in the previous table was checked against the Comparative Maternal and Newborn Survey for 2019, and the agreement between the two is reported here as a consistency test rather than as a second measurement of the same quantity.\n\nBoth figures were placed on the same reference definition before the comparison was drawn, since the survey counts a delivery at the place where it occurred while the administrative series counts it at the place of registration. Restated on the survey definition, the two differ by less than the margin the annex allows, and no adjustment was carried forward.\n\nThe survey contributed no new denominator to this section. Its own weighting is described in its technical report and is not restated here, and where the two sources disagreed at the regional level the national figure was retained because the regional cells of the survey are too thin to support the comparison.\n\nMention: \"Comparative Maternal and Newborn Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The facility delivery estimate was corroborated against the 2019 Comparative Maternal and Newborn Survey after both figures were restated on one reference definition.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:3e00a89869430886453bf5f1", "occurrence_id": "a9a1782ea94a6dd985c0cd22", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The distributional shares below were derived from the Nationwide Inequality Registry (NIR), offered here only to support the comparison the household series rather than as the section's principal evidence.\n\nRankings were built on the equivalised aggregate the registry stores for each record, using the scale the registry itself applies, and the top and bottom shares were read off that ranking without re-estimating the underlying totals. Because the registry holds administrative components rather than a consumption diary, the aggregate excludes own-produced consumption, which pulls the lower tail upward relative to a survey measure.\n\nThe derived shares come without intervals and are read as descriptive only. Where a record carried a component flagged as imputed, it was retained in the ranking but excluded from a second run of the same calculation, and the two runs are reported side by side so the effect of that choice is visible rather than buried.\n\nMention: \"Nationwide Inequality Registry (NIR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"Top and bottom distributional shares were derived from the Nationwide Inequality Registry (NIR) as a supporting comparison, with imputed components shown separately.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:ae82f3a41471c26fafb9081f", "occurrence_id": "f235528c687e28bbdabcaf53", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The retention rate quoted for the secondary cycle is taken from the Triennial Enrolment and Retention System of Slovakia, and it is reproduced here exactly as the system publishes it rather than recalculated from the enrolment records behind it.\n\nA pupil was counted as retained where the system held a continuation entry for the next grade at any school rather than only at the same school, which is the convention that makes this rate higher than the one the ministry circulates for internal use. Transfers between cycles were treated as continuation, and a withdrawal recorded without a reason field still closes the spell.\n\nNeither share was recomputed; the release prints both to one decimal and they are kept so. The system covers schools that report to the central register, so a provider operating outside that obligation is absent from the denominator, and the rate should not be read as a property of all secondary schooling in the country.\n\nMention: \"Triennial Enrolment and Retention System\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The secondary retention rate was quoted as published from the Triennial Enrolment and Retention System of Slovakia, without recalculation.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:be0cef5431270bd8fb5394da", "occurrence_id": "84c2064b6efdcebb1f0321ba", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The literacy estimate in the discussion was compared against the Rolling School Assessment Survey for 2021, and the comparison is presented as a check on the estimate rather than as a new measurement of learning.\n\nPupils were assessed with a short tool at the end of the school year, and an absent child was not substituted from the register, so the assessment denominator sits below the enrolment denominator used by the estimate it is checked against. The two were nonetheless placed on a common definition of proficiency before the figures were set side by side.\n\nWhere the two sources disagreed, the school was retained and the disagreement was carried into the annex rather than resolved by dropping either figure. No uncertainty bounds accompany the derived comparison, and it is read as descriptive.\n\nSchools were drawn in two strata and the tool was piloted outside the sampled districts before the main round, so the instrument itself was not first used where it is reported. That piloting is why the survey is used here as a check rather than as the estimate's basis.\n\nMention: \"Rolling School Assessment Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The literacy estimate was cross-checked against the 2021 Rolling School Assessment Survey after both were restated on one definition of proficiency.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:114e62b22c38b866cc260c1c", "occurrence_id": "bd8335654521fe6c42d85103", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Data sources.** The two coverage figures below are reproduced directly from the Nationwide Parcel Footprint Registry (NPFR), the 2019 release compiled by CEDLAS, and neither was recomputed for this section.\n\nThe first gives the share of cadastral parcels carrying a usable footprint and the second the share of those parcels whose footprint was matched to a registered owner. both carry the precision the source sets, and the gap between them is the part of the registry where geometry exists but the ownership link does not.\n\nParcels were retained where the boundary closed and the area fell inside the range the compiler considers plausible for the municipality, which excludes a small number of very large rural parcels from both figures. The definitions behind the two shares are left with the source itself, and no interval accompanies either.\n\nThe registry is refreshed by municipal submission rather than by a single national capture, and a municipality that submitted twice in the same cycle contributed only its later file. The two shares therefore describe the state of the submissions the compiler accepted, which is a narrower claim than coverage of the cadastre itself.\n\nMention: \"Nationwide Parcel Footprint Registry (NPFR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Both coverage shares were quoted directly from the 2019 Nationwide Parcel Footprint Registry (NPFR) compiled by CEDLAS.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:c617bc365f20e1a3f32e9e0b", "occurrence_id": "a42333167ad444c577074231", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Coverage and limitations.** The incidence rates below were computed from the Urban Occupational Safety Survey (UOSS) for 2017, and not a source of quoted numbers here.\n\nExposure was built from the hours each workplace reported rather than from headcount, so an establishment running several shifts contributes more exposure time than one of the same size running a single shift. Events were counted where the respondent could place them inside the reference period, and an event reported only as a hospital visit was carried at the same weight as any other.\n\nThe rates were taken after weighting to the establishment frame, and uncertainty was not carried through, so the comparison between the two size bands is descriptive. Small cells were left in the table rather than pooled, because pooling would have hidden the band the discussion turns on.\n\nReporting covered establishments with at least one employee, so a workshop working only with family labour appears in neither the exposure time nor the event count.\n\nMention: \"Urban Occupational Safety Survey (UOSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"Injury incidence rates were computed from the 2017 Urban Occupational Safety Survey (UOSS) using reported hours rather than headcount as the denominator.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:0c3b115f0562c58c9a373f33", "occurrence_id": "c8371da067ccb2f97630adf8", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The coverage figure for mapped structures in the second column is taken from the Representative Building Footprint System for 2022, and it is reproduced without adjustment because the purpose here is to report what the system holds.\n\nA structure was counted as covered where a footprint existed and the structure had been confirmed on the ground or in imagery of sufficient resolution, and the two cases are not distinguished in the published share. A structure demolished after capture was retained, since the system records the state at capture rather than the state at publication.\n\nNeither share was recomputed; decimal places come through unchanged. The system is maintained by successive capture campaigns rather than by a single sweep, so a comparison across regions mixes vintages of imagery, and the table should be read as coverage rather than as currency.\n\nEach campaign passed a topology check before publication, and a footprint that failed it was held back rather than corrected by hand, so the published share understates what was captured. The held-back cases are counted in the system's own quality report and are not restored anywhere in this table.\n\nMention: \"Representative Building Footprint System\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The mapped-structure coverage share was quoted as published from the 2022 Representative Building Footprint System.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:faa5e4f7e012e05233897054", "occurrence_id": "9cc9334918dc10e5b59d0b7c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Data sources.** The two transition shares reproduced below are taken directly from the Municipal Youth Transition Registry of Myanmar for 2016, and neither was recalculated for this section.\n\nThe first share gives the proportion of recorded young people who were in education or training at the last contact, and the second the proportion whose first recorded employment came more than a year after that contact. the table keeps the published rounding of both, and the second is computed only over those who left education.\n\nContact was attempted at the household last recorded, and a young person who had moved and could not be re-approached was left in the denominator rather than removed, which lowers the first share. The registry keys on the person rather than on the household, so a sibling pair contributes two records.\n\nRegistration was opened at the municipal office rather than at the household, so a young person who never approached an office has no record and appears in neither share. The registry is strongest where office contact is routine, which is the assumption the discussion makes when it uses these two figures together.\n\nMention: \"Municipal Youth Transition Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Both transition shares were quoted directly from the 2016 Municipal Youth Transition Registry of Myanmar, as published and without recalculation.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:e406ce5b7ad3ba7c026ecacf", "occurrence_id": "5184ba937c4b4749830ab0b2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Basis for the estimates.** The two counts below are reproduced directly from the Follow-Up Protracted Situation System (FPSS), the 2024 release maintained by the Federal Police, and no recomputation was applied to either.\n\nThe first count gives sites with at least one registration event in the reporting month and the second the individuals registered at those sites during it. neither share was restated at another precision, and the second is a flow rather than a stock, so it cannot be added across months to recover a population.\n\nAn office reporting a range had its range carried through rather than reduced to a midpoint, and the release prints the range where the estimating office judged a point figure unreliable. the definitions behind both figures stay with the source, and the movement recorded is the last move observed at the site rather than an intended destination.\n\nRegistration was recorded at the site on the day it happened, so the counts describe presence there rather than the population entitled to register, which the release states.\n\nMention: \"Follow-Up Protracted Situation System (FPSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Both monthly counts were quoted directly from the 2024 Follow-Up Protracted Situation System (FPSS) maintained by the Federal Police.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:1e77fc16b88c5c89d69358f5", "occurrence_id": "494f23bd011149f0fd2bb557", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The exposed-area figure used in the damage table was checked against the Harmonised Flood Extent Dataset (HFED), and the agreement between the two is reported as a test of the estimate rather than as a new measurement of extent.\n\nBoth were restated onto the same statistical geography before the comparison, since the extent product is published on a grid and the damage estimate is made over administrative units. Cells straddling a boundary were split by area rather than assigned wholly to one unit, and the same split was applied to both sources so the comparison could not be won by a boundary convention.\n\nWhere the two disagreed beyond the tolerance set in the protocol, the cell was flagged and carried into the annex with both values. No uncertainty bounds accompany either figure, and the resulting comparison is read as descriptive.\n\nExtent was derived where the flood was visible in both the imagery and the field check, and a cell confirmed by only one of the two was left outside the comparison. The dataset is harvested from several contributing products, so the check here is on agreement between sources rather than on the accuracy of either one.\n\nMention: \"Harmonised Flood Extent Dataset (HFED)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The exposed-area estimate was corroborated against the Harmonised Flood Extent Dataset (HFED) after both sources were restated on one statistical geography.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:685a1129fb9a8fca3d13ff89", "occurrence_id": "51bb2b10133a1233577570df", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Data sources.** The two shares reproduced below are taken directly from the Cross-Sectoral Enrolment and Retention Registry (CERR), and neither was recalculated from the enrolment records behind it for this section.\n\nthe first reports the share of registered pupils with a continuation entry for the next grade and the second the share of those continuations recorded at a different school from the one of origin. both are reproduced at source precision, and the second is a share of the first rather than of all enrolment.\n\nA continuation was accepted wherever the system held one, so a pupil who moved between cycles still counts as retained, and a withdrawal recorded without a reason field closes the spell at the last known grade. The registry keys on the pupil rather than on the school, which is why the two shares cannot be read as a measure of institutional stability.\n\nNothing else in this section rests on the registry, and neither definition is restated here.\n\nMention: \"Cross-Sectoral Enrolment and Retention Registry (CERR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Both enrolment continuation shares were quoted directly from the Cross-Sectoral Enrolment and Retention Registry (CERR), as published and without recalculation.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:011912ccf3cdd87cfe76844b", "occurrence_id": "66b735569e4707c768527e52", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Coverage and limitations.** The completion and placement rates below were computed from the Periodic Vocational Training System (PVTS) for 2018, which serves as and the section's own figures are built on it.\n\nProgrammes were grouped by the occupation code the system stores at entry rather than at exit, because a change of track during a programme would otherwise move a graduate into a category the graduate never trained for. A participant who enrolled in two programmes inside the reference period contributed once, at the earlier enrolment.\n\nPlacement was measured against any recorded activity rather than against a contract, so self-employment counts where the system holds a contact record for it. The derived rates are weighted to the programme frame and come without intervals, and the comparison between the two delivery modes is therefore presented as descriptive rather than as an estimate of difference.\n\nMention: \"Periodic Vocational Training System (PVTS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"Completion and placement rates were computed from the 2018 Periodic Vocational Training System (PVTS), grouping programmes by the occupation code held at entry.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:377d40090c4f34147d48b8ca", "occurrence_id": "941033e7b80f69e23e952e08", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Basis for the estimates.** The two inspection figures quoted in the discussion are reproduced directly from the Government Labour Inspection Survey for 2016, and neither was recomputed here.\n\nthe first states the proportion of inspected establishments found to have at least one outstanding requirement and the second the share of those establishments where the requirement concerned working hours rather than safety equipment. the original rounding is preserved on both, and the second is a share of the first.\n\nEstablishments were selected by the inspectorate's own scheduling rather than by a drawn sample, which is stated in the survey's own notes and is the reason the shares are quoted rather than extended. An establishment visited twice in the period contributed once, at its first visit, and an establishment that refused entry is recorded as inspected without a finding rather than excluded from the denominator.\n\nMention: \"Government Labour Inspection Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Both inspection shares were quoted as published from the 2016 Government Labour Inspection Survey, with the second expressed as a share of the first.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:becdf76a94df39243743c4af", "occurrence_id": "3cefc3b129c1f91a82d5ec6f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The trend line in the discussion of informal activity is drawn from the Government Informal Sector Index (GISI) for Cambodia, and the index is cited for direction rather than for any of its levels.\n\nThe index combines several administrative signals into a single series, and the weights it uses are revised periodically, so a movement in the line cannot be read as a movement in the underlying activity without knowing which revision produced it. This section therefore quotes no level from it and takes only the sign of the change between the two points the discussion needs.\n\nNone of the definitions the index uses are restated here, and the components behind the series are left with the source itself. Where the discussion needs a level, it uses the household series instead, and the index appears only as background to that level.\n\nThe index is also published at a national level only, so the discussion cannot lean on it for a claim about any single area inside Cambodia, and no regional breakdown is inferred from the national line at any point in the section.\n\nMention: \"Government Informal Sector Index (GISI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Government Informal Sector Index (GISI) for Cambodia is cited only for the direction of change, with none of its levels or definitions used in the discussion.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:8a0576948c4d964e8615fbfd", "occurrence_id": "1f36d38b8c4daf42207e43ea", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Data sources.** The two coverage figures below are reproduced directly from the Joint School Feeding System (JSFS), and both are reproduced exactly as filed.\n\nthe first gives the percentage of registered schools reporting a feeding day in the reference week and the second the share of enrolled pupils at those schools. both are given to the same single decimal as published, and the second is reported for the schools in the first rather than for all schools.\n\nA school was counted as feeding where it reported at least one day, so a school that served twice in the week carries the same weight as one that served five times, and the figures describe coverage rather than intensity. A school closed for maintenance during the week was excluded from both numerator and denominator rather than treated as a zero.\n\nThe system is compiled from returns the districts file themselves, and the definitions behind the two shares are left with the source itself.\n\nMention: \"Joint School Feeding System (JSFS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Both feeding coverage shares were quoted directly from the Joint School Feeding System (JSFS), with the second computed only over the schools in the first.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:ec59ed473665cbe00c0c1009", "occurrence_id": "61223f12af90741bb6b1362e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Data sources.** the two expenditure figures below come straight from the Periodic Education Finance Survey (PEFS) for Nairobi, 2019, and neither was restated for this section.\n\nThe first gives the per-pupil amount reported by the school and the second the portion of that amount the school identified as available for classroom materials rather than for staffing or facilities. precision matches the release exactly, and the second is a component of the first rather than a separate total.\n\nSchools reported for their own fiscal year, which does not coincide with the government year, and the survey does not align them; the figures are therefore quoted as reported rather than deflated or re-based. A school that reported in-kind receipts valued them at the amount it recorded, and those valuations were left unchallenged.\n\nthe source's own definitions are not reproduced here, and no interval accompanies either.\n\nMention: \"Periodic Education Finance Survey (PEFS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Both per-pupil expenditure figures for Nairobi in 2019 were quoted directly from the Periodic Education Finance Survey (PEFS) as the survey reported them.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:74050e286570ec7314056718", "occurrence_id": "1874750f3a732a1f87a17829", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Coverage and limitations.** The activity shares below were computed from the Local Skills and Employment Census for 2022, released by SHIP, which carries the argument rather than supplying an excerpt.\n\nOccupation was coded from the verbatim task description the respondent gave rather than from the title the respondent chose, because the titles in the region's small workshops recur with several different contents. A person working in two establishments was enumerated once, at the one where more hours were worked, and the second job does not appear in the shares below.\n\nOwn-account workers are counted as employed, following the convention the census applies, and the shares are weighted to the household frame built during listing. The derived figures come without intervals, so the comparison between the two occupational groups is presented as descriptive, and the cells below the size the protocol allows are printed rather than suppressed.\n\nMention: \"Local Skills and Employment Census\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"Activity shares were computed from the 2022 Local Skills and Employment Census compiled by SHIP, coding occupation from the verbatim task description rather than the reported title.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:3495d609360d60098b1a82e9", "occurrence_id": "f5fa12479803cd2ca3bf8ec2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Household hardship in the period is described at a general level in the Subnational Minimum Needs Survey (SMNS), and none of its values enter the estimates that follow.\n\nThe survey asked whether a household could meet a list of minimum items and did so as a yes or no entry per item, which makes it useful for describing what households say they lack but unsuitable for measuring how far below a threshold they fall. That is why it appears here as background and why no share from it is carried into the table.\n\nWhat the discussion takes from the survey is the ordering of the items rather than their frequency: the same few items are reported unmet across the areas it covers, and that ordering is consistent with the pattern the estimates produce. The instrument's own notes are not restated, and the definitions it uses for a household are left with the survey itself.\n\nCoverage follows the areas the survey happened to reach, and the release says nothing about the areas it did not, which is a second reason no share from it is carried forward here. The ordering above is the only property the discussion relies on.\n\nMention: \"Subnational Minimum Needs Survey (SMNS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Subnational Minimum Needs Survey (SMNS) is cited as background for which items households report unmet, with no values used in the estimates.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:47be8b2c001c571a06ae58b9", "occurrence_id": "2c43b08e6bf68a32e440b8ee", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Data sources.** The two coverage counts below are reproduced directly from the Midline Social Protection Census (MSPC), and neither was recomputed for this section.\n\nThe first counts households holding at least one active entitlement and the second counts the individuals covered by those entitlements. the published decimals reach the table unchanged, and the second cannot be summed across areas to recover a population because a household may hold entitlements under more than one programme.\n\nAn entitlement was treated as active where the census found a payment or an eligibility record inside the reference window, and a household whose record had lapsed without a recorded termination was still carried as active, since the alternative would have counted a paperwork failure as an exit. The census keys on the household, and an individual in two entitled households appears once.\n\nwhat the terms mean is left to the source.\n\nMention: \"Midline Social Protection Census (MSPC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Both coverage counts were quoted directly from the Midline Social Protection Census (MSPC), reproduced to one decimal place as published.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:a0a12a165ac404ac0f59368b", "occurrence_id": "68d52e2aa9e09fc83a386c71", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Coverage and limitations.** The deprivation shares below were computed from the Audited Child Wellbeing Survey for 2020, using the audit file released by New Cronos, and the survey is what the analysis works from rather than a source it leans on for a quotation, so the shares above stand or fall with the audit file it was computed from.\n\nDeprivation was counted at the item level, so a household lacking two of the listed items contributes two deprivations rather than one household, and the shares below are of items rather than of children. The audit file was used to correct the household weights, and no other field was altered.\n\nWhere a caregiver declined a module the household was retained with the module scored as missing rather than dropped, because dropping it would have removed whole neighbourhoods from the denominator. The derived shares are reported without intervals and are read as descriptive, and the two age groups are shown separately rather than combined.\n\nMention: \"Audited Child Wellbeing Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"Deprivation shares were computed from the 2020 Audited Child Wellbeing Survey using the audit file released by New Cronos to correct the household weights.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:2d7ced7f487461657275402c", "occurrence_id": "906b5693e5328490a4845445", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Referral practice in Sud and Nippes is described at a general level by the Standardised Patient Flow Registry (SPFR), and none of the counts it holds are used in the estimates that follow.\n\nThe registry records, for each service point that reports to it, whether referrals were logged and whether the receiving site confirmed them. It does so as a status entry rather than as a measured interval, so it can say that a flow exists between two sites but not how long a patient waited inside it, which is the quantity the discussion needs.\n\nCoverage across Sud and Nippes is uneven, since reporting is voluntary at the smaller sites, and the release describes neither the non-reporting sites nor the reason they were absent. For that reason the registry appears here as background to the pattern rather than as evidence of its size, and the definitions it uses for a completed referral are left with the registry itself.\n\nMention: \"Standardised Patient Flow Registry (SPFR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Standardised Patient Flow Registry (SPFR) for Sud and Nippes is cited as background on referral practice, with none of its counts used in the estimates.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:7aee15ab0c2f744358f7e6a7", "occurrence_id": "02ff1bbd6f5b924bd2628f1a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Exposure in the affected area is summarised at a general level by the Metropolitan Disaster Exposure Dataset for 2020, and the dataset contributes no value to the cost figures that follow.\n\nThe dataset assigns each structure an exposure class from the modelled extent of the hazard rather than from observed damage, and it is published as a screen for fieldwork rather than as a measurement of loss. That distinction is why the cost estimates do not rest on it, and why no class from it is carried into the table.\n\nWhat the discussion takes from the dataset is the shape of the exposed area: the classes cluster along the low ground between the two districts, and the fieldwork was weighted accordingly. The model behind those classes, the resolution at which it runs and the inputs it consumes are left with the dataset itself, and none of them are restated here.\n\nWhere the modelled extent and the field observations disagreed about a structure, the dataset kept the modelled class and flagged the case, so the classes describe the model rather than the ground at exactly those points. The cost estimates could not inherit that flag.\n\nMention: \"Metropolitan Disaster Exposure Dataset\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The 2020 Metropolitan Disaster Exposure Dataset is cited only to describe where exposure concentrated, contributing no value to the cost estimates.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:27c533ec8bfa19b09f71907a", "occurrence_id": "3234388347862922fa9472d8", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The readiness shares quoted in the discussion of service quality are reproduced from the Cross-Sectoral Service Delivery and Readiness Survey (CSDRS) for 2024, used alongside the main series rather than against it the administrative series rather than as its replacement.\n\nReadiness was asked of the clinician on duty rather than of the manager named in the record, which is the convention that lifts the shares relative to an administrative return, since the clinician knows which items are actually in use. Service availability covers public and not-for-profit sites alike, and private for-profit sites are outside the frame.\n\nStock-outs were recorded as the constraint reported most often rather than as the only one, and the survey does not weight its results to a facility frame, so the shares are quoted as the survey's own description of what it observed. Neither share was recomputed here, and neither number was expanded or truncated.\n\nMention: \"Cross-Sectoral Service Delivery and Readiness Survey (CSDRS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"Readiness shares were quoted as published from the 2024 Cross-Sectoral Service Delivery and Readiness Survey (CSDRS) as a supporting comparison to administrative returns.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:0885415b3cb3714e51dafd0c", "occurrence_id": "08d252ab26982d95bdc9a48c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Data sources.** below, the two vacancy figures are lifted from the Cross-Sectoral Hiring and Vacancies Report (CHVR), and neither figure was re-derived for this discussion.\n\nthe first records the fraction of reporting establishments with at least one vacancy open beyond the period they consider normal for the role, and the second gives the mean number of such vacancies at those establishments. both are quoted at the release's decimal accuracy, and the second is computed only over the establishments inside the first.\n\nReporting is by establishment rather than by site, so a firm with several locations files once, and a vacancy filled before the reference date was excluded even where it had been open for a long period. A firm that did not file is treated as having no vacancy rather than as missing, which the report states and which lowers the first figure.\n\nThe definitions behind both figures are left with the report itself, and no interval accompanies either.\n\nMention: \"Cross-Sectoral Hiring and Vacancies Report (CHVR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Both vacancy figures were quoted directly from the Cross-Sectoral Hiring and Vacancies Report (CHVR), with the mean computed only over establishments holding a vacancy.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:0585c21a28bfa94711ff3da6", "occurrence_id": "7d6781205127a6a745cdab87", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Basis for the estimates.** The two coverage figures below are reproduced directly from the Panel Immunisation Coverage Registry for 2021, compiled by UNODC, and neither was recomputed for this section.\n\nThe first gives the share of enrolled children with a recorded dose and the second the share of those children whose dose was recorded by a facility that reports electronically. the source's rounding is left untouched, and the second is a subset of the first rather than an independent measure.\n\nThe panel is a fixed cohort rather than a repeated draw, so a child who left the area is retained with its last recorded status, which holds the denominator steady and lowers the first figure relative to a survey that enumerates whoever is present. A dose recorded without a date was carried as unverified rather than discarded.\n\nno definition from that source is repeated above, and the discussion does not extend them to children outside the cohort.\n\nMention: \"Panel Immunisation Coverage Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Both immunisation coverage shares were quoted directly from the 2021 Panel Immunisation Coverage Registry compiled by UNODC.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:a097d8d0ccd5ce29bafab882", "occurrence_id": "1930181a05e4fe947c3a2021", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Post-harvest handling in the season is described at a general level in the Midline Storage and Loss Survey (MSLS) for 2022, and no share from that survey enters the loss estimates below.\n\nThe survey asked holders how grain was stored and for how long, and it recorded the practice rather than measuring the loss that followed, so it can describe where a method was common but not what that method cost. The estimates that follow are built from the measured samples instead.\n\nWhat the discussion takes from the survey is the ordering of the storage methods by how often each was reported, and that ordering is consistent with the pattern the measured samples produce. Coverage followed the zones the survey reached, and the release says nothing about the zones it did not, which is a second reason its shares are not carried forward here.\n\nThe module was asked of the holder rather than of another household member, and a holding worked by hired labour answered through the manager, which is the convention that makes the practice figures describe a decision rather than the labour that carried it out.\n\nMention: \"Midline Storage and Loss Survey (MSLS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The 2022 Midline Storage and Loss Survey (MSLS) is cited for which storage methods were common, with no shares used in the loss estimates.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:0d111d7bcdac20d9bf2af20d", "occurrence_id": "8b46087e5a4fa9f5967bbb79", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Data sources.** the pair of transition figures quoted here originates in the Longitudinal Youth Transition Index (LYTI), and neither was recalculated for this section.\n\nThe first is the index level at the wave the discussion treats as the baseline and the second the movement between that wave and the most recent one published. both retain one decimal, as the source gives them, and the second is a change in an index rather than a change in a counted quantity.\n\nThe index is rebuilt at each wave from the cohort still in contact, and the weights are revised at the same time, so a movement in the line mixes the change the index is meant to track with the change in how it is assembled. The report states this and does not attempt to separate the two.\n\nNeither figure is used here as a rate, and the definitions behind them are left with the index itself.\n\nAttrition between waves is reported by the index itself and is concentrated in the younger half of the cohort, so the movement quoted here describes those still in contact rather than the cohort as it was first drawn. The report notes this and does not impute the missing members back into the series.\n\nMention: \"Longitudinal Youth Transition Index (LYTI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Both transition figures were quoted directly from the Longitudinal Youth Transition Index (LYTI), with the second read as an index movement rather than a counted change.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:7856dffd981c542ec39ccd9c", "occurrence_id": "f0773a597fba34ddcdb5f9e3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Coverage and limitations.** The compliance rates below were computed from the Midline Labour Inspection Survey (MLIS), released by KNBS, the analysis rests on it directly rather than on any figure lifted from it.\n\nCompliance was assessed against the requirement the inspector cited at the visit rather than against the full obligation set for the sector, because the survey records only what was checked and not what could have been. A workplace visited more than once contributed at its last visit, where the earlier finding had already been acted upon.\n\nThe rates were weighted to the establishment frame underlying the listing, and the derived shares carry no bounds, so the comparison between the two sectors is presented as descriptive. Cells below the size the protocol allows are printed rather than pooled, and the reader is asked to treat them as indicative.\n\nEstablishments that refused entry are recorded as inspected without a finding rather than as missing, which raises the compliance rates by an unknown amount, and the survey does not attempt to correct for it. The rates below should be read as an upper bound for that reason.\n\nMention: \"Midline Labour Inspection Survey (MLIS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"Compliance rates were computed from the Midline Labour Inspection Survey (MLIS) released by KNBS, scored against the requirement the inspector cited rather than the full obligation set.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:0ea3dbdcfb40d36025ac58cd", "occurrence_id": "d9000aebb190b6911ccd7fd6", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Coverage and limitations.** The coverage shares below were computed from the Regional Social Protection Census for 2022, and it supplies the measurement rather than the quotation.\n\nA household was counted as covered where the census found an active entitlement of any programme, and the four programmes are not separated in the shares below because a household holding two would otherwise be counted twice. The alternative was applied in a second run and is reported in the annex.\n\nEnumeration followed the household rather than the individual, so a person staying in two entitled households during the window is covered once. The shares are weighted to the listing frame, and neither derived figure has an interval, so the comparison between the two regions is read as descriptive rather than as an estimate of difference.\n\nTwo districts were enumerated after a boundary revision and are reported on the revised geography, so the regional totals mix two vintages of statistical geography. The census states which districts are affected and does not restate their earlier figures onto the new boundary.\n\nMention: \"Regional Social Protection Census\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"Coverage shares were computed from the 2022 Regional Social Protection Census, counting a household once across programmes to avoid double-counting multiple entitlements.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:a1e11b3d3859abf4fea45ca5", "occurrence_id": "267f3d86aad8a6a65acb9979", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The remittance trend used in the discussion of household income is drawn from the Cross-Sectoral Migration and Remittance Survey for 2016, and the survey is quoted for direction rather than for the level it prints.\n\nRemittances were recalled over the previous months rather than observed, and the share of receiving households was computed over households that answered the migration module rather than over all enumerated households, which is stated in the survey's own notes and lifts the share relative to a full-enumeration denominator.\n\nBecause that denominator cannot be reconstructed from the release, the level is not carried into the table; what is used is the direction the survey reports between the two periods it covers, which is consistent with the administrative series the table rests on. The definitions behind the survey's own figures are left with the source itself.\n\nRecall was collected in the dry season, when migration for the harvest has already ended, so the share of receiving households carries a seasonal tilt that the survey does not attempt to remove. That is a third reason only the direction is used here.\n\nMention: \"Cross-Sectoral Migration and Remittance Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Cross-Sectoral Migration and Remittance Survey for 2016 is cited for the direction of the remittance trend, with its level left out of the table.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:4878b609e295db70f291070f", "occurrence_id": "ea44050f7227250189c42f9a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Enumerative Cadastral Frame System (ECFS) is used as an administrative source for the year 2022. Its coverage is treated as the reference basis for the tabulation presented in this annex, with the source designation retained in full to support traceability and consistent interpretation of the reported totals.\n\nTwo quantities are reproduced directly from the ECFS record. The source reports 31 grades and 184,620 cadastral units. These figures are presented without recalculation, interpolation, or adjustment, and constitute the principal evidence for the distribution summarized in the accompanying analysis.\n\nThe reported counts are used to establish the scale and internal structure of the cadastral frame. The first quantity indicates the number of grades represented, while the second indicates the corresponding number of cadastral units. Their joint presentation permits a basic consistency review of the tabulation while preserving the values as recorded in the cited source.\n\nNo additional estimation is applied to these observations. Subsequent discussion therefore relies on the two directly reported quantities and treats them as administrative reference values for 2022.\n\nMention: \"Enumerative Cadastral Frame System (ECFS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported values of 31 grades and 184,620 cadastral units from the Enumerative Cadastral Frame System (ECFS).\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:574c0bc8f809e5ffe7c1558b", "occurrence_id": "19ea0d9a0b24ea90315ef86d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The statistical review draws on the Standardised Informal Trade Collection, an other source produced by Asian Venture Capital Journal in 2020. The collection is used as a documentary reference for the assessment of reported informal activity. Its inclusion does not alter the classification framework applied in this report, and no additional interpretation is assigned to the material beyond the figures reproduced below.\n\nTwo figures are reproduced directly from the collection. The first is 14, referring to the recorded count in the relevant series, while the second is 27, referring to the corresponding reported amount. These values are retained in their published form and are not recalculated, adjusted, or combined with estimates generated for the present assessment.\n\nThe figures are presented to support traceability between the narrative findings and the underlying reference material. They should therefore be read as quoted observations rather than as a derived indicator or as evidence of change over time. The surrounding analysis relies on separately reviewed records, while the cited collection serves only to document the two numerical observations incorporated in this section.\n\nMention: \"Standardised Informal Trade Collection\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Standardised Informal Trade Collection and reproduces the figures 14 and 27 directly from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:8aad547aabd96cbfd2979ca6", "occurrence_id": "741f8dd73ee367de6ca79a47", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Triennial Shock Coping Index indicator for 2016 as its principal evidence on the capacity to absorb and manage adverse shocks. The source is treated as a structured measure for analytical purposes, with attention to the consistency of its observations and the comparability of its recorded dimensions. No individual observation is presented as a standalone finding.\n\nThe assessment derives an average level from the available observations and calculates the direction of change across the reporting intervals. This treatment places emphasis on the aggregate pattern rather than on selected entries. The resulting measure supports an appraisal of whether coping capacity appears to strengthen, weaken, or remain broadly stable over the period represented in the source.\n\nThe calculated average is used to establish a reference point for interpretation, while the estimated trend informs the assessment of movement over time. These derived results are considered alongside the implementation evidence and are used to identify areas requiring closer monitoring. The source therefore serves as the central quantitative basis for the assessment, while the reported conclusions rely on calculations from its observations rather than on a direct reproduction of its figures.\n\nMention: \"Triennial Shock Coping Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The assessment calculates an average and trend from the Triennial Shock Coping Index indicator for 2016.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:4fd4a8cc64d5aeda01ac69d9", "occurrence_id": "b8ddee65c5c920c89209de94", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Panel Clinic Quality Survey, a survey produced by Baier and Bergstrand, is used as the evidentiary basis for the service-readiness assessment. The source is treated as a reference record rather than as a basis for constructing additional indicators. Its reported observations are considered sufficiently specific for the limited claim made below.\n\nTwo figures are reproduced directly from the source. It reports that 31 clinics met the stated readiness threshold, while 68 percent of the assessed facilities satisfied the corresponding quality criterion. These values are retained in their reported form and are not recalculated, rounded, or combined with information from another source.\n\nTaken together, the figures support the narrower conclusion that minimum service capacity was present in a substantial share of the assessed facilities, although coverage was not universal. The evidence therefore supports continued attention to uneven implementation and to the operational conditions associated with quality compliance. No inference is made beyond the relationship documented by the source.\n\nMention: \"Panel Clinic Quality Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage takes the figures 31 clinics and 68 percent directly from the Panel Clinic Quality Survey to support a limited claim about service readiness.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:00e5f696ae1879efde0cf2ed", "occurrence_id": "e7d8c89c26c57aadcc1a75de", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The National Occupational Safety System is used as an administrative source for 2020 in the assessment of occupational safety reporting. Its records provide the central empirical basis for examining changes in reported incidents and related measures over the available observation period. The source is considered suitable for analytical use because it reflects routine reporting rather than a one-time inquiry.\n\nThe analysis does not reproduce individual entries or present isolated source figures as findings. Instead, reported observations are arranged into a consistent series, from which a general trend is calculated. An average is also derived to indicate the prevailing level across the observations and to reduce the influence of unusually high or low entries. These calculations support comparison within the source while avoiding claims that cannot be sustained by the recorded coverage.\n\nInterpretation remains subject to the quality and completeness of administrative reporting. Changes may reflect differences in notification practice, recording discipline, or the number of reporting units, in addition to changes in underlying safety conditions. The resulting trend and average are therefore used as summary indicators for implementation monitoring. They provide a structured basis for identifying movement in occupational safety outcomes and for determining whether subsequent reporting warrants closer review.\n\nMention: \"National Occupational Safety System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage derives a trend and an average from the National Occupational Safety System for 2020.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:aa7f9e9d178848aace0d5c23", "occurrence_id": "d1907ec2c45ff85ee767bbef", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Decennial Fuel and Energy Use Survey, produced by Latinobarómetro Corporation as a survey for 2015, is used as the principal evidentiary source for the assessment of fuel and energy use. Its definitions provide a consistent basis for interpreting reported consumption and expenditure measures, while limiting reliance on reconstructed estimates.\n\nTwo figures are quoted directly from the source. The survey records that 62.4 percent of observed households reported regular use of the principal fuel category, while 27.8 percent reported expenditure within the corresponding energy category. These values are reproduced without recalculation and constitute the central quantitative evidence for the comparison presented here.\n\nThe figures should be read as reported shares rather than as estimates of total demand. Differences in household circumstances may affect the relationship between reported use and expenditure, and the survey does not by itself establish causal explanations for the observed distribution. Nevertheless, the paired measures provide a sufficiently clear reference for monitoring subsequent implementation results and for checking the internal consistency of related reporting.\n\nMention: \"Decennial Fuel and Energy Use Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The author took the directly reported figures of 62.4 percent and 27.8 percent from the 2015 Decennial Fuel and Energy Use Survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:b74c59b212d329efa227a64b", "occurrence_id": "4a177af587fecc8606180dbe", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The employment profile draws on the Municipal Enterprise and Jobs System, an administrative source covering Africa in 2021. Its records provide the supporting basis for the reported distribution of municipal enterprises and associated jobs across participating reporting units. The source is used here to substantiate the assessment that enterprise activity is concentrated in a limited number of local economies.\n\nThe underlying records were assembled from routine returns, reviewed for internal consistency, and harmonised to address differences in terminology and reporting practice. Duplicate entries were removed where identification was possible, while unresolved observations were retained with appropriate caution. The resulting series was placed in the project archive to support subsequent verification and controlled reuse.\n\nInterpretation remains subject to the coverage and completeness of the source material. In particular, the reported employment profile should be read as an administrative representation of recorded activity rather than as a complete measure of all enterprise operations. This limitation does not affect the value of the dataset for comparing recorded patterns within the covered scope, provided that comparisons are made on a consistent basis.\n\nMention: \"Municipal Enterprise and Jobs System\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The passage records the collection, harmonisation, and archiving of the Municipal Enterprise and Jobs System for use in supporting an employment profile.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:aaaf4b71760caab4e7fa5f8c", "occurrence_id": "22dbc14b0ceddb99d27048e1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Departmental Enterprise and Jobs Registry database covering Rio de Janeiro is used as the principal empirical basis for the assessment of enterprise activity and employment conditions. Its records provide a consistent observation structure for examining the distribution of establishments, the scale of registered employment, and changes across the available reporting periods. No single record is treated as representative of the overall pattern.\n\nThe analysis derives aggregate counts from the database and calculates average values across the observed periods. It also estimates the direction and magnitude of change by comparing successive observations, rather than reproducing isolated figures from the registry. This procedure reduces the influence of short-term variation and supports a more stable reading of enterprise dynamics. The resulting measures are used to identify whether activity is expanding, contracting, or remaining broadly unchanged.\n\nInterpretation is limited to the coverage and definitions embodied in the Departmental Enterprise and Jobs Registry database. The calculated averages and trends therefore describe registered activity in Rio de Janeiro and should not be read as estimates of unregistered enterprises or employment outside the source framework. The results are presented as analytical indicators for implementation monitoring, with emphasis on comparability over time and consistency in the treatment of records.\n\nMention: \"Departmental Enterprise and Jobs Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses the Departmental Enterprise and Jobs Registry database to calculate average enterprise and employment counts and derive trends for Rio de Janeiro.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:27663a2ee8845032d936d84d", "occurrence_id": "a3ec7260b877a4fd6bd779e0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: For this assessment, the Baseline Administrative Boundary Census (BABC), a census covering Mexico City in 2017, is used as the reference source for the geographic baseline. Its administrative coverage is treated as the common basis for interpreting the distribution of recorded units and localities.\n\nThe census reports 31 administrative units and 184 localities. These two figures are reproduced directly from the source and are used without recalculation. They support the assessment that the baseline provides broad geographic coverage while retaining a sufficiently detailed administrative structure for subsequent comparison.\n\nThe reported counts should be read as the position established by the cited source rather than as an independently derived estimate. In the present analysis, they anchor the description of coverage and provide a fixed point for reviewing completeness, territorial differentiation, and the consistency of subsequent records. No adjustment is applied to either figure in the accompanying discussion.\n\nMention: \"Baseline Administrative Boundary Census (BABC)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The author takes the directly reported figures of 31 administrative units and 184 localities from the BABC census covering Mexico City in 2017.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:b20c7d956055abb74e3bac67", "occurrence_id": "e079c22ffb754ec6d72b7e6d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Continuous Settlement and Elevation Registry (CSER) database is treated as a reference for the 2015 observation used in the analysis. Its entries provide a consistent basis for checking whether the reported settlement count is within the range indicated by the underlying registry. The citation is retained in the assessment annex because the source supports a specific verification point rather than the full set of conclusions.\n\nFor this purpose, the CSER record is compared with an estimate obtained from a separate source. The comparison is limited to the settlement count and does not imply that the two sources were produced through identical procedures. Agreement within the stated tolerance is taken as supporting evidence for the estimate, while any discrepancy is recorded for reconciliation in the results table. No adjustment is applied solely because the registry reports a different value.\n\nThe registry is therefore used as a cross-check on one claim in the broader discussion of settlement coverage. It does not replace the estimate used for the main calculation, and it is not used to infer changes beyond the 2015 reference. The evidence is considered adequate for validation at the level of the reported aggregate, subject to the limitations noted in the assessment.\n\nMention: \"Continuous Settlement and Elevation Registry (CSER)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The source provides a 2015 reference for cross-checking a settlement-count estimate.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:aba8943b7aa1e914eb473592", "occurrence_id": "0e4a9d580b74e032207d6320", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Metropolitan Early Childhood Registry (MECR) database is referenced in the assessment as a background source for the 2021 reporting frame. Google Scholar is identified as producing the database. The citation is retained to establish the documentary setting of the discussion and does not serve as the basis for a reported estimate or classification.\n\nThe registry is not used to establish levels, distributions, changes, or relationships, and no indicator is derived from its entries. No values are transcribed, recalculated, or compared. Its role is limited to a brief reference supporting the description of available administrative evidence, with the substantive findings based on the analysis presented elsewhere in the report.\n\nThis treatment is intentional. The reference is incidental to the assessment and is not presented as evidence for a finding. Readers should interpret the entry as an orientation point concerning the existence of a relevant database, rather than as a source from which measurements, rates, or totals have been taken. The citation therefore has documentary relevance but no numerical contribution to the results.\n\nMention: \"Metropolitan Early Childhood Registry (MECR)\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The passage uses the 2021 MECR database, identified as produced by Google Scholar, for background only and takes no values from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:54744aedcbd4e213754162d5", "occurrence_id": "634adc483a16a54dfe430640", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Audited School Feeding Survey (ASFS) is a survey produced by the municipal government and covering China. It provides a direct statistical basis for assessing the reach of school feeding arrangements and the extent to which provision is operational across the surveyed system. The source is used as reported, without introducing adjustments to its coverage definitions or tabulation procedures.\n\nThe available evidence supports the conclusion that school feeding is established across a substantial share of the system, although delivery remains uneven. Two figures are quoted directly from the survey. It reports coverage of 84 percent of schools and records 1,260 schools with an active feeding arrangement. These observations support the broader finding that implementation has progressed beyond isolated provision while retaining material differences in institutional reach.\n\nThe figures are presented as reported results rather than as estimates derived from other administrative material. They are used to anchor the assessment of programme scale and to indicate the magnitude of the implementation base. No additional calculation is required for the claim made here, and the reported counts should be interpreted within the scope of the survey. The ASFS therefore serves as the principal documentary reference for this specific comparison and for the associated statement on coverage.\n\nMention: \"Audited School Feeding Survey (ASFS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage takes directly from the Audited School Feeding Survey (ASFS) the figures of 84 percent coverage and 1,260 schools with an active feeding arrangement.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:c702d327f7ebd02db8d46492", "occurrence_id": "42a93623148ebfceaece1e09", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Subnational Vocational Training System (SVTS) is used as an administrative source for 2020. It provides the principal numerical basis for the observation presented in this section and is treated as a direct record rather than as an estimate derived from secondary material.\n\nTwo figures are reproduced directly from the source. The record identifies 31 registered training centres and reports that 68 percent of enrolled participants completed the relevant training activity. These values are retained without adjustment, recalculation, or conversion, so that the reported evidence remains aligned with the underlying administrative record.\n\nThe figures support a restrained interpretation of system coverage and participation. The count of 31 indicates the recorded institutional base, while the proportion of 68 percent provides the corresponding measure of completion. Taken together, they establish the scale of provision and the observed level of participation for 2020. No additional inference is made from these observations, and the figures should not be read as a measure of outcomes beyond the coverage of the source.\n\nMention: \"Subnational Vocational Training System (SVTS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the figures 31 registered training centres and 68 percent completion directly from SVTS for 2020.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:898a908c62e11877d4f69d6a", "occurrence_id": "aca8fad0e8d0226015437989", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Endline Labour Force Survey (ELFS) provides the principal statistical basis for the labour market assessment. It is a survey produced by GAD, with the reference year recorded as 2018. The source is retained as the central evidence for deriving the indicators presented below, and its designation is used consistently throughout the analysis.\n\nThe reported observations are processed to calculate changes across the available categories and to derive an average level for the assessment period. These calculations provide a summary of the direction and distribution of labour market outcomes rather than reproducing individual figures from the source. The resulting measures are used to support interpretation of the observed pattern and to identify whether outcomes moved consistently or varied across categories.\n\nThe calculated trend and average are therefore treated as analytical results based on the Endline Labour Force Survey (ELFS), rather than as values quoted directly from the survey. This approach preserves the source as the evidentiary foundation while making the reported measures comparable within the assessment. All interpretations in this section should be read in conjunction with the calculated series derived from the 2018 observations.\n\nMention: \"Endline Labour Force Survey (ELFS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the 2018 Endline Labour Force Survey (ELFS), produced by GAD, to calculate a trend and an average.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:5516f02d8957511f4c97d166", "occurrence_id": "7feb2641ad06bada1ad19123", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Standard Water and Climate Registry database for Eastern Europe in 2017. The source is used as a reference point for reviewing reported access conditions and the recorded exposure of basic water services to climate-related disruption. Its coverage is treated as descriptive rather than as a basis for estimating changes over time.\n\nThe available evidence indicates that service continuity remains uneven across the monitored areas, particularly where maintenance capacity and seasonal water availability are constrained. The assessment reproduces two figures directly from the cited database: 42 percent of the recorded service areas reported an interruption during the reference period, while 68 percent reported having an operational contingency measure. These figures support the broader finding that formal preparedness is more common than uninterrupted service delivery.\n\nInterpretation is limited to the information recorded in the source and does not extend beyond its stated coverage. The figures are therefore presented as reported observations for the specified period, rather than as estimates of current conditions or evidence of a causal relationship. This treatment is consistent with the use of the registry as a supporting source within the wider assessment.\n\nMention: \"Standard Water and Climate Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported figures of 42 percent and 68 percent from the Standard Water and Climate Registry database.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:4c77a13eeb3b957068071aae", "occurrence_id": "3b8773da5f8f5bea870b45ca", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Continuous Ward Outline Census is used as the primary statistical reference for the administrative profile of Sri Lanka. Produced by IVAP and referring to 2019, this census provides a consistent basis for describing ward-level coverage and the distribution of recorded households. Its use is appropriate for the present assessment because the source is structured around direct enumeration rather than retrospective estimation.\n\nTwo figures are reproduced directly from the source. The census records 31 wards within the assessed coverage and reports 72,418 households across those wards. These values are presented without recalculation or adjustment, and they constitute the central quantitative evidence for the paragraph. The figures should therefore be read as reported totals from the cited source, rather than as derived indicators or projections.\n\nThe reported counts support a concise description of the scale of the administrative units under review. They also provide a reference point for subsequent comparison within the assessment, while avoiding assumptions about changes not documented in the source. No additional interpretation is applied to the figures at this stage. The Continuous Ward Outline Census is consequently retained as the evidentiary basis for the ward and household totals used here.\n\nMention: \"Continuous Ward Outline Census\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the source’s reported figures of 31 wards and 72,418 households.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:1c35bec651e70beb38098ac5", "occurrence_id": "3ba27bd2a37864e5ea743544", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Data sources.** The High-Frequency Land Cover Change Collection is recorded as an other source produced by UNDESA for Nairobi County and referring to 2024. It is used here as the primary factual basis for the land-cover observations presented in this paragraph. The source is cited in its stated form to preserve the connection between the reported measures and the underlying collection.\n\nTwo percentages are reproduced directly from the collection. The first indicates that 31.4 percent of the assessed area was associated with the principal recorded land-cover condition. The second indicates that 12.8 percent was associated with the corresponding secondary condition. These values are presented as reported measures rather than as estimates derived through further calculation.\n\nThe figures provide a concise basis for interpreting the distribution observed in Nairobi County. Their use is limited to describing the proportions reported for 2024, and no additional adjustment or projection is applied. This treatment supports consistent reference to the cited evidence within the assessment and avoids introducing measures not contained in the source.\n\nMention: \"High-Frequency Land Cover Change Collection\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported values of 31.4 percent and 12.8 percent from the cited collection.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:36edd19e8c0c8aa080dbab5b", "occurrence_id": "f06d344472b9cdae1abc7dc9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Cross-Sectoral Drought and Resilience Index (CDRI) as its principal indicator for Morocco in 2015. The source provides a structured basis for examining drought exposure and resilience conditions across the assessed dimensions. Its use is limited to the stated geographic and temporal coverage, with no inference made regarding conditions outside that scope.\n\nThe analysis does not reproduce a single index value as a headline result. Instead, it calculates an average from the component observations and uses the resulting series to identify the prevailing direction of change. This treatment supports comparison across the dimensions included in the indicator while reducing the influence of isolated observations. The derived measure is therefore an analytical output based on the source rather than a figure quoted directly from it.\n\nResults indicate differentiated resilience conditions within the covered assessment units. The calculated average is used to establish the central tendency, while the estimated trend provides an indication of whether observed conditions are improving or weakening over the period represented by the available evidence. Interpretation remains cautious where component values diverge, and the aggregate result is not treated as a substitute for dimension-level review. The source is retained as the central evidentiary reference for the assessment and for subsequent monitoring of drought-related resilience.\n\nMention: \"Cross-Sectoral Drought and Resilience Index (CDRI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses the 2015 CDRI indicator for Morocco to calculate an average and derive a trend rather than quote a source value.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:041cb017cae74e4bc5a4fb8e", "occurrence_id": "1482c363c0271b42185dc836", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Harmonised Minimum Needs Survey (HMNS) is used as the principal empirical basis for the assessment. It is a 2016 survey produced by SEDESOL and provides a consistent set of observations on household conditions and access to essential services. The material is treated as an analytical input rather than as a source for reproducing isolated reported figures.\n\nFor the purposes of this assessment, the HMNS records are consolidated into comparable groups and used to calculate average levels across the observed measures. The resulting averages provide a summary of the distribution of minimum needs and reduce the influence of unusually high or low individual observations. This procedure supports comparison across the reporting categories while retaining the underlying coverage of the survey.\n\nThe same calculations are used to derive a trend in the incidence of reported needs across the ordered categories available in the HMNS. The trend is interpreted as a directional pattern in the compiled evidence, rather than as a direct statement reproduced from the survey documentation. Differences between categories are therefore assessed through calculated means and changes in those means.\n\nThe findings should be read as estimates generated from the HMNS material. They describe the pattern obtained through aggregation and calculation, and not a verbatim transcription of any single HMNS result. This treatment provides the basis for the assessment of relative need and for the subsequent monitoring framework.\n\nMention: \"Harmonised Minimum Needs Survey (HMNS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the 2016 SEDESOL HMNS survey to calculate averages and derive a trend from its observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:f8f4bcf947c53142c139e41f", "occurrence_id": "99104349db42004657e6cdf7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Statistical Working Conditions Registry (SWCR) database is used as the central empirical source for the assessment of working conditions. Its records provide a consistent basis for examining the distribution of reported conditions across the observation set. The analysis treats the entries as comparable observations and applies the same inclusion rules throughout, with attention to coverage, completeness, and internal consistency.\n\nResults are not presented as a transcription of individual registry entries. Instead, the analysis aggregates the relevant observations, calculates average levels for each comparable period, and follows changes across the sequence. This approach reduces the influence of isolated records and supports a measured assessment of direction and persistence. Where the number of observations varies, averages are interpreted together with the breadth of the underlying records rather than as standalone estimates.\n\nThe resulting trend is used to inform the assessment of implementation performance and to identify areas requiring continued monitoring. Any apparent movement is considered meaningful only where it is supported by repeated observations and remains compatible with the registry’s internal structure. The SWCR database therefore serves both as the evidentiary basis for the analysis and as a check on the consistency of the reported pattern. Detailed calculations should be read as derived measures, not as values reproduced directly from the source.\n\nMention: \"Statistical Working Conditions Registry (SWCR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis calculates average levels and identifies a trend from successive observations in the SWCR database.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:8addf52c0b375005878dd522", "occurrence_id": "6616d0aeb915b007954e67fe", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Multi-Topic Arrival and Registration Report is the principal report used for the assessment of recorded arrivals and completed registrations. It provides the central evidence for examining changes in intake activity and the relationship between initial arrival and subsequent registration. The source is treated as an operational record rather than as a narrative account.\n\nThe analysis does not reproduce individual entries from the report. Instead, the recorded counts are grouped into comparable observation periods, and an average is calculated to establish the prevailing level of activity. The resulting measure reduces the effect of irregular reporting intervals and provides a stable basis for comparison across the series.\n\nA trend is then derived from the same observations by comparing the calculated period values. This approach indicates whether arrivals and registrations increased, declined, or remained broadly stable over the sequence examined. The interpretation therefore rests on the direction and magnitude of the calculated movement, rather than on any single reported figure.\n\nThe findings should be read as an analytical summary of the information contained in the Multi-Topic Arrival and Registration Report. Differences between arrivals and registrations may reflect timing between the two processes, incomplete follow-up, or variations in recording practice. Accordingly, the average and trend are used to support implementation monitoring and to identify patterns requiring further operational review.\n\nMention: \"Multi-Topic Arrival and Registration Report\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage derives an average and a trend in arrivals and registrations from the Multi-Topic Arrival and Registration Report.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:4365b746040a4c2f569881e1", "occurrence_id": "f04196fecc630577490eb7b3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Provincial Diagnostic Capacity Registry database is used as the sole evidentiary source for this assessment of diagnostic capacity. Its entries are treated as reported records rather than estimates generated for the present analysis. No adjustment, projection, or external validation has been applied to the figures reproduced below.\n\nTwo figures are taken directly from the registry. It records 31 facilities with an operational diagnostic function, while 68 percent of facilities report uninterrupted access to the required diagnostic service. These values are presented in their recorded form and are not combined into a derived indicator. The first figure establishes the observed scale of available capacity; the second indicates the reported continuity of access among the recorded facilities.\n\nThe results should therefore be read as a concise description of documented capacity at the time represented in the source. Differences in reporting completeness may affect the coverage of individual entries, and the figures do not by themselves establish service quality, geographic equity, or changes over time. They nevertheless provide the central quantitative basis for the assessment and support a focused review of facility functionality and access continuity. Any subsequent operational prioritisation should retain the distinction between the count of facilities and the reported percentage.\n\nMention: \"Provincial Diagnostic Capacity Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Provincial Diagnostic Capacity Registry database for two direct figures, 31 facilities and 68 percent reporting uninterrupted diagnostic access.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:32859480e6d288c1eb8b1ef9", "occurrence_id": "d38a38c3ad7d75cfe14b1d38", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: For the assessment, the Continuous Early Warning Registry (CEWR) database provides the evidentiary basis for the numerical statements concerning 2016. It is used as a recorded administrative source, with the figures retained in the form in which they appear in the database. No supplementary source is introduced for the values reported here.\n\nTwo figures are reproduced directly from the database. It records 31 cases in one reviewed group and 47 cases in a second reviewed group. These values are presented as reported counts, rather than as estimates, projections, or results derived through a separate calculation. Their inclusion is intended to preserve the direct quantitative content of the registry.\n\nThe figures support the assessment of recorded activity during the relevant period. They should therefore be read as observations contained in the CEWR database and not as a measure constructed from external information. The analysis does not adjust, combine, or otherwise transform the two counts.\n\nFor consistency, the source designation remains Continuous Early Warning Registry (CEWR) database throughout the assessment. The direct reproduction of 31 cases and 47 cases permits the numerical evidence to be checked against the cited source and limits interpretation to what is documented there.\n\nMention: \"Continuous Early Warning Registry (CEWR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage reports the directly reproduced figures of 31 cases and 47 cases from the CEWR database for 2016.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:d6c13cbf2330fec2ec31ea17", "occurrence_id": "22c65ab15bc5f85a79f88ce7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses Midline Fuel and Energy Use Microdata as the principal evidence for the analysis. This microdata source covers Timor-Leste and refers to 2019. It provides the underlying observations required to examine household fuel selection, energy use, and the distribution of reported consumption across the population. No additional source is used to establish the patterns described in this section.\n\nThe observations are transformed into comparable measures before interpretation. Average fuel use is calculated across the reported records, and corresponding average energy use is derived using the same analytical basis. These calculations provide a more stable representation than isolated responses and support comparison across the relevant use categories. The resulting measures are treated as estimates for assessment purposes rather than as individual reported figures.\n\nA trend is then derived from the calculated averages and the ordering of the observed use categories. The analysis considers whether fuel and energy use are concentrated, dispersed, or changing across the records, while avoiding claims that cannot be supported by the available observations. The findings therefore reflect patterns computed from Midline Fuel and Energy Use Microdata, rather than figures reproduced directly from the source. This approach also permits the assessment to identify broad differences in use without implying greater precision than the microdata support.\n\nMention: \"Midline Fuel and Energy Use Microdata\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses 2019 microdata for Timor-Leste to calculate average fuel and energy use and derive an overall trend.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:a2fb3413b131531b87968128", "occurrence_id": "51d8022e9cc6c8e2ddee11df", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis uses the Small-Area Parcel Footprint Collection (SPFC), classified as an other source referring to 2022. The collection provides parcel-level footprint observations suitable for aggregation and comparison. Its use in this assessment is limited to the evidence required for the summary measures reported below.\n\nIndividual records are not reproduced as headline results. Instead, the observations are grouped by the available reporting categories, and the mean footprint is calculated for each group. The resulting series supports an assessment of the direction and relative magnitude of variation within the observed distribution.\n\nThe calculated average provides a stable summary of the recorded footprints, while the derived trend indicates whether larger or smaller parcels are more prevalent across the reporting categories. These measures are used to support interpretation of the distribution and to identify broad differences that may be obscured by reliance on isolated records. The results should therefore be read as analytical estimates derived from the source rather than as figures directly quoted from it.\n\nMention: \"Small-Area Parcel Footprint Collection (SPFC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The assessment derives grouped averages and a distributional trend from the 2022 SPFC observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:d95ae8d715c27ed5df91fc18", "occurrence_id": "fb740f2469ddb4963e73c47f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: For the reference year 2017, the Integrated Protection Incident Collection (IPIC) is classified as an other source in the evidence base used for this assessment. The collection is cited as the underlying record for the two quantitative observations reported below, with no supplementary attribution or version information required for interpretation.\n\nTwo figures are reproduced directly from IPIC. The collection records 31 incidents during the reference period and identifies 14 affected persons in the corresponding records. These values are presented as reported and are not recalculated, aggregated with other materials, or interpreted as a time series.\n\nThe figures provide a limited but specific basis for assessing the scale of recorded protection concerns. They should be read as counts contained in the cited source rather than as estimates of total occurrence. Any comparison with subsequent reporting would require a separate observation period and a documented assessment of coverage and consistency.\n\nFor implementation purposes, the source is retained as a traceable reference for the two figures used in this annex. The reported counts support the immediate description of recorded incidents and affected persons, while broader conclusions remain subject to the scope and completeness of the underlying collection.\n\nMention: \"Integrated Protection Incident Collection (IPIC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the figures 31 incidents and 14 affected persons directly from IPIC for 2017.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:1c0646073428e6c3d8d84c37", "occurrence_id": "f951862a6cd1a4d48ad2a1d0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Coordinated Fuel and Energy Use Survey is a survey produced by the Brazilian Ministry of Labor for 2015. It is used here as the principal documentary basis for the assessment of reported household energy use. The source is treated as a statistical record for the stated period, with its published classifications retained for comparability.\n\nTwo figures are taken directly from the survey and reproduced without recalculation. It reports that 31 percent of households used the specified fuel category, while 18 percent reported the associated energy-use condition. These values provide the central evidence for the assessment and are presented as observed proportions rather than as estimates derived from a model.\n\nThe figures indicate a material difference between the two reported conditions, although the survey alone does not establish the causes of that difference. Interpretation therefore remains limited to the distributions recorded in the source. The assessment does not extend the results beyond the population represented by the survey or infer changes outside the stated reference period. Any comparison with operational targets should consequently preserve the original definitions and denominator used for each reported proportion.\n\nMention: \"Coordinated Fuel and Energy Use Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The assessment takes the directly reported figures of 31 percent and 18 percent from the Coordinated Fuel and Energy Use Survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:88939a26a40df62b45caec43", "occurrence_id": "87d6229e93f030d989735c7c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Panel Population Grid Registry is cited as a database produced by Twitter for 2022. It is retained in the assessment to establish the relevance of population coverage when interpreting the broader findings. The citation is therefore documentary and limited in scope.\n\nThe registry supports the statement that population-based evidence provides a useful frame for reviewing implementation conditions. This statement forms one element of a wider discussion concerning measurement, coverage, and interpretation. No observation, count, rate, or other value from the registry is incorporated into the assessment, and no indicator is recalculated from its contents.\n\nThe material is not treated as an operational input, a benchmark, or a basis for comparison. Its function is confined to substantiating the stated contextual point while the analysis relies on the evidence presented elsewhere in the report. This approach preserves a clear distinction between background documentation and the information used to reach the assessment conclusions.\n\nMention: \"Panel Population Grid Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The source provides contextual support for a claim about population coverage, while none of its database values are used.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:f2dbba66c1795364a746c675", "occurrence_id": "9903ed21e3b6640873ef86f0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Follow-Up Apprenticeship Survey provides the principal evidence for assessing apprenticeship outcomes in Lusaka. It is treated as a survey of participant conditions and post-apprenticeship status, with the available observations providing the basis for the analysis. The source is retained in its reported structure, while individual responses are reviewed for completeness and comparability before aggregation.\n\nThe analysis does not reproduce isolated responses as headline results. Instead, it calculates average values across the relevant observations and examines their movement across the ordered observation periods. This approach reduces the influence of exceptional cases and supports a more stable assessment of whether outcomes improved, remained broadly unchanged, or weakened. The resulting trend is used to inform interpretation of programme performance rather than to imply precision beyond the coverage of the source.\n\nResults derived from the survey indicate the general direction of change in participation, completion, and subsequent engagement. These measures are presented as analytical summaries, not as direct quotations from individual records. Where subgroup differences are material, they are considered alongside the overall average, subject to the limitations imposed by response coverage and the consistency of recorded information. The evidence therefore supports monitoring and operational judgement, while avoiding claims that cannot be substantiated through the observed records.\n\nMention: \"Follow-Up Apprenticeship Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives average outcomes and an overall trend in apprenticeship results from the Follow-Up Apprenticeship Survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:307591d7fcc5a7af9e907748", "occurrence_id": "012e800a94f1786ef483531e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment relies on the Recurrent Displacement Tracking Index (RDTI) as a consistent reference for interpreting reported displacement conditions. MOH produces the RDTI, which is used here as an indicator of the scale and persistence of needs recorded through the reporting system. The source is retained without adjustment to preserve the evidentiary basis of the assessment.\n\nFor the present analysis, two figures are reproduced directly from the Recurrent Displacement Tracking Index (RDTI). The record reports 31 households requiring follow-up assistance. It also reports that 18 percent of recorded households experienced a recurrent displacement event. These figures are presented as reported and are not recalculated, combined with other records, or used to derive a separate rate.\n\nThe two observations provide a limited but direct basis for interpreting the distribution of immediate needs and the recurrence of displacement. The first figure indicates the number of households identified for follow-up, while the second indicates the reported share associated with recurrence. Their use is therefore descriptive rather than predictive. Any comparison with subsequent reporting should preserve the definitions and coverage applied by MOH in the RDTI, since changes in recording practice could affect apparent movements in either measure.\n\nMention: \"Recurrent Displacement Tracking Index (RDTI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the RDTI figures of 31 households requiring follow-up assistance and 18 percent experiencing recurrent displacement.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:fe5fbff09f237e868f1bd861", "occurrence_id": "b4cc69d36eac0026ab578777", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Comparative Asset Ownership Survey (CAOS) is used as the principal survey source for Jordan in 2024. Its observations provide the central empirical basis for assessing patterns of household asset ownership across the analytical population. The source is treated as a consistent reference for constructing comparable measures rather than as a set of isolated results.\n\nThe analysis derives average ownership levels from the reported observations and compares those averages across the relevant groups. It also calculates a trend from the distribution of ownership measures, allowing the assessment to identify gradual changes in access to selected assets. Individual observations are therefore not reproduced as headline findings; they are incorporated into summary calculations that support interpretation of the broader pattern.\n\nThe resulting estimates are used to describe the direction and scale of change while limiting emphasis on short-term variation. This treatment improves comparability within the assessment and provides a more stable basis for monitoring household conditions. All calculated averages and trend statements in this section are derived from the Comparative Asset Ownership Survey (CAOS), with the source retained as the principal evidence for the analysis.\n\nMention: \"Comparative Asset Ownership Survey (CAOS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives average asset ownership levels and trends for Jordan in 2024 from the Comparative Asset Ownership Survey (CAOS).\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:cdf23a10bba6d6d73f82a15f", "occurrence_id": "b23d2c144dee0218eb3174bb", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Continuous Apprenticeship Survey is a survey produced by PWT and is used here as the documentary basis for the participation estimates. Its tabulated results provide a consistent reference for assessing the distribution of apprentices across the categories considered in the analysis. No adjustment is applied to the cited observations.\n\nThe evidence indicates a marked difference between the lower and higher parts of the distribution. The reported share is 31 percent for the first category and 68 percent for the comparison category. These two figures are reproduced directly from the Continuous Apprenticeship Survey and support the broader finding that participation is uneven rather than uniformly distributed.\n\nThe figures are presented as descriptive evidence and are not used to construct an additional indicator. Their role is limited to substantiating the paragraph’s assessment of the observed gap. The source therefore provides a narrow but sufficient basis for the statement, while the surrounding interpretation remains restricted to the values recorded in the survey.\n\nMention: \"Continuous Apprenticeship Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported figures of 31 percent and 68 percent from the Continuous Apprenticeship Survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:8919ef1a3f05bd56e7f1b202", "occurrence_id": "ed77d5353f56c030e9aac791", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Data sources.** The National Built-Up Surface Dataset is used as a geospatial source produced by University of Dayton for 2020. It provides a consistent spatial basis for examining the distribution of built-up surfaces and for relating observed settlement patterns to the implementation indicators assessed in this annex.\n\nThe analysis does not reproduce a reported figure from the dataset. Instead, it aggregates the mapped observations across the reporting units and calculates an average built-up surface measure. This derived measure is then compared with the corresponding implementation observations to identify the direction and relative strength of change.\n\nThe resulting trend supports the assessment that expansion of built-up surfaces is uneven across the units considered. The finding is used as one element of the broader assessment and is not treated as a standalone measure of project performance. Interpretation remains subject to the coverage and spatial resolution of the underlying observations, while the calculated average provides a transparent basis for comparison within the analysis.\n\nMention: \"National Built-Up Surface Dataset\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The passage derives an average and a trend assessment from the National Built-Up Surface Dataset rather than quoting a reported figure.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:0730c3df0e9d63ce5c6ce0af", "occurrence_id": "8fb6e8d304b975cb604d3091", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Enumerative Hiring and Vacancies System is treated as an administrative source produced by Hyperion for 2016. Its records provide the central empirical basis for the assessment of staffing movement and vacancy pressure. The material is used as a consistent series rather than as a set of isolated reported values.\n\nFor each observation, the analysis aligns recorded hiring activity with recorded vacancies and derives an average for the period under review. The resulting measure reduces the effect of short-term fluctuations and supports comparison across successive observations. This calculation indicates whether staffing demand generally increased, declined, or remained broadly stable.\n\nThe derived trend shows the direction and approximate scale of change relevant to implementation monitoring. Particular observations are retained to identify departures from the average, but interpretation rests on the calculated series as a whole. The source therefore functions as the principal evidence for assessing movement in hiring and vacancies, while the reported averages and trend are analytical results generated from its records.\n\nMention: \"Enumerative Hiring and Vacancies System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses Hyperion’s 2016 administrative records from the Enumerative Hiring and Vacancies System to calculate an average and derive a hiring and vacancy trend.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:ffe8c2728262149815ff38ec", "occurrence_id": "4075547ee0646da727e4c22f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The data source is the Annual Asset Ownership Index (AAOI), an indicator produced by Michelin for Vietnam and referring to 2015. It provides the principal evidence for the assessment of asset ownership patterns and is treated as a structured basis for calculation rather than as a passage from which isolated figures are reproduced.\n\nThe analysis derives an average across the asset measures reported in the indicator and uses the resulting values to identify the direction and relative strength of the observed trend. This approach reduces the influence of any single measure and provides a comparable summary for interpretation within the assessment. The calculation is based on the underlying information in the Annual Asset Ownership Index (AAOI), with attention to consistent treatment of the available measures.\n\nThe resulting summary is used to support the findings on asset ownership, while preserving the distinction between the reported indicator and the analytical results derived from it. No individual value is presented as a direct quotation from the source. Instead, the computed average and category-level trend provide the basis for describing the distribution and for identifying areas requiring further review. This treatment gives the source a central evidentiary role without implying that the calculated results were published directly by Michelin.\n\nMention: \"Annual Asset Ownership Index (AAOI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage identifies the Annual Asset Ownership Index (AAOI) as Michelin’s 2015 indicator for Vietnam and explains that an average and category-level trend are computed from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:f90c44bf7017150330c77a1d", "occurrence_id": "8c19e7de9e549ca5594efedc", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis uses the Cross-Sectoral Household Consumption Index as a common reference for interpreting household consumption patterns. The source is the 2021 indicator produced by Ghana Statistical Service and is treated as the central statistical basis for the comparison presented below.\n\nThe reported values of 47.8 percent and 62.4 percent are reproduced directly from the indicator. The difference between these figures provides the principal evidence for the observed variation across the consumption measure, without further adjustment or transformation in this assessment.\n\nThe figures are used to support a limited descriptive reading of the distribution. They indicate that the measured level is not uniform across the relevant household categories and that the higher reported value is materially above the lower one. No additional calculation is required to establish this distinction.\n\nAccordingly, the index is retained as a reference point for monitoring and interpretation. Its values are presented in their reported form to preserve comparability and to avoid introducing assumptions beyond those contained in the source.\n\nMention: \"Cross-Sectoral Household Consumption Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage directly reproduces 47.8 percent and 62.4 percent from the Cross-Sectoral Household Consumption Index.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:7a63eeb71ca394acb3a07fd1", "occurrence_id": "eca4e1761e1363adaa302210", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws its central evidence from the Clustered Inequality Registry database for 2015. The source is treated as a reference record for the distribution of observations across the relevant categories, with no adjustment to the values reproduced in the analysis.\n\nTwo figures are taken directly from the registry. The lower grouping is recorded at 31 percent, while the upper grouping is recorded at 47 percent. These values are presented as reported measures rather than as estimates generated through a separate calculation. Their inclusion permits the assessment to describe the observed difference between the two groupings without introducing an additional analytical series.\n\nThe figures support a limited interpretation. The upper grouping has the larger recorded share, exceeding the lower grouping by 16 percentage points when the two reported values are compared. This comparison is confined to the evidence contained in the cited database and is not used to infer change over time. The results therefore establish the distribution observed in 2015 while leaving broader causal interpretation outside the scope of this annex.\n\nFor purposes of reproducibility, the registry name and the cited year are retained in the text accompanying the figures. No supplementary source is required for the two values presented here, and the numerical statements should be read as direct reproductions of the database entries.\n\nMention: \"Clustered Inequality Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported values of 31 percent and 47 percent from the Clustered Inequality Registry database for 2015.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:a5043307eb2a7603accd2eda", "occurrence_id": "83156fb8c575e08ea64f5a8b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The principal data source is the Regional Early Grade Reading Survey (REGRS), a survey designed to support measurement of early reading performance. The source is used as the central evidence for the statistical assessment, with attention to comparable reading domains and consistent treatment of observations. No individual result is presented as a standalone reported figure.\n\nThe analysis derives mean performance measures from the REGRS records and calculates the direction and magnitude of change across the observed assessments. These calculations provide an aggregate view of reading outcomes and reduce the influence of isolated observations. The resulting trend is interpreted as an indication of overall movement in performance, while the average is used to describe the general level represented in the records.\n\nThe estimates are intended for implementation monitoring and should be read as summary measures rather than as direct reproductions of the survey entries. Differences in the number of observations can affect the precision of the averages and the apparent strength of the trend. Accordingly, the analysis applies the same calculation approach throughout and presents the derived results as evidence for tracking progress and identifying areas requiring further review.\n\nMention: \"Regional Early Grade Reading Survey (REGRS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses REGRS to calculate average reading performance and a trend across the observed assessments.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:8c05bf689e8e5f0f7fbc89c3", "occurrence_id": "9143f08028a20f04a1cf0c33", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Implementation monitoring uses the Midline Building Footprint Index, an indicator that provides a standardized measure for assessing the recorded presence of building footprints. The source is used as a direct reference for the coverage assessment presented below, rather than as a basis for recalculating or extending the underlying series.\n\nThe indicator reports that 68 percent of the assessed units contain a mapped building footprint, while 27 percent contain more than one mapped footprint. These two figures are reproduced directly from the Midline Building Footprint Index and are presented without adjustment. Together, they support the finding that coverage is substantial but uneven across the assessed units.\n\nThe results should be interpreted as an input to implementation monitoring. They provide a consistent numerical basis for reviewing the reach of the activity and for identifying areas where additional verification may be warranted. The figures do not, on their own, establish the quality, functionality, or continued use of the structures represented in the index.\n\nAccordingly, the indicator is used here to substantiate a limited claim concerning recorded physical coverage. Other dimensions of performance require separate evidence and should not be inferred from the two reported percentages.\n\nMention: \"Midline Building Footprint Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Midline Building Footprint Index to support a coverage assessment with figures of 68 percent and 27 percent.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:476139fe64d1967e30e79c7b", "occurrence_id": "d3d30b5fc9011b3d97f3c46b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Pilot Poverty and Vulnerability Microdata, a microdata source used to examine household welfare and exposure to economic shocks. The source is treated as the principal statistical reference for the indicators presented below, with no adjustment to the reported estimates.\n\nTwo figures reproduced directly from the source provide the central evidence for the assessment. The reported poverty incidence is 42.6 percent, while the share of households classified as vulnerable is 18.4 percent. These values are retained in their published form and are used to establish the scale of deprivation and the associated risk of deterioration in household conditions.\n\nThe figures should be read as descriptive measures rather than as evidence of change over time. They support the identification of households requiring closer monitoring and inform the selection of indicators for implementation reporting. Any subsequent comparison should preserve the definitions and population coverage applied in the source, since differences in classification may affect the apparent magnitude of poverty or vulnerability.\n\nThe results provide a concise basis for targeting analytical attention and for assessing whether implementation arrangements are responding to the conditions recorded in the underlying microdata. Further interpretation is subject to the limitations of the source and to the quality of the underlying responses.\n\nMention: \"Pilot Poverty and Vulnerability Microdata\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the directly reported figures of 42.6 percent poverty incidence and 18.4 percent household vulnerability from the Pilot Poverty and Vulnerability Microdata.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:7cd6e1f64b2f1813eeda7945", "occurrence_id": "00a18ec80e9941f1701fba05", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Biennial Living Standards Index (BLSI) is used as the principal indicator for the 2022 reference year in this assessment. It provides the observation base for monitoring household living conditions and for assessing movement in the measured outcome. The source is retained in its stated form to support a transparent and reproducible calculation.\n\nThe analysis does not reproduce an individual reported figure as the main result. Instead, the available observations are combined to obtain an average level, after which the direction and pace of change are assessed across the relevant sequence. This treatment gives greater weight to the underlying pattern than to any single observation and reduces the effect of short-term variation. The resulting measure is used as the central basis for judging whether conditions have improved, remained broadly stable, or deteriorated.\n\nInterpretation remains subject to the coverage and measurement properties of the indicator. The calculated average is therefore presented as an analytical summary rather than as a separate administrative record. Changes in the derived measure are read together with the consistency of the observations and the stability of the calculation procedure. This approach supports comparison within the assessment period while avoiding claims that cannot be established from the underlying series. The reported findings consequently reflect a computed trajectory derived from the BLSI evidence, rather than a direct restatement of isolated source values.\n\nMention: \"Biennial Living Standards Index (BLSI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives an average level and a direction of change from the 2022 BLSI observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:0979491a6c3b888dadfccfd9", "occurrence_id": "63ad6bd596b82962dec3377e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The statistical basis for the settlement profile is the Triennial Settlement and Elevation Registry, a database produced by the National Bureau of Statistics for Lao PDR and referring to 2018. The registry is used as the principal reference for the geographic observations presented in this assessment, with its coverage understood as a settled-area inventory rather than a population estimate.\n\nAvailable evidence indicates that service access varies across the covered settlements, particularly where physical elevation affects travel conditions and the placement of facilities. The registry reports 31 settlements and an elevation of 842 metres for the relevant observation. These two figures are reproduced directly from the source and support the broader finding that geographic constraints should be considered when interpreting differences in implementation results. They are not used to derive a trend or to estimate changes over time.\n\nInterpretation remains subject to the scope and structure of the registry. The entries provide a consistent basis for describing the distribution of settlements and recorded elevation, but they do not by themselves establish causation between terrain and service outcomes. Accordingly, the cited observations are treated as supporting evidence within the assessment, while conclusions concerning implementation performance are based on the wider body of documented results.\n\nMention: \"Triennial Settlement and Elevation Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage draws from the Triennial Settlement and Elevation Registry to report 31 settlements and an elevation of 842 metres for Lao PDR in 2018.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:33386126e59ad97d44a0b057", "occurrence_id": "43c8a78be4c7f9b60cc900d3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Midline Public Works Survey (MPWS) is the principal empirical source for the assessment of implementation performance in 2015. As a survey, it provides the basis for interpreting reported coverage and observed results while preserving the distinction between recorded information and subsequent analytical treatment. The evidence is used descriptively, with no inference beyond the information contained in the source.\n\nPreparation of the material involved assembling the available questionnaires and associated records, checking their internal consistency, and harmonising common measures across reporting units. Responses were reviewed for duplicate entries, incomplete observations, and incompatible coding, after which the retained material was organised in a stable archive. These procedures reduce the risk that apparent differences arise from changes in notation or record structure.\n\nResults derived from the MPWS are presented as a statistical baseline rather than as a projection. The archived survey supports summary comparisons of reported implementation outputs and beneficiary reach, with exclusions documented in the accompanying calculations. Interpretation remains limited to information directly supported by the source, and the harmonised file is retained as the central evidence for the assessment. Collection, quality review, harmonisation, and archiving are treated as one continuous evidence-management process.\n\nMention: \"Midline Public Works Survey (MPWS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The passage identifies the 2015 MPWS survey as central evidence and describes its collection, quality review, harmonisation, and archiving.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:89c2091a964701a7d677dc93", "occurrence_id": "f7a776b38c9a1830a56128a2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Coordinated Mental Health Registry (CMHR) database is used as a reference source for the assessment of reported mental health activity. Its records provide an additional basis for reviewing the plausibility of service estimates and for identifying material differences between reported activity and the corresponding analytical series. The source is cited in the supporting discussion of service utilisation.\n\nAn estimate derived from an independent source is cross-checked against the CMHR database. The comparison does not replace the underlying estimate, but it provides a separate measure against which the magnitude and direction of the reported result can be examined. The available entries show sufficient alignment to support the claim that the estimated level of activity is consistent with the observed pattern in the registry.\n\nThis evidence is considered alongside the broader assessment of reporting completeness, data quality, and continuity of service delivery. The CMHR comparison supports one claim within that assessment by reducing uncertainty around the estimate, while leaving the principal interpretation dependent on the wider body of evidence. No adjustment is applied solely on the basis of the registry comparison.\n\nMention: \"Coordinated Mental Health Registry (CMHR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the CMHR database to cross-check an independent estimate and support a claim about the consistency of reported mental health activity.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:659eed99b1ce4bb513799c56", "occurrence_id": "865a16d8e273c7e17b46e2de", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment considers the timing of movement-related decisions as a relevant operational issue. In this regard, information on reported intentions is treated as one contextual element supporting the broader finding that implementation planning benefits from timely indications of likely demand and pressure.\n\nThe Pilot Movement Intentions System (PMIS) is identified as an administrative source produced by ICRG for 2023. Its citation is included to situate the discussion and to support the general relevance of intention-based information to planning decisions. The source is not used to establish a numerical result, and no entries, totals, rates, or other reported quantities are reproduced in the assessment.\n\nThe reference therefore functions as supporting context for one claim within the wider analysis rather than as a basis for measurement or comparison. The conclusions presented here rely on the surrounding assessment and do not assign an independent estimate to the source. This treatment preserves the distinction between acknowledging a relevant information system and drawing findings from its recorded content.\n\nMention: \"Pilot Movement Intentions System (PMIS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The passage uses the 2023 PMIS attribution to ICRG as contextual support for a broader planning claim and reports no source values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:0fe2ff672b757acfc2942911", "occurrence_id": "798491eddb90db378f247702", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Rainfall conditions were reviewed as part of the environmental and population analysis for Cairo. The review uses the Harmonised Rainfall Anomaly Dataset, a geospatial source produced by UNHCR and IOM for Cairo in 2017. The source is treated as a reference series for interpreting departures from usual rainfall conditions and for identifying areas where service pressures may be affected by weather variability.\n\nTwo figures are reproduced directly from the dataset. It records 42.6 millimetres for the selected rainfall measure and 18.9 percent for the corresponding anomaly share. The figures are retained in their reported form and are not recalculated or combined with estimates from other material. Their inclusion provides a concise quantitative basis for the assessment of uneven rainfall conditions across the covered area.\n\nTaken together, the cited observations support the claim that rainfall variability should be considered when sequencing field activities and reviewing exposure to short-term environmental stress. They do not, on their own, establish causality or predict future conditions. Their function in this assessment is limited to documenting the reported measurements and anchoring the interpretation in a traceable geospatial record. This treatment is consistent with a cautious use of secondary data in operational planning.\n\nMention: \"Harmonised Rainfall Anomaly Dataset\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the 42.6 millimetres and 18.9 percent figures from the Harmonised Rainfall Anomaly Dataset to support an assessment of rainfall variability in Cairo.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:9a358e0f9471a92fd4fb9183", "occurrence_id": "0a18e9f60fa58efe3d863413", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Audited School Assessment Survey is a survey produced by FISE, the organisation that produces it. It is treated as the principal evidentiary source for the assessment of recorded school conditions and reported compliance. The source is used in its stated form, with no adjustment to the figures reproduced below.\n\nTwo figures are taken directly from the survey and provide the central quantitative basis for this assessment. The survey records 31 grades assessed and reports that 74 percent met the applicable minimum standard. These figures are presented as reported measures rather than as estimates derived from a separate calculation.\n\nThe results indicate that coverage extended across a defined set of grades while compliance remained below complete attainment. The 31 grades provide the denominator for interpreting the assessment scope, and the 74 percent result indicates the share meeting the stated standard. The figures should therefore be read together, since the first describes the assessed coverage and the second describes the recorded outcome within that coverage. No further inference is made regarding changes over time or conditions outside the survey evidence.\n\nMention: \"Audited School Assessment Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Audited School Assessment Survey for its directly reported figures of 31 grades assessed and 74 percent meeting the minimum standard.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:b2de1a4168e2ed0cf3ec97d2", "occurrence_id": "fb6457742f001bd65f6d05ec", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Urban Vocational Training Registry database for 2017 provides the administrative basis for the assessment of vocational training capacity. Its entries are treated as reported records for the reference period and are used to support the broader finding that the registry contains sufficiently specific information for describing participation and institutional coverage.\n\nThe assessment cites directly from the database the reported figure of 31 registered training centres and the reported figure of 684 enrolled participants. These figures are presented without adjustment or extrapolation and are retained in their recorded form. Together, they indicate the scale of the recorded training network and the number of participants associated with it during the reference period.\n\nThe evidence should nevertheless be interpreted as a description of recorded activity rather than as a complete measure of all training provision. Coverage may depend on the completeness of entries and on the consistency with which participating centres reported their information. No additional estimate is derived from the two cited figures, and no change over time is inferred. The source is therefore used for a bounded descriptive claim concerning the extent of provision and participation represented in the registry.\n\nMention: \"Urban Vocational Training Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The assessment took the directly reported figures of 31 registered training centres and 684 enrolled participants from the 2017 database.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:4d277fb9999b0dba1e883872", "occurrence_id": "3c1327fec24dd4ba61321a82", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Analytical Collective Bargaining Survey (ACBS) is the survey used for the quantitative observations presented in this section. It provides a consistent basis for describing reported bargaining outcomes and for checking the internal plausibility of the accompanying institutional assessment. The source is treated as a direct statistical record rather than as an interpretive commentary.\n\nThe available evidence indicates that collective bargaining coverage remains uneven across reporting units and that formal recognition does not uniformly translate into negotiated outcomes. Two figures are reproduced directly from the survey. The first is 68 percent for units reporting an active agreement, while the second is 14 percent for units reporting no negotiated adjustment during the reference period. Taken together, these figures support the assessment that the principal constraint concerns implementation and follow-through rather than the complete absence of bargaining arrangements.\n\nThe figures are presented without additional calculation or extrapolation. They are used to anchor the broader discussion of reporting consistency, institutional capacity, and the practical application of negotiated provisions. Differences in reporting completeness may affect comparisons, and the results should therefore be read as indicative of observed conditions within the survey frame. No causal inference is drawn from the two values, and no ranking is assigned to individual reporting units. The evidence nevertheless provides a sufficiently clear basis for monitoring whether formal agreements are accompanied by measurable implementation.\n\nMention: \"Analytical Collective Bargaining Survey (ACBS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the ACBS survey to report 68 percent with an active agreement and 14 percent with no negotiated adjustment.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:e267f77c6d0ca624ac4edcd4", "occurrence_id": "b4c8d39bb8ef2fe96dd1c313", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws its quantitative evidence from the Standard Crop and Livestock Dataset (SCLD), a geospatial source used to support comparison across agricultural observations. The source is retained as the common basis for the calculations reported in this annex, with records reviewed for internal consistency before aggregation. No separate institutional, geographic, or temporal attribution is required for the analytical use presented here.\n\nThe analysis calculates average levels and directional trends from the source rather than reproducing individual entries. Observations are grouped into comparable categories, and summary measures are derived to identify changes in crop and livestock activity. This treatment reduces the influence of isolated records and provides a stable basis for interpreting broad movements in the indicators.\n\nThe resulting measures are used to assess relative performance, variation, and the persistence of observed patterns. Where differences appear across categories, the assessment considers the calculated averages and trend estimates together rather than relying on a single observation. The findings therefore represent an analytical synthesis of the SCLD records and are not a direct transcription of the underlying dataset.\n\nThese results inform the implementation assessment by establishing a consistent quantitative reference for monitoring agricultural conditions. They are presented with appropriate caution, since aggregate measures may conceal variation among individual records. The calculated evidence nevertheless provides a sufficiently comparable basis for reporting progress and identifying areas requiring further review.\n\nMention: \"Standard Crop and Livestock Dataset (SCLD)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives comparable averages and directional trends in crop and livestock activity from the Standard Crop and Livestock Dataset (SCLD).\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:e4a6fd8649c7731c0506eec0", "occurrence_id": "332e425b63ebd9464f2c7297", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Verified Apprenticeship Index (VAI), an indicator produced by CDR for measuring apprenticeship conditions. The source provides a structured basis for examining changes in participation and completion across the observed records. Its inclusion supports one finding within the broader assessment of implementation performance.\n\nIndividual entries are not reproduced as reported figures. Instead, the available observations are consolidated into a simple average, while successive values are compared to establish the direction and consistency of the trend. This treatment reduces the influence of isolated movements and provides a summary measure suitable for comparison within the analysis.\n\nThe resulting pattern indicates a gradual strengthening of apprenticeship performance, although variation remains across the records. The interpretation is therefore limited to the movement and average derived from the source and does not present either measure as a direct value published by CDR. The indicator is used as supporting evidence rather than as a standalone measure of overall programme results.\n\nMention: \"Verified Apprenticeship Index (VAI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives an average and trend from the Verified Apprenticeship Index (VAI) indicator produced by CDR.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:8b27c8d93dad44414a2c0370", "occurrence_id": "df990ee69bf2235175e6dd5e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Follow-Up Disability and Support Survey (FDSS) is the survey covering the Amhara region and provides the principal evidence for the assessment of disability-related support conditions. Its observations are treated as a consistent basis for examining differences across the population and for identifying the direction of change in reported access and need. No separate institutional source is introduced for this purpose.\n\nThe analysis derives average levels from the survey observations and uses these averages to construct a broad trend. This approach reduces the influence of isolated responses and gives greater weight to the overall distribution of reported conditions. The resulting pattern is used to assess whether support needs appear to be declining, stable, or increasing across the covered population.\n\nThe estimates are interpreted cautiously, since an average can conceal variation among respondents and does not establish the cause of observed differences. Nevertheless, the FDSS provides the central empirical basis for the assessment, and the calculated trend is used in preference to reproducing individual survey responses. The findings therefore summarise the direction and general magnitude of change while retaining the survey’s coverage of the Amhara region.\n\nMention: \"Follow-Up Disability and Support Survey (FDSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses the Follow-Up Disability and Support Survey (FDSS) to calculate average levels and derive a trend for the Amhara region.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:27713aa15df438bfb3b23e1e", "occurrence_id": "5b43320390788769423b51c2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Continuous Unemployment Insurance Registry is used as the principal database for the 2020 observation. It provides the administrative basis for the coverage measure examined in this annex and is treated as the controlling source for the reported values.\n\nTwo figures are reproduced directly from the registry without adjustment. The recorded unemployment insurance coverage rate is 18.7 percent, while the corresponding number of insured persons is 142,600. These values are presented as reported and are not recalculated from secondary estimates.\n\nThe evidence indicates that the registry contains sufficient information for a focused assessment of coverage during the reference period. Interpretation remains limited to the measures documented in the source, and no additional assumptions are introduced regarding population composition or reporting completeness.\n\nFor monitoring purposes, the two reported figures provide a fixed basis for reviewing implementation results. Subsequent analysis should preserve the definitions and units used in the database so that comparisons remain consistent with the underlying administrative record.\n\nMention: \"Continuous Unemployment Insurance Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the 18.7 percent coverage rate and 142,600 insured persons directly from the Continuous Unemployment Insurance Registry for 2020.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:0f1c52b60916e68876e7659f", "occurrence_id": "9aec4cc16d42bc813e259fa5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis draws on the Pilot Working Conditions Survey, a survey produced by JP Morgan, as supporting material for the interpretation of working conditions. The source is introduced within the discussion of labour outcomes and is not treated as a component of the statistical series. Its inclusion provides background for the proposition that reported conditions should be read alongside differences in employment arrangements and access to workplace protections.\n\nThe survey does not supply observations, benchmarks, or parameters for the estimates presented in this report. No values from the source are incorporated into the tables, and no classification, weighting, or comparison is derived from it. Its function is confined to indicating that the measured outcomes have a wider interpretive setting, particularly where formal indicators may not capture the full range of conditions experienced by workers.\n\nThis limited use is consistent with the evidentiary structure of the assessment. The principal findings rely on the designated records and procedures described elsewhere, while the cited source supports a single contextual claim within the broader discussion. Accordingly, readers should regard the reference as explanatory background rather than as an additional statistical input or an independent basis for inference.\n\nMention: \"Pilot Working Conditions Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The source provided contextual support for one claim about interpreting working conditions, while none of its values were used in the analysis.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:e6c9037307a0dda42f2eeade", "occurrence_id": "afa52d4de71e346a16950cdc", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Provincial Rural Credit Survey is a survey covering Ecuador and provides the principal evidentiary basis for this assessment of rural borrowing conditions. Its use is limited to the values reported in the source, with no additional classification or recalculation applied in this passage. The source is treated as a direct statistical record for the indicators considered below.\n\nThe survey reports that 31 percent of rural households obtained credit during the reference period. It also records 14,200 borrowers across the observed coverage. These two figures are reproduced directly from the Provincial Rural Credit Survey and constitute the central evidence for the assessment. They indicate both the relative reach of formal or recorded borrowing and the corresponding number of participating borrowers, while avoiding an inference about changes over time.\n\nThe figures should be read as descriptive measures within the coverage of the survey. They do not, on their own, establish the quality of lending, the adequacy of loan amounts, or the persistence of access among households. For implementation purposes, however, they provide a consistent basis for reviewing outreach and for identifying whether subsequent monitoring should examine participation rates and borrower volumes separately. Any comparison with later administrative records would require compatible definitions and coverage before interpretation.\n\nMention: \"Provincial Rural Credit Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported figures of 31 percent of rural households obtaining credit and 14,200 borrowers from the Provincial Rural Credit Survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:71f3d6caf85d1e974900be7f", "occurrence_id": "65fd9bdec5a6a3148ce24b25", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Small-Area Nutrition Surveillance Index as an indicator produced by US Census for 2020. It is treated as the reference measure for interpreting the distribution of observed nutrition conditions across monitored units. No adjustment, interpolation, or supplementary classification is applied to the cited values.\n\nThe reported figures of 14.2% and 27.8% are reproduced directly from the indicator. Their inclusion supports the assessment that nutrition conditions remained uneven across the monitored units, while avoiding an inference about change over time. The figures are therefore used as evidence within the broader interpretation rather than as a separate estimate.\n\nThe measure is presented alongside the implementation findings to clarify the scale of the reported condition and to maintain consistency between the narrative assessment and the underlying statistical record. The citation is limited to the two values required for this comparison. All percentages are retained in their published form, and no additional transformation is introduced for the purposes of this annex.\n\nMention: \"Small-Area Nutrition Surveillance Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported values 14.2% and 27.8% from the 2020 Small-Area Nutrition Surveillance Index.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:53a5b9bce8f12a37f0850acf", "occurrence_id": "a8b7b141a2377862c0c4fc0e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Recurrent Diagnostic Capacity Report is the cited 2024 report for this assessment. It provides the central quantitative evidence for the review of diagnostic readiness and is treated as the authoritative basis for the two figures presented below. No recalculation or reconciliation with another source is applied.\n\nThe report records 68 operational diagnostic units within the assessed system. This figure is reproduced directly from the source and indicates the reported scale of functioning capacity at the time of measurement. The same report states that 42 percent of those units met the specified readiness threshold. This second figure is also taken directly from the report and is presented without adjustment.\n\nTaken together, the two reported values indicate that the availability of operational units was broader than the proportion meeting the readiness threshold. The difference is relevant for implementation planning because nominal coverage does not by itself demonstrate consistent capacity to deliver the required diagnostic functions. The evidence therefore supports continued attention to equipment availability, staffing, maintenance, and routine quality controls. These observations remain limited to the information contained in the cited report and should not be interpreted as an independently calculated trend or as evidence of change over time.\n\nMention: \"Recurrent Diagnostic Capacity Report\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Recurrent Diagnostic Capacity Report to present the directly reported figures of 68 operational diagnostic units and 42 percent meeting the readiness threshold.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:0928a45fa3a7b6f4a0a3d821", "occurrence_id": "756283f523b2e07b7c42ab37", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Metropolitan Ward Outline Dataset is used as the geospatial source for the boundary review. Its inclusion provides a single reference for checking the numerical scale of the administrative outline record and for supporting the tabulations presented in this annex. The source is cited in full in the evidence paragraph so that the reported counts remain traceable to the underlying material.\n\nTwo numerical observations are reproduced directly from the Metropolitan Ward Outline Dataset. The record contains 31 wards, and it identifies 684 outline features. These figures are presented as reported values rather than as results derived through comparison, aggregation, or extrapolation. They therefore serve as the central evidence for the descriptive statement concerning the structure and extent of the mapped record.\n\nThe reported counts support a limited interpretation. The number of wards establishes the unit used for the review, while the number of outline features indicates the volume of spatial material associated with that unit. No additional calculation is required for the purpose of this assessment. Subsequent tables may organise the evidence for presentation, but they do not alter the two figures reproduced above. This treatment preserves a clear distinction between values taken from the source and analytical observations made in the report.\n\nMention: \"Metropolitan Ward Outline Dataset\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes 31 wards and 684 outline features directly from the Metropolitan Ward Outline Dataset.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:83480fdfe16f742f1bf728ed", "occurrence_id": "484dcac4ce94a7fa2b33f2d7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The monitoring assessment uses the Follow-Up Soil Fertility Survey as a survey for the soil indicators reviewed in this annex. No transformation, reconciliation, or external classification is applied to the figures reproduced below. The source is retained under its stated title so that the numerical evidence can be checked against the underlying record.\n\nTwo figures are taken directly from the Follow-Up Soil Fertility Survey and reproduced here without rounding or recalculation. It reports that 31 percent of sampled plots meet the specified fertility criterion, while 14 percent fall below the corresponding threshold. These values constitute the central quantitative evidence for the assessment and are presented as reported, rather than as a derived comparison.\n\nThe reported proportions indicate a material distributional concern for soil management and support continued attention to input use, targeting, and verification during implementation. The findings should be interpreted as a description of the observed distribution, not as evidence of change over time. Subsequent operational decisions should preserve the distinction between the two reported figures and the analytical interpretation applied in this annex.\n\nMention: \"Follow-Up Soil Fertility Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Follow-Up Soil Fertility Survey to report 31 percent meeting the fertility criterion and 14 percent falling below it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:ef731d9497c9129bbdad9771", "occurrence_id": "81fcf510c58ba247406703e2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Urban Debt and Savings Survey is cited as a survey produced by Barth, Caprio, and Levine, with the 2016 observations serving as the reference evidence for the assessment. The source supports the finding that reported household balances and borrowing patterns should be interpreted as comparable measures rather than as directly interchangeable administrative totals.\n\nFor the analytical file, the underlying records were collected, harmonised, and archived before tabulation. Definitions were reviewed against the stated concepts, duplicate entries were removed, and classifications were aligned to permit consistent comparison across the reported observations. These procedures support the use of the source for descriptive assessment while limiting the interpretation to the information recorded in the survey.\n\nThe cited material is therefore used to substantiate one component of the broader results discussion, specifically the treatment of savings and debt indicators as structured observations suitable for comparison. It is not used to establish a causal relationship, and no inference is drawn beyond the scope of the recorded responses. The resulting evidence provides a documented basis for the reported assessment and preserves a traceable link between the analytical tables and the archived source material.\n\nMention: \"Urban Debt and Savings Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The passage uses the 2016 observations from the Urban Debt and Savings Survey as collected, harmonised, and archived evidence.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:6976290a6105b8c145f563e4", "occurrence_id": "22fb6b615cb6817c678bc73d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Record-Based Land Cover Dataset, produced by IIASA as a geospatial source for 2023. The dataset provides the evidentiary base for examining changes in recorded land-cover categories across the observation series. Its use is limited to the analytical purpose described below, and no additional attribution or geographic qualification is introduced.\n\nThe reported results are derived through aggregation of the underlying records. Category shares are consolidated and compared across successive observations, after which the assessment calculates an average level and an average rate of change. This procedure reduces the influence of isolated entries and gives greater weight to the pattern observed across the full set of records. The resulting measures therefore represent calculated summaries rather than figures reproduced directly from the dataset.\n\nThe analysis indicates a gradual redistribution among the recorded categories, with the direction and magnitude of change varying by category. The calculated averages support comparison of relative movement and provide a consistent basis for monitoring implementation results. Where differences are small, they are interpreted cautiously, since aggregation may conceal variation within individual records. The Record-Based Land Cover Dataset remains the central empirical input for these calculations, while the conclusions presented here reflect the transformations applied during analysis rather than a verbatim statement of the source.\n\nMention: \"Record-Based Land Cover Dataset\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the 2023 IIASA geospatial Record-Based Land Cover Dataset to calculate category averages and trends rather than reproduce individual figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:9cebecc7e1ff67986033dfff", "occurrence_id": "98e33a29c3798d94448390cb", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Inter-Agency Health Workforce Survey (IHWS) as its principal survey for examining health workforce availability and distribution. The survey provides observations that are sufficiently structured to support comparison across reporting points, while retaining the distinctions required for analysis of staffing categories and service functions.\n\nThe analysis does not reproduce individual survey entries as reported values. Instead, it calculates mean staffing levels for the principal categories and estimates the direction and magnitude of change across successive observations. These derived measures are used to identify persistent shortages, areas of relative stability, and differences in workforce composition. The resulting averages provide a more consistent basis for comparison than isolated responses, particularly where the number of reporting units varies between observations.\n\nInterpretation remains cautious because calculated measures reflect the coverage and internal consistency of the IHWS records. The assessment therefore gives greater weight to patterns that persist across several observations and treats isolated movements as indicative rather than conclusive. On this basis, the survey serves as the central evidence for the workforce trend analysis and for the aggregation of staffing indicators used in the implementation review.\n\nMention: \"Inter-Agency Health Workforce Survey (IHWS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives mean staffing levels and changes across successive reporting points from the IHWS.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:05208b73fa16f746927866b8", "occurrence_id": "1949627fafff31e1351a73b9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The evidence base for this section is the Rolling Facility Assessment Survey (RFAS), a survey used to provide a standardized account of selected service conditions. The source is treated as a direct reporting instrument, and its reported values are used without recalculation or interpretation beyond the scope of this assessment.\n\nThe survey records 31 grades across the assessed facilities. It also reports 47 percent for the specified indicator. These figures are reproduced directly from the RFAS and constitute the principal quantitative evidence for the findings presented here. The first figure describes the recorded distribution, while the second indicates the reported proportion associated with the indicator.\n\nThe results should therefore be read as a concise presentation of the measurements contained in the source. No additional adjustment has been applied to either figure, and no comparison with an external series is introduced. The evidence supports a limited description of observed conditions and provides a basis for the operational interpretation that follows. Any broader inference would require information beyond the two values reported in the survey.\n\nMention: \"Rolling Facility Assessment Survey (RFAS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the RFAS figures of 31 grades and 47 percent as direct evidence.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:d140a7d0441ac21fa44b60a8", "occurrence_id": "1f2232c53555d3d8e2f22260", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Sentinel Ward Outline Census, a census produced by the National Institute of for Punjab and referring to 2017. The source provides the underlying observations required to examine distributional change across the reporting units. Its use is limited to supporting the interpretation of variation in coverage and does not constitute a separate administrative benchmark.\n\nThe reported pattern is derived by calculating an average across the observations and by examining the resulting trend over the relevant units. This treatment reduces the influence of isolated values and permits comparison of the general direction of change. The resulting measure is therefore an analytical summary generated for this assessment, rather than a direct reproduction of any single figure contained in the source.\n\nThe evidence indicates a gradual change in the measured distribution, with differences between units remaining material for implementation planning. These findings are used to qualify the broader assessment of access and consistency, while recognising that an average may conceal local variation. The source is consequently treated as a supporting statistical basis for the claim, not as a standalone estimate of current conditions.\n\nMention: \"Sentinel Ward Outline Census\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The assessment uses the 2017 Punjab census from the National Institute of to calculate an average and identify a trend rather than quote individual figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:add93e7abe95bcdf18a3f6ba", "occurrence_id": "4bd7297268296aee169ee3dc", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Continuous Socioeconomic Survey (CSS) is used as a survey-based reference for assessing the internal consistency of the reported indicators. Its observations provide a separate basis for reviewing the level and distribution of the measure under consideration, while avoiding reliance on a single reporting stream.\n\nThe estimate presented in the main analysis is cross-checked against the corresponding information in the CSS. The comparison is intended to identify material divergence between the estimate and an independent source, rather than to replace the estimate used for the assessment. Differences are reviewed in relation to coverage, measurement definitions, and the timing of observation, insofar as these affect comparability.\n\nThe CSS results broadly support the direction of the reported finding and provide a reasonable check on its magnitude. Where small discrepancies occur, they are treated as evidence of normal variation between sources and not as grounds for revising the estimate in the absence of a clear methodological explanation. The source is therefore retained as corroborating evidence for this claim within the broader assessment.\n\nThis use of the CSS strengthens the evidentiary basis of the paragraph while preserving the distinction between the principal estimate and the supplementary verification. The survey is not used to infer additional results beyond the comparison described above.\n\nMention: \"Continuous Socioeconomic Survey (CSS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The CSS survey is used to cross-check a principal estimate and assess the consistency and magnitude of related indicators.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:1a2fa2006a392ffeea7d97b6", "occurrence_id": "fc938d3e7e8f1e627216d34f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Metropolitan Informal Sector Survey (MISS) is a survey produced by NCHS/CDC/WHO and provides the principal empirical basis for the assessment of informal economic activity. Its use in this note is limited to evidence reported within the source itself. The survey is treated as an authoritative reference for the indicators reproduced below, without additional adjustment or interpretation of its underlying coverage.\n\nTwo figures are taken directly from the survey. The first records that 47 percent of surveyed operators reported irregular access to working capital. The second records that 62 percent reported reliance on household resources during periods of reduced activity. These figures are presented as reported measures and are not recalculated for the purposes of this assessment.\n\nThe difference between the two reported proportions is retained as a descriptive feature of the evidence. It indicates that dependence on household resources was more widespread than irregular access to working capital, while not establishing a causal relationship between the two conditions. The figures therefore support a cautious characterization of financial vulnerability among informal operators.\n\nThe MISS evidence is used alongside the implementation findings to maintain consistency between the statistical note and the wider assessment. No extrapolation is made beyond the populations represented in the survey, and no additional estimates are derived from the two reported figures.\n\nMention: \"Metropolitan Informal Sector Survey (MISS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the MISS survey to report directly the figures of 47 percent and 62 percent concerning financial access and household resource reliance.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:5e6c8c769d1cf495b0a60dff", "occurrence_id": "842dc929d34dffe0457d03b7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Rolling Informal Trade Survey (RITS) is the survey used for the assessment of informal commercial activity. Its coverage and definitions are treated as the reference basis for the accompanying discussion. No adjustment has been applied to the reported values, and the figures are presented in the units used by the source.\n\nThe evidence reproduced from the survey consists of two reported values. RITS records that informal transactions represented 64.2 percent of observed activity, while 18,700 traders were included in the corresponding count. These values are taken directly from the Rolling Informal Trade Survey (RITS) and are not estimates generated for this report.\n\nThe first figure indicates the relative weight of informal transactions within the measured activity. The second indicates the size of the trader population captured by the survey. Read together, the values support a finding that informal exchange constitutes a substantial component of the assessed market and that the observed network includes a broad group of participating traders.\n\nThe results should nevertheless be interpreted within the measurement boundaries established by the survey. They provide a consistent basis for monitoring and comparison, but they do not by themselves establish causality, describe unobserved activity, or quantify transactions outside the survey frame. The two reported values are therefore retained as core descriptive evidence for implementation monitoring.\n\nMention: \"Rolling Informal Trade Survey (RITS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes directly from the Rolling Informal Trade Survey (RITS) the figures of 64.2 percent for informal transactions and 18,700 traders.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:c628bab924831abf1617d9c0", "occurrence_id": "4ff6eb2add13f8045280263a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Cross-Sectoral Teacher Absence System (CTAS) provides the principal empirical basis for this assessment. It is an administrative source covering Turkey and referring to 2015. The records are treated as a structured observation of reported teacher absence, with coverage interpreted consistently across the units represented in the system. No individual record is presented as a standalone result.\n\nTo support comparison, the analysis aggregates the observations and calculates an average rate for the covered units. It also derives a trend from the ordered observations, using the resulting summary measures to identify the direction and relative scale of absence. These calculations are used in preference to reproducing isolated entries, since a single entry would not describe the distribution represented in CTAS.\n\nThe resulting estimates are read as analytical summaries of the administrative evidence rather than as direct quotations from the system. Differences in the calculated average and trend are therefore discussed as patterns in the observed records, while their interpretation remains limited to the coverage of Turkey in 2015. This treatment supports a concise assessment of variation without extending the evidence beyond the source.\n\nMention: \"Cross-Sectoral Teacher Absence System (CTAS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses CTAS observations for Turkey in 2015 to calculate an average rate and derive a trend in teacher absence.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:f231187fff591a0ff9eb2893", "occurrence_id": "c4076a59f0c1b680725d5419", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Standard Living Standards Survey (SLSS) is used as the principal survey for the assessment of household conditions. Its observations provide the central empirical basis for the calculations reported below. The source is treated as a consistent body of evidence for comparing the selected measures across successive observation periods.\n\nThe analysis derives period averages from the SLSS rather than reproducing individual reported values. It then compares those averages to identify the direction and pace of change in household conditions. This procedure reduces the influence of isolated observations and supports a more stable interpretation of movements in the indicators. The resulting trend is used to assess whether observed improvements were sustained or limited to particular periods.\n\nThe calculated averages and the associated trend are presented as analytical results generated from the survey observations. They should therefore not be read as direct quotations from the source. Minor differences between individual observations and the reported averages are expected and reflect the aggregation procedure. On this basis, the SLSS supports the assessment of changes in living conditions while preserving a clear distinction between source observations and the estimates derived for this report.\n\nMention: \"Standard Living Standards Survey (SLSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses the SLSS to calculate period averages and derive a trend in household conditions rather than reproduce individual survey observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:ba79536ff7738591d6630149", "occurrence_id": "6f14ea95ad0cf21a1abea921", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The District Drought and Resilience Dataset (DDRD), a geospatial source produced by HZZO for 2023, provides the reference observations used in this assessment. Its coverage supports a consistent reading of drought exposure and resilience conditions across the reporting framework. The source is treated as a primary statistical input rather than as contextual background.\n\nTwo values are taken directly from the dataset. It records 31 drought exposure indicators and assigns a resilience classification to 68 percent of the assessed units. These figures are reproduced without recalculation and are used to establish the baseline for subsequent interpretation. The first value indicates the breadth of the measurement structure, while the second provides the principal proportion used in the assessment narrative.\n\nThe evidence should be read as a descriptive representation of recorded conditions. It does not by itself establish causality, measure change over time, or support inference beyond the observations contained in the source. Any later comparison should therefore retain the same definitions and reporting basis. On this basis, the DDRD supplies a bounded and auditable foundation for the results presented in this section.\n\nMention: \"District Drought and Resilience Dataset (DDRD)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes directly from the DDRD the figures of 31 drought exposure indicators and 68 percent of assessed units assigned a resilience classification.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:917d0b3be550009393807f15", "occurrence_id": "134b699b4401574c496e16fd", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: For the purposes of this assessment, the Longitudinal Deprivation and Hardship System is treated as an administrative source for 2024. Its records provide a consistent basis for reviewing the distribution of material hardship and for supporting the interpretation set out below. No adjustment or extrapolation is applied to the reported proportions.\n\nThe evidence indicates that hardship is not evenly distributed across the population represented in the records. The source reports that 41 percent of households experienced at least one recorded deprivation, while 18 percent experienced multiple forms of hardship during the reference period. These two figures are reproduced directly from the system and are used to substantiate the broader finding that a substantial minority faces compounded constraints rather than isolated shortfalls.\n\nThe results should be read as an assessment of recorded conditions, rather than as an estimate of unobserved needs. Differences in registration, reporting, and coverage may affect the observed distribution, but they do not alter the descriptive value of the reported proportions for implementation monitoring. The figures therefore provide a limited but relevant benchmark for judging whether planned measures are reaching households exposed to more than one hardship. Subsequent reviews may examine changes in the composition of deprivation, subject to the continued availability and comparability of administrative records.\n\nMention: \"Longitudinal Deprivation and Hardship System\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Longitudinal Deprivation and Hardship System’s directly reported 41 percent and 18 percent figures to support the assessment that hardship is unevenly distributed and often compounded.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:84373416efee507ea7b05120", "occurrence_id": "d834ab909526646c696f47b8", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Subnational Labour Inspection Census is a census referring to 2022 and provides the principal empirical basis for the assessment of labour inspection capacity. Its records are used to establish the distribution of inspection activity across reporting units, with attention to staffing, assigned functions, completed visits, and recorded follow-up. The source is considered sufficiently structured for comparative analysis, while differences in reporting coverage require cautious interpretation.\n\nThe analysis does not reproduce individual observations as headline results. It calculates the average number of inspectors and completed visits across reporting units and derives a workload trend by comparing units with lower, intermediate, and higher recorded activity. These calculations provide a more stable indication of operational variation than any single entry and support the identification of concentration in inspection functions.\n\nThe resulting pattern indicates that inspection capacity is unevenly distributed and that higher staffing levels are generally associated with greater recorded activity. The relationship is not treated as deterministic, since workload may also reflect assignment practices, reporting completeness, and the composition of inspected establishments. The calculated averages are therefore used as analytical measures rather than as administrative targets.\n\nFor implementation purposes, the findings support prioritisation of staffing reviews, harmonised monitoring, and further verification of unusually high or low observations. Subsequent assessments should preserve the same aggregation approach so that changes in average workload and the associated trend can be examined consistently over time.\n\nMention: \"Subnational Labour Inspection Census\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the Subnational Labour Inspection Census for 2022 to calculate average staffing and activity levels and derive a workload trend across reporting units.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:8ce9dc9aa1e7f18a1122978f", "occurrence_id": "39f5ba6b7bc26c4624d5c81e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The data source High-Resolution Working Conditions Index is used as an indicator of working conditions for the assessment. It provides a structured basis for examining variation across the observations retained for analysis. No individual entry is treated as a standalone finding, and the source is considered together with the wider evidence used in this section.\n\nFor analytical purposes, the observations are consolidated into summary measures. The assessment derives an average from the index and compares the resulting values across successive observations to identify the direction and consistency of change. This treatment reduces the influence of isolated movements and permits a clearer reading of the underlying pattern.\n\nThe derived trend supports the conclusion that working conditions exhibit gradual rather than uniform change over the period considered. The average is used to anchor this conclusion within the broader assessment, while the movement in the index provides an additional indication of whether the observed pattern is persistent. These calculations are presented as analytical results based on the cited indicator and should not be interpreted as values reproduced directly from the source.\n\nMention: \"High-Resolution Working Conditions Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives an average and a trend from the High-Resolution Working Conditions Index to support the assessment of changing working conditions.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:b11f7f5c090423ca0bdd2490", "occurrence_id": "44aff1bf8157d91cc1444cec", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Standard Coastal Erosion Index (SCEI) is an indicator produced by SEP for 2019. It is used in this assessment as the principal reference for describing the measured incidence of coastal erosion. No adjustment is applied to the reported values, and the cited observations are retained in their published form.\n\nTwo figures are reproduced directly from the SCEI and provide the central quantitative evidence for this assessment. The first figure is 14.2 percent, while the second figure is 18.7 percent. These values are presented as separate observations rather than as a derived estimate, and they should therefore be read as reported measurements from the same indicator.\n\nThe difference between the figures is noted only to clarify their relative magnitude. It is not treated as a calculated trend, forecast, or measure of change over time. The figures support the assessment of exposure and provide a consistent basis for the accompanying discussion of implementation priorities.\n\nFor reporting purposes, the Standard Coastal Erosion Index (SCEI) remains the applicable source for these two observations. The citation to SEP and 2019 identifies the indicator used in the analysis, while the direct reproduction of 14.2 percent and 18.7 percent preserves the evidentiary link between the narrative findings and the underlying statistical record.\n\nMention: \"Standard Coastal Erosion Index (SCEI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The assessment takes the directly reported figures 14.2 percent and 18.7 percent from the Standard Coastal Erosion Index (SCEI) produced by SEP for 2019.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:5346e2a6d089e916d7cdbd3c", "occurrence_id": "3899c59d3a5a361081c13650", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Nationwide Housing Conditions Index is used as an indicator produced by DMSP/OLS for 2024. It provides a structured basis for assessing variation in housing conditions across the observed units, while avoiding reliance on a single reported value. The source is incorporated into the analytical framework alongside the underlying administrative and survey evidence.\n\nFor this assessment, the reported observations are converted into a mean level and a simple trend measure. The calculation gives greater weight to the overall direction and average condition than to isolated changes, which may reflect temporary reporting or measurement effects. The resulting series is therefore interpreted as an analytical input rather than as a direct quotation of the source.\n\nThe calculated average supports the finding that housing conditions remain uneven, with improvement concentrated in units that already show stronger baseline performance. The trend measure also indicates that progress is gradual and does not remove the persistence of lower-condition units. These results inform the prioritisation of monitoring and the sequencing of implementation support. They should be read as an aggregate assessment derived from the indicator, not as a substitute for detailed verification at unit level.\n\nMention: \"Nationwide Housing Conditions Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives an average and trend in housing conditions from the 2024 Nationwide Housing Conditions Index.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:0a6a048ee42aade2b4118a99", "occurrence_id": "8bf1e7e39709d2aae7ed9b28", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Stratified Public Works Survey (SPWS) is the principal survey used for the assessment of participation and work intensity in Mozambique. Its observations provide the central empirical basis for examining changes in reported activity across the covered population. The source is treated as a structured statistical record, with attention to the comparability of measures and the consistency of reported responses.\n\nThe analysis does not reproduce individual survey entries as standalone results. Instead, observations are grouped according to the relevant analytical categories, after which average levels are calculated and compared to identify the direction and persistence of change. This approach reduces the influence of isolated responses and supports a more stable interpretation of movement in the indicators.\n\nThe resulting trend is therefore derived from the SPWS survey rather than presented as a direct quotation from it. Average values are used to support the assessment of implementation performance, while variation across observations is retained as an indication of distributional differences. Findings should consequently be read as analytical estimates based on the survey evidence, with conclusions limited to the population covered by Mozambique.\n\nMention: \"Stratified Public Works Survey (SPWS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses the SPWS survey covering Mozambique to calculate average levels and derive a trend in implementation-related indicators.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:2c45a5b9f7ea23aa3263f346", "occurrence_id": "0083637c2559ed49a6303ab1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Provincial Social Protection Survey is the principal survey used for the assessment of social assistance performance. It provides the central empirical basis for examining participation, receipt of support, and reported changes in household conditions. The information is treated as a consistent analytical source rather than as a set of isolated quotations.\n\nFor the present assessment, the reported observations are organized into comparable groups and used to calculate an average level of access and a trend in participation. These calculations summarize the direction and magnitude of change across the observed records, while reducing the influence of unusually high or low responses. The resulting measures are presented as analytical estimates derived from the survey, not as figures reproduced without processing.\n\nThe calculated average indicates the typical level of reported access, while the trend provides an indication of whether participation is expanding, stable, or declining. Together, they support interpretation of implementation progress and help identify areas requiring further review. The analysis remains subject to the coverage, response, and measurement limitations inherent in the source, and the results should therefore be read as evidence of broad movement rather than as a complete account of all eligible households.\n\nMention: \"Provincial Social Protection Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses the Provincial Social Protection Survey to calculate an average level of access and a trend in participation.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:9def637467f75c849a082357", "occurrence_id": "76a84533fd8e05e360494fb0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Longitudinal Job Search System (LJSS), produced by UNESCO-IBE, as an administrative source for the descriptive evidence presented in this annex. The record is referenced in its stated form, without adjustment, interpolation, or linkage to a separate reporting series. Its inclusion supports the identification of the observed search and transition categories used in the accompanying discussion.\n\nTwo figures are reproduced directly from the record. The assessment reports 31 recorded search episodes and 68 completed transitions, retaining the units and numerical precision provided by the source. These values are presented as documentary observations rather than as estimates generated through the analytical procedures applied elsewhere in the report.\n\nThe figures are used only to anchor the scale of the recorded activity and to clarify the basis of the subsequent narrative. No inference is drawn from their difference, and no interpretation of change is attached to the ordering of the values. The source therefore serves a limited evidentiary function within the annex, providing two reported quantities while leaving the broader analytical conclusions to the indicators and methods described in the main text.\n\nMention: \"Longitudinal Job Search System (LJSS)\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"source\", \"summary\": \"The passage takes 31 recorded search episodes and 68 completed transitions directly from the Longitudinal Job Search System (LJSS).\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:076a21aba7046bf1174fb6df", "occurrence_id": "4aebe9e47744d64062e6ae84", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Multi-Topic Hiring and Vacancies Census (MHVC) as the principal empirical source for labour market measurement. This census covers Slovensku and refers to 2016. Its reported observations provide the basis for examining the distribution of hiring activity and unfilled positions across the surveyed units.\n\nRather than reproducing individual values, the analysis calculates average levels from the MHVC observations and compares their movement across the relevant categories. The resulting trend is used to identify areas of relative expansion, stability, or contraction in recruitment demand. This treatment reduces the influence of isolated observations and supports a more consistent reading of the underlying pattern.\n\nThe derived averages are applied throughout the assessment as the central quantitative evidence. They inform the interpretation of differences in hiring intensity, the assessment of vacancy pressure, and the identification of groups requiring closer review. Findings are therefore presented as analytical results derived from the census, rather than as a transcription of figures appearing in the source. This approach also permits the evidence to be incorporated into the broader results framework without introducing unsupported estimates.\n\nMention: \"Multi-Topic Hiring and Vacancies Census (MHVC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the MHVC census covering Slovensku in 2016 to calculate comparative averages and derive a hiring and vacancy trend.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:33c58ff1d78890056125c8e9", "occurrence_id": "b4b68f06bb429725608a5b51", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Local Built-Up Surface Index is used as an indicator of built-up surface conditions in the assessment. It is produced by the General Statistics Office and is treated as a statistical reference for the observations reported below. The material is used in its stated form, with no supplementary attribution or adjustment introduced in this section.\n\nTwo observations are reproduced directly from the Local Built-Up Surface Index. The first records a value of 18.4 percent, while the second records a value of 22.7 percent. These figures are presented as reported entries from the indicator and are not recalculated or converted for the purposes of the assessment. Their inclusion provides the central quantitative evidence for the discussion of observed surface conditions.\n\nThe difference between the two reported values is retained as part of the evidence, but no additional estimate is derived from it. The figures support a concise description of the recorded situation and provide a transparent basis for review. Interpretation is therefore limited to the information conveyed by the cited indicator, while the underlying values remain identifiable and directly traceable to the named statistical source.\n\nMention: \"Local Built-Up Surface Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported values of 18.4 percent and 22.7 percent from the Local Built-Up Surface Index.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:e6341e6d61fe971b0f7306a0", "occurrence_id": "53dc7b6deb10593cae8f5a32", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Joint Debt and Savings Registry, a database produced by Booking.com. It is treated as the principal empirical input for the measurement of household debt and savings conditions. The registry supplies the underlying observations used in the statistical calculations presented below.\n\nThe analysis does not reproduce individual registry entries as reported values. Instead, it calculates average levels across the available observations and derives a trend from their ordered sequence. This approach provides a consistent indication of the direction and magnitude of change while limiting the influence of isolated observations. The resulting measures are used to support the assessment of financial conditions and to identify persistent movements in the indicators.\n\nThe computed averages and trend constitute the central evidence for the conclusions in this section. Interpretive statements are based on the aggregate measures generated from the registry, with attention to the consistency of movement across the observations. The results should therefore be read as analytical estimates derived from the database rather than as direct transcription of its entries.\n\nMention: \"Joint Debt and Savings Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage derives average levels and a trend from the Joint Debt and Savings Registry database produced by Booking.com.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:689b67fc3124a2cf01c82ffa", "occurrence_id": "f889c50357af5858826ac13f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment treats household welfare indicators as complementary measures rather than as stand-alone evidence. Differences in reported income and expenditure can reflect coverage, timing, and reporting conditions, and interpretation therefore requires attention to the type of evidence supporting each observation.\n\nThe discussion draws on the Monthly Household Income and Expenditure Microdata (MHIEM), a microdata source relating to 2023. The reference is included to provide contextual support for the relevance of household-level evidence in interpreting welfare indicators. No values from the source are used in the calculations, comparisons, or presentation of results.\n\nThis citation supports the broader methodological claim that statistical findings should be read alongside information on household measurement and reporting. The analysis consequently retains a cautious interpretation of observed differences and does not attribute changes to income or expenditure patterns on the basis of this source alone. The source functions as background evidence within the assessment, while the reported findings remain based on the evidence specified for each analytical result.\n\nMention: \"Monthly Household Income and Expenditure Microdata (MHIEM)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The 2023 MHIEM microdata provided contextual support for the relevance of household-level evidence, with no source values used.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:da0ca80433ec3d59ca3d3692", "occurrence_id": "b6368e9cefcbd4257536cba5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Standardised Apprenticeship Survey is the survey used for the quantitative assessment of participation and completion outcomes. It provides the principal evidence for the comparison presented in this annex, with the reported measures retained in their original form and without adjustment or extrapolation.\n\nTwo figures are taken directly from the source. The survey reports that 68 percent of registered apprentices completed the relevant training requirement, while 41 percent proceeded to a formal skills assessment. These values are presented as reported measures and are not recalculated from component observations.\n\nThe difference between the two figures indicates a substantial gap between completion of training and participation in assessment. This gap is relevant to implementation monitoring because it may affect the interpretation of recorded attainment and the reliability of completion-based indicators. The figures should therefore be read together rather than treated as interchangeable measures of performance.\n\nThe evidence supports continued attention to the transition from training participation to assessment. Further analysis may examine the administrative and operational factors associated with this transition, but such analysis is outside the scope of the present note. The Standardised Apprenticeship Survey remains the cited basis for the two figures used here.\n\nMention: \"Standardised Apprenticeship Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the survey’s directly reported figures of 68 percent completion and 41 percent participation in formal skills assessment.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:42ff1428574758f928506958", "occurrence_id": "6b8b1fb67f3796780bcb80c5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis draws on the Stratified Household Income and Expenditure Census, a census conducted in 2021. The source provides the central empirical basis for assessing differences in household resources and spending patterns across the population covered by the survey framework.\n\nRather than reproducing individual entries, the assessment calculates average income and average expenditure for each household stratum. These derived measures are used to identify the direction and relative strength of the relationship between resources and consumption. The resulting comparison provides a more stable basis for interpretation than isolated observations.\n\nThe calculated averages indicate a consistent upward trend in expenditure alongside higher income levels, although the increase is not uniform across strata. The analysis therefore treats the gradient between income and expenditure as the principal finding, while recognising that variation within strata remains material. This approach supports the use of the census for distributional analysis without presenting the underlying observations as direct estimates of a single household.\n\nResults are reported in aggregated form to preserve comparability among strata and to limit undue weight being assigned to atypical cases. The derived trend is used for assessment purposes and should not be interpreted as a direct quotation of any one value recorded in the source.\n\nMention: \"Stratified Household Income and Expenditure Census\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage derives stratum-level average income and expenditure and uses them to estimate a cross-stratum upward trend.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:159aa9778cff897d1f642fcf", "occurrence_id": "875385336827083b48473756", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis draws on the Periodic Household Consumption Survey, a survey conducted in 2016. The source provides the central empirical basis for assessing household consumption patterns and associated distributional indicators. Its observations are treated as a consistent analytical set rather than as isolated reported figures.\n\nThe assessment calculates an average across the available observations and derives a trend from their relative movement. This procedure supports comparison among consumption categories while reducing the influence of irregular individual entries. The resulting measures are used to describe the direction and general level of household consumption, with interpretation limited to the information contained in the survey.\n\nThe calculated average is presented as an analytical summary, not as a value reproduced directly from the source. Similarly, the trend reflects the treatment of the underlying observations and should not be read as a separately published estimate. Results are therefore reported with appropriate caution, particularly where the number of observations is limited or where category-level variation is substantial. The source remains the principal evidence for the calculations presented in this assessment.\n\nMention: \"Periodic Household Consumption Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives an average and a trend from the 2016 survey observations rather than reproducing individual figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:452a6e302856bca856c688fc", "occurrence_id": "14f5a65632a8edbbe877065e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Municipal Time Use Survey (MTUS) is the 2023 survey used for the descriptive evidence in this assessment. Its reported measures provide a consistent basis for considering the distribution of daily activities across surveyed households and for identifying the scale of unpaid work relevant to service planning. No adjustment or extrapolation is applied to the two figures presented below.\n\nThe first figure, reported directly in the MTUS, is that 47.8 percent of respondents undertook unpaid household work on the reference day. This proportion is used as an indicator of the breadth of such responsibilities within the surveyed population. It is presented without transformation and should not be interpreted as an estimate for populations outside the survey coverage.\n\nThe second figure reported directly in the same source is 6.2 hours per day devoted to unpaid household work by respondents who recorded that activity. Taken together, the two figures indicate both the prevalence of unpaid work and the time intensity associated with it. They are therefore used as the central empirical reference for the assessment of time-related constraints and are reproduced as published in the Municipal Time Use Survey (MTUS).\n\nMention: \"Municipal Time Use Survey (MTUS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported figures of 47.8 percent and 6.2 hours per day from the 2023 MTUS survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:2a53a78d42a171f83d4a8e3d", "occurrence_id": "ccc5f7544e7700dcda1c3c69", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Local Enterprise and Jobs Census is the principal statistical input for this assessment. It is a census produced by UN DESA and is used to establish a consistent basis for examining enterprise activity and employment. The source is treated as an analytical dataset rather than as a repository of isolated figures.\n\nThe assessment computes changes across successive observations and derives average levels for the main measures. These calculations provide an indication of the direction, pace, and stability of enterprise and job creation. Individual entries are not reproduced as stand-alone findings. Instead, the underlying observations are aggregated so that the results reflect general movement and reduce the influence of unusually high or low records.\n\nThe resulting trend supports interpretation of implementation performance and provides a reference for identifying areas requiring closer review. The calculated averages are also used to distinguish persistent patterns from short-term variation. This approach permits the assessment to draw on the full coverage of the Local Enterprise and Jobs Census while maintaining a clear separation between source data and reported analytical results. Any differences observed across measures are therefore interpreted through the computed series rather than through selected values taken directly from the census.\n\nMention: \"Local Enterprise and Jobs Census\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The assessment derives trends and average levels from the Local Enterprise and Jobs Census to evaluate enterprise activity and employment.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:08b015d013d334f0ac941399", "occurrence_id": "0cb16e58be5f28ab36ebedb4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Data sources.** The Monthly Landslide Hazard Collection is an other source produced by SEP for the year 2016. It provides a monthly sequence of hazard observations suitable for assessing variation over time. The collection is used here as the principal empirical basis for the calculation, rather than as a set of isolated values reproduced in the text.\n\nThe observations were reviewed for continuity and arranged in their reported monthly order. A simple average was calculated across the available monthly records, and the resulting measure was compared with the corresponding position in the sequence. This treatment gives greater weight to the overall pattern than to any single monthly observation and reduces the influence of short-lived fluctuations.\n\nThe resulting trend indicates the direction and relative stability of hazard conditions during the period covered by the collection. The analysis therefore reports derived summaries, including the average level and the change across the sequence, while retaining the source’s monthly structure for verification. Individual observations are used to support the calculation and are not presented as independently quoted findings. Differences between adjacent months are interpreted cautiously, since they may reflect temporary variation within an otherwise gradual movement. The derived measures provide the basis for monitoring progress and for identifying whether subsequent implementation activity coincides with a lower, higher, or broadly stable pattern.\n\nMention: \"Monthly Landslide Hazard Collection\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses SEP’s other source, the Monthly Landslide Hazard Collection for 2016, to calculate an average and assess a time trend rather than quote individual observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:383d5cb89f75e110e64ad35d", "occurrence_id": "d4efcd2cb60bf2bb0b0a40ab", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Rolling Facilities and WASH System, treated as an administrative source for facility status and service continuity. Its records provide a structured basis for examining reported changes across the observation period, while limitations in coverage and reporting frequency are considered in the interpretation of results.\n\nFor the relevant indicator, the analysis derives a trend from successive observations and calculates an average rather than reproducing individual reported figures. This approach reduces the influence of isolated fluctuations and supports a more stable reading of overall performance. The resulting measure is used to support the finding that continuity has improved gradually, although the pace of improvement is not uniform across reporting intervals.\n\nThe calculation is presented as one element of the broader assessment and is not interpreted as a complete measure of system quality. Differences in recording practices, delayed updates, and occasional gaps may affect comparability. The estimate is therefore used to establish direction and general magnitude, with the underlying records retained for verification and subsequent monitoring.\n\nMention: \"Rolling Facilities and WASH System\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The analysis takes a calculated trend and average for service continuity from the Rolling Facilities and WASH System.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:1dbf206b39ab1c628db8a92b", "occurrence_id": "8af1fd4a054a1b8693f98b72", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Provincial Enterprise and Jobs Registry (PEJR) is a database produced by UNESCO-IBE and is used here as the principal source for the reported employment evidence. Its entries provide a consistent basis for describing the recorded scale of enterprise activity and associated jobs. The source is cited by its full title and acronym to maintain traceability within the assessment record.\n\nTwo figures are taken directly from this database. The recorded activity comprises 31 enterprises and 62 jobs. These values are reproduced without adjustment and are treated as the central quantitative evidence for the paragraph. The first figure indicates the number of enterprises represented in the registry, while the second indicates the associated jobs recorded in the same source.\n\nThe figures should therefore be read as registry observations rather than as estimates derived from a separate calculation. They support a concise account of the scale captured by the source and permit the results to be presented in a transparent manner. No additional classification is introduced in this note, and the reported values remain linked to the cited database. This treatment is consistent with an implementation report that identifies the evidence base and states the corresponding counts in ordinary prose.\n\nMention: \"Provincial Enterprise and Jobs Registry (PEJR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the figures of 31 enterprises and 62 jobs directly from the Provincial Enterprise and Jobs Registry (PEJR).\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:b9e2b6861be703e3db4026c6", "occurrence_id": "213be472844c012376c0eb9f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The High-Resolution Graduation Programmes System (HGPS) is used as an administrative source for the assessment. Its records provide a direct basis for the limited quantitative statements presented in this section. No adjustment, projection, or inferred change is applied to the reported counts.\n\nThe source reports 31 programme records in the reviewed material. It also reports 47 completion records. These figures are reproduced directly and are used to establish the scale of the recorded activity. They should therefore be interpreted as reported administrative counts rather than as estimates generated through the assessment.\n\nThe two figures support a concise description of coverage and completion within the available record. Their use is restricted to verification of the documented totals, and no additional numerical conclusion is drawn from them. Any difference between programme records and completion records is retained as reported, without attribution to factors not documented in the source.\n\nMention: \"High-Resolution Graduation Programmes System (HGPS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the figures 31 programme records and 47 completion records directly from the HGPS.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:6ffabd06bb549d2b595c571f", "occurrence_id": "97a6d5c9c73e7457ce62e864", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Rapid Education Finance Registry (REFR) is used as the principal database for the financing analysis. Its records provide the central evidence for examining the distribution and movement of education expenditures across reporting periods. The registry is treated as an input to analysis rather than as a source of isolated figures reproduced in the text.\n\nThe assessment calculates period-specific averages from the recorded values and compares those averages to identify the direction and consistency of change. This procedure reduces the influence of unusually large or small entries and supports a more stable interpretation of financing patterns. The resulting trend is therefore derived from the underlying records, not presented as a direct quotation from the database.\n\nThe calculated measures are used to inform the assessment of expenditure performance and to indicate whether observed financing movements are sustained or irregular. Interpretation remains conditional on the completeness and internal consistency of the records. Where variation is material, the analysis retains the variation in the calculated results rather than replacing it with a single representative figure.\n\nThe findings should consequently be read as analytical estimates based on the Rapid Education Finance Registry (REFR), with the database serving as the evidentiary basis for the reported averages and trend assessment.\n\nMention: \"Rapid Education Finance Registry (REFR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives averages and a trend assessment from the Rapid Education Finance Registry (REFR) database.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:19fa0ad241987ed1fb8f4f70", "occurrence_id": "37c19297d1d1b865b6e200b4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Endline Transport Network Dataset is used as the principal geospatial source for assessing the internal consistency of reported transport access. Produced by WALS, it supplies a structured representation of network continuity, route availability, and connections relevant to the implementation assessment. Its use permits the analysis to retain a common spatial basis when comparing reported access conditions across the reporting units.\n\nThe dataset is used to cross-check an access estimate derived from a separate source. The comparison focuses on whether the estimated level of access is consistent with observable network characteristics, rather than treating either measure as a complete account of service availability. Where the two sources indicate similar patterns, the estimate is considered more credible for analytical purposes. Divergences are retained as quality signals and are reviewed before aggregation.\n\nResults from this cross-check inform interpretation of the reported outcome and support cautious use of the estimate in subsequent analysis. The evidence does not establish causality, and the network representation cannot capture all operational or seasonal factors affecting access. It nevertheless provides an independent basis for identifying implausible values, assessing the direction of apparent disparities, and documenting the limits of the underlying estimate. The dataset is therefore treated as central corroborating evidence within the assessment, while the separately derived estimate remains the principal measure used for reporting.\n\nMention: \"Endline Transport Network Dataset\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The assessment uses network continuity and access indicators from the Endline Transport Network Dataset to cross-check an independently derived access estimate.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:acd2667e750c0b200dc7e4a6", "occurrence_id": "c8fc3798bda1233f4f30ac32", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Standard Poverty and Vulnerability Survey (SPVS) is a survey produced by the World Inequality Lab, with the observations used in this assessment referring to 2021. It provides the principal empirical basis for the discussion of household welfare and vulnerability. The source is treated as a direct statistical reference rather than as a basis for additional estimation.\n\nTwo values are reproduced directly from the SPVS. The survey records that 42 percent of surveyed households were below the relevant monetary threshold, while 18 percent reported a simultaneous exposure to a specified vulnerability condition. These figures are presented without recalculation, interpolation, or adjustment, and they should therefore be read as reported measures from the source.\n\nThe results provide a concise indication of the scale of welfare pressure captured by the survey. The first figure describes the prevalence of poverty under the assessment threshold, and the second identifies the proportion facing the associated vulnerability condition. Taken together, the values support the use of the SPVS as the central evidence for this section and establish a consistent reference point for subsequent interpretation.\n\nMention: \"Standard Poverty and Vulnerability Survey (SPVS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported figures of 42 percent and 18 percent from the SPVS.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:db15732fabe448b838aef585", "occurrence_id": "ec33d369e2aa4100054ff2e6", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Municipal Business Demography Survey (MBDS), a survey covering Slovensku for 2022 and available in Wave 5, is the principal empirical source for this assessment. The records are treated as observations of local business activity and are retained at the level required for comparison across reporting units. No individual observation is presented as a headline result.\n\nUsing the MBDS records, the analysis calculates mean levels across the observed units and derives a trend from their ordered values. This procedure gives greater weight to the distribution of observations than to any single reported figure. The resulting measures indicate the direction and average scale of business activity represented in the survey, while limiting the influence of unusually high or low cases. Calculations are performed on the available responses after basic consistency checks, with missing entries excluded from the relevant denominator. The output therefore summarises the evidence through computed measures rather than reproducing figures from the source.\n\nThe estimates are used to support interpretation of changes in business presence and operating continuity. They should be read as analytical results derived from the Municipal Business Demography Survey (MBDS), not as administrative totals or independently verified counts. The accompanying tables report rounded averages and directional movements, while the underlying survey remains the basis for assessing coverage and comparability. This treatment preserves a clear distinction between source material and the measures generated for the assessment.\n\nMention: \"Municipal Business Demography Survey (MBDS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses Wave 5 of the MBDS survey for Slovensku in 2022 to calculate average levels and derive a business activity trend.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:0ca09f51e236a426ea92d7ec", "occurrence_id": "432434f8a07e34dbbebe520b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Within the broader assessment, implementation performance is interpreted through institutional capacity, service delivery, and the conditions affecting participation in value chains. This framing supports the claim that observed implementation results should be read alongside the structure of local economic activity, rather than treated as isolated administrative outcomes.\n\nThe District Value Chain Census (DVCC), a census produced by the Samoa Bureau of Statistics, is cited to provide background for that claim. Its role is limited to establishing the relevance of value-chain conditions to the interpretation of implementation evidence. No estimates, counts, rates, or other values from the source are used in the assessment, and no calculation or comparison is derived from it.\n\nAccordingly, the citation functions as supporting context within a broader discussion, not as the evidentiary basis for a measured result. Findings reported in this assessment rely on the accompanying analytical material and on the stated implementation record. The DVCC reference helps clarify why economic linkages are considered in the interpretation, while leaving the reported results unchanged and avoiding any inference from the source beyond that limited purpose.\n\nMention: \"District Value Chain Census (DVCC)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The source contributes only contextual support for considering value-chain conditions, with no numerical values taken from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:2b380d3434f041fd98fb14d8", "occurrence_id": "07c4107128e41b30b71f4210", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Integrated Mechanisation Registry (IMR) database for 2022 was retained as the principal evidence for the assessment. Its records provide a structured basis for describing mechanisation assets, service availability, and reported use. The source is treated as an administrative record rather than as an independently modelled estimate.\n\nThe assessment assembled the material through a controlled process of collection, harmonisation, and archiving. Entries were reviewed for duplicate observations, inconsistent terminology, and incomplete reporting before being consolidated into a common analytical file. Where records differed in format, the underlying information was preserved while presentation was standardised for comparison.\n\nThe resulting archive supports the tabulations presented in this report and establishes a traceable basis for subsequent verification. The IMR database was not extended with external estimates, and no additional assumptions were introduced to fill unreported observations. Its retained structure permits the evidence to be revisited if corrections are identified during implementation monitoring.\n\nInterpretation remains subject to the completeness and internal consistency of the registry. Accordingly, the figures derived from the source are used to document reported conditions and to support operational assessment, rather than to imply precision beyond the records retained in the archive.\n\nMention: \"Integrated Mechanisation Registry (IMR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The passage identifies the 2022 IMR database and describes its collection, harmonisation, and archiving as the assessment’s central evidence.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:fe5ef9c7c18d52ef996d5fc4", "occurrence_id": "8ea40d406454a7616cb2e300", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Comparative Household Income and Expenditure Survey (CHIES) is a survey produced by Planning Commission, GoI and is used as the underlying source for the distributional assessment. Its observations provide a consistent basis for examining household income and expenditure patterns within the analytical framework.\n\nFor this assessment, the available observations are consolidated and used to calculate an average level and a direction of change across the sequence of reported values. The resulting measure is therefore derived from the source rather than presented as a transcription of individual figures. This approach reduces the influence of short-term variation and supports comparison across the analytical categories used in the report.\n\nThe calculated trend supports the broader finding that household resource conditions exhibit a gradual and uneven pattern of change. Interpretation remains subject to the coverage and measurement conventions of the survey, and the derived estimates are used as indicative evidence rather than as standalone outcomes. The source is consequently cited for the basis of the calculation, while the report retains responsibility for the aggregation and interpretation of the results.\n\nMention: \"Comparative Household Income and Expenditure Survey (CHIES)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses CHIES to calculate an average and trend that support a broader assessment of household resource conditions.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:336f7280f80d1537bd42512c", "occurrence_id": "c17f1c69b4c4a0df2e0568c9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Clustered Housing Conditions Survey, a survey for 2021, provides the principal household-level evidence used in this assessment. It is cited to support the finding that housing conditions remain uneven across the observed population, while preserving the reported values without recalculation or adjustment. The source is used as a direct statistical reference rather than as a basis for deriving a time series.\n\nTwo figures are reproduced directly from the survey. It reports that 62 percent of households occupy dwellings with adequate structural conditions, while 18 percent of households report serious deficiencies affecting basic habitability. These values support the broader assessment that improvements in housing quality have not benefited all households to the same extent. The figures are presented in the form recorded by the source and should not be interpreted as an estimate of change between periods.\n\nThe evidence is relevant for implementation monitoring because it identifies the scale of housing-related constraints affecting household welfare. At the same time, the reported shares should be read as descriptive measures for 2021. They do not establish causality, and they do not by themselves indicate whether observed conditions resulted from income, tenure, infrastructure access, or other factors. Further analysis may therefore be required before the figures are used to define eligibility thresholds or resource allocations. For present purposes, the survey provides a consistent factual basis for the assessment claim and the two quoted measures.\n\nMention: \"Clustered Housing Conditions Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage takes the 62 percent and 18 percent housing-condition figures directly from the Clustered Housing Conditions Survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:d8dd2069ca8ba058896c386e", "occurrence_id": "ec4232865aa519e8915eb85f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Civil Labour Force Registry (CLFR) database covering Malawi in 2022 as a background reference for the labour market setting. The source is noted to establish the relevant statistical domain and period, while the substantive findings presented below are based on the evidence described in this assessment.\n\nNo observations, estimates, classifications, or tabulations are taken from the CLFR database for the analysis. Its mention does not support a comparison, calculation, or interpretation of labour market outcomes. The reference is retained to indicate the availability of a national administrative source relevant to the subject under review.\n\nThe discussion therefore remains independent of the reported contents of the registry. Readers should understand the citation as a documentary indication of scope rather than as an evidentiary input. Any conclusions, where presented, derive from the materials and procedures specified in the assessment and should not be attributed to the CLFR.\n\nMention: \"Civil Labour Force Registry (CLFR)\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The passage uses the Malawi 2022 CLFR database only to frame the discussion and takes no numerical values from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:c03fe9f91ce79f65cc37b524", "occurrence_id": "5e81d9e07aeb1672122ab035", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Analytical Minimum Needs Index (AMNI) indicator for 2019 provides the central evidence for the assessment of minimum needs conditions. It is treated as a structured measurement source for identifying relative levels of deprivation and for supporting comparison across the observations available in the underlying record. No single AMNI figure is taken as the result of the assessment.\n\nThe analysis calculates an average from the AMNI observations and uses the resulting series to estimate a trend over the period under review. This approach reduces the influence of isolated movements and provides a more stable basis for interpreting changes in minimum needs conditions. The derived average is therefore an analytical measure produced for this assessment, rather than a value reproduced directly from the source.\n\nThe resulting trend is used to assess whether conditions have improved, remained broadly stable, or weakened over time. Interpretation is limited to the information captured by the indicator and does not imply coverage beyond the recorded observations. The calculated measures are presented as evidence for implementation monitoring and as a basis for identifying areas requiring continued attention. In this use, the AMNI functions both as a reference framework and as the principal input to the quantitative assessment.\n\nMention: \"Analytical Minimum Needs Index (AMNI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The assessment uses the 2019 AMNI indicator to calculate an average and derive a trend rather than reproduce a single source value.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:61234ba526685c71dd0d9b67", "occurrence_id": "f2dc96a4b02105064352e276", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Nationwide School Assessment Survey is a survey used as the principal reference for the outcome measures reviewed in this annex. It provides a consistent basis for interpreting reported differences in learner performance and participation. The source is treated as administrative evidence for the purposes of comparison, while no additional adjustments are introduced in the figures reproduced below.\n\nThe assessment supports the broader finding that participation and achievement remain uneven across the assessed population. Two figures are quoted directly from the survey. It reports that 31 percent of assessed learners reached the specified competency threshold, while 68 percent participated in the relevant assessment activity. These figures are presented as reported and are not combined into a derived indicator. Their use is limited to illustrating the scale of the implementation gap identified in the preceding analysis.\n\nInterpretation should therefore focus on the reported levels rather than on changes over time. The figures indicate that participation is materially higher than demonstrated competency, a difference that is consistent with the implementation concern discussed in this section. The survey evidence also provides a basis for maintaining the existing monitoring emphasis on assessment coverage, learner progression, and the reliability of school-level reporting. No inference is made beyond the two reported figures, and the source is cited solely to substantiate the claim concerning the continuing gap between participation and measured competency.\n\nMention: \"Nationwide School Assessment Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Nationwide School Assessment Survey for the directly reported figures of 31 percent reaching competency and 68 percent participating.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:a29cfb59c236506f8ec55337", "occurrence_id": "e44d438a8fbe33335a235ba5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Annual Transport Network Dataset (ATND) is treated here as a geospatial source produced by FAPRI for 2017. Its network observations provide the underlying evidence for the accessibility assessment, while the interpretation remains at an aggregate level. The dataset is cited to support the reported pattern in service reach and connectivity, not to reproduce individual records or present the source as a complete account of observed conditions.\n\nFor this purpose, the analysis calculates an average across the recorded network units and uses the resulting distribution to identify a broad trend in transport availability. The reported measure therefore reflects a derived summary of the source rather than a direct quotation of any single value. This treatment reduces the influence of unusually high or low observations and permits comparison among the analytical groups used in the assessment.\n\nThe resulting pattern is used as one supporting element in the discussion of access constraints. It is read together with the assessment’s other evidence, and no causal inference is assigned to the dataset alone. The calculated average is retained for consistency in the results tables, while the underlying observations remain available for quality review and replication of the analytical procedure. This distinction is material because the conclusion concerns the direction and general magnitude of the pattern, not an isolated observation.\n\nMention: \"Annual Transport Network Dataset (ATND)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the 2017 FAPRI geospatial Annual Transport Network Dataset (ATND) to calculate an average and identify a transport trend.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:96ab111d158ee9eda323406b", "occurrence_id": "430832d08dba5e8713ace9c3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Municipal Deprivation and Hardship Registry (MDHR) is used as the principal database for the assessment of household hardship. Its entries provide a consistent basis for reviewing the distribution of deprivation indicators and for identifying the scale of reported need. No adjustment or external calibration is applied to the figures cited below.\n\nThe registry records a deprivation rate of 42.6 percent among the households covered by the assessment. This figure is taken directly from the MDHR database and is presented as reported, without recalculation or extrapolation. It provides the main measure of the prevalence of hardship in the reviewed records.\n\nThe same source reports 18,740 households meeting the relevant hardship criteria. This second figure is also reproduced directly from the registry and is used to indicate the associated caseload. Taken together, the two reported values establish the quantitative basis for the assessment and support the use of the registry as the central evidence for the findings presented in this section.\n\nMention: \"Municipal Deprivation and Hardship Registry (MDHR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the 42.6 percent deprivation rate and the figure of 18,740 households directly from the MDHR database.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:e4def80c4137b6e199efbba9", "occurrence_id": "c62e66bd5680705853a1db2d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Implementation progress is assessed against the operating conditions relevant to drought preparedness and resilience. The assessment considers institutional coordination, continuity of essential functions, and the capacity to maintain planned activities during periods of climatic stress. These dimensions provide the basis for interpreting implementation performance within the broader results framework.\n\nThe Coordinated Drought and Resilience Registry database is cited in relation to the 2023 reporting period as contextual support for the observation that drought risk requires coordinated resilience measures. The reference serves to corroborate that general characterization within the discussion of implementation conditions. No figures, classifications, or other recorded values from the database are incorporated into the assessment, and no calculation or comparison is derived from it.\n\nThe cited source therefore supports one contextual claim in a broader paragraph concerning preparedness and continuity. Findings reported elsewhere are based on the documented implementation evidence and the indicators defined for the assessment. This approach preserves a clear distinction between contextual reference material and the evidence used to assess progress, while retaining the source as a relevant background reference for the reported operating environment.\n\nMention: \"Coordinated Drought and Resilience Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The source provided contextual support for one broader claim, while none of its recorded values were used.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:62a946dce93bccb559141d8a", "occurrence_id": "f6f2cadc1056a7070df392b5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Harmonised Living Standards Survey is used as a survey for the descriptive statistics presented in this annex. It provides a consistent basis for reviewing household conditions and for checking the internal coherence of the indicators used in the assessment. The source is cited in passing and is not used to establish a broader causal interpretation.\n\nTwo figures are taken directly from the survey. The reported coverage rate is 62.4 percent, while the corresponding rate for the second group is 37.6 percent. These values are reproduced as published, without recalculation, interpolation, or adjustment. Their inclusion is intended to document the magnitude of the reported observations rather than to imply a time series or a comparison across successive observations.\n\nThe remaining analysis relies on the definitions and tabulations set out in the assessment materials. Differences between the two reported figures should therefore be read as descriptive results within the cited source. No additional assumptions are introduced concerning population scope, measurement error, or the construction of the underlying indicators. The figures are retained in the narrative to support traceability and to permit straightforward reconciliation with the source material during review.\n\nMention: \"Harmonised Living Standards Survey\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported figures of 62.4 percent and 37.6 percent from the Harmonised Living Standards Survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:53dd13f7ff227980074b834d", "occurrence_id": "2a90d19dde082392a04aea05", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Civil Time Use Survey is a survey produced by ATO covering Viet Nam and referring to 2021. It provides the principal empirical basis for the time allocation assessment, with attention to the distribution of daily activities across the observed population. The material is treated as a statistical input for aggregate analysis rather than as a descriptive quotation.\n\nThe assessment calculates average time devoted to selected activities from the underlying observations and compares those averages across relevant population groups. These calculations are used to identify the prevailing allocation of time and to distinguish broad patterns from isolated results. No single reported observation is treated as representative in itself; the interpretation relies on the resulting summary measures.\n\nA trend is also derived by arranging the available observations into comparable categories and examining the direction and relative magnitude of change. This approach supports a consistent reading of time-use differences and reduces the risk that unusual individual responses will determine the findings. The calculated averages and trend provide the central evidence for the assessment of participation, unpaid activity, and the balance between principal forms of daily work.\n\nThe resulting indicators should be read as analytical estimates derived from the survey material. They are used to support implementation monitoring and to establish a reproducible basis for subsequent comparison, while retaining the coverage and reference period specified for the Civil Time Use Survey.\n\nMention: \"Civil Time Use Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage describes deriving average time-use measures and a trend from the Civil Time Use Survey rather than quoting individual figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:d1237e0e2592934bfc35c92a", "occurrence_id": "a3d09d834266cb8426636705", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Cross-Sectoral Storage and Loss Registry is a database produced by Oxford and referring to 2023. It was incorporated into the evidence base after records were collected from the relevant reporting streams, harmonised across differing descriptions, and archived for subsequent review. The source is cited here to support the assessment of reporting consistency rather than to establish a separate programme activity.\n\nThe registry indicates that storage and loss information is not uniformly recorded across the materials reviewed. Differences in terminology, coverage, and record completeness affect the direct comparability of observations and require caution when interpreting aggregate results. This finding supports the broader assessment that improvements in routine information management are necessary for reliable monitoring.\n\nFor implementation purposes, the archived material provides a traceable basis for checking classifications and documenting adjustments made during consolidation. The resulting evidence was used to verify the consistency of the underlying records and to retain a reproducible account of the information considered. No inference is drawn beyond the stated assessment of record quality and comparability.\n\nMention: \"Cross-Sectoral Storage and Loss Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The passage uses the 2023 Oxford database to support a claim about uneven reporting consistency and to document the collection, harmonisation, and archiving of records.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:da2836f0da75d0fbe1b2ff84", "occurrence_id": "79888377947afa2bc9d5ae59", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Civil Nighttime Lights Index (CNLI) indicator as a supplementary measure of observed activity. It is applied to support the finding that changes in recorded luminosity provide a consistent signal for comparing relative levels of activity across the units examined. The measure is treated as an analytical input rather than as a direct substitute for household, enterprise, or administrative statistics.\n\nRecords were collected from the cited series, checked for completeness, and harmonised to a common structure before analysis. Differences in naming, coding, spatial granularity, and observation status were reconciled through documented procedures, while unresolved entries were retained with their original qualification. The resulting archive preserves the source values and the processing decisions needed to reproduce the summary tables. No inference is drawn from isolated changes where coverage or comparability is uncertain.\n\nWithin the assessment, the indicator supports a narrow empirical claim concerning the association between measured luminosity and reported activity. It is therefore used alongside other evidence and is not presented as an independent estimate of welfare, population, or service access. The archived series provides a transparent basis for checking the direction and consistency of the reported pattern, subject to the limitations noted in the methodological discussion.\n\nMention: \"Civil Nighttime Lights Index (CNLI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The passage identifies the CNLI indicator and records its collection, harmonisation, and archiving for use in the assessment.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:c9b8338390fc2b5fd820764b", "occurrence_id": "8d47b111725025e32a855e71", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis draws on the Ministerial Hiring and Vacancies Survey (MHVS), a survey produced by UNHCR RBWCA. The source provides the principal empirical basis for assessing movements in reported hiring activity and unfilled positions. Its use is limited to the information required for the indicators presented in this assessment.\n\nRather than reproducing individual observations, the analysis aggregates the available records and calculates average levels for the relevant categories. It also derives a trend from the sequence of observations, allowing changes in staffing conditions to be assessed in a consistent manner. The resulting measures are therefore analytical estimates based on the survey evidence, rather than figures copied directly from a single record.\n\nThe calculated averages and trend are used to support interpretation of staffing pressures, recruitment performance, and the persistence of vacancies. Differences between categories are considered together with the direction of change, while isolated fluctuations are not treated as decisive evidence. This approach reduces the influence of exceptional observations and provides a more stable basis for monitoring implementation. The MHVS remains the central source for these calculations, and the derived results are reported as summary measures in the accompanying analysis.\n\nMention: \"Ministerial Hiring and Vacancies Survey (MHVS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives average hiring and vacancy levels and a trend from the Ministerial Hiring and Vacancies Survey (MHVS).\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:198d7b84df05fa515533adec", "occurrence_id": "62cb8437bf0e481b74bdfe0b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses Harmonised Inequality Microdata as the principal evidence for the distributional analysis. This microdata source is produced by ENIA and refers to 2021. It provides the underlying observations required to examine differences across households and to maintain a consistent basis for summary calculations.\n\nThe results reported here are calculated from the source rather than reproduced as isolated values. Observations are grouped according to the relevant welfare distribution, after which means are derived for each group and compared across the distribution. The resulting series is used to estimate the direction and magnitude of the trend, while the overall average provides a single measure for monitoring purposes. This treatment reduces the influence of individual observations and permits comparison of broad patterns within the assessment framework.\n\nThe calculated average indicates the central position of the distribution, whereas the derived trend identifies whether disparities are widening or narrowing over the analytical sequence. These measures are used to support implementation monitoring and to inform interpretation of changes in beneficiary conditions. No single observation is treated as representative of the aggregate result. Instead, the analysis relies on the complete set of available records in Harmonised Inequality Microdata, with procedures applied consistently to preserve comparability and reduce avoidable variation in the reported indicators.\n\nMention: \"Harmonised Inequality Microdata\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses ENIA's 2021 microdata to calculate a distributional average and derive a trend.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:0743fc69d4cfd634af423eae", "occurrence_id": "757f0a11f82cc4bc04382fa1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Cross-Sectoral Early Childhood Survey (CECS) is used as a survey covering Sindh to inform the assessment of participation in early childhood services. Its measures are treated as descriptive evidence for comparing reported access across households and for identifying differences in service engagement. The source is cited in the analysis because it provides a consistent basis for interpreting the distribution of reported participation.\n\nThe broader assessment indicates that engagement remains uneven. In the CECS, 62 percent of households reported that a child had participated in an organised early childhood activity, while 48 percent reported access to a formally structured service. These two figures are reproduced from the survey without recalculation and are used to support the finding that participation is not uniform across the covered population.\n\nThe figures should be read as reported household measures rather than as estimates of all children or as evidence of change over time. They provide a limited but relevant empirical basis for the paragraph’s conclusion on unequal service reach. Other findings in the assessment are interpreted separately and are not attributed to the CECS.\n\nMention: \"Cross-Sectoral Early Childhood Survey (CECS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage takes the 62 percent and 48 percent figures directly from the CECS to support the finding of uneven early childhood service participation in Sindh.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:f560a9f9162125dd242830f9", "occurrence_id": "d139413f82c2d5d2c7cfb305", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Baseline Disability and Support Survey (BDSS) is a 2020 survey and serves as the principal data source for the analysis. Its observations provide the basis for assessing reported access to support, unmet needs, and variation across the recorded population. The source is treated as an analytical dataset rather than as a set of isolated reference figures.\n\nFor each indicator, the available observations are consolidated into summary measures and compared across the relevant categories. The resulting calculations identify the direction of change and provide an average level for the observed responses. This approach reduces the influence of individual responses and permits a consistent assessment of the distribution represented in the survey. The trend is therefore derived from the underlying BDSS observations, while the average reflects the combined values retained for analysis.\n\nInterpretation remains subject to the coverage and response structure of the source. The calculated measures should be read as descriptive estimates of the conditions recorded by the BDSS, not as independent administrative totals. They are used to establish the analytical baseline, support comparison between indicators, and inform the assessment of support requirements. No single reported observation is treated as sufficient evidence of a general pattern; the findings rely on the aggregate movement and average calculated from the survey data.\n\nMention: \"Baseline Disability and Support Survey (BDSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses the 2020 BDSS survey to calculate an aggregate trend and average rather than reproduce individual figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:c17b36a5f0732aacff3710a0", "occurrence_id": "1adb90ece8aa900ee20728b5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Harmonised Sea Level Dataset as the geospatial source for observations referring to 2015. The dataset provides the evidentiary basis for examining variation in recorded sea levels across the analytical series. Its role is limited to supporting the calculation described below, rather than serving as a source of isolated values reproduced in the text.\n\nFor each observation interval, the recorded values are assembled into a consistent analytical series and reviewed for continuity. The analysis calculates an average level and derives a trend from the resulting sequence. This procedure reduces the influence of short-term fluctuation and provides a summary measure suitable for comparison across reporting intervals. The calculated measures are used to identify the direction and approximate persistence of change, while avoiding interpretation of individual observations as independent findings.\n\nThe resulting estimates indicate a gradual movement in the measured series, with variation around the calculated average. The trend is treated as an analytical result generated from the dataset and not as a value directly stated by the source. Limitations arise from the temporal coverage and regularity of the underlying observations, and the estimates should therefore be read as indicative measures for monitoring rather than as precise forecasts. Further updates can apply the same calculation procedure when additional observations become available.\n\nMention: \"Harmonised Sea Level Dataset\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis computes an average and derives a sea-level trend from the Harmonised Sea Level Dataset for 2015.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:ab15c252d20c0c616247457f", "occurrence_id": "38de07bf27ddb8b84d006c66", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment treats the Coordinated Vegetation Condition System as an administrative source for 2022. It is used as the documentary basis for the vegetation condition observations presented in this annex. No adjustment, interpolation, or consolidation is applied to the cited observations, and the extraction preserves the units and precision reported in the source. The source is therefore used as a direct evidentiary record rather than as a basis for a separate calculation.\n\nTwo figures are reproduced directly from the Coordinated Vegetation Condition System. The reported share in the first observation is 47.3 percent, while the corresponding share in the second observation is 62.1 percent. These values are presented as reported and are not converted into an index, rounded to whole numbers, or interpreted as estimates generated by the assessment. Their inclusion provides a transparent link between the narrative finding and the underlying administrative record.\n\nThe difference between the two figures is noted only to clarify the scale of the reported variation. It is not presented as a trend, since the cited material supplies the observations used here without requiring an additional series. The figures support the conclusion that vegetation condition is uneven across the observations covered by the source, while the limits of the available evidence are retained in the interpretation. Future updates should preserve the same treatment of source values to maintain comparability.\n\nMention: \"Coordinated Vegetation Condition System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the 2022 administrative source and reproduces the figures 47.3 percent and 62.1 percent.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:a083a2c281b76c007f813529", "occurrence_id": "39af418dae50315d9d424dc3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The built-up surface estimate was reviewed against the Standard Built-Up Surface Survey, a survey produced by Maddison. The source was treated as an independent reference for assessing the plausibility of the reported extent and for identifying material divergence between measurement approaches.\n\nThe comparison supports the claim that the estimated built-up surface is within a reasonable range of the corresponding reference measure. Differences were retained where they could reflect divergent definitions, observation procedures, or treatment of marginal areas. No adjustment was applied solely on the basis of the comparison, since the survey is used for validation rather than as a replacement for the underlying estimate.\n\nThe cross-check provides a limited but relevant assurance for the broader assessment. It indicates that the principal result is not dependent on a single measurement stream, while also preserving the uncertainty associated with differences in coverage and classification. The finding should therefore be interpreted as corroborative evidence rather than as an independent recalculation of the estimate.\n\nMention: \"Standard Built-Up Surface Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The author used the Standard Built-Up Surface Survey from Maddison to cross-check an independent estimate of built-up surface extent.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:2408f86c14d9fe1f35e90cb4", "occurrence_id": "6f5ae7e359dcdc963f3ef28e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Municipal Primary Care Index, an indicator produced by the UK Cabinet Office for 2018, is used to support the assessment of basic service coverage. The source provides a consistent reference for interpreting differences in reported access and for locating the position of primary care within the wider service profile. Its use is limited to the figures stated below, without further adjustment or extrapolation.\n\nThe assessment reports that 62 percent of households were within the defined access measure and that 47 percent received the specified level of routine contact. These two figures are taken directly from the Municipal Primary Care Index and provide the numerical basis for the observation that coverage remained uneven across the service system. They are presented as reported values rather than as estimates generated through the assessment.\n\nThe evidence indicates that aggregate availability should not be treated as equivalent to effective access. The gap between the two reported figures supports a cautious interpretation of service performance and indicates the relevance of local implementation conditions to results. The figures are therefore used to substantiate a limited claim concerning coverage, while broader conclusions remain subject to the evidence reviewed elsewhere in the assessment.\n\nMention: \"Municipal Primary Care Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Municipal Primary Care Index to report the directly sourced figures of 62 percent and 47 percent.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:1f14cb7b52c79b8f0e9c7fb6", "occurrence_id": "a105fbf2d8cc071630294d3f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Cross-Sectoral Immunisation Coverage Survey is treated as the principal survey for 2020 in assessing reported immunisation coverage. Its results provide a reference point for reviewing the reasonableness of the coverage estimate used in the assessment and support a consistent interpretation of the available evidence.\n\nThe estimate derived from routine reporting records is cross-checked against the findings of the Cross-Sectoral Immunisation Coverage Survey. The comparison indicates whether the reported level is broadly supported by an independent measurement and helps identify material differences requiring further review. The survey evidence is therefore used as the central benchmark rather than as a supplementary narrative reference.\n\nWhere the two measures are aligned, the estimate is retained for analytical purposes, subject to the limitations of the underlying reporting process. Where a discrepancy is observed, the survey result provides a basis for qualifying the estimate and for avoiding unwarranted precision in the presentation of coverage. This approach improves the transparency of the assessment while preserving the distinction between the reported estimate and the source used to test it.\n\nMention: \"Cross-Sectoral Immunisation Coverage Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The author uses the 2020 survey results as a benchmark to cross-check an immunisation coverage estimate derived from routine reporting records.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:1fcd66ce4b2d8700eaf667d8", "occurrence_id": "9163b934a44b547eeb7bde56", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Clustered Non-State Education Index (CNEI) is treated as an indicator for the assessment of non-state education activity. The 2024 reference year is used to align the index with the period covered by the project’s analytical work. The indicator is produced by UNTAD and provides a structured basis for reviewing patterns that are otherwise difficult to observe consistently across reporting systems.\n\nThe assessment uses the CNEI as the principal external benchmark for cross-checking an estimate derived from a separate source. This comparison is intended to test the plausibility of the estimate rather than to replace the underlying calculation. Where the two measures show similar direction and scale, the estimate is considered sufficiently robust for planning purposes. Where material differences arise, the discrepancy is recorded for further review and is not resolved through mechanical adjustment.\n\nThe evidence from the CNEI is therefore interpreted alongside the project estimate, with attention to coverage, definitions, and possible differences in reporting practice. Its value lies in providing an independent reference point for quality assurance and in identifying areas where the estimate may require qualification. The resulting assessment retains the separately derived estimate while documenting the cross-check supplied by the CNEI indicator produced by UNTAD for 2024.\n\nMention: \"Clustered Non-State Education Index (CNEI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the 2024 CNEI indicator produced by UNTAD to cross-check a separately derived estimate.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:f8a20f06d8a90279d7476810", "occurrence_id": "a8afa85f8abbf41af7954a43", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Biennial Deprivation and Hardship Registry (BDHR) database provides the principal empirical basis for the assessment. Its records are treated as a consistent body of observations on deprivation and hardship, permitting comparison across successive observation periods. The database is used to establish the analytical baseline and to support the interpretation of changes in reported conditions.\n\nThe analysis does not reproduce individual registry entries as standalone findings. Instead, the observations are aggregated and reviewed for their direction and persistence over time. A central tendency is calculated from the available records, and the resulting average is used to distinguish broad movement from isolated variation. This approach reduces the influence of irregular observations and supports a more stable reading of the underlying pattern.\n\nThe resulting trend is presented as an analytical estimate derived from the BDHR database rather than as a direct transcription of any single record. It provides the basis for assessing whether hardship is broadly increasing, declining, or remaining stable across the observation sequence. Interpretation is limited to patterns supported by the aggregated evidence, with attention to the possibility that changes in coverage or reporting intensity may affect individual observations. The calculated measures therefore function as summary evidence for implementation monitoring and not as independently verified counts of all affected persons.\n\nMention: \"Biennial Deprivation and Hardship Registry (BDHR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives an average and an overall trend from the BDHR database rather than reproducing individual records.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:0adad8a66ddcb74f19f7af8c", "occurrence_id": "4237a082aaa4b97a3e74e9f9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Civil Public Works Index, an indicator produced by the Department of National Planning for 2018, is used as the reference series for the assessment. Its inclusion provides a consistent basis for examining movement in civil public works activity while avoiding reliance on isolated observations. No additional source classification is required for the calculation presented here.\n\nThe analysis derives an average level and a directional trend from the available observations. These results are treated as summary measures for comparison across the reporting period, rather than as reproduced values from the source. The procedure reduces the influence of short-term variation and supports a more stable interpretation of sector performance.\n\nOn this basis, the reported change is assessed as gradual and broadly consistent with the underlying series. The derived measures support the finding that activity evolved over time without establishing a causal explanation. They are therefore used to inform the assessment of implementation progress and to frame the interpretation of related outputs. The source is retained in the evidence record so that the calculation can be reviewed and repeated using the same reference material.\n\nMention: \"Civil Public Works Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives an average and a directional trend from the Civil Public Works Index rather than quoting individual figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:2a199ea9170e8417a1e8f126", "occurrence_id": "5886eda443c3e72969ebded5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Standard Informal Sector Survey is the survey covering Gabon used for the statistical observations in this assessment. It provides the principal evidence for describing participation in informal economic activity and for identifying the scale of reported engagement. No adjustment or supplementary source is applied to the two figures reproduced below.\n\nThe survey reports that 63.4 percent of the relevant population is engaged in informal activity. This figure is presented directly as reported in the source and is used to establish the magnitude of participation. It should therefore be read as a source observation rather than as an estimate derived through comparison with another dataset.\n\nThe same source reports that 18.7 percent of informal operators have access to the specified form of support. This second figure is also taken directly from the survey and provides the principal indication of coverage within the assessed population. Together, the two reported proportions support the assessment of the sector’s scale and the limited reach of formal support arrangements. Their use is descriptive and does not imply a recalculation, projection, or causal interpretation.\n\nMention: \"Standard Informal Sector Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Standard Informal Sector Survey for Gabon and reports its directly sourced figures of 63.4 percent and 18.7 percent.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:9d36cd02a8b1735cebaa310b", "occurrence_id": "bab9524c46a16967801a458f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The principal reference is the Stratified Literacy and Numeracy Index, an indicator produced by IFPRI and applied here as the common basis for interpreting reported learning conditions. Its construction permits the results to be read across the relevant population groups without introducing an additional adjustment in this assessment. The source is treated as the authoritative record for the two values reproduced below.\n\nThe index reports 62.4 percent for literacy attainment and 47.1 percent for numeracy attainment. These figures are taken directly from the source and are presented without recalculation, rounding, or transformation. The difference between the two reported measures is therefore a feature of the underlying record rather than an estimate generated for this report.\n\nThe evidence indicates a higher recorded level for literacy than for numeracy. This comparison is descriptive and does not establish a causal relationship between the measures. It is used to support the assessment of relative performance and to identify the area requiring closer review in subsequent implementation work.\n\nThe two values should be read as reported indicators rather than as independently verified observations. Their use in this annex is limited to documenting the position reflected in the source and maintaining consistency between the narrative assessment and the cited statistical evidence.\n\nMention: \"Stratified Literacy and Numeracy Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage retains the source-reported figures of 62.4 percent for literacy attainment and 47.1 percent for numeracy attainment.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:acce37171f9ebd6eceecddc4", "occurrence_id": "fd63c1fbef0b00e904e45a18", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Government Built-Up Surface Registry as the principal reference for the reported built-up surface. It is a database produced by INCAJU and referring to 2020. The source is treated as an administrative record and is cited without adjustment to its coverage or internal classification.\n\nTwo figures are reproduced directly from the database. The first records 42 hectares, while the second records 58 hectares. These values provide the central quantitative evidence for the comparison presented in this note and are retained in their reported form.\n\nThe figures are not recalculated, extrapolated, or combined with observations from another source. Their purpose is to establish the magnitude of the recorded surface and to support a restrained interpretation of the difference between the two reported values. Any assessment of change should therefore remain limited to the information contained in the registry and should not imply a broader estimate than the source supports.\n\nThe use of the Government Built-Up Surface Registry also permits the results to be traced to a single database and a clearly stated reference year. This improves consistency in review and reduces the risk that the two quoted measurements are interpreted as estimates generated by the assessment itself.\n\nMention: \"Government Built-Up Surface Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported figures of 42 hectares and 58 hectares from the Government Built-Up Surface Registry.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:7b485fbde755f392cd01c040", "occurrence_id": "9003b0f5758b565dce95ba36", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Data sources.** The Monthly Camp Management Survey (MCMS) is the principal survey used for the assessment of recurring management conditions. It is produced by Refinitiv Eikon and is treated as a consistent source for assembling observations across reporting periods. The assessment relies on the information contained in the survey rather than reproducing individual entries or presenting isolated reported figures.\n\nThe underlying observations are reviewed for comparability and consolidated into summary measures. A simple average is calculated to describe the central level of the reported conditions, while successive observations are ordered to identify the direction and persistence of change. This procedure reduces the influence of short-term variation and provides a basis for assessing whether recorded conditions improved, deteriorated, or remained broadly stable.\n\nThe resulting trend is used as the principal quantitative indication in this assessment. Interpretation remains limited to patterns supported by the MCMS material, with attention to changes in the average and to the consistency of movement across observations. The calculated measures therefore serve an analytical purpose and should not be read as direct quotations from the source. Any divergence between individual observations and the calculated average is retained as part of the assessment of variability, rather than treated as an error or omitted from consideration.\n\nMention: \"Monthly Camp Management Survey (MCMS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage derives an average and a trend from the Monthly Camp Management Survey (MCMS) to assess changes and variability.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:85d9e61fef07ce6569b7966c", "occurrence_id": "9e5ff00f054b6a3a6594bbba", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Subnational Cash Transfer Registry Survey (SCTRS) is used as the principal survey for assessing changes in registered cash transfer activity. Its records provide a consistent basis for comparing observations across reporting periods, while retaining the underlying variation that is relevant to implementation monitoring. No supplementary source is introduced for this calculation.\n\nThe analysis derives a trend from the SCTRS observations by ordering the reported values and examining their movement across successive periods. It also computes an average over the available observations, allowing the assessment to distinguish the general level from short-term increases or declines. These calculations are used in preference to reproducing isolated figures, since a single observation would not adequately represent the pattern in the registry.\n\nThe resulting measures indicate the direction and central tendency of recorded activity and support a restrained interpretation of implementation progress. The trend is treated as descriptive rather than causal, and the average is used as a summary measure rather than as a benchmark for individual records. Differences between observations remain material where the registry shows uneven activity, but the calculated measures provide the principal quantitative basis for this assessment.\n\nMention: \"Subnational Cash Transfer Registry Survey (SCTRS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses SCTRS observations to calculate a trend and average for registered cash transfer activity.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:ef53e9f229e7695d1dba3184", "occurrence_id": "98d5e2a24bcac8704173104f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses subnational evidence to support the finding that reported food security conditions varied across local areas. The evidence is interpreted alongside the implementation record, with attention to the consistency of coverage and the treatment of observations that were not available for all areas.\n\nThe Subnational Food Security System is an administrative source covering Uganda. Records were collected from local reporting channels, harmonised to support comparison across administrative areas, and archived for subsequent review. The resulting series provides a structured basis for examining the distribution of food security indicators without implying that all local observations were recorded with equal completeness.\n\nFor the purposes of this assessment, the source supports the claim that geographic differences were material to implementation planning. Its archived records were used to check whether the pattern described in the wider analysis was also visible below the national level. Inconsistencies in reporting were retained as a limitation rather than corrected through unsupported assumptions.\n\nThe evidence therefore informs interpretation of local variation while remaining supplementary to the broader implementation findings. It is used as a documented statistical input, with the collection and harmonisation procedures treated as part of the evidence trail.\n\nMention: \"Subnational Food Security System\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The passage reports that the Subnational Food Security System supplied collected, harmonised, and archived administrative records covering Uganda.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:77e0d713a316d46be993ddeb", "occurrence_id": "d6b2b0c7e6d35ea7f1e1fa00", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Decennial Migration and Remittance Census is the principal census used for the assessment of migration-linked transfers and household receipts. Its records provide a consistent basis for examining reported amounts, participation rates, and the distribution of receipt categories across successive reference periods. The material is treated as the central empirical input to the analysis rather than as a set of isolated figures.\n\nFor each indicator, the analysis computes a three-period moving average and compares the resulting values across the observation sequence. This procedure reduces the influence of short-term fluctuations and supports identification of the underlying trend in reported transfers. The same approach is applied to participation and receipt measures, allowing changes in incidence to be considered alongside changes in amounts.\n\nThe resulting estimates indicate whether observed movements reflect a sustained pattern or a temporary departure from the prevailing level. Where the number of observations is sufficient, the assessment also derives an overall average and measures the direction of change between the initial and final observations. These calculations are used to support interpretation of the findings and to establish a transparent basis for subsequent monitoring. No individual entry is treated as representative without reference to the calculated averages and the trend derived from the census.\n\nMention: \"Decennial Migration and Remittance Census\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives moving averages, overall averages, and trends from the Decennial Migration and Remittance Census.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:53833ca8437145e95fc2d52b", "occurrence_id": "6c90d93d4a00ecc07a2d5d64", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Small-Area Skills and Employment System (SSES) is used as an administrative source for the assessment of local labour-market conditions. Its records provide a structured basis for reviewing reported participation, employment status, and skill distribution across the units considered in the analysis. The source is cited as supporting evidence for the interpretation of the results.\n\nThe estimate presented in the main analysis is cross-checked against SSES to assess whether the observed level is consistent with independently recorded administrative information. This comparison is used to support one claim concerning the reliability of the estimate, rather than to replace the principal source or establish a separate time series. Differences between the two measures are treated as a basis for qualification where coverage, timing, or reporting practices may not be fully aligned.\n\nThe comparison indicates that the estimate remains within a plausible range when assessed against the administrative records. SSES therefore provides a corroborating reference for the stated finding, while the estimate drawn from the other source remains the basis for the reported result. No broader inference is made from the comparison, and the administrative material is not used to attribute changes to a specific intervention.\n\nMention: \"Small-Area Skills and Employment System (SSES)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the Small-Area Skills and Employment System (SSES) as an administrative source to cross-check an independently derived estimate.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:f68607a71f0f67ba6411ee7e", "occurrence_id": "88a64ae0e66df4b81701f475", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Staffing conditions are treated as a material determinant of service continuity, particularly where recruitment and retention pressures affect the distribution of qualified personnel. The evidence is therefore used to support the broader assessment of workforce capacity, rather than as a separate measure of institutional performance.\n\nThe assessment draws on Metropolitan Teacher Workforce Microdata, a microdata source produced by IRI Worldwide. The record is used as a documentary basis for two figures reproduced in this discussion. It reports that 68 percent of teachers remain in the profession beyond the initial placement period and records 4,216 teachers within the reviewed workforce. These figures are taken directly from the source and are cited without recalculation.\n\nThe two observations indicate both the scale of the personnel base and the extent of continued participation. They support the finding that workforce stability is uneven and that aggregate staffing totals should not be interpreted as evidence of uniform coverage or sustained availability. The figures are consequently presented as descriptive evidence within the wider implementation assessment. They do not, on their own, establish causality, measure instructional quality, or determine whether staffing levels are adequate for all service requirements. Their principal value is to anchor the assessment in a consistent administrative record while maintaining a clear distinction between reported quantities and subsequent interpretation.\n\nMention: \"Metropolitan Teacher Workforce Microdata\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses 68 percent and 4,216 teachers from Metropolitan Teacher Workforce Microdata to support the assessment of workforce stability.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:2cef1968009c64586588e8b0", "occurrence_id": "9117bc290078e7e69d6180c2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis uses the Verified Labour Inspection Survey, a survey covering 2024, as its principal empirical basis. The source is treated as a consolidated record of inspection activity and associated labour conditions. Its entries provide the observations required to examine differences in inspection intensity and reported compliance across the period covered.\n\nThe estimates presented in this annex are calculated from the underlying observations rather than reproduced as individual source figures. A mean level of inspection activity is derived for the reporting period, and the sequence of observed results is used to identify the prevailing trend. This treatment reduces the influence of isolated fluctuations and provides a consistent basis for comparison.\n\nThe calculated average indicates the general level of activity, while the trend measure indicates whether performance strengthened, weakened, or remained broadly stable over time. These measures are interpreted jointly because a stable average may conceal uneven movement within the period. The analysis therefore gives greater weight to the direction and persistence of change than to any single reported observation.\n\nThe findings should be read as summary indicators derived from the survey evidence. They support monitoring of implementation progress and identification of areas requiring further review, but they do not establish causality. Any subsequent assessment should consider operational explanations for deviations from the calculated pattern and verify whether they persist in later reporting.\n\nMention: \"Verified Labour Inspection Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives an inspection average and a directional trend from the Verified Labour Inspection Survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:94c52ff35f176fca52ca8e25", "occurrence_id": "7f1ae0f0de6c4c5fdbed5392", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Biennial Social Protection Survey (BSPS) as a supporting statistical source. This survey was produced by the Department of Defense for 2019 and is cited in the analysis as the basis for two reported proportions. No adjustment, reclassification, or extrapolation is applied to the figures presented below.\n\nDirect extraction from the survey gives 31 percent of respondents indicating receipt of the relevant support, while a second direct figure records 68 percent indicating awareness of the corresponding provision. These values are reproduced directly from the BSPS and are not estimates generated through the assessment. Their inclusion is intended to preserve the reported evidence without implying additional precision.\n\nTaken together, the two observations support the broader finding that access and awareness were not uniform across the population covered by the survey. The difference between 31 percent and 68 percent is presented as descriptive evidence rather than as a measure of change over time. The figures therefore provide a limited but transparent basis for interpreting implementation performance, while the survey remains the sole cited source for this specific statement.\n\nMention: \"Biennial Social Protection Survey (BSPS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the BSPS to report 31 percent and 68 percent as direct evidence of uneven access and awareness.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:9317f438773801345ca1ab7c", "occurrence_id": "851d3fb0a0e55c5d9940e9a5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Combined Informal Trade Registry, a database used to establish the reported scale of informal commercial activity. The source is treated as the primary evidentiary record for the indicators discussed in this annex, and no adjustment is applied to the figures reproduced from it.\n\nTwo numerical observations are taken directly from the database. It records 31 registered traders within the reported set and 47 recorded transactions. These figures are presented as reported values rather than as estimates derived through subsequent calculation. Their inclusion provides a direct indication of the observed participation level and transaction volume captured by the registry.\n\nThe evidence remains limited to the entries contained in the Combined Informal Trade Registry. Accordingly, the figures should be read as administrative counts within the recorded coverage, not as a measure of all activity. The assessment uses them to support the descriptive findings and to maintain traceability between the narrative and the underlying record.\n\nMention: \"Combined Informal Trade Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Combined Informal Trade Registry to report 31 registered traders and 47 recorded transactions directly from the database.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:1fe86bcc2c1c9f2393594fec", "occurrence_id": "cbcca1830947829bd5e46d90", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Small-Area Minimum Needs Registry (SMNR) database is used as the principal statistical reference for the assessment. Its structure permits a consistent reading of minimum-needs indicators across the units represented in the registry, while limiting interpretation to values explicitly recorded in the source. No adjustment, projection, or supplementary estimate is introduced in the figures reported below.\n\nThe assessment reproduces two figures directly from the SMNR database. The first is 31 grades recorded in the relevant classification. The second is 68 percent, reported as the corresponding share in the registry. These figures are presented as source values rather than as results of a calculation undertaken for this report.\n\nTaken together, the two observations provide the central quantitative basis for the accompanying discussion. The count of 31 grades indicates the breadth of the recorded classification, while the 68 percent figure indicates the reported proportion associated with the selected measure. Their use is limited to descriptive assessment, and no causal interpretation is attached to either value.\n\nThe SMNR database is therefore retained as the controlling reference for this passage. Any subsequent revision of the underlying registry would require a corresponding review of the figures reproduced here and of the conclusions that depend on them.\n\nMention: \"Small-Area Minimum Needs Registry (SMNR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported figures of 31 grades and 68 percent from the SMNR database.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:385847a284df2d72813c1157", "occurrence_id": "50a341b178b3b7be99917fc0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Stratified Primary Care Survey (SPCS), a survey produced by the Kazakhstan Ministry of Transport in 2022. The source provides the central empirical basis for the analysis and is treated as the principal input to the measurement exercise. Its observations are used in their reported structure, with no additional source introduced for the estimates presented here.\n\nThe analysis calculates weighted averages across the surveyed units and derives a comparative trend from the distribution of observations. These calculations are used to identify the direction and relative magnitude of change within the source material, rather than to reproduce isolated reported figures. The resulting measures support interpretation of service coverage, utilization, and observed variation across the analytical categories.\n\nThe calculated averages are used for consistency checks and for the construction of summary indicators in the implementation assessment. Where dispersion is material, the analysis retains the underlying differences rather than replacing them with a single unqualified value. The findings should therefore be read as derived measures based on the survey observations, with the SPCS serving as the central evidence for the reported pattern.\n\nMention: \"Stratified Primary Care Survey (SPCS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage reports weighted averages and a derived trend calculated from the 2022 SPCS survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:279913db68c733762a6d1d82", "occurrence_id": "ef6adc3e9f4b8324ff9cd25c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Panel Learning and Equity Registry (PLER) database provides a structured reference for assessing patterns in learning and equity across China. Its records are used here to establish a consistent comparison point for the indicators discussed in this assessment. The material is treated as supporting evidence rather than as a replacement for the principal estimate.\n\nAn estimate obtained from a separate source is cross-checked against the PLER evidence. The comparison is used to determine whether the reported direction and relative differences remain plausible when examined through an alternative body of records. Agreement between the two sources strengthens the interpretation, while any divergence is retained as a qualification on the reported finding.\n\nThe PLER evidence also assists in reviewing the internal coherence of the broader analysis. It provides a basis for examining whether observed disparities are concentrated among particular learner groups and whether the pattern is sufficiently stable to support the accompanying conclusion. No adjustment is made solely on the basis of this comparison, and the cross-check is presented as a validation measure within the wider assessment.\n\nMention: \"Panel Learning and Equity Registry (PLER)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The PLER database is used to cross-check an independent estimate and assess the consistency of learning and equity patterns in China.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:87624bbe69f4ca46e5677a07", "occurrence_id": "dc084fa00433520f5b4adce2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The implementation assessment considers continuity in youth transition support as a relevant operational issue. The Recurrent Youth Transition Registry database is used to support the contextual observation that transition pathways warrant attention in implementation planning. Its role is limited to substantiating that observation within the wider discussion of delivery risks and monitoring arrangements.\n\nThe assessment does not extract, reproduce, or analyse any values from the database. No counts, rates, categories, or derived measures from the registry are incorporated into the findings, and no comparison is made on the basis of its records. The source therefore provides background relevance rather than an evidentiary basis for a quantified result.\n\nThe broader paragraph draws on administrative and programme considerations to describe the need for continuity, referral, and follow-up during implementation. Within that discussion, the registry is retained as a contextual reference supporting one claim about the importance of youth transitions. The conclusions and proposed actions remain grounded in the assessment evidence presented elsewhere in the report.\n\nMention: \"Recurrent Youth Transition Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The passage uses the Recurrent Youth Transition Registry database to support one contextual claim about youth transitions without taking any values from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:057188ee12b602c0dd9479fd", "occurrence_id": "25dab1d484ed81ba1f9d5595", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Data sources.** The Metropolitan Debt and Savings Registry (MDSR) is cited as a database relevant to the interpretation of financial conditions discussed in this assessment. Its inclusion provides background for the discussion of debt and savings records and helps situate the institutional considerations examined in the surrounding analysis. No location, production arrangement, or temporal designation is assigned to the registry in this document.\n\nThe reference supports one contextual claim within a broader paragraph concerning the need to consider both liabilities and accumulated resources when reviewing financial resilience. The registry is not used to establish a benchmark, quantify an outcome, or substantiate a distributional comparison. Its contents are therefore not incorporated into the calculations, classifications, or conclusions reported in the assessment.\n\nThis treatment preserves a clear distinction between documentary context and analytical evidence. The findings rely on the methods and materials specified elsewhere in the report, while the MDSR reference indicates only that debt and savings information constitutes a relevant area for consideration. No values from the database are reproduced, transformed, or interpreted, and no additional attributes are inferred from its title.\n\nMention: \"Metropolitan Debt and Savings Registry (MDSR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Metropolitan Debt and Savings Registry (MDSR) is used solely to provide contextual support for one claim, with no database values taken into the analysis.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:49a66340aafe1284f8f57705", "occurrence_id": "b2af18293b111a0029458145", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The principal empirical input is the Regional Targeting and Payments Survey, a survey used to examine access to assistance, registration outcomes, and the regularity of payment receipt. The material is treated as the central evidence for assessing distributional performance and operational reach. No additional source is used for the estimates presented here.\n\nThe assessment does not reproduce individual observations from the survey. It aggregates the reported measures across comparable population groups and derives average rates for registration, eligibility confirmation, and successful payment. These averages provide a consistent basis for examining whether access and delivery outcomes improve, remain stable, or deteriorate across successive observations.\n\nThe resulting trend is used to inform interpretation of implementation performance. Changes in the average payment rate are read alongside changes in registration and confirmation rates, allowing the analysis to distinguish coverage constraints from delivery constraints. The estimates are therefore analytical measures derived from the survey rather than figures presented as direct quotations from it.\n\nThe findings support a cautious assessment of progress. Variation between observations remains relevant, but the calculated averages reduce the influence of isolated cases and provide a clearer indication of the prevailing pattern. This approach also permits comparisons across recipient groups without introducing an additional data source or an unsupported assumption about the causes of observed change.\n\nMention: \"Regional Targeting and Payments Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage derives average registration, eligibility confirmation, and payment rates and uses them to assess trends in implementation performance.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:84991f8c62656c8314d65fab", "occurrence_id": "aa17fdf14728700a276a26ee", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Harmonised Enterprise and Jobs Survey is used as the sole evidentiary source for the indicators discussed below. It is a 2024 survey, and its reported results are treated as the reference values for this assessment. No adjustment, projection, or supplementary source is applied to the two observations reproduced here.\n\nThe survey records 31 establishments meeting the specified reporting criteria. It also reports a participation rate of 47 percent. These two figures are taken directly from the source and are presented without recalculation. Their function is to establish the observed scale of participation and the corresponding share reported by the survey, rather than to support a comparison across periods.\n\nThe figures should therefore be interpreted as descriptive results for the reference observation covered by the survey. They provide a concise basis for monitoring implementation performance and for identifying whether subsequent reporting remains consistent with the underlying record. Any later assessment should preserve the definitions and counting conventions applied in the source so that changes in the reported values are not confused with changes in measurement.\n\nMention: \"Harmonised Enterprise and Jobs Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the figures 31 establishments and a 47 percent participation rate directly from the 2024 Harmonised Enterprise and Jobs Survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:54dd8fc1bf233c6821f9ee94", "occurrence_id": "3dcfb9713117e977801b4aa4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Regional Early Warning Report (REWR), a report covering 2024, is used as the sole documentary basis for the numerical statements in this section. The source is cited in full to preserve the provenance of the observations and to distinguish reported values from subsequent interpretation. No additional source is introduced for the measures presented below.\n\nThe assessment reproduces two figures directly from the report. The first records 68 percent for the indicator under review, while the second records 14 reported events during the period addressed by the source. These values are retained in their published form and are not recalculated, combined, or converted into a separate estimate. Their function is to establish the principal empirical reference for the assessment.\n\nTaken together, the figures indicate a substantial level of exposure alongside a limited number of recorded events. The difference between the measures should not be interpreted as a trend, since the available evidence consists of the two reported values only. The results therefore support a concise description of conditions documented in 2024, while leaving broader changes over time outside the scope of this note. The citation to the Regional Early Warning Report (REWR) should accompany any reuse of these figures.\n\nMention: \"Regional Early Warning Report (REWR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported figures of 68 percent and 14 reported events from the Regional Early Warning Report (REWR) for 2024.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:02e9401eb76aa99a498c6807", "occurrence_id": "9988d4b5bcb9c8d5aac52dca", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Coverage results were reviewed against the Baseline Immunisation Coverage Registry database to assess the internal consistency of the reported immunisation estimate. The registry provides a structured reference for recorded coverage observations and permits comparison with the estimate obtained through a separate source. This comparison was undertaken as a verification measure rather than as a basis for revising the underlying estimation method.\n\nThe comparison indicates broad agreement in the direction and relative level of coverage across the reporting categories. Differences remain in selected categories and may reflect differences in denominator construction, reporting completeness, or the timing of registration. These discrepancies do not materially affect the claim that coverage remains uneven across the assessed population.\n\nThe registry evidence therefore supports the use of the independent estimate for analytical purposes, while providing a separate check on its plausibility. Results should be interpreted with due regard to the scope of the recorded observations and to possible variation in completeness between categories. No adjustment has been applied solely on the basis of the cross-check.\n\nMention: \"Baseline Immunisation Coverage Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The source was used as a database to cross-check an immunisation coverage estimate from a separate source.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:bf790840466c17660729918f", "occurrence_id": "ecdbf1417f91d35139e14964", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Clustered Market Access Registry (CMAR) database for Azerbaijan covering 2015. The source is treated as the reference point for the market-access observations reported in this section, with its entries read as recorded rather than adjusted for subsequent classification changes.\n\nTwo figures are reproduced directly from the database. The registry records 31 market clusters and 14 peripheral access points. These values are cited to support the assessment that market access was uneven across the observed network, with a substantial concentration of activity in established clusters and more limited reach in peripheral locations.\n\nThe figures are presented as descriptive evidence and are not used to derive a time series, estimate a growth rate, or infer changes outside the coverage of the source. Their function is narrower, namely to substantiate the paragraph’s finding that the distribution of access was concentrated rather than uniform. Interpretation therefore remains bounded by the geographical and temporal scope of the database.\n\nTaken together, the cited entries provide a concise statistical basis for the assessment of spatial variation. No additional adjustment is applied to the reported counts, and the values remain attributable to the Clustered Market Access Registry (CMAR) database as consulted for Azerbaijan in 2015.\n\nMention: \"Clustered Market Access Registry (CMAR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage takes the figures of 31 market clusters and 14 peripheral access points directly from the CMAR database for Azerbaijan in 2015.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:8f07cf5523dc8ec63663074c", "occurrence_id": "f1e35e011992c12248ce7596", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Departmental Elevation Model Registry (DEMR) database for 2022 is referenced in this assessment to establish the availability of a relevant administrative record. The citation is limited to the role of the registry as background for the discussion and does not introduce measurements, classifications, or other values from the database.\n\nWithin the broader analysis of information management, the reference supports the observation that elevation-related records form part of the evidence environment considered during implementation. This observation is presented alongside discussion of reporting procedures, quality assurance, and the interpretation of operational information. The cited material therefore serves a contextual function and is not treated as an analytical input.\n\nNo calculation, comparison, or inference is derived from the DEMR database in this passage. Its inclusion indicates that a recognised record was available for consideration while leaving the substantive assessment dependent on the evidence described elsewhere in the report. This approach preserves a clear distinction between background documentation and data used directly to substantiate findings.\n\nMention: \"Departmental Elevation Model Registry (DEMR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The DEMR database provided contextual support for one broader claim, and no source values were used.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:1b515c73788ae0a9f88cbd08", "occurrence_id": "3a0f668200cf5e6caf4cc355", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis draws on the Experimental Minimum Needs Microdata (EMNM), a microdata source produced by Prio/Uppsala. The material provides the central empirical basis for assessing changes in minimum needs indicators across the reporting sequence. Its records are treated as observations for statistical processing rather than as isolated figures for direct quotation.\n\nFor each indicator, the analysis calculates an average across the available observations and compares the resulting values over successive reporting intervals. The resulting trend is used to identify broad movement, persistence, and variation in minimum needs conditions. This approach reduces the influence of individual observations and supports a consistent comparison of aggregate outcomes.\n\nThe calculations apply the same inclusion rules throughout and retain the underlying scale of each indicator. Missing observations are excluded from the relevant average, while the number of usable records is reviewed as a check on comparability. The findings therefore represent derived measures based on the Experimental Minimum Needs Microdata (EMNM), not values reproduced directly from a source table. Interpretation is limited to patterns supported by the calculated averages and trends, with attention to changes in the number of observations contributing to each result.\n\nMention: \"Experimental Minimum Needs Microdata (EMNM)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives comparable averages and trends in minimum needs indicators from the Experimental Minimum Needs Microdata (EMNM).\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:98c05e9722e02fe2b82baaee", "occurrence_id": "fb7e6e508540afcd6e0d0241", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment of service delivery indicates that access to irrigated production remains uneven across the surveyed units. This constraint is associated with limited uptake of improved inputs and with differences in the reliability of water provision. The finding is supported by two figures reproduced directly from the Verified Irrigation and Inputs Report, a report referring to 2018. The report records that 68 percent of the assessed units received irrigation services during the reference period, while 42 percent obtained the specified inputs through the monitored channels.\n\nThese figures are used to substantiate the broader conclusion that physical access to water has not translated into consistent access to complementary production materials. The gap between the two measures is therefore treated as an implementation concern rather than as evidence of a change over time. No additional calculation is applied to the reported values, and the figures are presented in the form in which they appear in the source.\n\nThe result has implications for supervision and for the interpretation of output indicators. Where irrigation is available but input access is less consistent, reported production outcomes may reflect constraints outside the operation of water infrastructure. Future monitoring should maintain separate observation of these conditions and should document the basis for any consolidation of service coverage. The cited report is accordingly retained as the supporting reference for this assessment.\n\nMention: \"Verified Irrigation and Inputs Report\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Verified Irrigation and Inputs Report for 2018 and directly takes the figures of 68 percent receiving irrigation services and 42 percent obtaining specified inputs.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:66d050fc823bec2575c6ec1b", "occurrence_id": "ba8ae496e2f02a85e866b344", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment considers the structure and documentation of transport networks as part of the broader review of service delivery conditions. This framing recognizes that network records can provide useful background for interpreting institutional arrangements, maintenance responsibilities, and the limits of administrative reporting.\n\nThe Subnational Transport Network Registry (STNR) database produced by the University of Dayton is cited to support this general observation. Its inclusion indicates that an established reference exists for describing transport-network information, but the discussion does not draw on any recorded values, categories, or measurements from the source. The citation therefore provides documentary support for one contextual statement within the wider assessment rather than evidence for a calculation or comparison.\n\nThe analysis remains focused on the interpretation of implementation conditions and does not attribute findings to the registry beyond that limited supporting role. No substantive result in the assessment depends on extraction, aggregation, or transformation of information from the cited database.\n\nMention: \"Subnational Transport Network Registry (STNR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The source is used solely to support a general contextual statement about transport-network documentation, with no values taken from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:65ec8d9e2e73108ac49a5d37", "occurrence_id": "372f7eb016ee9ee1ac73a334", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Endline Wage Structure System (EWSS) is treated as an administrative source for the assessment of wage conditions associated with 2017. Its records provide the central empirical basis for summarising the distribution of reported wages and for examining variation across the wage structure. The source is used in accordance with its recorded coverage, without introducing supplementary institutional or geographic classifications.\n\nThe analysis computes an average from the underlying observations and derives a trend across the ordered wage categories. These measures are calculated from the source records rather than reproduced as isolated reported figures. The resulting summaries therefore describe the central level and directional pattern evident in EWSS, while retaining the distinctions present in the recorded wage structure.\n\nInterpretation is limited to the information supported by the source. The calculated average is used as a summary measure, and the computed trend is used to assess how wage levels vary across the structure. Together, these results provide a concise basis for monitoring wage differentiation and for comparing the internal pattern of the observations within the assessment framework. Any apparent movement is consequently understood as a feature of the calculated series, not as a separately stated value in the source.\n\nMention: \"Endline Wage Structure System (EWSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the 2017 EWSS records to calculate an average wage level and a trend across the wage structure.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:44bcbba47bda58b3b60ec140", "occurrence_id": "9fe2b8e9b16ab308cac97de4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Linked Flood Extent Registry database for 2024 is used as the sole cited source for the numerical observations in this section. It provides the reference values applied in reviewing the distribution of recorded flood exposure and supports the interpretation presented below. No adjustment or extrapolation is applied to the cited values.\n\nThe assessment indicates that recorded exposure is unevenly distributed across the observations reviewed. The registry reports 14.6 per cent in the higher-impact group and 3.1 per cent in the lower-impact group. These figures are reproduced directly from the database and are used to substantiate the finding that the reported burden is concentrated rather than evenly dispersed.\n\nThe comparison is descriptive and does not imply a change over time. It is intended to clarify the scale of the difference within the reported records and to provide a consistent numerical basis for the accompanying assessment. The cited proportions should therefore be read as registry observations for 2024, rather than as estimates generated through a separate calculation.\n\nMention: \"Linked Flood Extent Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the directly reported values of 14.6 per cent and 3.1 per cent from the 2024 Linked Flood Extent Registry database to support a claim of unevenly distributed exposure.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:7c5a078c02cc0c42b05e705e", "occurrence_id": "8ef8f8d1ee159a2b6dae2e0b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis uses the Endline Informal Trade Survey, a 2018 survey produced by NCHS/CDC/WHO, as a supplementary source for assessing reported informal trade activity. The source provides a distinct reference point for checking the internal consistency of the estimate presented in this section.\n\nThe estimate is derived from a separate body of administrative information and is not calculated from the survey. The survey is used to cross-check the magnitude of the estimate and to identify whether the reported level is broadly consistent with observed trading patterns. This comparison supports the claim that informal activity constitutes a material component of the measured total.\n\nThe comparison is interpreted cautiously because the two sources may differ in coverage, reporting practices, and measurement conditions. Accordingly, the survey evidence is treated as corroboration rather than as a replacement for the principal estimate. Any divergence is noted as a potential indication of underreporting, sampling variation, or differences in the definition of activity. On this basis, the cited source strengthens one conclusion within the broader assessment while leaving the remaining results dependent on the primary evidence and the stated analytical assumptions.\n\nMention: \"Endline Informal Trade Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Endline Informal Trade Survey is used to cross-check the principal estimate and support the assessment of informal trade activity.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:c16b16a2c614842561db1f62", "occurrence_id": "b906b1305df1e56f0dc7f34b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment relies on the Government Targeting and Payments System (GTPS) as an administrative source for examining changes in recorded programme activity. The system provides a consistent basis for comparing observations across reporting periods and for checking the internal continuity of the series. Its records are used as the central empirical input to the analysis.\n\nThe results are derived by calculating period averages and measuring the direction of change across the resulting sequence. This approach reduces the influence of isolated entries and permits the assessment to distinguish sustained movement from short-term variation. The reported pattern therefore reflects calculations applied to the GTPS records rather than a reproduction of selected observations.\n\nThe estimated average is used to establish the typical level observed in the available records, while the calculated trend indicates whether activity increased, declined, or remained broadly stable over successive periods. These measures support interpretation of implementation performance and provide a basis for identifying deviations that may warrant further review. All conclusions in this section remain conditional on the completeness and consistency of the administrative records available in GTPS.\n\nMention: \"Government Targeting and Payments System (GTPS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis calculates period averages and a trend from Government Targeting and Payments System (GTPS) records.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:50d93acb84df2b61306f199d", "occurrence_id": "1b1b09af4a44674ca4c3a3b5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Integrated Labour Force Census is used as a census source for the labour-market indicators considered in this assessment. Its observations provide a consistent basis for examining participation, employment, and the distribution of work across the reporting series. No additional attribution is required for the interpretation presented here.\n\nThe analysis derives an average level and a trend from the observations, rather than reproducing individual figures as reported values. Successive records are reviewed for direction and persistence, and the resulting measures are used to reduce the effect of short-term variation. This treatment supports comparison across the analytical period while retaining the underlying structure of the source.\n\nOn this basis, the assessment finds that labour-market participation has not evolved uniformly. The calculated pattern indicates that changes in employment have been accompanied by differences in the pace of adjustment across population groups. This finding is used as one supporting claim in the broader discussion of labour-market performance and does not constitute a separate estimate of population size or employment totals.\n\nThe results should therefore be read as analytical measures derived from the Integrated Labour Force Census, with the average and trend serving to clarify the direction of change. They are not presented as direct quotations from the source and should not be interpreted as replacing the source observations.\n\nMention: \"Integrated Labour Force Census\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the Integrated Labour Force Census to derive an average and a trend supporting a claim about labour-market participation.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:94465c6b27f64be23fa01f3d", "occurrence_id": "3ae8ac49587f08c7156b99c5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Monthly Wage Structure Index as an indicator produced by Oxford Economics. It is treated as a supporting statistical source for the discussion of wage conditions and is cited without adjustment to its published measurement. The source is retained in the evidence record so that the reported values remain traceable within the assessment framework.\n\nTwo figures are carried over directly from the indicator, namely 112.4 and 118.7. Their inclusion supports the broader finding that wage movements warrant continued attention in the review of economic conditions. No additional transformation is applied to these values, and they are presented as reported rather than recalculated for the purposes of this assessment.\n\nThe evidence is used narrowly to substantiate the relevant statement and does not serve as the sole basis for the wider interpretation. Taken together with the surrounding analysis, the figures provide a concise quantitative reference for the observed movement in wage structure. The citation therefore supports the paragraph while preserving a clear distinction between the reported measurements and the assessment drawn from them.\n\nMention: \"Monthly Wage Structure Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The assessment takes the figures 112.4 and 118.7 directly from the Monthly Wage Structure Index.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:b4089e02135a0079404bf47b", "occurrence_id": "ed57065f26085702dda7c493", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Provincial Drought and Resilience Dataset, a geospatial source produced by UNHCR, IOM, was used as the principal empirical input for the assessment. Its records were reviewed for internal consistency and arranged to support comparison across the observed reporting sequence. The dataset is treated as evidence for measurement, rather than as a narrative description of drought conditions.\n\nThe analysis derives average exposure levels and a direction of change from the underlying observations. These calculations provide a smoothed account of variation and reduce the influence of isolated entries. The resulting measures are used to identify whether reported conditions show sustained deterioration, relative stability, or gradual improvement. No individual record is presented as a standalone estimate of the overall position.\n\nQuality checks examined the completeness of the observations, the comparability of categories, and the presence of values that could distort the calculated measures. Where coverage was uneven, the derived averages were interpreted cautiously and used alongside the broader pattern. The findings therefore reflect a summary constructed from the dataset, with emphasis on the trend and average generated through analysis rather than on figures reproduced directly from the source.\n\nMention: \"Provincial Drought and Resilience Dataset\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives comparative averages and a direction of change from the Provincial Drought and Resilience Dataset.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:1cc906d6f8855722f13f21f0", "occurrence_id": "b238e8af2d4cfcfd7fe92f42", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: For the 2020 reference, the Diagnostic Terrain and Elevation Collection (DTEC) is used as an other source in the supporting evidence. The collection provides an independent basis for reviewing the magnitude of the estimate reported in the accompanying analysis, while remaining separate from the series from which that estimate was obtained.\n\nThe estimate is cross-checked against the DTEC material to assess whether the reported pattern is consistent with an alternative measurement basis. This comparison is applied to the claim concerning the observed level and direction of change. It does not replace the primary estimate or extend the analysis beyond the period under review.\n\nThe comparison indicates broad agreement sufficient to retain the claim in the main assessment. Differences between the two sources are treated as a qualification on precision rather than as evidence of a contradictory result. The DTEC reference is therefore used as corroborating evidence within the paragraph, with the principal interpretation continuing to rely on the separate underlying series. No additional adjustment is made solely on the basis of this cross-check.\n\nMention: \"Diagnostic Terrain and Elevation Collection (DTEC)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the 2020 Diagnostic Terrain and Elevation Collection (DTEC) as an other source to cross-check and corroborate an estimate from a separate series.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:a45d88bfa46dea85200fb085", "occurrence_id": "668af63b687817af68c62514", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Harmonised Labour Inspection System is used as an administrative source for 2015 and provides the central evidence for this assessment. Its entries are treated as reported administrative records rather than as estimates generated through a separate analytical exercise. The source is cited in the form in which the relevant observations are recorded, with no adjustment to the reported counts.\n\nThe record shows 31 completed inspections and 74 recorded non-compliances. These two figures are taken directly from the Harmonised Labour Inspection System and are reproduced to establish the scale of activity and the corresponding number of identified cases. They are presented as separate counts because they describe related, but not identical, aspects of inspection performance.\n\nThe figures should therefore be read as a limited measure of recorded enforcement activity. The number of completed inspections indicates the volume of work registered during the reference period, while the number of recorded non-compliances indicates the cases identified through that work. The comparison does not imply that every inspection produced one case or that the observations represent all underlying conditions.\n\nFor reporting purposes, the source supports a transparent description of the observed records. Any interpretation beyond these two figures would require additional evidence, including information on coverage, inspection selection, and the classification of non-compliances. In the absence of such evidence, the reported counts are retained without extrapolation.\n\nMention: \"Harmonised Labour Inspection System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The author takes the directly reported figures of 31 completed inspections and 74 recorded non-compliances from the 2015 Harmonised Labour Inspection System.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:c2533a4a94e204c257549770", "occurrence_id": "73d14fef3546803980d79376", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Enumerative Mixed Movement Survey, a survey covering Dominican Rep., is used as the principal evidentiary basis for the assessment of reported movement patterns. Its observations are treated as descriptive measures of the population reached by the enumeration exercise and are not extended beyond the coverage indicated in the source. The material is cited in the text to support the estimates presented below.\n\nTwo figures are reproduced directly from this source. It records that 41 percent of surveyed persons reported movement within the period covered by the enumeration. It also records that 27 percent identified a return movement as their most recent transition. These values are presented without recalculation, adjustment, or aggregation from other material, and they constitute the central quantitative evidence for the assessment.\n\nThe difference between the two reported shares is noted as an indication of the composition of observed movement, rather than as a measure of change over time. Interpretation remains limited to the persons and locations represented in the survey. The figures therefore support a cautious reading of movement profiles and provide a consistent basis for the associated narrative findings. Any comparison with administrative records would require evidence not contained in the cited source.\n\nMention: \"Enumerative Mixed Movement Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported figures of 41 percent for movement and 27 percent for return movement from the Enumerative Mixed Movement Survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:177cfbbbe2b2a1a9ee3d4368", "occurrence_id": "a8bc7d64421d420d19964a45", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Panel Living Standards Survey is the principal empirical source for this assessment and is a survey covering Guatemala. Its observations provide the basis for examining changes in living conditions across the records available for analysis. The source is treated as a consistent longitudinal input rather than as a collection of isolated headline figures.\n\nThe analysis calculates average levels across relevant household observations and compares those averages over successive observation points. This procedure yields a measured trend, with changes assessed from the underlying distribution of records rather than inferred from a single reported value. The resulting estimates are used to describe the direction and relative pace of change in welfare conditions.\n\nSeveral checks are applied before aggregation, including review of coverage, removal of unusable records, and alignment of comparable observations. These steps are intended to limit the influence of irregular responses and maintain comparability within the analytical series. The estimates should therefore be interpreted as derived measures based on the survey evidence, not as figures reproduced directly from the source.\n\nThe calculated averages and trend inform the assessment of distributional movement and support the interpretation of changes in household welfare. They are presented as analytical results generated from the Panel Living Standards Survey and should not be read as independent administrative counts.\n\nMention: \"Panel Living Standards Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The assessment calculates comparable averages and a welfare trend from the Panel Living Standards Survey covering Guatemala.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:ee421bad243049c0fb3a9ee2", "occurrence_id": "90780a4fd347196ff8e04f9e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Stratified Graduation Programmes Registry (SGPR) database as its principal source for the summary presented below. The database provides a consolidated record of programme characteristics and reported outcomes, with entries treated as the reference point for the accompanying statistical note.\n\nTwo figures are reproduced directly from the registry. It records 31 programme entries and reports 68 participant records within the material reviewed. These values are presented without adjustment or extrapolation and are used to anchor the description of coverage and reported participation.\n\nThe figures should be read as registry counts rather than as estimates generated through the assessment. They support a limited descriptive interpretation of the documented programme portfolio and provide a transparent basis for comparison with the narrative findings. No inference is made beyond the information recorded in the SGPR database.\n\nMention: \"Stratified Graduation Programmes Registry (SGPR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the figures 31 programme entries and 68 participant records directly from the SGPR database.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:f8a426feac69e7c4b379bfd8", "occurrence_id": "451774f39c3ab84e8e9ea3c5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Midline Service Delivery and Readiness Index indicator, produced by Geostat, was incorporated into the assessment evidence base. It provides a structured measure of service readiness and supports the interpretation of differences in facility preparedness across the reporting framework. The source is treated as an input to the analysis rather than as a substitute for direct verification.\n\nAvailable records were collected and reviewed for completeness before being harmonised with the assessment’s common definitions. Inconsistent labels were aligned, duplicate entries were resolved where documentation permitted, and the resulting files were archived with the supporting material. These procedures improved the comparability of observations and reduced the risk that differences in recording practice would be interpreted as differences in readiness.\n\nOn this basis, the indicator supports the finding that service preparedness is uneven across the assessed units. The finding is presented with appropriate caution because the index reflects recorded readiness dimensions and may not capture all operational constraints. The archived material remains available for subsequent quality checks and for reconciliation with related assessment evidence.\n\nMention: \"Midline Service Delivery and Readiness Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The passage records the collection, harmonisation, and archiving of the Midline Service Delivery and Readiness Index indicator to support a service-readiness finding.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:d95d28f9ec1aa5426429030e", "occurrence_id": "8715c349e783e17ee8280eba", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Recurrent Cash Transfer Registry Survey is the principal reference for the beneficiary and payment observations used in this assessment. It is a 2016 survey and is treated as the direct source for the two quantitative statements reported below. No adjustment, interpolation, or subsequent classification is applied to the cited values.\n\nThe survey records that 31 percent of registered households received a transfer during the reference period. It also reports that 68 percent of registered households had a complete payment history. These figures are reproduced as reported and provide the central evidence for assessing the consistency of registry coverage and payment recording.\n\nThe difference between the two reported shares indicates that the presence of a household in the registry did not uniformly correspond to a complete payment record. This finding is relevant to the interpretation of beneficiary counts and should be considered when registry totals are compared with operational outputs. The evidence supports a cautious reading of apparent coverage levels, particularly where payment history is used as an indicator of effective access. The survey figures are therefore retained in their original form for traceability and are not presented as a constructed estimate.\n\nMention: \"Recurrent Cash Transfer Registry Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Recurrent Cash Transfer Registry Survey’s reported figures of 31 percent and 68 percent to assess registry coverage and payment-record completeness.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:97835274f2dde2f244cf93a2", "occurrence_id": "83a125f5412ee64ffd02b56c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Midline Household Consumption Survey (MHCS) is the principal survey used for the consumption analysis, with 2024 identifying the year to which its observations refer. The source provides the underlying household-level measures required for a consistent assessment of consumption conditions.\n\nThe analysis uses these observations to calculate average consumption levels and to identify the direction and pace of change across the observation period. Results therefore represent derived summaries of the survey evidence rather than isolated figures reproduced from the source. The calculated averages are used to establish the central tendency of consumption and to reduce the influence of individual observations.\n\nThis treatment supports comparison across analytical groups and provides a stable basis for interpreting changes in welfare. The resulting trend is reported as an analytical estimate generated from the MHCS data, with attention to internal consistency in the calculation of averages and changes. No additional source is required for the estimates presented in this section.\n\nMention: \"Midline Household Consumption Survey (MHCS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses the 2024 MHCS survey to calculate average consumption levels and derive a trend from its household observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:e0214e82861719224e6dc63d", "occurrence_id": "283f0df47d962d6f7ddeef81", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis draws on the Small-Area Referral Pathways Registry (SRPR), a database referring to 2022. The source provides structured observations on referral movement, completion status, and recorded delays. Its coverage supports comparison across the pathway categories used in the assessment, while preserving a consistent basis for aggregation.\n\nThe reported results are derived measures rather than reproduced entries from the registry. Mean referral duration is calculated from the available observations, and the distribution of completed pathways is aggregated to identify the direction and scale of change across successive stages. These calculations reduce the influence of individual records and provide a summary measure suitable for implementation monitoring.\n\nThe resulting trend is used to assess whether referral performance is becoming more timely and whether completion remains concentrated within particular pathway groupings. The average is also used as a reference point for interpreting variation in the reported indicators. Where the number of observations is limited, the findings are treated as indicative and are not extended beyond the coverage represented in the SRPR database.\n\nThe analysis therefore uses the registry as the central empirical basis for estimating aggregate performance. The presentation retains the underlying distinction between observed records and measures calculated from them, allowing the assessment to report a comparable trend and average without implying that either measure appears directly in the source.\n\nMention: \"Small-Area Referral Pathways Registry (SRPR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The author calculates an average and a referral-performance trend from the 2022 SRPR database.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:d8ac2e6407ff7613cc936150", "occurrence_id": "b217405a90784654d319bac1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Recurrent Return Intention Survey is used as the reference survey for the assessment of reported return preferences in Somalia. The source covers Somalia and refers to 2017. Its results are treated as descriptive evidence for the population and period addressed, rather than as a measure of subsequent movement or verified return.\n\nThe broader assessment indicates that return intentions were divided across respondents. Two figures are reproduced directly from the survey. The first indicates that 62 percent of respondents expressed an intention to return, while the second records that 38 percent did not express such an intention. Taken together, these figures support the assessment that interest in return was material but not shared by all respondents.\n\nThe figures are presented without further calculation and are retained in their reported form. They therefore provide a concise basis for interpreting the distribution of stated intentions, while leaving unresolved the timing, feasibility and outcome of any prospective return. The evidence should consequently be read as an account of reported preferences, with the survey serving to substantiate the specific claim concerning the balance between expressed intention and its absence.\n\nMention: \"Recurrent Return Intention Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported figures of 62 percent expressing an intention to return and 38 percent not expressing such an intention from the Recurrent Return Intention Survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:c59a870323d417eea39f9dc9", "occurrence_id": "869296a0906d6a561ad2a4f7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Coordinated Vocational Training Registry (CVTR) database provides the underlying observations for the analysis of participation and completion patterns. It refers to 2016 and is identified as a source produced by PISA, the organisation responsible for maintaining the registry. The source is used to support the assessment of whether observed changes are consistent across reporting categories.\n\nThe results presented in this section are derived through calculation rather than reproduced as individual registry entries. Specifically, the available observations are combined to obtain an average level and to estimate a trend over the relevant reporting sequence. This treatment reduces the influence of isolated values and provides a more stable basis for interpreting variation in vocational training activity. The resulting measure is used as one supporting element in the broader assessment of system performance.\n\nThe calculated pattern indicates a gradual change in the reported activity, although the direction and magnitude should not be interpreted as a direct statement about every participating unit. Differences in reporting completeness may affect the aggregate result, and the database should therefore be read together with the methodological qualifications accompanying the analysis. On this basis, the source supports the claim that participation changed over time, while the precise contribution of individual observations remains outside the scope of this assessment.\n\nMention: \"Coordinated Vocational Training Registry (CVTR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The passage identifies the 2016 Coordinated Vocational Training Registry (CVTR) database produced by PISA and explains that an average and trend were calculated from its observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:89f126e0fa01a42b9bc940dd", "occurrence_id": "db13e03a0d322b6ed3bb1148", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Baseline Rural Livelihoods System is treated as an administrative source covering Latvia and produced by Badan Pusat Statistik or BPS. Its records provide a fixed reference for describing household conditions and for checking the internal consistency of the accompanying assessment. No adjustment has been applied to the reported observations in the extracts used here.\n\nThe source records 31 households within the observed livelihood profile. This figure is reproduced directly and is used to indicate the scale of the population represented in the assessment. The same record reports that 18 percent of households received livelihood support during the observation period. That proportion is also reproduced directly, without recalculation or extension to households outside the source coverage.\n\nTaken together, these figures establish the narrow evidentiary basis for the paragraph. The household count identifies the number represented, while the reported proportion indicates the recorded reach of support. They are presented as source values rather than as estimates derived from comparison, and no additional interpretation is assigned to them. The evidence therefore supports a descriptive statement about the recorded baseline only, with conclusions limited to the population and measures contained in the cited system.\n\nMention: \"Baseline Rural Livelihoods System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Baseline Rural Livelihoods System figures of 31 households and 18 percent receiving livelihood support.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:56607e9fb6e340b4a20dddf3", "occurrence_id": "ce5fd0c5a3aed4a20a2af670", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Household Living Standards Survey (HLSS) is treated as the principal survey for assessing changes in household welfare. Its observations provide the basis for the analytical series used in this assessment, rather than serving as isolated reported figures. The analysis draws on reported household conditions, access to basic services, and selected expenditure measures to establish a consistent basis for comparison.\n\nFor each measure, the available observations are combined to calculate average levels and to identify the direction and pace of change across successive observations. This approach reduces the influence of short-term variation and permits the assessment to distinguish persistent movement from isolated departures. The resulting trend estimates are used as the central evidence for interpreting changes in living standards.\n\nThe calculations also examine whether the observed movement is broadly shared across household groups. Differences in average outcomes are considered alongside the aggregate trend, with attention to the distribution of gains and shortfalls. No single observation is treated as representative on its own. The assessment therefore relies on the pattern generated from the HLSS and reports the derived averages and trends as analytical results. This treatment supports measured conclusions while avoiding claims that cannot be sustained by the available observations.\n\nMention: \"Household Living Standards Survey (HLSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the HLSS to calculate average welfare measures and assess trends across successive observations and household groups.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:acd7391e597a99d31c3b6b0a", "occurrence_id": "81c004a5a771f7704036d83a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The administrative source used for the displacement analysis is the Verified Displacement Tracking System (VDTS), produced by GGDC. It provides a structured record of reported movements and related observations. The source is cited here to support the assessment of changing displacement conditions, while the surrounding analysis draws on the wider evidence assembled for implementation monitoring.\n\nRecords from VDTS were collected through the reporting process, checked for internal consistency, harmonised across reporting units, and archived for controlled reuse. The harmonisation preserved the distinctions required for aggregation while reducing duplication and variation in naming. Missing or inconsistent entries were retained in the archive with their status documented, rather than replaced by inferred values. This treatment supports a reproducible account of the observations used in the analysis.\n\nOn this basis, the assessment finds that displacement was uneven across the observed reporting categories and that the recorded burden was concentrated in a subset of them. The conclusion is supported by the archived VDTS records and should be read as a description of reported observations, not as a complete estimate of all movements. The source therefore provides an administrative basis for the claim while leaving interpretation subject to the coverage and recording practices documented during compilation.\n\nMention: \"Verified Displacement Tracking System (VDTS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The passage records GGDC’s administrative VDTS and the collection, harmonisation, and archiving of its records to support an assessment finding.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:176b3950e015e5561f5c9f8c", "occurrence_id": "74062c1b1e25721579b0bd3d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Annual Hiring and Vacancies Survey is the central data source for this assessment. It is a survey covering 2018 and is used to establish the distribution of hiring activity and unfilled positions across the observations available for analysis. The source is retained as the principal empirical basis for the indicators reported below.\n\nThe analysis does not reproduce isolated observations as reported figures. Instead, it consolidates the underlying records and calculates an average level together with a trend across the relevant categories. This approach reduces the influence of individual fluctuations and provides a consistent basis for assessing the direction and scale of labour demand. The resulting measures are used in preference to a simple transcription of values from the survey.\n\nThe calculated indicators support interpretation of changes in recruitment pressure and vacancy incidence within the project monitoring framework. They should be read as derived measures, rather than as direct quotations from the source. Any differences among categories reflect the aggregation procedure and the distribution of observations in the survey. The estimates are therefore suitable for comparison within this assessment, while remaining dependent on the coverage and internal consistency of the underlying records.\n\nMention: \"Annual Hiring and Vacancies Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage derives a trend and an average level from the Annual Hiring and Vacancies Survey covering 2018.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:299acc61f7a659fafee84a22", "occurrence_id": "7c4cd82324da6a54e811fa11", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Longitudinal Maternal and Newborn Survey is used as the principal empirical basis for this assessment. It is treated as a 2019 survey of maternal and newborn outcomes, and its observations provide the sequence required for comparison across the analytical period. The source is retained in its stated form, with no adjustment to its title or attribution.\n\nRather than reproducing individual observations, the analysis aggregates the available measures and derives an average level for each relevant interval. These interval means are then compared to estimate the direction and approximate pace of change. The resulting series is used to distinguish persistent movement from isolated variation, while preserving the common basis supplied by the survey.\n\nResults are presented as derived statistics, not as direct quotations from the underlying records. The trend measure and the pooled average support the assessment of change in maternal and newborn outcomes and inform interpretation of implementation performance. Because the calculations are based on the same longitudinal evidence, comparisons remain internally consistent, although they should be read as analytical summaries of the source rather than as additional observations.\n\nMention: \"Longitudinal Maternal and Newborn Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives interval means, a pooled average, and a trend from the Longitudinal Maternal and Newborn Survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:f0ff251b3fc078063904e200", "occurrence_id": "95893c5f13cfdcb566299297", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Biennial Service Delivery and Readiness Survey (BSDRS) is the principal survey used for the service-readiness observations reported in this assessment. The source refers to 2016 and provides a standardized basis for describing conditions across the observed service units. Its results are used here as reported, without adjustment or extrapolation.\n\nTwo figures are taken directly from the source. The survey records a readiness rate of 74 percent for the core service package and 62 percent for the supporting inputs required for routine delivery. These values are presented as separate reported measures and should not be interpreted as a calculated change, composite index, or estimate derived from other observations.\n\nThe figures provide a concise indication of the gap between nominal availability and the presence of the inputs needed for effective operation. The difference between the two reported rates is retained only as a descriptive contrast; no additional trend is inferred. Interpretation should therefore remain limited to the conditions measured by the BSDRS in 2016 and to the definitions applied in that survey.\n\nMention: \"Biennial Service Delivery and Readiness Survey (BSDRS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The author uses the BSDRS survey for 2016 and reports its two direct figures of 74 percent and 62 percent.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:f9a9cd89edd390b270986cc9", "occurrence_id": "1b6f1531f14e5b1e10a90032", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Annual Working Conditions Survey is used as the relevant survey for Tigray Region and refers to 2023. It provides a reference point for the presentation of selected working-conditions indicators in this assessment. The source is retained as a descriptive input and is not used to establish a broader time series or to support causal interpretation.\n\nTwo figures are quoted directly from the source. The first records that 47 percent of respondents reported access to paid leave, while the second records that 62 percent reported receiving written information on employment terms. These values are reproduced as reported and are not recalculated, weighted, or combined with other observations. Their inclusion is intended to preserve the numerical content of the source within the supporting evidence for the assessment.\n\nThe figures should therefore be read as survey results for the stated coverage and reference period. They describe reported conditions among the surveyed population and should not be interpreted as estimates for areas or periods outside that coverage. Any comparison with subsequent monitoring should use the same definitions and population basis. The source is cited here for transparency and traceability of the two reported values, while the main discussion remains focused on implementation status and the identification of operational follow-up needs.\n\nMention: \"Annual Working Conditions Survey\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"source\", \"summary\": \"The passage reproduces two figures from the Annual Working Conditions Survey, namely 47 percent and 62 percent, for Tigray Region in 2023.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:235022f45e6c920be454c5f0", "occurrence_id": "27a0334247b06e56fde527fd", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Biennial Labour Inspection Survey is used as the principal reference for this assessment. It is a survey covering Ghana and referring to 2016. The source provides the central quantitative basis for the observations presented in this annex.\n\nTwo reported counts are reproduced directly from the source without adjustment. The survey records 31 completed inspections and 47 identified non-compliance cases. These figures are presented as reported values and are not recalculated, extrapolated, or combined with other administrative material.\n\nThe first count indicates the observed scale of inspection activity during the reference period. The second records the number of cases in which non-compliance was identified. Read together, the figures provide a limited but direct indication of enforcement activity and the incidence of detected violations. They should therefore be interpreted as descriptive measures of the coverage represented in the source, rather than as estimates of total activity or prevalence.\n\nFor purposes of this assessment, the two values are retained as the principal evidence associated with the cited source. No adjustment is applied for possible differences in reporting practice, case classification, or inspection coverage. The resulting interpretation remains confined to the information documented in the survey and to the reference period specified above.\n\nMention: \"Biennial Labour Inspection Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes 31 completed inspections and 47 identified non-compliance cases directly from the Biennial Labour Inspection Survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:08f9c43fcac67ce54ccab5a4", "occurrence_id": "f7eec0fec6f30af524de7266", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Ministerial Maternal and Newborn System, an administrative source maintained for routine monitoring of maternal and newborn services. The source is treated as the primary evidentiary basis for the observations presented in this section, and its entries are cited in their reported form.\n\nTwo figures are reproduced directly from the system without recalculation or adjustment. It records 68 percent of reported deliveries as having received the specified service, while 1,240 maternal and newborn contacts are recorded for the same reporting period. These values are presented as reported measures and are not used to derive a rate, comparison, or time series.\n\nThe figures provide a limited but direct indication of service coverage and recorded workload. Interpretation remains subject to the completeness and consistency of routine reporting, particularly where records may not capture all service contacts. No supplementary source is introduced for the estimates cited here, and the assessment therefore treats the two reported values as descriptive evidence rather than as a basis for broader population inference.\n\nMention: \"Ministerial Maternal and Newborn System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes 68 percent and 1,240 maternal and newborn contacts directly from the Ministerial Maternal and Newborn System.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:27238a3f2c469abbc954b4d8", "occurrence_id": "c7cb2ea7056df1c481528a12", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Data sources.** The Harmonised Coastal Erosion Collection (HCEC) is classified as an other source produced by IMSS. It is used as the principal empirical input for the assessment of coastal change, with the extraction restricted to the observations required for the analytical series. No supplementary institutional or geographic identifiers are introduced in this note, in order to preserve a clear link between the calculations and the cited collection.\n\nThe reported indicators are derived from the underlying observations rather than reproduced as isolated values. In particular, the analysis calculates average erosion levels and evaluates the direction and consistency of change across successive observations. This treatment reduces the influence of individual entries and provides a more stable basis for comparison within the assessment framework.\n\nThe resulting measures are used to support interpretation of exposure, persistence, and variation in erosion conditions. Differences between calculated averages are considered together with the estimated trend, while unusually high or low observations are retained for quality review rather than removed solely because they affect the summary measure. The collection therefore serves as the central evidence for the quantitative assessment, and the reported findings should be understood as analytical results generated from it.\n\nMention: \"Harmonised Coastal Erosion Collection (HCEC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives average erosion levels and trend measures from the Harmonised Coastal Erosion Collection rather than reproducing individual observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:400c44cc20c07d984a45efa9", "occurrence_id": "0832806d15b4b72242776b92", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The National Parcel Footprint Dataset is used as the principal reference for the parcel-level evidence presented in this assessment. It is a geospatial source referring to 2018 and is treated as the baseline record for describing the observed footprint. No adjustment, interpolation, or subsequent classification is applied to the figures reproduced below.\n\nTwo figures are taken directly from the source. The dataset reports 18,642 parcels and an associated footprint of 7,416 hectares. These values are reproduced as stated because they provide the central quantitative basis for the assessment of parcel coverage. They should therefore be read as reported totals rather than as estimates generated through the present analysis.\n\nThe figures support a bounded description of the recorded parcel structure during the reference year. The parcel count indicates the number of mapped units represented in the source, while the area figure indicates the corresponding extent reported for those units. Together, they establish the scale of the evidence used in this annex and provide a consistent basis for subsequent interpretation. Any comparison with later administrative material would require separate validation and is outside the scope of this passage.\n\nMention: \"National Parcel Footprint Dataset\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported figures of 18,642 parcels and 7,416 hectares from the 2018 geospatial National Parcel Footprint Dataset.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:13dd02b33b4631fb3d82aac8", "occurrence_id": "769e6899af952da82b90d414", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis uses the administrative Harmonised Business Demography System as the principal reference for the assessment of business population movements. Its records provide a consistent basis for reviewing the number of active units, new entries, exits, and continuing units observed in the reporting framework. The source is treated as the central evidence for the descriptive results rather than as supplementary background material.\n\nThe recorded counts and derived rates are compared with an estimate obtained from a separate statistical source. This comparison is intended to identify material differences in level, direction, and coverage, while avoiding the presumption that either series is error free. Where the two measures are broadly aligned, the administrative evidence strengthens confidence in the estimate. Where they diverge, the discrepancy is retained for review and is considered in interpreting the reported result.\n\nThe cross-check also supports a more cautious reading of changes in business activity. Differences may reflect registration practices, reporting thresholds, timing, or the treatment of units that are inactive but remain recorded. Accordingly, the Harmonised Business Demography System is used to validate the broad pattern and to test the plausibility of the independent estimate, while the estimate remains separately identified in the assessment. The resulting comparison provides a transparent basis for judging the robustness of the findings and for documenting limitations in the available evidence.\n\nMention: \"Harmonised Business Demography System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the administrative Harmonised Business Demography System for recorded counts and rates and to cross-check an independent estimate.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:e47f50c95f647aecb34622bb", "occurrence_id": "6eb9be4c020f95ae68a18489", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Multi-Topic Vocational Training System, an administrative source maintained for routine monitoring of vocational training activity. The source provides a consistent body of observations on participation and provision, allowing the analysis to focus on movement over time rather than on isolated reported values. Its use is appropriate for establishing broad patterns, while the absence of contextual identifiers limits interpretation of differences across reporting periods.\n\nThe values were reviewed in sequence and converted into a simple trend measure, with an average calculated to indicate the prevailing level of activity. This treatment reduces the influence of short-term fluctuations and supports comparison between periods without presenting individual entries as independently verified estimates. The resulting measures are used as the central evidence for assessing whether participation expanded, remained stable, or declined.\n\nThe findings should be read as indicative statistical results derived from recorded administrative observations. They describe the direction and general magnitude of change captured by the system, but they do not establish the causes of that change. Potential differences in registration practice, reporting completeness, or classification may affect the measured pattern. Accordingly, the trend and average are presented as monitoring indicators for implementation review and as a basis for identifying issues requiring further examination.\n\nMention: \"Multi-Topic Vocational Training System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the Multi-Topic Vocational Training System administrative source to derive a trend and an average across reporting periods.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:11ccf779369e5aa71f076782", "occurrence_id": "e47755cd7148bcbd48487a1f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Rural credit access is treated in this assessment as one element of the broader conditions affecting productive activity and household resilience. The evidence indicates that access is not uniform across rural communities, and that administrative capacity, distance, and the structure of local demand may influence the reach of financial services. This observation is used to frame the assessment rather than to establish a quantified result.\n\nThe Cross-Sectoral Rural Credit Dataset, a geospatial source produced by the Development Strategy Center and referring to 2022, is cited in support of this general observation. Its inclusion provides contextual grounding for the statement that rural credit conditions differ across localities. No values, rates, counts, or other numerical entries from the dataset are incorporated into the analysis, and no calculation or comparison is derived from it.\n\nThe assessment consequently relies on the cited source for background relevance within the discussion of uneven access, while the substantive findings remain based on the evidence presented elsewhere in the report. This treatment limits the citation to one supporting claim and avoids attributing precision to information that has not been directly analysed. The distinction is maintained to preserve a clear separation between contextual reference and reported results.\n\nMention: \"Cross-Sectoral Rural Credit Dataset\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The passage uses the 2022 geospatial Cross-Sectoral Rural Credit Dataset produced by the Development Strategy Center only to support a general claim about uneven rural credit access, without using its values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:4fd49af50cff49399220b991", "occurrence_id": "eaa231806a0a2cea2a26128d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Joint Crop and Livestock Survey (JCLS) provides the principal survey basis for the assessment of agricultural conditions in Syria. Its observations constitute the central empirical input for reviewing production patterns, cultivated activity, and livestock conditions. The source is used as a consistent basis for comparison across the observations available in the assessment.\n\nThe analysis does not reproduce isolated entries from the survey. Instead, it combines the reported quantities to calculate average levels and to derive the direction and pace of change across successive observations. These calculations support a measured interpretation of movement in agricultural activity and reduce the risk that an isolated high or low observation is treated as representative. The resulting averages and trends are presented as analytical measures derived from the JCLS material.\n\nThe findings should therefore be read as estimates produced through aggregation of the survey evidence rather than as direct transcription of individual records. Differences in coverage, reporting conditions, and the composition of observed units may affect comparability, and the calculated results are used primarily to identify broad tendencies. On this basis, the JCLS serves both as the cited source and as the underlying evidence from which the assessment derives its summary measures.\n\nMention: \"Joint Crop and Livestock Survey (JCLS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage describes calculated averages and trends derived from the Joint Crop and Livestock Survey (JCLS) for Syria.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:260ac985fd7303ad8d3a749c", "occurrence_id": "96f801486636a93f7e973b58", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Independent Wage Structure Survey (IWSS), a survey, is used as the principal empirical source for the assessment of wage structure. Its observations provide the basis for examining the distribution of reported pay across occupational categories and for identifying changes in the relative position of those categories. The source is treated as an analytical dataset rather than as a set of isolated figures.\n\nThe analysis aggregates the observations into comparable measures and calculates average wage levels for the categories under review. It then derives a trend from those averages, allowing the assessment to distinguish broad movement from short-term variation in individual records. This approach reduces the influence of exceptional observations and supports a consistent comparison across the wage structure.\n\nResults based on the IWSS are presented in summary form and are interpreted alongside the stated methodological limitations. The derived averages are used to assess dispersion, relative ranking, and the direction of change, while individual values are not treated as representative on their own. The survey therefore functions as the central evidence for the statistical assessment, with conclusions based on calculated patterns rather than direct reproduction of selected observations.\n\nMention: \"Independent Wage Structure Survey (IWSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the IWSS to calculate average wage levels and derive a trend from the survey observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:e2df014fa6d34e2c9ce23717", "occurrence_id": "c2e3f88872bb4ba64de537dd", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Multi-Topic Clinic Quality Index (MCQI) indicator for 2018 as a reference point for interpreting reported service conditions. The index is treated as a descriptive measure rather than as a target or a basis for estimating change. Two figures are reproduced directly from this source, with 68 percent recorded for the first measure and 74 percent for the second measure.\n\nThese values support the assessment’s broader finding that observed service performance was uneven across the measures reviewed. The lower figure indicates a material gap in one area of provision, while the higher figure suggests comparatively stronger performance in another. The contrast is used to qualify the narrative and to avoid presenting overall quality as uniform.\n\nThe figures are reported without additional adjustment, aggregation, or calculation. They are retained in their published form to preserve comparability with the underlying indicator and to permit verification against the cited source. Interpretation is therefore limited to the scope of the two measures and does not imply that the values represent a comprehensive assessment of all service dimensions. In this annex, the source provides supporting evidence for one claim within the wider discussion, rather than constituting the sole basis for the assessment conclusion.\n\nMention: \"Multi-Topic Clinic Quality Index (MCQI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage takes the 68 percent and 74 percent figures directly from the 2018 MCQI indicator to support a claim about uneven service performance.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:35c2412dff40a2cd65fd21fc", "occurrence_id": "0f2f177500f781ff201a9a95", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment relies on the Periodic Transport Network Census (PTNC), a census referring to 2021 and used here as the documentary basis for a limited statement on network condition. The source is retained in its reported form so that the supporting evidence remains traceable within the implementation record.\n\nWithin the broader discussion of maintenance performance, the paragraph reproduces directly from the PTNC two reported figures, namely 68 maintained links and 43 links requiring intervention. These observations support the claim that serviceability is uneven across the network, while avoiding any inference beyond the information contained in the census. The figures are presented as quoted results rather than as estimates generated for this assessment.\n\nThe evidence is therefore sufficient to substantiate the stated operational concern, but it does not by itself establish the causes of the observed condition. Interpretation remains limited to the scope of the census and to the categories reported there. Subsequent monitoring should preserve the same definitions when reviewing whether maintenance coverage has improved and whether the identified intervention requirement has been reduced.\n\nMention: \"Periodic Transport Network Census (PTNC)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the 2021 PTNC census and directly reproduces figures for 68 maintained links and 43 links requiring intervention.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:39ba724ec54e04381347b649", "occurrence_id": "ed7d675e54d826d985693ba9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the District Inequality Index as an indicator produced by IVAP. The source provides a structured basis for reviewing differences in observed conditions across administrative units and supports interpretation of the distributional evidence presented in this section. Its inclusion is limited to the verification of one substantive claim and does not replace the principal estimate used for the analysis.\n\nThe reported estimate is cross-checked against the District Inequality Index because the two measures arise from different sources and provide partially independent views of inequality. The comparison indicates that the direction of the observed disparity is consistent across the available evidence. This result strengthens the assessment of uneven outcomes while avoiding an inference that the measures are identical or directly interchangeable.\n\nThe cross-check is therefore treated as corroborative rather than as a basis for recalculation. Differences between the estimates may reflect distinctions in coverage, construction, or measurement procedures. The finding is retained at the level of the broader paragraph, where it supports the conclusion that inequality remains material and varies across the units considered. No additional weighting is applied to the external indicator, and the principal estimate remains the reference for the reported assessment.\n\nMention: \"District Inequality Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the District Inequality Index, an IVAP indicator, to cross-check an independent inequality estimate.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:0e5f513120f8efffff3be38f", "occurrence_id": "e7356240858892242dae8be2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Comparative Health and Household Dataset as a geospatial source covering Moldova. The material is applied to support the interpretation of differences in household access to basic health services across the areas represented in the analysis. Its geographic structure permits comparison of reported conditions without extending the findings beyond the population covered by the source.\n\nTwo figures are reproduced directly from the dataset. It records that 68 percent of households reported access to a regular health facility, while 42 percent reported that travel time constrained use of services. These values are retained as reported and are not recalculated in this assessment.\n\nTaken together, the figures support the broader finding that physical availability does not ensure effective access. The difference between reported facility access and reported travel constraints indicates that service coverage should be interpreted alongside geographic reach and household circumstances. This observation is used to qualify the assessment of service equity and to identify the need for more detailed local review where access appears uneven.\n\nThe cited material is therefore treated as supporting evidence for the paragraph rather than as a basis for a separate estimate. No inference is drawn about conditions outside Moldova, and the two reproduced values remain linked to the scope and coverage of the Comparative Health and Household Dataset.\n\nMention: \"Comparative Health and Household Dataset\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported figures of 68 percent and 42 percent from the Comparative Health and Household Dataset to support a claim about geographic access to health services in Moldova.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:11b1dfb3decfcf6da9c5f642", "occurrence_id": "19b7b5bbd30f78abb744ab1a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Provincial Rural Credit Registry (PRCR) database is used as the central empirical source for the assessment. Its records provide the basis for examining the distribution and evolution of rural credit observations across the reporting sequence. The source is cited to establish the evidentiary basis of the analysis, while the reported results are derived measures rather than direct reproductions of individual entries.\n\nThe observations are grouped into comparable reporting intervals, and an average is calculated for each interval. These averages are then arranged in sequence to identify the underlying trend, including the direction of movement and the degree of continuity between intervals. This approach reduces the influence of isolated observations and supports interpretation of the general pattern contained in the database.\n\nThe resulting estimates are presented as analytical measures generated from the Provincial Rural Credit Registry (PRCR), not as figures attributed to a single record. Differences between successive averages are reviewed to distinguish sustained movement from short-term variation. The assessment therefore relies on the PRCR database both for the underlying observations and for the construction of the summary trend used in the discussion. No inference is made beyond the pattern supported by the calculated averages.\n\nMention: \"Provincial Rural Credit Registry (PRCR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage explains that interval averages and a trend are calculated from the PRCR database rather than quoted as individual registry figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:799c114d806b8f83c85d98ad", "occurrence_id": "7a5e7fbfb45f6d7708160d9d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Monthly Labour Force Index indicator is used as the principal time-series input for the labour market assessment. It is treated as a recurring statistical measure rather than as a single reported observation. The analysis derives a moving average from the available monthly sequence and compares successive periods to identify direction and persistence. No isolated value is used to represent the overall position.\n\nFor analytical consistency, the calculation applies the same treatment to each observation in the sequence. Short-term fluctuations are therefore moderated, while sustained changes remain visible in the resulting measure. The estimated average is reviewed alongside the direction of movement, allowing the assessment to distinguish a temporary variation from a broader adjustment in labour market conditions. This approach also limits the risk that an unusually high or low observation will determine the interpretation.\n\nThe resulting trend supports the broader finding that labour market conditions changed gradually rather than through a single abrupt movement. The evidence is used to qualify the assessment of participation and employment pressures, not to present a standalone statistic. The calculated measure provides a concise basis for comparison across reporting periods and informs the discussion of implementation risks. It is consequently incorporated as supporting evidence within the wider assessment, with emphasis placed on the pattern generated from the series rather than on any individual observation.\n\nMention: \"Monthly Labour Force Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the Monthly Labour Force Index indicator to calculate a moving average and directional trend supporting the broader labour market assessment.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:c92cf3561528532997447ce0", "occurrence_id": "776c495fa5415b9a4c027868", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis uses the Panel Primary Care Survey (PPCS) as a survey source for Thailand. The material is treated as an observational reference for primary care access, service use, and the distribution of reported needs. Its coverage provides a basis for examining whether the estimate presented in the surrounding analysis is consistent with evidence reported through a separate source.\n\nThe estimate is cross-checked against the PPCS results at the level of the reported population and service indicators. This comparison is intended to identify material divergence rather than to combine observations into a new measure. Where the two sources point in the same direction, the finding is considered more robust for the purposes of the assessment.\n\nDifferences are retained as part of the quality review. They may reflect variation in definitions, respondent coverage, timing, or the construction of the underlying measures. The PPCS evidence therefore supports one claim concerning the scale and distribution of primary care use, while the principal estimate remains drawn from the other source.\n\nThe result should be read as corroborating evidence within the broader paragraph, not as a replacement for the main estimation procedure. No adjustment is applied solely on the basis of the comparison, and the survey material is used to document plausibility and indicate areas requiring cautious interpretation.\n\nMention: \"Panel Primary Care Survey (PPCS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the Panel Primary Care Survey (PPCS) survey for Thailand to cross-check a separate estimate of primary care use.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:708cefe7bd8c52d71996300d", "occurrence_id": "de29bb5fd354e0c4d485ae36", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The evidence base for this assessment is the Ministerial Skills and Employment Survey (MSES), a survey covering Sulaymaniya in 2018. The source provides the principal observations used to examine participation, skills, and employment conditions within the covered population. Its relevance is retained through the analysis rather than through reproduction of isolated reported figures.\n\nThe analysis computes average indicators across the reported groups and derives a comparative trend from the distribution of skills and employment outcomes. These calculations provide a more stable basis for interpretation than any single observation, particularly where reported values vary across categories. The resulting measures are used to identify broad differences in participation and skill attainment within Sulaymaniya.\n\nThe findings therefore reflect analytical use of the MSES evidence for 2018. The survey supports comparison of group-level outcomes and the calculation of summary measures, while the derived averages and trend provide the basis for the assessment narrative. No additional source is required for the interpretation presented in this section.\n\nMention: \"Ministerial Skills and Employment Survey (MSES)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses MSES to compute average employment indicators and a comparative skills trend for Sulaymaniya in 2018.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:49b07a06107e9ed7914028cb", "occurrence_id": "0c9137d100a40a6c4141ea08", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Follow-Up Land Cover Change Dataset, a geospatial source produced by ISTAT, as its principal empirical basis. The dataset provides a consistent body of observations for examining changes in land-cover conditions and supports the construction of comparable measures across the assessment period.\n\nThe analysis does not reproduce individual dataset entries as reported results. Instead, observations are grouped according to the analytical categories used in the assessment, after which average levels and changes are calculated. The resulting measures indicate the direction and approximate pace of change, while reducing the influence of isolated observations and recording variation across categories. This treatment provides a more stable basis for interpretation than reliance on selected entries.\n\nThe calculated trend is used to inform the assessment of overall land-cover dynamics and to identify categories requiring closer review. Average values are presented as derived indicators rather than as direct quotations from the dataset. Any limitations arising from classification differences, incomplete observations, or changes in recording practice are considered when interpreting the results. The Follow-Up Land Cover Change Dataset therefore serves as the central evidentiary input for the quantitative discussion, with conclusions based on calculations performed on its observations.\n\nMention: \"Follow-Up Land Cover Change Dataset\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives average land-cover levels and an overall change trend from the Follow-Up Land Cover Change Dataset.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:33984605e2fa80ec330e6874", "occurrence_id": "fb9e1afaaef240b8b8a7ace3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment examines the distribution of household access indicators and uses a limited set of observations to support the interpretation of service disparities. The evidence is intended to inform a broader finding on uneven conditions across the areas covered by the assessment.\n\nTwo reported values are taken directly from the Continuous Output Area Census, a census covering Panama in 2022. The recorded proportion is 47.8 percent, while the corresponding count is 12,640 households. These figures are reproduced without adjustment and are used as descriptive evidence rather than as an estimate derived from the assessment.\n\nThe results support the broader conclusion that access remains uneven among households. The proportion indicates the scale of the reported condition, and the accompanying count provides a direct indication of its observed population base. Interpretation is therefore limited to the information represented in the cited source, with no extrapolation beyond the reported coverage.\n\nThe two values are presented together to maintain consistency between the percentage measure and the associated household count. They provide a concise reference point for the assessment narrative and are retained in the form in which they appear in the source.\n\nMention: \"Continuous Output Area Census\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage takes 47.8 percent and 12,640 households directly from the Continuous Output Area Census covering Panama in 2022.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:bc324b1e3cb83cc807ab375e", "occurrence_id": "893c76a5226aa1e00bf10135", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Clustered Wage Structure System (CWSS) as its principal administrative source for examining wage structure. The material provides consistent observations across occupational groupings and supports comparison of remuneration patterns without relying on isolated reported figures. Its use permits the analysis to retain a common basis for the calculations presented below.\n\nThe results are derived by aggregating the available observations and calculating average wage levels for each grouping. Changes across the ordered observations are then assessed to identify the direction and regularity of the underlying trend. Individual values are therefore treated as inputs to a broader statistical calculation rather than reproduced as standalone findings. This approach reduces the influence of exceptional observations and provides a more stable basis for interpretation.\n\nThe resulting averages indicate the relative position of the groupings, while the calculated trend shows whether wage levels move consistently or display limited variation. The estimates are used to support discussion of distributional differences and to inform the assessment of implementation outcomes. All calculations reported in this section are based on the Clustered Wage Structure System (CWSS), with the administrative record serving as the central evidentiary basis for the analysis.\n\nMention: \"Clustered Wage Structure System (CWSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives average wage levels and a trend from the Clustered Wage Structure System (CWSS).\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:48ed580c9d9bd21931f888be", "occurrence_id": "830c813f8415a63e75aa1591", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Joint Targeting and Payments Report is used as the principal statistical source for 2018. The report provides the reference values applied in this assessment and supports a consistent reading of the recorded caseload structure. Its use is limited to the observations required for the present comparison.\n\nTwo figures are reproduced directly from the report. The first records 31 grades within the classification used for payment determination. The second records 68 percent for the corresponding share reported in the same material. These values are retained as stated, without recalculation or adjustment, and constitute the central evidence for the assessment presented here.\n\nThe figures are used to anchor the interpretation of coverage and classification performance. The count of 31 grades indicates the breadth of the applicable structure, while the 68 percent figure indicates the reported proportion associated with the payment measure. Taken together, they provide a concise basis for reviewing implementation status and for identifying areas requiring routine verification.\n\nNo additional estimates are derived from these observations. The assessment therefore treats the quoted values as documentary evidence and maintains their original units and wording for purposes of traceability.\n\nMention: \"Joint Targeting and Payments Report\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes 31 grades and 68 percent directly from the Joint Targeting and Payments Report for 2018.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:a6c4d50e6772c4b49260fcd6", "occurrence_id": "a70c8af5da8910f9bb377138", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Decennial Living Standards Microdata (DLSM), a microdata source referring to 2018. The records provide the central empirical basis for the welfare analysis and are used to examine variation across observed units. All references to the source in this assessment retain its stated name and acronym.\n\nThe analysis derives average welfare measures directly from the records rather than reproducing a statistic reported in the source. It also constructs a trend measure by ordering the relevant observations and comparing their calculated averages. This treatment supports a consistent reading of changes in the distribution and reduces reliance on isolated figures.\n\nThe resulting estimates are used in the assessment of coverage, disparities, and overall performance. They are presented as analytical outputs generated from the underlying observations, with interpretation limited to patterns supported by the calculated averages and trend. The DLSM material therefore functions as the principal evidence for the quantitative discussion, while the text distinguishes derived results from values that may appear in the source itself.\n\nMention: \"Decennial Living Standards Microdata (DLSM)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The assessment calculates average welfare measures and a trend indicator from the 2018 DLSM microdata.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:47246ca9efa8cc274624e23c", "occurrence_id": "28da1ee23ba73ed09aab4b74", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Baseline Teacher Workforce System, an administrative source produced by TFDD, as its principal evidence on the teacher workforce. The records are used to establish a consistent analytical series for staffing levels, recruitment flows, and workforce attrition. This treatment supports assessment of underlying movement in the workforce rather than reproduction of isolated reported figures.\n\nThe analysis calculates average staffing levels from the available records and derives a trend in workforce growth across successive reporting periods. The resulting measures are used to examine whether changes in recruitment and attrition are associated with a sustained expansion or contraction of the workforce. Estimates are interpreted as analytical summaries of the source records and not as additional entries in the system.\n\nQuality review considered the internal consistency of the reported observations and the comparability of the measures used in the calculations. The derived averages and trend provide the basis for implementation monitoring, while the limitations of the underlying records are taken into account when assessing changes in staffing. No inference is made beyond the information required for the workforce analysis.\n\nMention: \"Baseline Teacher Workforce System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives average staffing levels and a workforce growth trend from the Baseline Teacher Workforce System.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:a8d1146fb71872478a3eeafc", "occurrence_id": "317093d9d67101f5767404d1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment finds that apprenticeship participation is substantial but uneven across the recorded pathways. Differences in enrolment are relevant to implementation planning because they affect the distribution of training places, supervision requirements, and completion support. The finding is supported by two figures reported directly in the Independent Apprenticeship Registry (IAR) database.\n\nThe database records 1,842 active apprentices. It also reports that 57 percent of registered apprentices were enrolled in craft-based pathways. These figures are presented as recorded, without adjustment or further estimation, and are used to anchor the assessment of participation patterns. They indicate that the overall volume of activity coexists with a marked concentration in a limited group of pathways.\n\nThe evidence does not, by itself, establish the reasons for this concentration or its implications for completion outcomes. It nevertheless provides a consistent reference point for reviewing the adequacy of placement capacity and related support measures. Subsequent monitoring should retain the same definitions when comparing administrative counts, since changes in classification could affect interpretation even where the underlying activity remains stable. The registry figures are therefore used as a descriptive baseline for implementation review rather than as an independent measure of programme effectiveness.\n\nMention: \"Independent Apprenticeship Registry (IAR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the IAR database for the directly reported figures of 1,842 active apprentices and 57 percent enrolled in craft-based pathways.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:547581fe84ecd94ec33c6fe6", "occurrence_id": "dea1d8f2f1271384577d4ad0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The District Cash Transfer Registry Survey is the principal empirical source for this assessment of Tanzania. It is a survey produced by IVAP and is treated as the central evidence for describing changes in registry-based cash transfer coverage. The material is used for analytical estimation rather than for reproduction of isolated reported figures.\n\nThe analysis consolidates the observations into comparable reporting periods and calculates average coverage across those periods. It then derives the direction and approximate pace of change from the resulting series. This approach reduces the influence of short-term variation and provides a consistent basis for assessing whether recorded coverage is expanding, stable, or declining.\n\nThe calculated averages indicate the underlying level of registry coverage, while the estimated trend provides the principal measure of movement over time. These results are used to support the assessment of implementation performance and to identify changes requiring further review. Individual observations are therefore interpreted within the aggregate pattern, with attention to continuity and comparability across the survey record.\n\nThe findings should be read as estimates generated from the District Cash Transfer Registry Survey rather than as direct quotations from the underlying entries. The survey remains the sole basis for the quantitative comparison presented in this section, and the reported conclusions follow from the averages and trend measures calculated from its observations.\n\nMention: \"District Cash Transfer Registry Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives multi-period averages and a trend in cash transfer registry coverage from the District Cash Transfer Registry Survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:cfd6058b2ff36b3a05d5af23", "occurrence_id": "6efd9a5c92f94d4830564a10", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Departmental Land Cover Change Index (DLCCI) as the principal indicator for examining changes in land cover conditions. The source provides the central evidence for the analytical series, with observations interpreted as comparable measures of change rather than as isolated reporting entries.\n\nThe analysis derives a mean level from the available DLCCI observations and calculates a trend across the sequence. This treatment supports an assessment of the direction and consistency of change while reducing the influence of individual fluctuations. The resulting measures are used to inform implementation monitoring and to identify whether observed movements are broadly sustained.\n\nIndividual source values are therefore not reproduced as stand-alone findings. Instead, the calculated average and trend provide the basis for interpretation, with the underlying indicator retained as the reference for subsequent review. Any apparent variation is considered alongside the full sequence, allowing the assessment to distinguish persistent movement from short-term deviation.\n\nMention: \"Departmental Land Cover Change Index (DLCCI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the Departmental Land Cover Change Index (DLCCI) indicator to calculate an average and a trend for implementation assessment.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:c0fc6d3c1814ea1c9b5a396f", "occurrence_id": "edbb983f8c3b949d7424a5c1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The District Early Warning Registry (DEWR) database is used as the principal empirical source for the assessment. It refers to 2015 and provides a structured set of observations on early warning conditions across the reporting units. The entries are treated as comparable observations for analytical purposes, subject to the limitations inherent in administrative recording. No additional source is required to establish the descriptive baseline presented here.\n\nThe assessment does not reproduce selected registry entries as stand-alone findings. Instead, the observations are aggregated and processed to obtain an average level across the reporting units and to identify the direction of change within the recorded series. The resulting trend is used as the central evidence for interpreting the persistence or variation of reported conditions. This procedure gives greater weight to the distribution of observations than to any single reported value and reduces the risk that an isolated entry will determine the narrative.\n\nThe calculated measures are presented as analytical outputs derived from the DEWR database, rather than as figures copied directly from the source. Interpretation remains limited to the information contained in the registry, and the calculations are intended to support monitoring, prioritisation, and subsequent verification. In this use, the database serves both as the evidentiary base and as the basis for a transparent summary of observed patterns.\n\nMention: \"District Early Warning Registry (DEWR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The author uses the DEWR database for 2015 to calculate an average and identify a trend.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:5c14652e73b6b5b57806d630", "occurrence_id": "58d1899b20c80bef172f2bf3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Analytical Child Wellbeing Census (ACWC) is cited as a 2021 census providing general background for the assessment of child wellbeing. Its inclusion supports the broader observation that child wellbeing should be considered alongside service access, household resilience, and developmental outcomes when interpreting implementation conditions.\n\nThe reference is used to situate that observation within an established body of evidence rather than to provide measurements for the analysis. No numerical finding, indicator value, or estimate from the ACWC is used to establish the observation, and no result is reproduced from the census.\n\nThe citation therefore serves a limited evidentiary function within the wider discussion. It reinforces the relevance of child wellbeing as a cross-cutting consideration, while the conclusions presented in this section rest on the assessment framework and the evidence described elsewhere in the report. The ACWC is not used for comparison, aggregation, or calculation, and its inclusion does not alter the interpretation of the reported implementation findings.\n\nMention: \"Analytical Child Wellbeing Census (ACWC)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The ACWC provides contextual support for treating child wellbeing as a relevant cross-cutting consideration, without contributing any of its values to the analysis.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:ec8e14691cc70fbb13078a44", "occurrence_id": "d5148367cbfa34024abe9b2d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Administrative evidence is used to assess whether the estimated movement in unemployment is consistent with observed access to income support. The Audited Unemployment Insurance Registry (AUIR) is cited as a database containing records of submitted claims, accepted cases, and benefit payments. Its coverage of registered claimants provides a direct reference for interpreting the scale of the reported change, while the limitations of registration-based evidence are recognised.\n\nThe estimate under review is drawn from a separate source and is not replaced by registry information. Instead, AUIR counts are used to cross-check the claim that unemployment declined over the period examined. Agreement between the estimated direction and the movement in accepted claims supports the plausibility of that claim, although differences in eligibility, reporting, and timing may produce gaps between the two measures. The comparison is therefore treated as corroboration rather than as a recalculation.\n\nRegistry evidence also indicates the number of individuals receiving support and the volume of claims processed. These observations help distinguish a change in labour-market conditions from a change in access to benefits or administrative processing. The results should be read as a consistency check within the wider assessment, with the separate estimate remaining the basis for the headline measure. No inference is made about persons outside the registration system, and the registry is not used to establish total unemployment.\n\nMention: \"Audited Unemployment Insurance Registry (AUIR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The passage uses beneficiary counts and claim totals from AUIR to cross-check a separate unemployment estimate.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:b1f7a560294e66df58567174", "occurrence_id": "6f584d78568d5835b87e5d6f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Audited Primary Care Survey (APCS) is used as the principal empirical source for this assessment. It is a survey covering primary care service conditions, and the 2017 reference year provides the basis for the comparison presented below. The source is treated as the direct record of reported conditions rather than as an estimate derived from administrative information.\n\nTwo figures are reproduced directly from the APCS. The survey records that 68 percent of facilities reported uninterrupted availability of essential services, while 42 percent reported the presence of the required staffing complement. These values constitute the central evidence for the assessment of operational readiness and are presented without recalculation or adjustment.\n\nThe difference between the two reported proportions indicates that service continuity exceeded staffing adequacy in the observed system. This pattern is relevant to implementation planning because nominal service availability may not correspond to sufficient personnel capacity. The figures should therefore be read as descriptive measures for the 2017 period and not as evidence of change over time.\n\nUse of the APCS in this manner preserves the distinction between directly reported results and subsequent interpretation. The assessment relies on the two quoted values to identify an implementation constraint while avoiding assumptions about conditions outside the survey evidence.\n\nMention: \"Audited Primary Care Survey (APCS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the APCS figures of 68 percent for uninterrupted service availability and 42 percent for the required staffing complement.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:bdcdbcc35ea8789b30962b3a", "occurrence_id": "9ca3073d6d758cd2b89272d5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Comparative Water Stress Collection (CWSC), classified as an other source within the evidence base. The collection is used to support the analysis of water stress indicators across the observation series. Its records are treated as comparable observations after routine checks for missing entries, inconsistent units, and changes in coverage. No individual observation is presented as a reported headline figure.\n\nThe analysis derives a direction of change from the sequence of recorded values and calculates an average level for the period under review. The resulting trend is used to identify whether water stress has intensified, remained broadly stable, or declined over the available observations. The average provides a benchmark against which deviations in individual observations are assessed. These calculations constitute the central quantitative basis for the assessment and are preferred to reliance on isolated values.\n\nResults should be interpreted as an analytical summary of the source data rather than as a direct reproduction of the collection. The derived measures support comparison across the observation series and inform the prioritisation of monitoring attention. Their use is subject to the completeness and internal consistency of the underlying records, and the assessment therefore retains the source as the principal reference for subsequent verification.\n\nMention: \"Comparative Water Stress Collection (CWSC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The assessment uses the Comparative Water Stress Collection to calculate an average and derive a trend from its observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:181aa082a35abeebba180b7a", "occurrence_id": "a9906e7065e54a0895e830a9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Provincial Rainfall Anomaly Registry is a database produced by Deloitte for 2019. It provides the principal observational basis for the rainfall assessment and is treated as the underlying record for estimating departures from expected conditions. The registry is used as a consistent source for the analysis rather than as a set of isolated reported figures.\n\nThe assessment computes the mean rainfall anomaly across the recorded observations and derives a directional trend from the resulting series. This procedure reduces the influence of individual extreme observations and permits comparison of the general movement represented in the database. The calculated average is therefore an analytical result obtained from the registry, not a value reproduced from its presentation.\n\nThe resulting measure indicates the prevailing pattern required for implementation planning and supports interpretation of variation over the observation period. Particular attention is given to the sign and persistence of the estimated trend, while the calculated average is retained as the summary indicator for monitoring purposes. No additional adjustment is applied beyond the aggregation and trend calculation described above.\n\nMention: \"Provincial Rainfall Anomaly Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives a mean rainfall anomaly and directional trend from the 2019 database produced by Deloitte.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:398e7d81869439912f15ef1f", "occurrence_id": "23943306cf150a58935d48e3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: MEC produces the Recurrent Immunisation Coverage Index (RICI), an indicator used to assess the continuity of routine immunisation performance. The source provides a consistent basis for examining movement in coverage across the observed reporting sequence, while limiting reliance on isolated observations. Its use in this assessment is therefore aligned with the need to identify sustained changes rather than short-term variation.\n\nThe analysis derives a trend from the RICI observations by comparing successive values and estimating the direction of change over the available sequence. It also calculates an average across those observations to provide a summary measure for the assessment period. These derived measures are used as the central evidence for interpreting performance, rather than reproducing individual index values as standalone findings.\n\nThe resulting estimates support a cautious assessment of implementation progress. A rising calculated trend, when accompanied by an average above the relevant operational baseline, is interpreted as evidence of improving continuity. Conversely, a declining trend or a low average indicates the need for closer review of delivery conditions and follow-up actions. The results should be read as analytical summaries of the RICI data and not as independent measurements of service quality.\n\nMention: \"Recurrent Immunisation Coverage Index (RICI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses MEC’s RICI indicator to calculate an average and derive a trend in immunisation coverage.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:55a5c3ac17bf51689667711e", "occurrence_id": "efe2af4c8a7c4ded6e3ad489", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Rolling Labour Force Survey is the principal source for the labour-market assessment. It is a survey, and its observations referring to 2018 are retained as the evidentiary base for the calculations presented below. The source is used in its stated form, with no additional institutional attribution or geographic qualification in this note.\n\nRather than reproducing individual reported values, the analysis converts the available observations into a consistent sequence and calculates an average for the period represented in the survey. A smoothed trend is then derived from that average, allowing short-term variation to be distinguished from the underlying direction of change. This procedure gives greater weight to the pattern across observations than to any single reported result.\n\nThe resulting measure is used to assess changes in labour-force participation and employment conditions. Interpretation remains limited to movements supported by the source, and no inference is made where the sequence does not provide sufficient continuity. The calculated average and trend therefore serve as the central evidence for the assessment, while individual observations are retained only to check internal consistency and identify irregular movements. This treatment supports comparability within the analysis and avoids overstating isolated fluctuations.\n\nMention: \"Rolling Labour Force Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives a calculated average and smoothed labour-market trend from the 2018 observations in the Rolling Labour Force Survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:52c440b647ab502d009d8816", "occurrence_id": "82b490e4992fd87e38e2de52", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Sentinel Forced Population Index is used as the principal indicator for assessing changes in reported displacement pressure during 2023. The source provides a structured basis for comparison across successive observations and supports the interpretation of movements in the measured population burden. Its inclusion is limited to the analytical claim concerning directional change.\n\nTo reduce the influence of short-term variation, the assessment derives an average from the observations and uses the resulting series to identify the underlying trend. Individual entries are therefore not reproduced as reported figures. The calculated measure is used to support comparison within the results framework and to determine whether the observed pattern is increasing, stable, or declining over the period represented by the source.\n\nThe derived trend is interpreted cautiously because aggregation can conceal differences between observations and because the indicator records a defined measurement rather than a complete account of all affected persons. The analysis consequently treats the result as evidence for a broader finding, not as a standalone estimate. Any conclusion drawn from the calculation remains subject to the coverage and consistency of the underlying entries in the Sentinel Forced Population Index.\n\nMention: \"Sentinel Forced Population Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses the Sentinel Forced Population Index indicator for 2023 to calculate an average and derive a trend rather than reproduce individual figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:defc4e6b213e2ddc0af85b3e", "occurrence_id": "1aa033596aaf0363fdb4478b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Integrated Cadastral Frame Registry (ICFR) database produced by CIMP as its principal empirical source. The registry provides a structured record of cadastral units, tenure classifications, and registration status, allowing changes in the recorded portfolio to be examined on a consistent basis. Its use supports comparison across successive observations without introducing additional administrative sources.\n\nThe analysis does not reproduce isolated entries from the registry. Instead, it groups the recorded units by observation period and calculates average changes in registration coverage, classification incidence, and unresolved cases. These calculations provide the basis for identifying the direction and approximate pace of change in the cadastral system. The resulting trend is therefore derived from the distribution of records in the ICFR database rather than presented as a direct quotation from it.\n\nThe derived averages indicate gradual expansion in the share of units with complete registration, accompanied by a smaller but persistent stock of records requiring clarification. Variation between observation periods remains material, particularly for classifications affected by incomplete documentation. The results are consequently interpreted as an analytical measure of administrative progress rather than as a definitive count of land rights. Further monitoring should apply the same aggregation procedure so that subsequent results remain comparable with the evidence generated from the Integrated Cadastral Frame Registry (ICFR).\n\nMention: \"Integrated Cadastral Frame Registry (ICFR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses the Integrated Cadastral Frame Registry (ICFR) database to calculate average changes in registration coverage, classifications, and unresolved records.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:64f1ab88f4257d872857cc38", "occurrence_id": "0762550781ce5ddf0ae16aa6", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis uses the Verified Elevation Model Census as a supporting source for the assessment. It is a census covering Korea and referring to 2016. The source is retained for the relevant geographic and temporal coverage, while its individual entries are processed within the analytical series.\n\nValues drawn from the census are consolidated into comparable observations before calculation. The resulting measures are used to derive an average level and to identify the direction of change across the period under review. These calculations provide evidence for the assessment of variation and support the interpretation of the principal result.\n\nThe source therefore functions as an input to the broader analysis rather than as a table of figures reproduced without adjustment. Differences between observations are examined consistently, and the derived trend is reported with the accompanying qualification that it reflects the aggregation procedure applied to the census material. This treatment supports a measured conclusion while limiting the risk of attributing calculated quantities directly to the source.\n\nMention: \"Verified Elevation Model Census\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the 2016 Korea census to calculate an average and identify a trend.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:cd8e7f970918e55c81db1763", "occurrence_id": "a7715df2e2ca62c3ff2a00c5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Experimental Poverty and Vulnerability Survey (EPVS) is a survey produced by LinkedInEconomicGraph and is used here as the underlying source for a limited assessment of changing exposure to poverty-related risk. Its observations provide a consistent basis for examining movement in the selected indicators while recognising the experimental nature of the instrument.\n\nFor the present assessment, the reported values are transformed into an average level and a directional trend. This calculation supports the finding that vulnerability shows a gradual reduction across the observations considered. The result is used as one element in the broader assessment, alongside the interpretation of household conditions and the appraisal of implementation progress. Individual survey entries are not reproduced as standalone evidence, since the analytical purpose is to summarise the pattern represented by the source.\n\nThe derived measures should be interpreted as indicative rather than as a complete estimate of welfare conditions. They are sensitive to the coverage and design of the survey and therefore provide supporting evidence for the assessment rather than a definitive measurement. The EPVS nevertheless offers a structured basis for comparing the selected indicators and for identifying the direction and approximate scale of change relevant to the report’s conclusions.\n\nMention: \"Experimental Poverty and Vulnerability Survey (EPVS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The passage derives an average and a directional trend from the EPVS to support a broader assessment claim.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:599436599d96d472239cc669", "occurrence_id": "9d82c867207f0fbd20e662b3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Socioeconomic analysis for the implementation assessment draws on Standardised Socioeconomic Microdata, a microdata source for 2020. The material is used to establish a consistent basis for examining household characteristics and selected welfare indicators. Its role is evidentiary rather than descriptive, since the assessment transforms the observations into summary measures for reporting.\n\nObservations are reviewed for completeness and grouped into analytically comparable categories before calculation. The assessment computes means for the principal indicators and derives trend measures across the relevant indicator sequence. These calculations provide a compact representation of differences in observed conditions and reduce the influence of isolated records. Reported results therefore represent analytical outputs generated from the source, not figures reproduced directly from its tables.\n\nInterpretation is limited to measures supported by the available microdata. The resulting averages are used to compare the distribution of outcomes across the defined groups, while the calculated trends indicate the direction and relative persistence of change within the analytical series. Where the number of observations is limited, findings are treated as indicative and are not used to support broader attribution. This approach preserves a clear distinction between source observations and estimates produced for the assessment.\n\nMention: \"Standardised Socioeconomic Microdata\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The assessment derives average indicators and trend estimates from the 2020 microdata rather than quoting individual observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:327d45346338cddef1716040", "occurrence_id": "b6a8bbcc8a7e94b449cbae67", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Pilot Working Conditions Registry (PWCR) is a database produced by the Chinese Wind Energy Association for 2021. It provides a structured reference for assessing reported working conditions and related employment measures. The registry is treated as supporting evidence for the specific comparison presented below, rather than as a substitute for the principal statistical series.\n\nThe estimate is cross-checked against the PWCR database to assess whether the reported level is consistent with the conditions recorded in the registry. This comparison is used to support the claim concerning the direction and approximate magnitude of the observed result. Differences between the estimate and the registry are not interpreted as errors, given the distinct purposes and measurement procedures associated with the two sources.\n\nThe cross-check strengthens the assessment of the relevant claim within the broader analysis while maintaining the estimate as the primary basis for interpretation. The PWCR material is therefore used selectively and solely to test the plausibility of the reported finding. No adjustment is made to the estimate on the basis of this comparison, and the registry does not provide the sole basis for any conclusion.\n\nMention: \"Pilot Working Conditions Registry (PWCR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the 2021 Pilot Working Conditions Registry (PWCR) database, produced by the Chinese Wind Energy Association, to cross-check an independent estimate.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:54f708631a65bdc7b4fff284", "occurrence_id": "bc456376873ecc9ef2312054", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Subnational Apprenticeship Survey is used as the reference source for the assessment of apprenticeship participation. It is a 2018 survey and is treated as the primary descriptive basis for the indicators discussed below. The source provides a consistent point of reference for interpreting reported participation, completion, and enterprise engagement, while the assessment remains limited to the observations recorded in the source.\n\nTwo figures are reproduced directly from this survey to support the broader finding that participation is uneven across the observed apprenticeship system. The survey reports that 47 percent of participating establishments offered structured supervision, while 2,186 apprentices were recorded as receiving an allowance. These figures are quoted without recalculation and are used only to anchor the interpretation of the scale and distribution of participation.\n\nTaken together, the reported values indicate that numerical participation does not by itself establish equivalent access to support arrangements. The higher-level assessment therefore distinguishes between the number of apprentices recorded and the conditions under which training is delivered. This distinction is relevant for implementation monitoring, since aggregate counts may conceal differences in supervision and material support. The survey evidence is accordingly cited as a direct empirical basis for the claim, while no additional inference is drawn from the two figures alone.\n\nMention: \"Subnational Apprenticeship Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Subnational Apprenticeship Survey's directly reported figures of 47 percent and 2,186 apprentices to support the assessment of uneven participation and differing support conditions.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:ae8d4f0d2e188837183eacb6", "occurrence_id": "f9f6bdb7118ce1e09aa8a9c8", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis uses the Quarterly Targeting and Payments Survey (QTPS), a survey produced by DANE in 2020, as its principal empirical source. The survey provides repeated observations on targeting outcomes, payment receipt, and reported access conditions. These observations are used to assess consistency across reporting periods and to support interpretation of programme performance.\n\nThe assessment derives the central results through calculation rather than by reproducing isolated figures from the source. Changes in the principal indicators are combined into a trend measure, while reported payment outcomes are used to calculate an average for the analytical period. This treatment reduces the influence of individual observations and provides a more stable basis for comparison.\n\nThe resulting estimates are presented as analytical measures derived from the QTPS records. They are used to identify the direction and approximate scale of change, while retaining the underlying survey as the evidence base. No inference is made beyond the indicators available in the survey, and the calculated trend and average are interpreted as descriptive results for implementation monitoring.\n\nMention: \"Quarterly Targeting and Payments Survey (QTPS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage explains that a trend and an average are calculated from the 2020 QTPS rather than reported as isolated survey figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:a0461c4718377207c18219bf", "occurrence_id": "dccd0c1a93c083f0f9a56298", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Metropolitan Multidimensional Poverty Survey is a 2021 survey produced by INVALSI and is retained in the evidence base as background documentation. Its inclusion identifies the reference used to frame the discussion of deprivation measurement and does not extend the analytical scope of the results presented below. The source is treated as a supporting reference for one statement concerning the relevance of multidimensional conditions to programme monitoring.\n\nThe surrounding assessment relies on the indicators and observed records described elsewhere in this report. No estimate, distribution, rate, or other numerical result from the Metropolitan Multidimensional Poverty Survey is reproduced, recalculated, or combined with the findings. The citation therefore provides documentary orientation rather than an input to the reported calculations. This distinction is maintained to avoid implying that the source was used to establish the magnitude, direction, or coverage of any reported outcome.\n\nThe reference supports the interpretation that poverty-related constraints may affect several dimensions of welfare and that monitoring arrangements should remain attentive to their joint occurrence. That proposition is stated at a general level and is used to connect the analytical narrative to the implementation discussion. All quantitative conclusions in this section are based on the evidence specified for the assessment, while the cited survey remains a contextual anchor for the single interpretive claim noted above.\n\nMention: \"Metropolitan Multidimensional Poverty Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The passage uses the 2021 INVALSI survey to support one contextual claim and reports no values drawn from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:220555f59b09eed5f648fb63", "occurrence_id": "5346a78a982413f2e8a644f3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Biennial Socioeconomic Census (BSC) is treated as the principal census source for the assessment. Its 2016 records provide the base observations for reviewing household conditions and service-related indicators. The source is retained in its stated form, and no adjustment is made to the terminology used in the underlying records.\n\nTo support comparison, the analysis aggregates the relevant BSC observations and computes an average for each indicator set. It also traces the direction of variation across the recorded categories, producing a distributional trend rather than reproducing isolated entries. These calculations give greater weight to the structure of the observations than to any single reported figure and provide a consistent basis for the summary measures presented here.\n\nThe resulting measures are used to identify broad differences in observed conditions and to assess whether patterns are concentrated or dispersed. Interpretation remains limited to information contained in the census, with care taken not to infer changes that the records cannot establish. The average and trend are therefore presented as analytical outputs derived from the 2016 BSC material, not as direct quotations from the source.\n\nMention: \"Biennial Socioeconomic Census (BSC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses the 2016 BSC census to calculate an average and identify a distributional trend.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:52789e9302768cd483388716", "occurrence_id": "842ebf928d773e630b122ab9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The District Health Workforce System is used as an administrative source covering Ukraine in 2020. It provides the principal evidence for the staffing position described in this section and is treated as a direct record of reported workforce conditions.\n\nTwo values are reproduced directly from the system. The reported establishment comprises 31,482 staffed posts, while 2,146 posts are recorded as unfilled. These figures are presented without recalculation and are retained in their original units to preserve the distinction between positions occupied and positions remaining vacant.\n\nThe evidence indicates a measurable staffing gap within the reported service structure. The unfilled posts represent a material proportion of the recorded establishment, although the figures should not be interpreted as a measure of service quality or workload. They support a descriptive assessment of workforce availability and provide a fixed reference for consideration of staffing constraints.\n\nInterpretation is limited to the coverage and definitions applied in the source. No adjustment is made for reporting completeness, subsequent personnel movements, or differences in local classification. The two reported values are therefore used as administrative observations rather than as estimates generated through modelling.\n\nMention: \"District Health Workforce System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes 31,482 staffed posts and 2,146 unfilled posts directly from the District Health Workforce System.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:63d1d1aecb8a1f275c7e2c4c", "occurrence_id": "7c957040ef722e41fda443fe", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Record-Based Access and Documentation Index is identified as the 2021 indicator used for the assessment of access and documentation conditions. It provides a consistent reference for interpreting reported differences in the availability and usability of administrative records. No adjustment to the source classification is applied in this assessment.\n\nTwo figures are taken directly from the indicator and used as the central evidence. The reported access rate is 68 percent, while the corresponding documentation rate is 42 percent. These values are reproduced as reported and are not recalculated from other material.\n\nThe difference between the two figures indicates a material gap between the presence of access arrangements and the availability of supporting documentation. This gap is relevant to implementation monitoring because improved access does not necessarily imply complete or reliable records. The findings therefore support continued attention to record quality, retrieval procedures, and documentation coverage in subsequent reporting.\n\nMention: \"Record-Based Access and Documentation Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported values of 68 percent and 42 percent from the Record-Based Access and Documentation Index.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:42ed47cf6c8743fb171d59b5", "occurrence_id": "0fdd939400e2cc80f3a23bbb", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Household Building Footprint Dataset is used as the geospatial source for the 2020 reference period. It provides the primary empirical basis for the building-count assessment and is retained in its reported form for the purposes of this annex. No adjustment is applied to the source figures before their inclusion in the baseline tabulation.\n\nThe report reproduces two figures directly from the dataset. The source records 31,486 household building footprints and 4,218 associated structures. These values are presented as observed quantities rather than as estimates generated through an additional calculation. Their inclusion permits the assessment to distinguish the reported inventory from later analytical measures and to maintain a clear line of evidence between the source and the results.\n\nThe figures support the implementation review by establishing the scale of the recorded building stock and the number of structures linked to household footprints. They should be read as source-level observations for the stated reference period. Any subsequent comparison, aggregation, or rate calculation is to be undertaken separately and should not be attributed to the dataset itself. On this basis, the dataset is suitable for documenting the initial inventory used in the assessment, subject to the standard interpretation that reported coverage reflects the content available in the source.\n\nMention: \"Household Building Footprint Dataset\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The report takes 31,486 household building footprints and 4,218 associated structures directly from the Household Building Footprint Dataset.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:896d4ee4bced1b5d6eb0f940", "occurrence_id": "b55092cb842e206937e72ec4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Provincial Water Body Dataset (PWBD), a geospatial source maintained for reference in the implementation record. The dataset provides a consistent basis for describing the recorded water-body inventory and for identifying the scale of the monitoring requirement. Its use in this section is limited to supporting the interpretation of coverage and status.\n\nTwo figures reproduced directly from the PWBD are relevant to this assessment. The source records 18 registered water bodies and reports that 42 percent of recorded sites fall within the monitored category. These figures are cited without recalculation and are used to support the broader finding that coverage remains uneven across the reported inventory.\n\nThe evidence does not, by itself, establish the causes of the observed distribution or indicate whether individual records have changed over time. It nevertheless provides a documented quantitative reference for implementation review, allowing the reported level of coverage to be considered alongside operational priorities and the need for continued verification. Any subsequent planning should retain the same definitions and counting basis so that comparisons remain interpretable.\n\nMention: \"Provincial Water Body Dataset (PWBD)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage takes the figures 18 registered water bodies and 42 percent of recorded sites in the monitored category directly from the PWBD.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:dd968bc443e09988ce3cb50b", "occurrence_id": "b613ef02ddf1d393ff01aa86", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Provincial Household Consumption Survey is the principal statistical reference used for the assessment of household consumption conditions in Yemen in 2017. This survey is used as a direct evidentiary source rather than as background material, and its reported measures are retained without adjustment. The source provides a consistent basis for presenting observed household outcomes in a form suitable for comparison across the assessment.\n\nTwo figures are quoted directly from the source. The survey records that 42 percent of households reported difficulty meeting routine consumption needs, while 17 percent reported that such difficulty persisted throughout the reference period. These values are reproduced as reported and are not recalculated, inferred from other indicators, or combined with estimates from another source. Their inclusion is intended to preserve the incidence recorded by the survey.\n\nTaken together, the figures indicate a substantial level of consumption stress among surveyed households, with a smaller but material proportion experiencing continued pressure. The difference between the two percentages should not be interpreted as a change over time, since the cited evidence refers to a single survey year. The results therefore support a cautious assessment of household vulnerability and provide a transparent quantitative basis for the related discussion. No broader trend is inferred from these two observations.\n\nMention: \"Provincial Household Consumption Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The analysis takes the figures of 42 percent and 17 percent directly from the Provincial Household Consumption Survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:5490325fb01535e5d552a984", "occurrence_id": "75c4d6d99ef54d44cdd7708c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Audited Disease Surveillance Registry is used as the principal database for assessing changes in reported disease events. Its entries provide the central empirical basis for the analysis of notification levels, reporting continuity, and variation across successive reporting periods. No separate source is introduced for the calculations presented below.\n\nThe analysis derives period-specific rates from the registry records and compares their movement over time. It also computes an average across the observed periods to provide a stable reference against short-term fluctuations. These results are calculated from the underlying entries rather than reproduced as isolated registry figures, and they therefore support interpretation of the direction and magnitude of change.\n\nThe resulting trend indicates whether notification activity is increasing, declining, or remaining broadly stable, while the calculated average provides a benchmark for assessing deviations in individual periods. Where changes are pronounced, the assessment treats them as signals for further operational review rather than as evidence of a causal relationship. The registry is consequently used as an analytical foundation for monitoring performance and identifying areas requiring follow-up.\n\nMention: \"Audited Disease Surveillance Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives period-specific rates, a trend, and an average from the Audited Disease Surveillance Registry database.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:a2e1866c411be9871ad31610", "occurrence_id": "9c8486ade33fd02b8afbe11d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment incorporates the Follow-Up Climate and Flood Risk System as an administrative source for 2024. Its records provide a structured basis for examining reported exposure, preparedness measures, and the distribution of identified risks across the reporting units. The source is used within the statistical review to support interpretation of changes in observed conditions and to inform the monitoring framework.\n\nRather than reproducing individual entries, the analysis calculates an average across the available observations and derives a trend from their ordered sequence. These calculations reduce the influence of isolated records and provide a consistent measure for comparison within the assessment period. The resulting indicators are treated as analytical estimates derived from the source, not as figures stated directly by the system.\n\nThe analysis indicates a gradual strengthening of recorded risk management, although the direction and magnitude of change vary among the observed units. This finding supports the broader conclusion that implementation capacity is improving while remaining uneven. The derived average is also used to identify whether reported conditions are concentrated around a common level or affected by a limited number of higher observations. Interpretation remains cautious because administrative records may reflect differences in reporting practice as well as differences in underlying conditions.\n\nMention: \"Follow-Up Climate and Flood Risk System\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the 2024 administrative source to calculate an average and derive a trend supporting the assessment of changing risk management.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:9bda5b0021fb2be0cd508f53", "occurrence_id": "0f1495b40662c8aeaf2b2aa2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Civil Hiring and Vacancies Survey, a 2022 survey produced by University of Dayton, provides an independent reference for the assessment of hiring activity and unfilled positions. Its reported measures are used to cross-check the estimate derived from a separate administrative source, with attention to the direction and approximate scale of the result rather than to exact correspondence between the series.\n\nThe comparison supports the interpretation that the observed movement in vacancies is consistent with the broader employment assessment. Differences in coverage, reporting practice, and timing may account for residual variation. The survey is therefore used as a corroborating reference within the analysis, and not as the sole basis for the estimate or for any allocation decision.\n\nOn this basis, the cited material strengthens the evidentiary basis for the vacancy assessment while preserving the distinction between independently reported measures. The resulting statement is limited to the consistency of the indicators and does not imply that the two sources measure identical populations or reporting units.\n\nMention: \"Civil Hiring and Vacancies Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The work taken from the source is an independent cross-check of hiring activity and vacancy estimates.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:bede1d7c13469b28bece4979", "occurrence_id": "c5d15543f276c3b4fbc9b823", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The data source used for this assessment is the Inter-Agency Unemployment Insurance System (IUIS), an administrative source covering Armenia in 2015 and produced by TurkSTAT. Its records provide the principal basis for the employment-related observations presented below.\n\nTwo values are reproduced directly from the source. The reported number of persons registered as unemployed is 18,420, while the corresponding rate among the covered labour force is 7.4 percent. These figures are retained in their stated form and are not recalculated for this assessment.\n\nThe first value indicates the recorded volume of unemployment within the system, and the second provides the associated proportion for the same reporting frame. Taken together, they support a concise description of the observed position while preserving the distinction between an administrative count and a percentage measure.\n\nInterpretation remains limited to the coverage and definitions embedded in the source. The figures should therefore be read as reported administrative measures for Armenia in 2015 rather than as estimates constructed from a separate survey or an independently derived series. This treatment maintains traceability to the cited record and avoids introducing adjustments that are not documented in the available material.\n\nMention: \"Inter-Agency Unemployment Insurance System (IUIS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage reports 18,420 registered unemployed persons and a 7.4 percent rate taken directly from the IUIS record.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:403c61202ac8555e6c90b9e5", "occurrence_id": "475c333e6bd1953708abd6a0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment relies on the Urban Agricultural Household Census (UAHC), a census produced by CDR, as the principal data source for measuring conditions among urban agricultural households. The source is treated as the evidentiary base for the analysis and is retained in its cited form throughout the assessment.\n\nThe reported results are derived through calculation rather than transcription. Household observations are consolidated to obtain mean values for the periods represented in the census, and the resulting sequence is used to identify the direction and approximate pace of change. This approach reduces the influence of isolated observations and permits comparison on a consistent basis. The assessment therefore presents an estimated average and a trend derived from the UAHC, rather than reproducing individual entries or selecting isolated figures.\n\nInterpretation remains limited to the information supported by the census. Differences in the calculated averages are read as evidence of movement in the measured conditions, while short-term variation is not treated as a structural shift unless it is reflected in the broader pattern. The resulting estimates support implementation monitoring and provide a reproducible basis for subsequent assessment. Any inference beyond the calculated trend should be treated with caution, given the scope of the cited source and the absence of supplementary evidence.\n\nMention: \"Urban Agricultural Household Census (UAHC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives mean values and a directional trend from the UAHC rather than quoting individual census entries.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:6b8aee5e4ef5d49db9987e2e", "occurrence_id": "3da4968661d6ce3ee2dc05e4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Local Irrigated Area Registry is an MRC database used to support review of reported irrigation coverage. It provides a structured reference for comparing the extent of irrigated activity recorded through routine reporting with estimates generated through a separate analytical approach. The registry is treated as a supporting source rather than as the sole basis for the assessment.\n\nThe estimate presented in the analysis indicates a broader level of irrigated area than would be inferred from a direct reading of the primary reporting series. This difference is material to interpretation, but it does not alter the overall finding that irrigated production remains concentrated within the areas captured by the assessment. The Local Irrigated Area Registry is used to cross-check this estimate and to test whether the implied scale is consistent with independently recorded coverage.\n\nThe comparison shows a sufficiently close correspondence for the estimate to be retained for analytical purposes, subject to the limitations of the underlying records. In particular, the registry may not capture all changes in operational status or reporting completeness at the time of assessment. Its role is therefore confirmatory. The resulting evidence supports the claim on the extent of irrigated activity while preserving a distinction between the estimated measure and the registry-based reference. This treatment is consistent with standard verification practice in implementation reporting, where an independent database is used to examine the plausibility of a principal estimate without replacing the underlying measurement procedure.\n\nMention: \"Local Irrigated Area Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the MRC Local Irrigated Area Registry database to cross-check an independent estimate of irrigated area.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:2ff3b7f134b8c95e677f51c5", "occurrence_id": "551579757108156eabcc8631", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Follow-Up Fuel and Energy Use Survey is identified as a survey produced by the National Statistics Office for Comoros. It is cited as a supporting reference for the discussion of household energy conditions and the institutional basis of the statistical record. Its inclusion indicates the availability of relevant background material, while preserving a clear distinction between contextual documentation and evidence used in the assessment.\n\nThe discussion does not draw levels, rates, totals, or other numerical findings from this source. No comparison, projection, or calculation is based on the survey, and no result is attributed to it. The reference therefore serves to situate the subject matter within the broader statistical environment rather than to substantiate a particular conclusion.\n\nAnalytical statements in this section rely on the evidence presented directly in the assessment and on the definitions applied there. The cited survey remains a passing reference, retained for traceability and subject orientation. This treatment avoids implying that the source supplied a measured value, a change over time, or an explanatory finding for the purposes of the present report.\n\nMention: \"Follow-Up Fuel and Energy Use Survey\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The passage identifies the Follow-Up Fuel and Energy Use Survey as a contextual source from the National Statistics Office for Comoros and draws no values or findings from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:3ff9568e72300031cbe09e80", "occurrence_id": "da0e1f8477c0d5d1be573215", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Data sources.** The Sentinel Deprivation and Hardship Survey is a survey produced by Barth, Caprio, and Levine covering Uzbekistan in 2015. It provides the principal empirical basis for the hardship assessment and is used to establish a consistent reference for comparing deprivation across the observed population.\n\nThe analysis does not reproduce isolated observations from the source. Instead, it derives average hardship indicators and calculates the direction of change across the available measures. This treatment reduces the influence of exceptional household outcomes and supports a summary interpretation suitable for implementation monitoring. The resulting trend is therefore an analytical estimate based on the survey material rather than a direct transcription of reported values.\n\nThe calculated average indicates a persistent concentration of hardship among households facing multiple deprivations. Differences between individual measures remain relevant, but the combined pattern provides the more stable basis for assessing progress and identifying areas requiring additional attention. Results should be read as evidence of the prevailing distribution of hardship represented in the source, with the derived averages serving as the principal measure used in this assessment.\n\nMention: \"Sentinel Deprivation and Hardship Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The assessment derives average hardship indicators and a trend from the Sentinel Deprivation and Hardship Survey covering Uzbekistan in 2015.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:1385940914fbf2aaa43abf41", "occurrence_id": "5d75fc86b48af8c014840699", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Panel Settlement and Elevation Registry (PSER) database is the principal empirical source for the settlement and elevation assessment of Dar-es-Salaam, with the cited records referring to 2016. The source is used to establish a consistent base for comparing recorded settlement conditions and elevation characteristics within the covered area.\n\nThe analysis does not reproduce individual registry entries. Instead, it groups the available observations, calculates average elevation measures, and derives a settlement trend from the resulting distribution. These calculations provide the central evidence for interpreting differences across settlement groupings and for identifying the direction and magnitude of observed variation. The reported measures therefore represent analytical results obtained from PSER rather than figures transcribed directly from the database.\n\nThe computed average is applied as a summary indicator, while the derived trend supports assessment of spatial differentiation in settlement conditions. Results are interpreted with attention to the coverage and structure of the source records, since aggregation can reduce the visibility of local variation. This treatment allows the registry to inform the assessment without implying that each reported measure is an individual observation. The source is consequently used as a basis for calculation, comparison, and internal consistency checks throughout the analysis.\n\nMention: \"Panel Settlement and Elevation Registry (PSER)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The assessment uses PSER records for Dar-es-Salaam referring to 2016 to calculate average elevation measures and derive a settlement trend.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:46785be476e4b53eb5b931d8", "occurrence_id": "05a817dfd001b08cf5011343", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis uses the Standard School Assessment Survey, a 2019 survey produced by the World Justice Project. The source provides the central evidence for assessing variation in reported school conditions and related performance measures. Its observations are treated as comparable inputs for calculation rather than as isolated figures reproduced in the text.\n\nMean values are calculated across the available assessment measures, with each measure receiving consistent treatment before aggregation. The resulting averages provide a summary of the observed level of performance and reduce the influence of individual responses. The analysis also examines the direction of change across the ordered measures to identify a broad trend in the reported results.\n\nThese calculations are used to support interpretation of the assessment pattern and to inform the discussion of implementation performance. No single observation is treated as determinative. Instead, the reported position reflects the average derived from the survey records and the direction indicated by the combined measures. This approach allows the findings to remain concise while retaining the underlying variation captured by the source.\n\nMention: \"Standard School Assessment Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives mean values and a broad directional trend from the 2019 Standard School Assessment Survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:04c09370c99f22c38b9d956a", "occurrence_id": "bd9c7cf8307cfe97bf95ed75", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Implementation progress is assessed against the operational requirements for housing condition monitoring. The analysis indicates that routine inspection, maintenance prioritisation, and referral of unresolved cases remain necessary for consistent service delivery. This conclusion reflects the administrative arrangements examined in the project and does not depend on a numerical series.\n\nFor contextual support, the Integrated Housing Conditions Registry database is referenced as a background source for the observation that housing conditions require regular monitoring and structured follow-up. The registry is not used to extract values, construct indicators, compare periods, or calculate results for this assessment. Its mention therefore supports one statement within the broader discussion of implementation arrangements and should not be interpreted as evidence of a separate statistical analysis.\n\nOn this basis, the assessment focuses on procedures that can be verified through project records, including the assignment of responsibilities, the recording of reported defects, and the closure of corrective actions. These elements provide the relevant basis for judging whether monitoring arrangements are functioning as intended. The cited source remains supplementary and does not alter the findings, thresholds, or conclusions presented in this annex.\n\nMention: \"Integrated Housing Conditions Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The source supplied only contextual support for the claim that housing conditions require regular monitoring and structured follow-up.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:741dc0199a4701280048770f", "occurrence_id": "9fc142e863179777d3b7479a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Independent Asset Ownership Survey, a World Bank-sponsored survey covering Fiji and referring to 2017, provides the principal empirical basis for the assessment. Its observations are treated as a consistent body of evidence on household access to selected productive and durable assets. The source is used in accordance with its stated coverage, while interpretation remains limited to the information required for the present analysis.\n\nRather than reproducing individual observations, the analysis calculates average ownership levels across the reported asset groups and derives a comparative trend from those averages. This approach reduces the influence of isolated household responses and provides a summary measure suitable for assessing the distribution of ownership. The resulting indicators are therefore analytical estimates derived from the survey, not figures presented as direct quotations from it.\n\nThe calculated measures are used to identify the relative concentration of ownership and the direction of change across the asset groups included in the source. Differences in the averages are interpreted cautiously, since they describe the surveyed population and reflect the scope of the underlying observations. No adjustment is made beyond the aggregation required to obtain the reported averages and trend. On this basis, the survey supports a concise assessment of asset ownership patterns and provides the central evidence for the findings presented in this section.\n\nMention: \"Independent Asset Ownership Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the Independent Asset Ownership Survey for Fiji and 2017 to calculate average ownership levels and derive a comparative trend.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:5ade91753c8bdbc88e62fb92", "occurrence_id": "9e48ab1de94ea243e9f9cc91", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Data sources.** The Sentinel Adaptation Finance Survey (SAFS) is cited in this assessment as a survey relevant to the interpretation of adaptation finance reporting. Its inclusion provides background for the discussion of how financing activities are identified and classified across the project framework.\n\nThe reference supports the contextual observation that adaptation finance is commonly examined through structured reporting categories rather than through a single undifferentiated total. This observation is used to frame the assessment of reporting practices and to clarify the basis for the narrative that follows.\n\nNo figures, estimates, classifications, or other values from the survey are reproduced or applied in the analysis. The cited material therefore does not determine a result, provide a denominator, or serve as an input to any calculation presented in this document. Its role is limited to supporting the stated background claim within the broader assessment.\n\nAll findings reported below are based on the evidence and procedures described in the relevant sections of the assessment. The citation is retained to document the basis for the contextual framing, while the conclusions remain dependent on the project’s own review and verification processes.\n\nMention: \"Sentinel Adaptation Finance Survey (SAFS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The passage takes no values from SAFS and uses the survey only to support a contextual statement about adaptation finance reporting.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:23362db33143004d7ca87c37", "occurrence_id": "d5c6eb3333913bd963a14f54", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The cited source is the Audited Water Body Index (AWBI), an indicator referring to 2016. It is used as the quantitative basis for a limited statement on recorded water body conditions. The source is retained in its stated form, and no adjustment is applied to the cited observations.\n\nThe assessment quotes directly from the source two figures, namely 42 percent and 68 percent. These values are presented as reported measurements rather than as estimates generated for this assessment. Their inclusion supports the broader finding that recorded conditions are uneven across the monitored set and that aggregate interpretation should therefore remain cautious.\n\nThe evidence is used for corroboration within the paragraph, not as a separate analytical series. The two reported figures illustrate the range relevant to the claim and provide a transparent basis for the accompanying interpretation. No additional calculation, projection, or comparison is derived from the AWBI material.\n\nOn this basis, the cited indicator provides a narrowly defined reference for the assessment. Its evidentiary role is limited to supporting the stated observation, while the surrounding discussion addresses implementation performance using the broader body of evidence available to the report.\n\nMention: \"Audited Water Body Index (AWBI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The assessment directly quotes 42 percent and 68 percent from the Audited Water Body Index (AWBI) indicator for 2016 to support a cautious finding on uneven recorded conditions.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:e131b3c85327234281948895", "occurrence_id": "eca2a2af015aae5c79f4ce5f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Continuous Minimum Needs Registry database as its principal administrative source for describing reported needs and the corresponding coverage of basic assistance. The database is treated as a structured record for reference, rather than as a basis for estimating changes over time or generating additional indicators. Its entries are cited in the form in which they are recorded.\n\nTwo figures are reproduced directly from the Continuous Minimum Needs Registry. The record identifies 31 households requiring support, while a separate entry reports 68 percent coverage among the households included in the register. These figures constitute the central evidence for the assessment of unmet needs and service reach. No adjustment, aggregation, or trend calculation has been applied to either value.\n\nThe figures should be read as registry observations and not as estimates of conditions outside the recorded population. Their use in this annex is therefore limited to establishing the scale of recorded need and the reported level of coverage. Subsequent discussion relies on these observations only where the relevant population is defined consistently with the database. This treatment preserves the traceability of the evidence and avoids implying a degree of precision that the underlying entries do not support.\n\nMention: \"Continuous Minimum Needs Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage reports 31 households and 68 percent as figures reproduced directly from the Continuous Minimum Needs Registry database.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:8eee7148a36449ac62c02703", "occurrence_id": "a9ed4b36dd5f061474f5be36", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis uses the Comparative Living Standards Survey, a survey produced by the Economist journal in 2016, as its principal empirical source. The source provides the observations required to assess changes in living standards and to establish a consistent basis for comparison across the measures considered in this annex.\n\nThe reported series are processed to calculate average levels and the direction of change over the available observations. These derived results, rather than isolated figures reproduced from the source, provide the central evidence for the assessment. The calculation reduces the influence of individual observations and supports interpretation of the general movement in the indicators.\n\nResults are presented in aggregated form and should therefore be read as analytical estimates based on the survey record. Minor variation in individual observations does not alter the overall pattern identified by the calculated averages and trend measures. The findings are used to inform the assessment of distributional conditions and to support subsequent interpretation of changes in living standards.\n\nMention: \"Comparative Living Standards Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses survey observations to calculate average levels and a trend in living standards.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:5f6f39ac60efa64cb43051b8", "occurrence_id": "0081d317a5c445becc8d0f01", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Data sources.** The analysis draws on the Rolling Output Area Dataset, a geospatial source produced by FAO in 2020. The dataset provides the central empirical basis for the assessment, with successive observations used to examine changes in the reported measure over time. No individual observation is treated as sufficient evidence of performance, and interpretation is based on the pattern formed across the available records.\n\nFor the purposes of this assessment, the observations are arranged in their recorded sequence and reviewed for consistency. A mean is calculated across the usable records, while the direction and rate of change are estimated from the same series. This approach reduces the influence of isolated fluctuations and provides a more stable indication of underlying movement. The resulting average is used as the reference level, and the estimated trend is used to assess whether outcomes are improving, declining, or remaining broadly unchanged.\n\nThe findings therefore represent a calculation derived from the Rolling Output Area Dataset rather than a reproduction of any single reported value. Differences between individual records are retained in the analysis, but conclusions rely on the aggregate pattern. This treatment supports comparison across reporting periods and provides a transparent basis for monitoring progress. Results should be read as analytical estimates generated from the source records, with the average and trend serving as the principal measures for interpretation.\n\nMention: \"Rolling Output Area Dataset\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives an average and an estimated trend from the Rolling Output Area Dataset.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:019a830239804ddd3a720525", "occurrence_id": "be8132dea1a30ec28aa3937c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Independent Unemployment Insurance Survey is used as the survey source for the analysis. Its reported measures are retained without recalculation, and the source is treated as evidence on participation in the unemployment insurance system. No adjustment is applied to the two observations used below.\n\nTwo figures are reproduced directly from this source. The first is 31 percent for respondents indicating that they had received a payment, while the second is 14 percent for respondents indicating that a claim had been rejected. These values are presented as reported, rather than as estimates derived from comparison with another dataset. Together, they provide the principal quantitative basis for the observations in this section.\n\nThe difference between the two reported shares is noted only to clarify the distribution of responses. It should not be interpreted as a measure of change, since the cited material supplies separate figures rather than a time series. The results nevertheless indicate that receipt of support was more frequently reported than rejection of a claim among the respondents covered by the survey. This finding is used to frame the implementation assessment, while the underlying response definitions remain those applied in the source.\n\nMention: \"Independent Unemployment Insurance Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the reported figures of 31 percent and 14 percent directly from the Independent Unemployment Insurance Survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:cf399460ad1f1d5c0843ef5a", "occurrence_id": "899788f6ef38d930d43e8ddb", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Record-Based Health Workforce Index is used as the principal evidence for assessing changes in recorded health workforce availability. It is a 2023 indicator that provides a consistent basis for examining successive reporting intervals and for distinguishing sustained movement from isolated variation. The source is treated as an analytical series rather than as a set of individual values to be reproduced.\n\nThe assessment derives an average across the available observations and calculates the direction and pace of change over time. This procedure reduces the influence of short-term fluctuations and supports comparison of the underlying pattern within the reporting period. The resulting measures are used to inform the interpretation of workforce adequacy and the identification of areas requiring closer monitoring.\n\nNo single observation is presented as representative of the full series. Instead, the computed average and trend provide the central basis for the findings reported in this annex. Where movement is limited, the analysis records a broadly stable pattern; where the calculated series changes materially, the assessment treats that movement as evidence of a shift in recorded workforce availability. The approach therefore preserves the evidentiary role of the Record-Based Health Workforce Index while relying on summary measures generated from its observations.\n\nMention: \"Record-Based Health Workforce Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives an average and an overall trend from the Record-Based Health Workforce Index.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:27e671c187ee21559a94e01c", "occurrence_id": "7fb569e886a4bcc0dded88f9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Regional Building Footprint Census is a 2024 census produced by IEG and is used as the supporting source for the building stock estimates presented in this assessment. Its tabulations are treated as reported observations rather than as a basis for recalculation or projection.\n\nThe assessment quotes directly from the census two figures that support the finding of uneven building coverage. The source records 31 percent for structures meeting the stated inclusion criteria, while 68 percent is reported for structures identified through the complementary enumeration. These values are reproduced without adjustment and are cited to anchor the comparison within the broader discussion of data completeness.\n\nThe resulting interpretation remains limited to the evidence available in the source. The figures indicate a material difference between the two recorded groups, but they do not establish the causes of that difference or imply change over time. Accordingly, the assessment uses them to qualify the reliability of aggregate building counts and to identify the need for cautious treatment of disaggregated results.\n\nNo additional weighting is applied to either observation. The cited values therefore function as documentary evidence for the assessment’s conclusion concerning coverage and comparability.\n\nMention: \"Regional Building Footprint Census\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported figures of 31 percent and 68 percent from the Regional Building Footprint Census to support a finding on uneven building coverage.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:3610344af2bdc0a545f9a645", "occurrence_id": "507847cb225dd0d48e26b306", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment treats informal trade as a continuing component of household economic activity. The available evidence indicates that participation is broad, while the scale of individual operations remains uneven. This pattern supports the conclusion that interventions should address both access to markets and the limited capacity of small operators to absorb temporary shocks.\n\nThe Household Informal Trade Census is a census produced by APIS and is used as the direct statistical basis for two figures in this assessment. It reports that 42.6 percent of surveyed households engaged in at least one informal trading activity, while 18.4 percent reported that such activity constituted their principal source of trading income. These figures are reproduced without recalculation and are cited to substantiate the broader assessment of the sector’s reach and economic significance.\n\nThe evidence should nevertheless be interpreted with appropriate caution. Participation does not establish the regularity, profitability, or stability of the activity, and the reported shares do not measure the value of goods exchanged. The source remains useful for establishing the relative prevalence of informal trade and for indicating the scale of households potentially affected by measures directed at registration, market access, and basic business support. On this basis, the assessment gives greater weight to the direction and magnitude of participation than to precise estimates of aggregate economic output.\n\nMention: \"Household Informal Trade Census\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The author takes the 42.6 percent participation figure and the 18.4 percent principal-income figure directly from the Household Informal Trade Census produced by APIS.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:7a4942e12dbf1f3fad529561", "occurrence_id": "fc3c766a677c0382283d442a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The District Supply Chain Survey is treated as the principal empirical source for the assessment. It is a survey produced by WB/Makerere/FASID and is used to characterize the distribution of reported service conditions across the observation set. The source is retained in its original form for traceability, while derived measures are calculated for analytical use.\n\nThe analysis computes average values across the reported observations and compares those averages across the relevant categories. It also estimates a directional trend by arranging the observations in their recorded sequence and examining the change in the resulting summary measure. Individual entries are therefore not presented as standalone findings; they serve as inputs to aggregate calculations intended to reduce the effect of isolated observations.\n\nThe resulting averages and trend estimates provide the central evidence for assessing consistency, variation, and the direction of change. Interpretation is limited to patterns supported by the source, with no inference beyond the reported observations. Where the number of observations differs between categories, the calculation is based on the available entries and is reported as an analytical summary rather than as a direct quotation from the source. This approach supports comparison while preserving a clear distinction between recorded information and measures generated for the assessment.\n\nMention: \"District Supply Chain Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives average values and a directional trend from the District Supply Chain Survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:6aa1a482c36ab5ce5e079231", "occurrence_id": "4aa02d0dbb693904657ef214", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Linked Nighttime Lights Index is an indicator produced by oDesk for 2022 and is used as a supporting measure in the assessment of observed economic activity. Its inclusion provides a standardized reference for interpreting variation in the underlying observations, while avoiding reliance on a single administrative measure.\n\nThe two figures reproduced in this assessment, 63.4 and 71.8, are taken directly from the indicator. The first figure is associated with one observation and the second with another observation in the source material. No transformation, aggregation, or recalculation has been applied to either value. Their use is therefore limited to representing the levels reported by the cited source.\n\nRead together, the figures support the broader finding that activity was uneven across the observations considered. The difference between the reported levels is sufficient to qualify an interpretation of uniform performance, but it does not by itself establish the causes of the variation. The results should accordingly be read as descriptive evidence supporting the assessment rather than as a standalone estimate of change.\n\nMention: \"Linked Nighttime Lights Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Linked Nighttime Lights Index values 63.4 and 71.8 to support a finding of uneven activity.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:02be982ed404b563e3472fd5", "occurrence_id": "fea11975796cc2304b7c447e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Follow-Up Public Works Index (FPWI) is the principal data source for this assessment. It is an indicator covering Sub-Saharan Africa, with the Third Version referring to 2019. The index provides the central evidence for examining changes in public works activity across the covered reporting units and supports a consistent basis for comparison.\n\nThe analysis does not reproduce individual observations as reported values. Instead, the observations are ordered and aggregated to derive an overall trend, while an average is calculated to reduce the influence of isolated movements within the series. This treatment permits the assessment to distinguish the general direction of activity from short-term variation and irregular reporting.\n\nThe resulting measure is used to describe the level and movement of public works activity represented by the source. Interpretation remains limited to the coverage and reference period of the index, and the derived trend is presented as an analytical summary rather than as a direct quotation from the underlying records. The FPWI therefore functions both as the documented reference and as the quantitative basis for the assessment.\n\nMention: \"Follow-Up Public Works Index (FPWI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses the Follow-Up Public Works Index (FPWI), an indicator for Sub-Saharan Africa in 2019 and its Third Version, to derive an average trend rather than quote individual observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:ee0dd86ef12c1dbfac843c3e", "occurrence_id": "e2df03efdcade53f00644565", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Analytical Inequality Index as an indicator of distributional conditions and employs it as the principal quantitative basis for this review. The source is treated as a structured measure suitable for monitoring change across successive observations. Its reported values are therefore read as inputs to analysis rather than as findings reproduced without adjustment.\n\nFor the purposes of the assessment, the observations are organised in sequence and examined for persistence, movement, and dispersion. A mean is calculated across the available observations to provide a stable summary of the level recorded by the indicator. The resulting average is used to reduce the influence of short-term variation and to support comparison within the analytical series.\n\nThe assessment also derives a trend from the same sequence by evaluating the direction and regularity of change between observations. This calculation provides the central evidence for judging whether inequality is rising, declining, or remaining broadly stable. No individual figure is presented as a standalone conclusion; interpretation rests on the combined average and trend generated from the Analytical Inequality Index.\n\nThese procedures support a cautious reading of the evidence. The average describes the general position of the series, while the trend indicates its underlying movement. Together, they inform the assessment of distributional performance and provide a consistent basis for subsequent monitoring and reporting.\n\nMention: \"Analytical Inequality Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives an average and a directional trend from the Analytical Inequality Index indicator.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:4ab42abe70a8440687fd2f88", "occurrence_id": "831c139b293a5ab02630dc7b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment indicates that programme coverage was broadly consistent with the planned implementation sequence, although the available administrative estimate remains subject to reporting gaps. This finding is used to support the conclusion that delivery systems were functioning across the reviewed implementation period.\n\nThe estimate is cross-checked against the Periodic Graduation Programmes Microdata, a microdata source produced by WBG. The comparison provides an independent check on the reported number of participating households and on the direction of observed change. The source is not used to replace the administrative series, since the two sources may differ in coverage, timing, and recording practice.\n\nThe cross-check supports the claim that recorded participation increased during implementation, while also indicating that the magnitude of the increase should be interpreted with caution. Differences between the sources are treated as a measurement consideration rather than as evidence of a substantive inconsistency. On this basis, the assessment retains the administrative estimate for programme monitoring and uses the Periodic Graduation Programmes Microdata to test its plausibility.\n\nThe result is consistent with the broader evidence on delivery performance. It supports continued attention to reconciliation procedures, documentation of reporting completeness, and periodic review of the underlying records before future estimates are used for resource allocation or outcome assessment.\n\nMention: \"Periodic Graduation Programmes Microdata\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The author uses Periodic Graduation Programmes Microdata from WBG to cross-check the reported participation estimate and support a cautious finding on implementation coverage.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:a7ec7e2883c1e6d32186414e", "occurrence_id": "f092e8bb812778d9415b0ee3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The District Vocational Training Registry database is used as the principal quantitative source for this assessment. Produced by ILO for South Africa, the First Collection provides a consistent basis for describing the recorded training profile and for identifying the scale of provision represented in the registry. The source is treated as an administrative reference rather than as an estimate derived from the assessment.\n\nTwo figures are reproduced directly from the registry. It records 31 trainees and 14 workshops in the relevant entries. These values constitute the central evidence for the comparison presented here and are retained without recalculation or adjustment. Their use is intended to preserve the reported distribution and to avoid introducing assumptions about coverage, classification, or completeness.\n\nThe figures should therefore be read as recorded observations within the District Vocational Training Registry database. They support a concise description of participation and provision, while the limitations of the source remain applicable to interpretation. No further aggregation is applied, and the evidence is presented in the form in which it appears in the First Collection.\n\nMention: \"District Vocational Training Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the directly reported figures of 31 trainees and 14 workshops from the District Vocational Training Registry.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:d1766d8aac4ca5a9e9828cf0", "occurrence_id": "314b4351fc7e45c6da0b6676", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Rolling Asset Ownership Index is an indicator produced by TRMM and constitutes the principal empirical input for the assessment. It provides a structured basis for examining changes in reported asset ownership across successive observations. The source is used as a consistent measurement series, with attention to the comparability of its categories and the continuity of its observations.\n\nThe analysis does not reproduce isolated entries from the index as standalone findings. Instead, the observations are consolidated to calculate an average level and to identify the direction and persistence of the underlying trend. This treatment reduces the influence of short-term variation and provides a summary measure suitable for comparison across reporting periods.\n\nThe resulting estimates are used to support the assessment of access to and accumulation of productive assets. Particular attention is given to whether movements in the calculated series are sustained rather than attributable to a single observation. The Rolling Asset Ownership Index therefore serves as the central evidence for the quantitative discussion, while the computed average and trend provide the basis for interpretation and implementation monitoring.\n\nMention: \"Rolling Asset Ownership Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage explains that the assessment uses the Rolling Asset Ownership Index to calculate an average and derive a trend rather than reproduce individual observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:dbe2bbceb678e6354f20854b", "occurrence_id": "670f7af2f2418cb75c296551", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Clustered Land Cover Registry (CLCR) database to support the classification of observed land-cover categories. The source is treated as a structured reference for distinguishing persistent classes from areas subject to recorded change, thereby informing the interpretation of the estimates presented in this section.\n\nThe underlying records were collected from administrative and technical returns, then harmonised to improve consistency in category definitions, naming conventions, and treatment of incomplete observations. During preparation, duplicate entries were reviewed, incompatible labels were aligned, and supporting metadata were retained with the corresponding records. The resulting archive provides a controlled basis for comparing classifications across the analytical file.\n\nUse of the CLCR database does not replace field verification or the quality review applied to the wider assessment. It instead provides a documented evidentiary layer for the claim that the reported distribution reflects a consistent application of land-cover categories. Where records remain incomplete, the affected observations are retained for transparency and are not treated as evidence of a substantive change. This approach supports reproducibility while limiting interpretation to the coverage and quality of the archived material.\n\nMention: \"Clustered Land Cover Registry (CLCR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The passage uses the CLCR database as a collected, harmonised, and archived source supporting consistent land-cover classification.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:2b7e29ba979ef9e20a5cf981", "occurrence_id": "1632770ff322759918b07590", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The principal reference for the boundary review is the Quarterly Administrative Boundary System (QABS), an administrative source produced by the International Water Association. The material is used as a fixed reference for tabulating boundary structures and checking the internal consistency of the accompanying records. Its use does not imply interpretation beyond the figures reproduced below.\n\nTwo figures are reproduced directly from this source and retained in their reported form. It records 31 administrative grades and 84 boundary segments. The first figure provides the count of grades represented in the system, while the second gives the count of segments associated with those grades. No recalculation, projection, or adjustment is applied to either value.\n\nTogether, these figures provide the quantitative basis for a descriptive assessment of structural coverage. The counts are not treated as a performance indicator, since grades and segments represent different units of observation. The results therefore support a descriptive reading only, with the QABS record serving as the evidentiary basis for the two reported quantities. Subsequent interpretation should preserve the distinction between the two measures and avoid inferring changes not recorded in the source.\n\nMention: \"Quarterly Administrative Boundary System (QABS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the QABS figures of 31 administrative grades and 84 boundary segments.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:65854f914720921889bcd79a", "occurrence_id": "4aa073023403cf7904e57665", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Continuous Temperature Records Registry (CTRR), a database, as the sole source for the temperature observations cited in this note. The reference is to the 2023 registry, and the source is retained in its stated form to support traceability of the reported values.\n\nTwo readings are reproduced directly from the database without recalculation or interpolation. The first figure is 18.4 degrees Celsius, while the second figure is 23.7 degrees Celsius. These values constitute the principal quantitative evidence considered in the assessment.\n\nThe figures are presented as recorded observations rather than as derived estimates. No additional adjustment is applied, and no inference is made beyond the values shown in the CTRR. This treatment limits the discussion to the evidence available in the cited record and avoids introducing assumptions not supported by the source.\n\nFor reporting purposes, the two readings provide a concise basis for documenting the observed temperature conditions. Their inclusion permits verification against the Continuous Temperature Records Registry (CTRR) database and preserves a direct link between the narrative assessment and the underlying record.\n\nMention: \"Continuous Temperature Records Registry (CTRR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage reports the two CTRR figures, 18.4 degrees Celsius and 23.7 degrees Celsius, taken directly from the 2023 database.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:2b650b4443660b46f05aec72", "occurrence_id": "dc7b2bbaf909c32fe2a77309", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Departmental Enrolment and Retention Registry database is the administrative source for the 2020 indicators presented in this annex. It provides the reference point for assessing participation and continuity, while limiting the analysis to the measures recorded in the registry. No adjustment has been made to the reported values.\n\nThe enrolment rate was 68 percent, while the retention rate was 74 percent. Both figures are reproduced directly from the database and support the broader assessment that participation was maintained beyond initial registration. The relationship between the two indicators is used here to substantiate that assessment, rather than to derive an additional series or estimate.\n\nThe results should therefore be read as reported administrative measures. They provide a concise basis for monitoring enrolment and retention and for identifying whether the recorded pattern is consistent with continued engagement. Interpretation remains subject to the coverage and definitions applied in the registry, but the two cited figures are sufficient for the limited claim made in this paragraph.\n\nMention: \"Departmental Enrolment and Retention Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Departmental Enrolment and Retention Registry database to report 68 percent enrolment and 74 percent retention for 2020.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:1fd4b592890e3fdf572bb421", "occurrence_id": "0987c88fd49f17b2db4ddf5d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Integrated Diagnostic Capacity Survey (IDCS) is treated here as the relevant 2019 survey for the capacity assessment. Polity IV, identified as the organisation that produces it, provides the source record used for the comparison. The source is cited in full to keep the evidentiary basis of the assessment clear and to distinguish reported observations from subsequent interpretation.\n\nThe assessment quotes two figures directly from the source, namely the 31 grades recorded for the diagnostic scale and the 68 percent share reported for satisfactory coverage. Taken together, these observations support the broader conclusion that diagnostic capacity remains uneven, while avoiding any recalculation or extension of the reported results. The figures therefore function as direct evidence for the assessment rather than as inputs to a separate statistical estimate.\n\nThe findings are presented with limited interpretation. The reported grades indicate the structure of the assessment, and the reported percentage provides a concise measure of coverage within the same evidentiary framework. No adjustment is applied to either figure, and no additional classification is inferred from the survey record. This treatment preserves the distinction between information contained in the Integrated Diagnostic Capacity Survey (IDCS) and conclusions drawn in the surrounding analysis.\n\nMention: \"Integrated Diagnostic Capacity Survey (IDCS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage identifies the 2019 Integrated Diagnostic Capacity Survey (IDCS), produced by Polity IV, and uses its directly reported figures of 31 grades and 68 percent to support a broader capacity finding.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:37449494633eb0cd67beef7f", "occurrence_id": "1b704c1db467b6b8f5707df4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Multi-Topic Early Grade Reading Survey is used as the principal evidence for the assessment of early reading performance. It is a survey produced by OECD-PISA and is treated as a structured source for examining patterns across the observed learning measures. No single reported observation is presented as representative of the overall result. The analysis instead derives a trend and an average from the underlying responses, allowing the assessment to distinguish broad movement from isolated variation.\n\nThe derived trend indicates the direction of change across the measured reading outcomes, while the average provides a summary of the level attained over the analytical set. These calculations are used to support implementation judgments concerning consistency, progression and the distribution of performance. The resulting interpretation therefore reflects the combined evidence in the survey rather than a selected quotation or an individual response.\n\nThe findings should be read as an analytical synthesis of the source material. Differences among observations may reflect variation in learning conditions, respondent characteristics or measurement circumstances, and the computed indicators do not remove that underlying variation. Nevertheless, the common calculation provides a stable basis for monitoring results and for identifying areas requiring further implementation attention. The survey is consequently retained as the central empirical reference for this assessment.\n\nMention: \"Multi-Topic Early Grade Reading Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage derives a trend and an average from the Multi-Topic Early Grade Reading Survey rather than quoting individual observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:627daf3748a1a0ff82b97c00", "occurrence_id": "1a46df19a4f6d4900582752b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment considers classroom practice as one dimension of implementation monitoring. This dimension is relevant to the interpretation of findings because observed practice can mediate the relationship between planned activities and reported results. The discussion therefore distinguishes contextual evidence from material used in estimation and does not infer levels, changes, or comparisons from the source described below.\n\nThe Endline Classroom Practice System is cited as an administrative source that situates classroom practice within the broader monitoring framework. Its inclusion supports the observation that implementation quality is considered alongside outputs and outcomes. No value from this source is incorporated into the estimates, and the citation is not used to establish a distribution, calculate a difference, or substantiate a trend. Its role is limited to reinforcing the relevance of classroom practice as a subject of assessment.\n\nThe remaining analysis relies on the indicators and procedures specified for the assessment. Interpretive statements are kept proportionate to the available evidence, with contextual references separated from numerical results. This treatment reduces the risk that a descriptive source will be read as an input to the statistical calculations or as an independent basis for conclusions about performance. The source is consequently retained in the evidence narrative, while the analytical tables continue to report only measures directly required for the assessment.\n\nMention: \"Endline Classroom Practice System\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The passage uses the Endline Classroom Practice System to support a contextual claim about classroom practice, without taking values from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:b75985d1259c0578967996da", "occurrence_id": "ee2a0cd0cd8baf6a7c23bc6b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Subnational Population Grid Registry database produced by Rhodes University for 2017 is used as the principal statistical reference for the assessment. Its gridded observations provide a consistent basis for reviewing the distribution of population across reporting units, while limiting the need for further interpretation of the underlying counts. The source is treated as a descriptive record rather than as a basis for projecting subsequent change.\n\nThe reported totals are reproduced directly from the database. The first observation records 31,482,000 persons, while the second records 4,217,000 persons. These figures constitute the central evidence for the comparison presented in this assessment and are retained in their reported form to preserve traceability to the cited source. No adjustment is applied to either value within the analysis.\n\nThe difference between the two observations is therefore presented as a contrast in recorded population totals, not as an estimated rate or an independently calculated series. Interpretation remains limited to the information contained in the registry, with attention to the stated unit of observation and the possibility that grid-based counts may conceal variation within individual reporting units. The figures should accordingly be read as source values supporting the assessment rather than as forecasts.\n\nMention: \"Subnational Population Grid Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the 2017 Subnational Population Grid Registry database produced by Rhodes University and reproduces 31,482,000 and 4,217,000 as its two principal figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:0f8782174d88a0f35861c62a", "occurrence_id": "5040786000592c167a258b04", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Baseline Aquaculture Output Survey, a survey produced by UNU-WIDER, as the central empirical input for measuring changes in aquaculture output. The source is treated as a consistent body of observations for analytical purposes, rather than as a set of isolated figures to be reproduced in the narrative.\n\nOutput observations are reviewed for internal consistency and arranged to support comparison across reporting units. The assessment then derives an average level from the observations and calculates the direction and relative pace of change over the available sequence. This treatment reduces the influence of individual fluctuations and provides a more stable basis for interpreting sector performance.\n\nThe resulting trend is used to inform the assessment of production dynamics, implementation progress, and the plausibility of observed changes. Individual observations remain available for verification, but the conclusions presented here rely on the calculated summary measures. The approach therefore gives priority to the pattern generated by the survey as a whole, while avoiding unsupported inference from any single reported observation.\n\nMention: \"Baseline Aquaculture Output Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives an average and a production trend from the Baseline Aquaculture Output Survey rather than reproducing individual figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:7611e9a57104eda0cb3b5dd2", "occurrence_id": "ba0f3a98c3efa10e6ab2dd07", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Stratified Informal Trade Census (SITC) is a census produced by the Department of National Planning for Guatemala in 2022. It provides the principal statistical basis for the observations reported in this section, with the source designation retained to support verification and consistent interpretation.\n\nTwo figures are reproduced directly from the census. The first records 31,482 informal trading units, indicating the measured scale of activity within the covered population. The second reports that 64.2 percent of the recorded operators were women. These values are presented as reported in the source and are not adjusted through additional estimation.\n\nThe figures provide a concise indication of both the volume and composition of informal trade in Guatemala. Their use is limited to descriptive reporting, and no inference is made beyond the scope of the census coverage. The source therefore serves as the reference point for the numerical statements in this assessment.\n\nMention: \"Stratified Informal Trade Census (SITC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Stratified Informal Trade Census (SITC) to report 31,482 informal trading units and 64.2 percent women operators in Guatemala in 2022.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:9dba336d8caec089a082c0d5", "occurrence_id": "a3fe5cc9d626a632d99ab9c7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Subnational Disability and Support Survey (SDSS) is the survey produced by ITU and serves as the principal evidence base for the assessment of disability-related support conditions. Its observations are treated as structured administrative evidence, with attention to consistency across reporting units and to the comparability of responses. The source is cited here to establish the basis of the analysis and not to reproduce isolated entries.\n\nThe assessment derives an average from the SDSS observations and uses the resulting measure to calculate a trend across the available reporting sequence. This approach reduces the influence of irregular responses and places emphasis on the direction and scale of change. The calculated results are used to assess whether support conditions are becoming more consistent, whether disparities persist, and whether implementation priorities require adjustment.\n\nInterpretation remains limited to the information contained in the Subnational Disability and Support Survey (SDSS). No additional assumptions are introduced regarding unobserved populations or conditions outside the survey framework. The derived average and trend therefore function as summary measures for monitoring purposes, while the underlying survey remains the central evidence for the assessment. Results should be read as analytical estimates generated from the source rather than as direct reproductions of individual survey records.\n\nMention: \"Subnational Disability and Support Survey (SDSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses ITU’s SDSS survey to calculate an average and derive a trend for implementation monitoring.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:71d937bc1f6af6638abc8d67", "occurrence_id": "018f3749d6b0b0786bcfd307", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis treats multidimensional deprivation as a condition reflected across several non-monetary dimensions, rather than as a single monetary outcome. This framing supports the finding that observed disadvantage is distributed unevenly among population groups and requires coordinated reading of measures.\n\nFor the supporting evidence, the assessment draws on the Cross-Sectoral Multidimensional Poverty Collection (CMPC), classified as other in the documentary inventory. The records were collected through a structured review, harmonised to improve comparability across measures, and archived with the accompanying documentation. These procedures provide a traceable basis for using the collection within the evidence set while preserving distinctions among the underlying measures.\n\nThe CMPC is used here to substantiate the broader interpretation that multidimensional poverty cannot be adequately represented by one indicator alone. It informs the consolidation of evidence on overlapping deprivations and helps identify where separate measures refer to related aspects of household welfare. The collection is therefore treated as supporting evidence for the assessment, not as a substitute for the analytical framework or for subsequent validation.\n\nIn implementation terms, the recordkeeping approach strengthens reproducibility and facilitates review of the evidence used in reporting. Any comparison drawn from the material remains subject to the definitions and coverage of the archived records, and findings are stated with appropriate caution.\n\nMention: \"Cross-Sectoral Multidimensional Poverty Collection (CMPC)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The passage uses harmonised and archived CMPC evidence on overlapping deprivations to support the claim that multidimensional poverty requires more than one indicator.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:6ad66415c1ff8d34a343294e", "occurrence_id": "1a62f53040e1e33c85bfe9bb", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Diagnostic Informal Trade System, an administrative source produced by the organisation Reinhart & Rogoff and referring to 2017. The source provides the principal empirical basis for examining recorded informal trade activity and for establishing comparable observations across the reporting framework.\n\nIndividual entries are not reproduced as reported figures. Instead, the assessment aggregates the observations and derives an average for the relevant measures, allowing the distribution of activity to be considered without attributing undue weight to isolated records. A trend is calculated from the grouped observations to identify the direction and relative stability of reported activity over the analytical sequence.\n\nThe resulting estimates are used to support the implementation assessment and to inform interpretation of operational performance. Differences between individual observations are retained in the calculations, while the presentation focuses on summary movements and average levels. This treatment reduces the risk that exceptional entries will determine the findings and provides a consistent basis for comparison within the available evidence.\n\nThe source is therefore used as a structured administrative record from which analytical measures are derived, rather than as a set of figures reproduced directly. The calculated trend and average constitute the central evidence for the conclusions presented in this section.\n\nMention: \"Diagnostic Informal Trade System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The assessment derives a trend and an average from the Diagnostic Informal Trade System rather than reproducing individual figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:ec7fa3499be673ac37b13058", "occurrence_id": "de536be302cdba34fb406311", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Harmonised Debt and Savings Survey is used as the principal survey source for the assessment of household financial conditions. Its observations provide the central evidence for examining the movement of debt and saving outcomes across the analytical series. The material is read as a consistent body of survey responses, with attention to comparability across observations and to the treatment of missing or irregular records.\n\nThe reported results are derived measures rather than direct reproductions of individual entries. For each analytical interval, the observations are aggregated to obtain an average, and the resulting sequence is examined to identify the direction and persistence of the trend. This procedure limits the influence of isolated responses and places emphasis on the underlying pattern. The assessment therefore presents the source through calculated summaries, while retaining the survey as the basis for all reported movements.\n\nInterpretation is restricted to changes supported by the observed sequence. In particular, the analysis distinguishes a sustained trend from a temporary fluctuation and treats the average as a descriptive measure rather than a forecast. No inference is made beyond the information contained in the survey responses. On this basis, the derived trend and average are used to support the implementation assessment and to guide the reading of subsequent results.\n\nMention: \"Harmonised Debt and Savings Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives an average and a trend from the Harmonised Debt and Savings Survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:e9b079334101472341493663", "occurrence_id": "15ce711c0382013633825401", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Representative Remote Sensing Dataset, a geospatial source used to support measurement of observed conditions across the analytical frame. Its records provide a consistent basis for comparing reported indicators and for identifying changes in their distribution over time. The dataset is treated as the principal empirical input for this part of the assessment, while interpretation remains limited to patterns that can be supported by the available observations.\n\nThe analysis does not reproduce isolated entries as headline results. Instead, the observations are cleaned, aligned, and aggregated to derive an average level and a trend for the indicators under review. This approach reduces the influence of individual anomalies and permits comparison between successive observation intervals. The resulting measures are therefore analytical estimates derived from the source rather than figures transcribed from it.\n\nThe derived trend indicates the direction and relative pace of change, while the calculated average provides a summary measure against which variation can be assessed. Together, these measures form the central evidence for the findings presented in this section. Limitations relate principally to the completeness and comparability of the underlying observations, and conclusions are consequently stated at the level supported by the aggregated results. Further disaggregation is not undertaken where the source does not provide a sufficiently consistent basis for interpretation.\n\nMention: \"Representative Remote Sensing Dataset\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage derives an average and a trend from the Representative Remote Sensing Dataset rather than reproducing individual entries.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:81e9979cc17ae590393d3180", "occurrence_id": "6a74fe9ca6f3fd90a4854e0d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Standardised Soil Fertility Survey (SSFS) is used as the principal survey for the soil fertility observations considered in this assessment. Its tabulations provide a consistent basis for reporting the distribution of fertility conditions across the sampled plots. The evidence is presented in the form available in the survey, without additional classification or adjustment.\n\nTwo figures are reproduced directly from the SSFS. The survey reports that 47 percent of sampled plots fall within the adequate fertility category, while 22 percent fall within the low fertility category. These percentages are retained as reported and are treated as the central quantitative evidence for the assessment of soil conditions.\n\nThe difference between the two reported shares indicates that adequate fertility is more common than low fertility within the observations covered by the survey. No further calculation is required for the purpose of this note, and the figures should not be interpreted as a measure of change over time. Their function is descriptive, providing a concise statistical reference for the discussion of fertility status and for the interpretation of related implementation results.\n\nMention: \"Standardised Soil Fertility Survey (SSFS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the 47 percent adequate-fertility figure and the 22 percent low-fertility figure directly from the SSFS.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:f6ff5353a5d8687cfed8e603", "occurrence_id": "2e37a8e8dc06af8e3fc82d41", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis draws on the Combined Digital Learning System, an administrative source produced by Google. The source provides the central evidentiary basis for assessing changes in recorded digital learning activity. Its entries are used as observations for analytical purposes rather than reproduced as reported figures.\n\nRecords were reviewed for consistency and grouped into comparable observation periods. The assessment then computes a period-to-period trend from the underlying entries and derives an average across the available observations. These calculations support a measured interpretation of variation and reduce the risk that an isolated entry is treated as representative of overall performance.\n\nThe resulting measures indicate the direction and general level of activity captured by the system. The trend is considered alongside the average to distinguish sustained movement from short-term fluctuation. No single record is assigned interpretive weight beyond its contribution to the aggregate calculation, and the findings are therefore presented as summary evidence rather than as direct quotations from the source.\n\nThis treatment supports monitoring of implementation progress using a consistent administrative record. The calculated measures provide a basis for comparison within the assessment and will be updated when additional observations become available.\n\nMention: \"Combined Digital Learning System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses the Combined Digital Learning System to calculate a period-to-period trend and an average from its administrative records.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:6bd24dae3025d785e9e412d3", "occurrence_id": "1c8ec61cc9d9e5fb2c7dd323", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Verified Working Conditions Survey (VWCS) is used as a survey for the assessment of reported workplace conditions. The VWCS is treated as the primary evidentiary source for the indicators presented in this note, and its reported values are retained without adjustment or extrapolation.\n\nTwo reported percentages are reproduced directly from the survey. The first indicates that 68 percent of respondents reported access to written employment terms. The second indicates that 31 percent reported receiving information on workplace safety procedures. These figures are presented as reported and are not recalculated from supporting variables.\n\nThe results provide a limited but clearly defined basis for assessing the prevalence of formal communication and safety information. Their use is therefore descriptive rather than predictive. Differences between the two percentages should be read as a contrast in reported conditions within the survey evidence, not as a measure of change over time or as an estimate generated through external modelling. The source supports the comparison because both figures are taken directly from the same survey and expressed using the same reporting basis.\n\nMention: \"Verified Working Conditions Survey (VWCS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the 68 percent and 31 percent findings directly from the VWCS.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:69686078e2b1613069aa98b9", "occurrence_id": "60cd141e862183bcb9e5ba44", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Baseline Asset Ownership Survey, a survey produced by IUCN for 2017. The source provides the underlying observations used to examine patterns in asset ownership and to support the related interpretation of household economic conditions. Its inclusion is limited to this analytical purpose within the broader assessment.\n\nThe reported result is derived through calculation rather than reproduction of individual entries. Observations from the survey are consolidated to obtain an average level and to identify the direction of change across the measured categories. This treatment reduces the influence of isolated responses and provides a consistent basis for comparison within the analysis.\n\nThe resulting trend is used to support the conclusion that asset ownership conditions changed unevenly across the assessment period. The calculated average is presented as an analytical measure, not as a figure directly quoted from the source. Accordingly, the finding should be read as an interpretation based on the survey evidence and not as a separate administrative count.\n\nMention: \"Baseline Asset Ownership Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives an average and a directional trend from the 2017 survey evidence rather than quoting individual observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:83d20a8064877aeef8dd0f86", "occurrence_id": "afa34731565afd9cb5a8af84", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The evidence base for this assessment is the Midline Disability and Support Survey (MDSS). CEPAL produces this survey, which is used here as the principal source for examining reported disability and support conditions. The source is treated as a statistical input to the assessment rather than as a narrative account, and its designation is retained in full to support traceability.\n\nThe analysis derives a sequence of comparable estimates from the MDSS observations and calculates an average across the available reporting units. It also calculates the direction and magnitude of change across the ordered observations, thereby producing a trend measure from the source. These derived measures, rather than isolated figures reproduced from the survey, provide the central basis for assessing variation in reported conditions.\n\nResults are interpreted at an aggregate level. The calculated average summarises the general level observed in the survey, while the trend indicates whether reported conditions move upward or downward across the observations. This treatment supports comparison within the source and limits reliance on any single response or isolated estimate. No separate claims are made beyond the patterns supported by the calculated measures.\n\nMention: \"Midline Disability and Support Survey (MDSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses the MDSS survey produced by CEPAL to calculate an aggregate average and a trend from its observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:f9f91e43e7eec11674022cdb", "occurrence_id": "7398f724bb26c18ee6ed5f5b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Linked Collective Bargaining Survey, a survey produced by Economist covering Zambia. The source is used as the principal evidence for describing the reported reach of collective bargaining. Its identification is retained in the text to support traceability and to distinguish the cited observations from analytical judgments made in this assessment.\n\nTwo figures are reproduced directly from the source without recalculation or adjustment. The first indicates that 31 per cent of employees were covered by a collective agreement, while the second indicates that 18 per cent were covered in the comparison reported by the survey. These values provide the central quantitative basis for the assessment of bargaining coverage and are presented as reported rather than derived estimates.\n\nThe difference between the two figures is therefore treated as an observed contrast within the cited material, not as a trend calculated from additional observations. Interpretation remains limited to the scope of the reported measures. In particular, the figures should not be read as evidence of changes over time, causal effects, or conditions outside the population represented in the survey. Their value for implementation monitoring lies in providing a clearly identified reference point for subsequent review of collective bargaining conditions.\n\nMention: \"Linked Collective Bargaining Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes 31 per cent and 18 per cent directly from the Linked Collective Bargaining Survey covering Zambia.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:f85d69cba78368fa4fa446f8", "occurrence_id": "b53c2221ef547ad9e49d9c97", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis uses the Household Deprivation and Hardship Survey (HDHS), a survey designed to support measurement of material deprivation and household hardship. The source provides the underlying observations for the assessment of changes in reported conditions across the observation period. Its use is limited to the evidence required for the related analytical claim.\n\nTo reduce the influence of short-term variation, the analysis derives an average from the available observations and calculates a trend across the successive periods. The resulting measure is used to indicate the general direction and level of hardship, rather than to reproduce individual survey responses. This treatment supports comparison within the analytical framework while preserving the distinction between reported observations and derived results.\n\nThe derived average indicates the central level observed over the period, while the trend measure identifies whether conditions improved, deteriorated, or remained broadly stable. These outputs are interpreted alongside the wider assessment and are not presented as additional observations from the HDHS. The approach therefore provides a concise statistical basis for the claim concerning the persistence and direction of household hardship.\n\nMention: \"Household Deprivation and Hardship Survey (HDHS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives an average and a trend measure from the Household Deprivation and Hardship Survey (HDHS) to assess the level and direction of household hardship.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:62998eef1126320672d56cb6", "occurrence_id": "a99bb062d5f6ff8c5ed67678", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Record-Based Disability and Support Registry (RDSR) database is used as the administrative source for the 2019 reference observations. It provides a consolidated record of persons identified as requiring support and of the support measures recorded against those cases. The source is treated as the primary evidentiary basis for the limited quantitative statements presented in this section.\n\nThe assessment quotes two figures directly from this database. It records 31 individuals as having an assessed need for support, while 14 individuals are recorded as receiving a support measure. These figures are reproduced without recalculation or adjustment and are used to establish the observed scale of registered need and recorded assistance.\n\nThe figures should be interpreted as registry counts rather than as estimates of the total population with disabilities or support requirements. Coverage depends on entry into the administrative system and on the completeness of the information submitted for each individual. The difference between the two counts therefore indicates the relationship between recorded need and recorded support within the database, but does not by itself establish the reason for that difference.\n\nFor monitoring purposes, the registry remains suitable for reporting the number of cases recorded and the number linked to a support measure. Any subsequent comparison should retain the same definitions and counting conventions so that changes in administrative coverage are not mistaken for changes in underlying need.\n\nMention: \"Record-Based Disability and Support Registry (RDSR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the 2019 RDSR figures of 31 individuals with an assessed support need and 14 individuals receiving a recorded support measure.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:6c4591d3144cf40e4c506a96", "occurrence_id": "216db21c4eef5be9f2280878", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Data sources.** The Decennial Mechanisation Registry (DMR) is treated as the principal database for the verification exercise. Its records provide an independent reference against which the estimate used in the assessment is cross-checked. The comparison is intended to identify material divergence in the reported scale of mechanisation and to assess whether the estimate remains within a plausible range.\n\nThe DMR database is reviewed for internal consistency before the comparison is made. Attention is given to the distribution of recorded units, the presence of repeated observations, and the treatment of incomplete entries. These checks do not alter the estimate drawn from the separate source. They provide a structured basis for determining whether differences reflect measurement conditions or a substantive discrepancy.\n\nThe cross-check indicates that the estimate is broadly compatible with the pattern observed in the registry, subject to the limitations of administrative recording. The DMR is therefore used as corroborating evidence rather than as a replacement series. Where the two sources diverge, the assessment retains the independently derived estimate and records the divergence as a matter for interpretation. This approach supports a cautious reading of the results and limits conclusions to findings that are supported by both the estimate and the database.\n\nMention: \"Decennial Mechanisation Registry (DMR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the Decennial Mechanisation Registry (DMR) database to cross-check an independently derived estimate and assess its consistency.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:52cea33b274e03a97661232f", "occurrence_id": "1b9fb37705a87aaaac713f25", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Verified Minimum Needs Survey (VMNS), a survey covering the United States in 2016, constitutes the principal empirical basis for the assessment. Its observations are used to establish a consistent measure of minimum needs across the reporting period and to support comparisons among population groups. The source is treated as the underlying evidence for the calculations rather than as a collection of isolated figures.\n\nThe analysis computes average need levels from the VMNS observations and derives a trend by aggregating records into comparable intervals. This procedure reduces the influence of individual responses and provides a summary measure suitable for monitoring changes in household requirements. Reported values therefore represent calculated results based on the survey distribution, not direct reproductions of selected entries.\n\nThe resulting estimates are used to assess the direction and approximate scale of change in minimum needs. Interpretation remains subject to the coverage and measurement properties of the source, but the computed average and trend provide the principal reference for the assessment. All subsequent comparisons rely on these derived measures and retain the survey as their evidentiary foundation.\n\nMention: \"Verified Minimum Needs Survey (VMNS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives trend estimates and average need levels from the 2016 United States VMNS survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:63c41f5319829dac8313de21", "occurrence_id": "de132949f4710aef49426574", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The discussion draws on the Standardised Child Wellbeing Registry (SCWR) database to establish the relevance of child wellbeing measures to the assessment framework. The source is used for contextual support within the broader analysis and does not provide the basis for any calculation, comparison, or reported estimate in this section. Its inclusion indicates that child wellbeing is recognised as a relevant dimension of implementation performance.\n\nThe associated claim is limited to the general applicability of wellbeing considerations in interpreting service results. No numerical, categorical, or disaggregated values from the database are used, reproduced, or inferred. The analysis therefore remains focused on the implementation evidence presented elsewhere in the report, while acknowledging the wider measurement framework in which that evidence may be considered.\n\nThis treatment is consistent with a cautious use of administrative and monitoring resources. The cited database provides background for interpretation rather than an independent result, and no conclusion is drawn from an unreported registry value. The distinction is maintained to avoid implying coverage, change, or performance levels that are not documented in the material reviewed.\n\nMention: \"Standardised Child Wellbeing Registry (SCWR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The source provides contextual support for a child wellbeing claim, while no values are taken from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:50295d2b50ea4e735b8c395f", "occurrence_id": "46a04e4b39ec87676eb862d9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Multi-Topic Working Conditions Survey is used as the principal statistical reference for this assessment. It is a survey produced by the CCCM Cluster for Thailand and refers to 2018. The source provides a standardized account of reported working conditions and supports a narrowly defined comparison of the measures presented below.\n\nThe evidence retained from the source consists of two reported figures. Thirty-one percent of respondents indicated that working conditions met the specified threshold, while 68 percent reported conditions outside that threshold. These percentages are reproduced directly from the survey record, without recalculation or interpolation. Their difference is therefore presented as an observed contrast within the source rather than as an independently derived estimate.\n\nThe figures should be read as descriptive indicators of the surveyed population. They do not, on their own, establish a causal explanation for the gap or permit conclusions beyond the coverage and definitions applied in the survey. Their value for implementation monitoring lies in providing a consistent reference point against which subsequent observations may be considered, subject to comparable measurement and interpretation. Any use in planning should retain the original denominator and response definitions to avoid overstating the precision of the comparison.\n\nMention: \"Multi-Topic Working Conditions Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Multi-Topic Working Conditions Survey to report directly the 31 percent and 68 percent figures for Thailand in 2018.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:0e1ee4a7f6ac662361a09437", "occurrence_id": "d1c2201aaf8be0514d0c52cf", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Record-Based Clinic Quality Index (RCQI) is used as the indicator for 2017. It provides a bounded measure of recorded clinic quality and is treated as the principal evidentiary basis for the assessment. No supplementary source is introduced for the values cited below.\n\nThe first value reproduced from the Record-Based Clinic Quality Index (RCQI) is 62 percent. This figure is retained in its reported form and is used to establish the observed level of quality in the assessment. It is not recalculated, adjusted, or combined with an external estimate.\n\nThe second value taken directly from the same indicator is 74 percent. Considered together, the two reported figures show a material difference between the measured dimensions covered by the index. The difference is presented descriptively and does not constitute a computed trend or an estimate of change over time.\n\nThese observations support use of the Record-Based Clinic Quality Index (RCQI) as a consistent reference for interpreting recorded performance. The figures are cited as published values, while conclusions remain limited to the evidence contained in the indicator for 2017. The assessment therefore relies on direct reproduction of the two values rather than on additional processing or attribution.\n\nMention: \"Record-Based Clinic Quality Index (RCQI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the RCQI indicator for 2017 and takes the values 62 percent and 74 percent directly from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:dd9b65b63bc16ef71c0fc6f4", "occurrence_id": "ec4c47c1eb3206a927f8d450", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Subnational Parcel Footprint Index (SPFI) is an indicator produced by UNESCO for structured comparison of parcel footprints. It is cited as the data source for the limited quantitative evidence used in this assessment. The source is retained in its stated form to support traceability and to avoid introducing adjustments that are not documented in the present analysis.\n\nThe assessment considers parcel footprint intensity as one element of the broader review of spatial structure and service access. On this basis, the available evidence indicates that footprint conditions are not uniform across the units considered. Two figures are reproduced directly from the Subnational Parcel Footprint Index (SPFI), namely 42.6 and 17.3. These values are used as reported and are not recalculated, combined, or extended into a separate series.\n\nThe figures provide a concise empirical basis for the related finding that differences in parcel configuration may affect the interpretation of aggregate results. They are therefore presented as supporting evidence for that claim rather than as a complete measure of performance. The indicator should be read alongside the qualitative findings, since the two reported values alone do not establish the causes of the observed variation. No further inference is drawn from the source beyond the limited comparison required for this assessment.\n\nMention: \"Subnational Parcel Footprint Index (SPFI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the directly reported SPFI figures 42.6 and 17.3 to support the finding that parcel footprint conditions vary across the units considered.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:a4119f74743b079113edadf1", "occurrence_id": "76f4a956aa4365237cc29c2d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Data sources.** The Coordinated Clinic Quality Survey (CCQS) is a survey identified in its Fifth Phase and is retained in the assessment as a contextual reference. The citation establishes the designated stage of the source, while no numerical or categorical content from the survey is incorporated into the analysis. Its mention is therefore limited to orientation within the evidence base.\n\nThe findings presented in this annex rely on the assessment framework and the explanatory text accompanying the results. The cited material is not used to substantiate a comparison, establish a distribution, or support an estimate. No counts, rates, scores, rankings, or institutional attributions are drawn from the CCQS reference. This treatment avoids implying that the cited survey provides an evidentiary basis for conclusions set out elsewhere in the document.\n\nThe reference is retained to signal continuity with background documentation and to permit readers to distinguish contextual material from evidence used in the assessment. It does not enter the tabulations, calculations, or interpretation of observed results. Accordingly, the discussion proceeds without reproducing, transforming, or comparing any content attributed to the cited source. The citation has a documentary function only and should not be read as a result, an indicator, or a basis for inference.\n\nMention: \"Coordinated Clinic Quality Survey (CCQS)\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"No values or analytical findings were taken from the Coordinated Clinic Quality Survey (CCQS); it was retained solely as background documentation.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:6c3a76f1a74860fa3532d71b", "occurrence_id": "d0771eb868087125f712f5fa", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis uses the Joint Nutrition Surveillance Survey as the principal statistical basis for the assessment. This survey was produced by PRIO/Uppsala and refers to 2020. The source is retained in its reported form to support a transparent reading of the available evidence.\n\nTwo figures are reproduced directly from the source, namely 31.4 percent and 18.7 percent. They are presented without recalculation, adjustment, or conversion, so that the numerical record remains consistent with the cited material. The figures provide the central quantitative reference for the discussion that follows.\n\nThe evidence is used narrowly for documentation and comparison within the assessment. No additional estimate is derived from the two reported values, and no inference is made beyond what the survey records. This approach limits interpretive expansion while preserving the utility of the source for monitoring and review.\n\nTaken together, the cited figures establish a concise statistical basis for the relevant findings. Their direct reproduction also permits subsequent reviewers to distinguish reported evidence from analytical interpretation and to verify that the assessment does not introduce an independent calculation.\n\nMention: \"Joint Nutrition Surveillance Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported figures of 31.4 percent and 18.7 percent from the 2020 Joint Nutrition Surveillance Survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:544517208f4bdb04d8edc895", "occurrence_id": "d4c54a73f12cdc5bf92cc9f6", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Rural Terrain and Elevation System (RTES) as an administrative source for the characterization of terrain conditions relevant to service access. The material was collected through a structured records process and retained in the project archive. Its inclusion provides a consistent basis for relating observed differences in accessibility to underlying physical conditions.\n\nThe records were harmonised before analysis to align geographic references, measurement conventions, and coverage across the reporting units. Quality checks addressed duplicate entries, incomplete observations, and inconsistencies in the representation of elevation and terrain categories. Where records could not be reconciled, they were retained separately and excluded from aggregate comparisons rather than adjusted through unsupported assumptions.\n\nThe resulting archive supports the interpretation of access patterns presented in the assessment and strengthens the reproducibility of the underlying evidence. RTES is used here to substantiate the claim that variation in terrain conditions is associated with differences in the feasibility and cost of reaching services. The source is not treated as a measure of service performance; it provides the physical reference layer against which the broader findings are assessed.\n\nMention: \"Rural Terrain and Elevation System (RTES)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The passage uses RTES for collecting, harmonising, quality-checking, and archiving administrative terrain and elevation records.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:04521b675fccae22c502adbd", "occurrence_id": "f62a5c5b306635356f941890", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Urban Inequality Survey is treated as the supporting source for the assessment of distributional conditions. It is a 2020 survey, and its reported results are used to substantiate the broader finding that unequal access remains a material constraint on welfare improvement. No adjustment, projection, or recalculation is applied to the cited observations.\n\nThe source records that 42 percent of surveyed households reported difficulty meeting essential expenditure, while 18 percent reported that assistance had become less accessible. These two reported values are reproduced directly from the Urban Inequality Survey and are presented as evidence within the broader assessment rather than as independently derived estimates.\n\nThe figures should be interpreted as descriptive measures for the population covered by the survey. They support the conclusion that aggregate improvements may coexist with persistent differences in household conditions. The analysis therefore uses the source to qualify the overall result and to indicate the scale of the distributional concern, while avoiding claims that cannot be established from the reported observations alone.\n\nMention: \"Urban Inequality Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Urban Inequality Survey to report directly the figures of 42 percent and 18 percent as evidence of persistent inequality.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:cdb8af53c05e74c9d660d1f0", "occurrence_id": "7db42f8a74486e77195e5a8b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Sentinel Heat and Cold Extremes Index (SHCEI) indicator provides the sole quantitative basis for this assessment. It is used to document the occurrence of exceptional temperature conditions and to support a concise comparison of heat-related and cold-related exposure. No supplementary source is required for the figures presented below.\n\nThe evidence comprises two values reproduced directly from the indicator. The recorded heat-related measure is 42 days, while the corresponding cold-related measure is 17 days. These figures are retained in their reported form and are not recalculated, smoothed, or combined into a derived measure. Their purpose is to establish the observed difference between the two categories using the source’s own reported values.\n\nThe resulting interpretation is limited to the information conveyed by these observations. Heat-related conditions account for the larger reported count, with a difference of 25 days relative to the cold-related count. This comparison is descriptive and should not be read as an estimate of causality, persistence, or future incidence. The SHCEI indicator is therefore cited as the direct evidentiary basis for the two values and for the limited comparison set out in this note.\n\nMention: \"Sentinel Heat and Cold Extremes Index (SHCEI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the SHCEI indicator and reports 42 heat-related days and 17 cold-related days directly from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:1c321290b3bd124bb6e13745", "occurrence_id": "8a04446d8f012b1f88c0d9cb", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Joint Demographic and Health Survey is identified in the supporting material as a 2015 survey relevant to the demographic setting considered in this assessment. The reference is retained to indicate the broader information base available for interpretation and to support a consistent reading of the surrounding narrative.\n\nNo numerical result, estimate, distribution, or other reported value from the Joint Demographic and Health Survey is used in the analysis presented here. The source is mentioned briefly and does not provide a basis for comparison, calculation, attribution, or inference. Its inclusion therefore has a documentary function rather than an analytical one.\n\nThe assessment relies on the evidence and procedures described in the main sections of the document. The cited survey is not used to establish a level, measure a change, validate an outcome, or supplement a calculation. This treatment limits the possibility that contextual reference could be read as evidence drawn from the source itself and preserves a clear distinction between background material and the results reported here.\n\nMention: \"Joint Demographic and Health Survey\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The Joint Demographic and Health Survey was used only for contextual background, and no values or analytical results were taken from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:8603992b5abe115633a6a094", "occurrence_id": "e51310729e94d7607bbe103a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The principal evidence for the assessment is the Linked Child Wellbeing Survey, a 2022 survey of child wellbeing outcomes. The source provides the observations used to establish comparable measures across the reporting period, while the present analysis treats those observations as inputs to estimation rather than as figures for direct reproduction.\n\nFor each outcome, the analysis calculates average levels and examines movement across successive observation periods. This approach reduces the influence of isolated responses and supports a more stable assessment of direction and magnitude. The resulting measures are used to identify whether wellbeing conditions improved, deteriorated, or remained broadly unchanged.\n\nThe computed averages and changes form the basis for the findings reported in this annex. They are interpreted alongside the consistency of responses and the distribution of observed outcomes, with attention to differences in the strength of the measured movement. No single observation is treated as representative on its own.\n\nThis treatment preserves the evidentiary role of the Linked Child Wellbeing Survey while ensuring that conclusions rest on derived patterns rather than on selected quotations from the underlying material. The estimates should therefore be read as analytical summaries of the survey evidence and not as independently collected administrative statistics.\n\nMention: \"Linked Child Wellbeing Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses the 2022 survey to derive average child wellbeing levels and trends rather than reproduce individual observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:84df30f3f686dc9f3bb7a80e", "occurrence_id": "c16d607964725a6c32a94b92", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Government Land Cover Change Registry database for Morocco covering 2019 as the underlying reference for land-cover conditions. The material is treated as a source for structured comparison across recorded classes and administrative reporting units, while the interpretation remains limited to the information required for the present assessment.\n\nThe analysis derives an average distribution from the registry entries and calculates a comparative trend across the recorded land-cover classes. These calculations provide an aggregate basis for assessing the relative prominence and direction of change represented in the underlying records, rather than reproducing individual observations as reported figures. The resulting measure is used to support the assessment of pressure on land resources and to inform the selection of activities for subsequent monitoring.\n\nThe derived indicators should therefore be read as analytical summaries of the cited database. They support one claim concerning the broad pattern of land-cover change, but they do not constitute an independent estimate of conditions outside the registry’s coverage. Interpretation is consequently confined to the documented entries and to the aggregation procedure applied in this assessment. This approach maintains comparability within the evidence base while avoiding unwarranted precision in the presentation of isolated records.\n\nMention: \"Government Land Cover Change Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses the Government Land Cover Change Registry database for Morocco and 2019 to derive an average and comparative trend rather than quote individual records.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:2eff13e99e430027ff18a4eb", "occurrence_id": "1e029226ed542112741cac13", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Nationwide Health and Household Survey (NHHS) provides a survey-based comparator for the coverage estimate reported in this section. Its results are used to cross-check the estimate derived from a separate records-based source, with attention to the direction and approximate scale of any difference. The comparison supports the claim that reported coverage is broadly consistent across the two sources.\n\nThe NHHS evidence is not treated as a replacement for the primary estimate. Differences may reflect variation in respondent reporting, coverage definitions, recall, and the treatment of incomplete observations. Accordingly, the comparison is used as a plausibility check rather than as a basis for recalculating the reported indicator.\n\nWithin the broader assessment, this cross-check provides supporting evidence for the interpretation of the coverage result. The survey findings are considered sufficiently aligned with the separate estimate to indicate that the principal conclusion is not dependent on a single measurement process. Any residual discrepancy is retained as a qualification to the estimate and should be considered when interpreting changes in service access or household outcomes.\n\nMention: \"Nationwide Health and Household Survey (NHHS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the NHHS survey to cross-check a coverage estimate derived from a separate records-based source.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:ac35d82711ac14ce3a2bdead", "occurrence_id": "0892eb3c8c144b0265d0a1c9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Enumerative Early Grade Reading System (EEGRS) is used as an administrative source for 2022. Its record provides the principal quantitative basis for the assessment presented in this annex, and the source name and acronym are retained to support traceability. No adjustment is applied to the reported values at the point of citation.\n\nThe evidence is used to establish the scale of observed reading performance and to indicate the distribution relevant to implementation monitoring. The figures are treated as reported observations rather than as estimates generated through further calculation. This approach preserves the distinction between information drawn from the source and interpretation undertaken in the assessment.\n\nTwo figures are reproduced directly from the record. The first indicates that 68 percent of assessed pupils reached the stated benchmark, while the second records 42 percent for the corresponding measure. Taken together, these values provide the central quantitative reference for the discussion. They indicate a substantial difference between the two reported measures and support a cautious interpretation of progress, without implying a trend beyond the information contained in the source.\n\nMention: \"Enumerative Early Grade Reading System (EEGRS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the 2022 EEGRS record and retains its reported values of 68 percent and 42 percent.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:9a5be33513603592c3ca0844", "occurrence_id": "5af3bcad26fb7cbcfa9a5aae", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Provincial Enrolment and Retention Index is used as the principal indicator for the assessment period associated with 2016. It provides the central evidentiary basis for reviewing enrolment participation and retention performance, while avoiding the introduction of estimates not documented in the source. The index is cited as a consolidated reference for the underlying records and is treated as sufficient for the descriptive analysis presented here.\n\nProject records indicate that the underlying material was collected through a structured compilation process, with entries reviewed for comparability before inclusion. Differences in terminology, coverage, and recording practice were harmonised to establish a consistent analytical series. The resulting files were archived with the supporting documentation, permitting subsequent verification of the figures and the procedures applied to them.\n\nThe assessment therefore relies on the Provincial Enrolment and Retention Index as a maintained evidence base rather than as an independent projection. Its use supports a transparent account of the information available for review and preserves the distinction between recorded observations and analytical interpretation. Any limitations arising from incomplete records, inconsistent reporting, or changes in classification are considered in the discussion of data quality, without altering the source designation or the recorded reference to 2016.\n\nMention: \"Provincial Enrolment and Retention Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The passage uses the Provincial Enrolment and Retention Index as a 2016 indicator and records its collection, harmonisation, and archiving as the central evidence base.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:1a961a82af1955b0c2ebefcc", "occurrence_id": "4e6b133dbcf25b536265b297", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Decennial Hiring and Vacancies System (DHVS) is treated as the administrative source for the staffing evidence presented in this assessment. Its records provide the principal basis for the observations below, with no supplementary source introduced for the two reported quantities.\n\nThe figures reproduced directly from the DHVS are 31 grades and 14 vacancies. These totals are retained as reported, rather than recalculated or inferred from related administrative entries. The first figure indicates the number of grades represented in the record, while the second indicates the number of vacancies recorded alongside them.\n\nTaken together, the figures provide a compact measure of the staffing structure and the outstanding recruitment requirement. Their use is limited to descriptive reporting, and no causal interpretation is assigned to the difference between the two counts. This treatment preserves the evidentiary status of the DHVS and avoids presenting derived estimates as observed values.\n\nThe source is therefore central to the paragraph because both numerical statements originate in the same administrative record. The figures should be read as registry totals for the coverage represented in the underlying data, subject to the completeness and classification conventions applied within that record.\n\nMention: \"Decennial Hiring and Vacancies System (DHVS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the DHVS figures of 31 grades and 14 vacancies as directly reported administrative totals.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:cefadf30337fc27003b4b738", "occurrence_id": "6670d54e53f77001b10d2903", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Panel Housing Conditions Survey (PHCS) as its principal empirical source. MEC produces the PHCS, a survey designed to record household housing conditions through a consistent set of observations. The source is treated as the basis for measurement rather than as a collection of isolated quotations.\n\nThe analysis calculates average values from the available observations and compares those averages across successive observations. This procedure provides an indication of the underlying trend in housing conditions while reducing the influence of unusually high or low individual results. The resulting measures are used to assess changes in occupancy, basic amenities, structural condition and housing adequacy.\n\nThe calculated averages are presented as analytical estimates derived from the PHCS material. They should therefore be read as summary measures of the observed distribution, not as figures reproduced directly from a single reported entry. The trend assessment relies on the direction and consistency of the calculated results across the source material.\n\nThis treatment supports a concise assessment of housing conditions and permits the findings to be integrated with the report’s broader monitoring framework. Interpretation remains limited to the information captured by the PHCS and does not extend beyond the scope of its observations.\n\nMention: \"Panel Housing Conditions Survey (PHCS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses PHCS observations to calculate average housing-condition indicators and identify their trend.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:fb904b445b8c46649ac64784", "occurrence_id": "4edaac398d6aea63d538e593", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment relies on the Departmental Land Cover Change Dataset (DLCCD), a geospatial source produced by DANE. For 2024, the DLCCD provides the principal comparative evidence for interpreting the reported change in land cover. Its tabulations are used as an external benchmark rather than as a replacement for the estimate obtained through the assessment procedure.\n\nThe estimate derived from a separate source is cross-checked against the patterns and totals recorded in the DLCCD. The analysis examines whether the direction and approximate magnitude of the estimate remain consistent with the cited dataset. Differences are retained for review where they exceed the tolerance applied in the assessment, since such differences may reflect definitional boundaries, measurement conditions, or the aggregation of observations. This treatment gives the source a direct evidentiary role in validating the estimate.\n\nOn this basis, the results are presented with the estimate as the primary reported measure and the DLCCD as the corroborating reference. The cross-check does not imply that the two measures are identical. It indicates instead whether the estimate falls within a plausible range supported by an independent geospatial record. This qualification is retained in the interpretation of results, while the source remains sufficiently consistent to support the assessment conclusion.\n\nMention: \"Departmental Land Cover Change Dataset (DLCCD)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The analysis takes a 2024 land-cover benchmark from the DLCCD and uses it to test an independently derived estimate.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:196c02fb10a9fbc566cde7f7", "occurrence_id": "031099d2568b07cb2478b5cc", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Baseline Water Stress System as an administrative source produced by Landsat. The material provides a consistent basis for describing pressure on available water resources and for locating differences in observed stress across the analytical units. Its use is limited to supporting the interpretation of the reported results, rather than serving as a substitute for local verification.\n\nThe records were collected into the project archive and harmonised before review. Differences in naming, structure, and coverage were reconciled through documented checks, while observations that could not be aligned were retained separately for quality control. The resulting series was treated as a reference layer for comparison with the other evidence considered in the assessment.\n\nOn this basis, the analysis finds that areas with higher recorded water stress require closer attention during implementation and monitoring. The source does not, by itself, establish the causes of that stress or determine the adequacy of proposed measures. It provides a comparable statistical signal that strengthens the broader assessment and supports prioritisation of follow-up work. All transformations applied during consolidation were archived with the underlying records to permit review and replication.\n\nMention: \"Baseline Water Stress System\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The passage records the collection, harmonisation, and archiving of the Landsat administrative water-stress source for comparative assessment.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:2b53790877740936b4b5975d", "occurrence_id": "c4ac091248c3fd25a4ce3e23", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Combined Child Wellbeing Survey (CCWS) as a supporting source for the interpretation of child wellbeing outcomes. The 2016 survey provides an independent reference point for the reported estimate and is treated as evidence for one claim within the wider discussion. Its use is limited to comparison of the level indicated by the estimate with the level recorded in the survey.\n\nSpecifically, the estimate is cross-checked against the corresponding CCWS observation to assess whether the direction and approximate magnitude are consistent. This comparison does not replace the principal estimate, nor does it establish a separate time series. It provides a reasonableness check on the claim that the observed outcome is materially different from the relevant benchmark. Where the two sources diverge, the estimate remains identified as the primary basis, while the survey result is retained as corroborating evidence.\n\nThe evidence is therefore read with appropriate caution. The CCWS measure may reflect differences in coverage, definitions, and reporting conditions relative to the source used for the estimate. Nevertheless, its inclusion strengthens the assessment by showing that the claim is not dependent on a single observation. No inference is drawn beyond the specific comparison, and the cited survey is not used to support unrelated findings.\n\nMention: \"Combined Child Wellbeing Survey (CCWS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The author uses the 2016 CCWS survey to cross-check an independent estimate of child wellbeing.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:de0cf26b13509e7695b6f630", "occurrence_id": "0d31043d3fd17cb04c4c6822", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Follow-Up Administrative Boundary Index (FABI) as an indicator of changes in administrative boundary conditions. The source is retained in the analytical record under its stated title, with no additional designation required for interpretation. Its observations are treated as a consistent basis for assessing movement across reporting periods.\n\nFor the analysis, the underlying FABI observations are summarised through an average and a calculated trend. These measures provide a general indication of the direction and level of change, while reducing the influence of isolated observations. The calculation is performed on the source values themselves and does not reproduce individual readings as reported results.\n\nThe resulting pattern supports the assessment claim that administrative boundary conditions show a measurable progression over time. This finding is used as one element within the wider review and is considered alongside the other evidence presented in the report. The source therefore informs the conclusion through derived summary measures rather than through the repetition of a particular recorded value.\n\nMention: \"Follow-Up Administrative Boundary Index (FABI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses FABI observations to calculate an average and trend supporting the assessment claim.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:0e8afa1015fa58753363e02f", "occurrence_id": "8efed04eb04236fae9a65b64", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses a consolidated measure of minimum needs to support interpretation of service access and household vulnerability. The measure is treated as an input to the analysis rather than as a complete account of living conditions, and its limitations are considered when comparing results across reporting units.\n\nThe Small-Area Minimum Needs Survey, a 2022 survey, provided the underlying observations for this purpose. The project team collected the available records, harmonised definitions and coding across reporting units, and archived the resulting files with supporting documentation. Quality checks addressed duplicate records, incomplete entries, and inconsistent classifications. Where the original material did not permit a reliable adjustment, the observation was retained in its recorded form and identified for review.\n\nThe resulting evidence supports the assessment finding that minimum needs are unevenly distributed and that aggregate results may conceal substantial variation among reporting units. The source therefore informs the interpretation of the consolidated measure, while the assessment avoids presenting it as a direct proxy for all dimensions of welfare. This treatment also establishes a reproducible basis for subsequent monitoring and for reconciliation with administrative records when they become available.\n\nMention: \"Small-Area Minimum Needs Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The author used the Small-Area Minimum Needs Survey to collect, harmonise, archive, and interpret evidence on variation in minimum needs.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:dd919a73a7c54d27f0825748", "occurrence_id": "e1c4cf6c4f82b311d6ebf264", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Ministerial Ward Outline System is cited as an administrative source produced by UNIDO for 2017. Its identification is retained in the evidence base because the system provides a structured reference for interpreting the distribution of administrative responsibilities. No alternative source is introduced for this part of the assessment.\n\nThe analysis reproduces two figures directly from the system, namely 31 wards and 8 service areas. These values are presented as reported, without adjustment or derivation, and are used to anchor the accompanying discussion of coverage. The figures therefore function as documentary evidence rather than as outputs of a separate calculation.\n\nTaken together, the reported counts support the assessment that the framework distinguishes several layers of administrative coverage rather than presenting a single undifferentiated total. The distinction is relevant to implementation review because it permits the scale of the operating structure to be read alongside the broader findings. The citation is consequently limited to the evidence required for this claim, while the interpretation remains within the scope of the assessment.\n\nMention: \"Ministerial Ward Outline System\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the directly reported figures of 31 wards and 8 service areas from the 2017 UNIDO administrative source to support a claim about differentiated coverage.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:f08ba5a400f40a295407205e", "occurrence_id": "a4698a0d8f52d9c13becf64d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Small-Area Host Community Registry is cited as a database produced by Univer sity of Minnesota. The source provides a consistent reference for interpreting reported differences in household conditions and service access across the assessed units. Its entries are treated as documentary evidence rather than as a basis for extrapolation beyond the recorded observations.\n\nIn the wider assessment, the registry supports the finding that host communities display material variation in household size and access to basic services. Two figures quoted directly from the source support this claim. It records 1,842 households in the relevant reporting set and 6,417 individuals associated with those records. These values are used as reported, without recalculation or projection, and are presented solely to anchor the interpretation of the broader pattern.\n\nThe figures should not be read as an estimate of all host communities. They define the numerical base for the paragraph and permit the assessment to distinguish between the scale of recorded households and the number of recorded individuals. This limited use is consistent with a statistical note that prioritises traceability, clarity, and restraint when citing external evidence.\n\nMention: \"Small-Area Host Community Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Small-Area Host Community Registry for two directly quoted figures, 1,842 households and 6,417 individuals.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:7d8c686c5c69ceaaf434be19", "occurrence_id": "b69ceff44dc237881d1647f7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis relies on the Follow-Up Targeting and Payments Survey, a survey produced by DoFE in 2017. The material is treated as the principal evidence for assessing changes in targeting and payment outcomes. Its observations are used to establish the direction and stability of movement across the reporting period, while preserving the distinction between measured responses and derived indicators.\n\nThe results presented here are calculated from the underlying observations rather than reproduced as isolated figures. A central tendency is obtained for the relevant outcome, and comparable averages are examined to identify whether performance improved, declined, or remained broadly unchanged. This approach reduces the influence of individual responses and provides a consistent basis for interpreting the distribution of reported outcomes.\n\nThe resulting trend is used to support the assessment of implementation performance. Differences between groups are considered alongside the overall average, with attention to whether the pattern is persistent across observations. The analysis does not treat a single reported value as representative of the full result. Instead, the survey evidence is aggregated and reviewed for internal consistency before being used in the assessment. This treatment supports a cautious interpretation of changes in targeting accuracy and payment regularity, while retaining the underlying survey as the principal empirical basis.\n\nMention: \"Follow-Up Targeting and Payments Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives averages and a trend from the Follow-Up Targeting and Payments Survey rather than reporting isolated observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:8e6a7c2d2dbcfbb3f7c806f3", "occurrence_id": "e6cb0ab0b0dbc413000124e7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Statistical Education Finance System (SEFS) is identified as an administrative source produced by NISR for 2015. Its records provide the principal empirical basis for the financial assessment and are treated as structured reporting evidence rather than as narrative information.\n\nThe analysis does not reproduce individual observations from the source. It aggregates the available entries and calculates an average across reporting units, thereby reducing the influence of unusually high or low values. This procedure supports comparison of the distribution of recorded amounts and provides a consistent basis for interpreting the level of reported financing.\n\nThe resulting average is used to assess the broad pattern of resource allocation and to identify material variation within the recorded observations. Calculations are retained at the level required for verification, while the main text presents only the derived measure and its interpretation. No adjustment is introduced beyond the aggregation required to obtain the average, and the result should therefore be read as a statistical summary of the records contained in the Statistical Education Finance System (SEFS).\n\nMention: \"Statistical Education Finance System (SEFS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the 2015 SEFS administrative records produced by NISR to calculate an average rather than reproduce individual observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:9ee5fb7b195221125e11f0a8", "occurrence_id": "8d7fb3dc5f8442c1cc554ce4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Diagnostic Deprivation and Hardship Survey as its principal survey source. The material is treated as the direct evidentiary basis for the deprivation profile presented below. No adjustment is applied to the reported proportions, and the wording retains the scope of the source. The evidence is therefore used to establish the observed distribution of hardship rather than to support a separate estimate.\n\nThe survey reports that 42 percent of households experienced deprivation under the measured conditions. It also reports that 18 percent faced hardship at the threshold used in the instrument. These figures are reproduced directly from the source and are presented as reported, rather than derived through comparison, projection, or recalculation. Their inclusion provides the central quantitative basis for the assessment of material conditions.\n\nTaken together, the results indicate that deprivation affected a substantially larger share of households than the recorded hardship measure. This difference is reported for descriptive purposes and should not be interpreted as evidence of change over time or as a causal relationship. The findings support a focused reading of the burden recorded by the survey, while the underlying definitions and thresholds remain those established in the source. Further analysis should preserve the same denominators and classifications to maintain comparability within the assessment.\n\nMention: \"Diagnostic Deprivation and Hardship Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported figures of 42 percent and 18 percent from the Diagnostic Deprivation and Hardship Survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:6324beb9ad8c89d201be22cc", "occurrence_id": "fb8bd0c3a476daed413aa7a8", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment considers attendance conditions as one element affecting implementation performance. The Representative Teacher Absence Survey (RTAS) is cited as a survey that provides background on the general issue of teacher attendance. Its inclusion supports the statement that patterns of attendance are relevant when interpreting operational results and assessing the reliability of reported implementation progress.\n\nThe reference is limited to this contextual function. No observations, estimates, rates, counts, or other values from RTAS are incorporated into the calculations, comparisons, or findings presented in this assessment. The discussion therefore does not derive a measure of absence, establish a change over time, or attribute a specific magnitude to the attendance issue.\n\nThis treatment is consistent with the scope of the analysis, which relies on the evidence presented in the assessment itself for all reported results. RTAS serves only to situate one claim within the broader discussion of implementation conditions. It does not determine any conclusion, alter any reported result, or provide a basis for an additional indicator.\n\nMention: \"Representative Teacher Absence Survey (RTAS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The passage uses the Representative Teacher Absence Survey (RTAS) only to support a contextual claim about teacher attendance and takes no values from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:8e5ea4e676327bc5bebb0785", "occurrence_id": "c1bf69c3e77b2f50b64635eb", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Monthly Remote Sensing Census is used as the census for the 2024 reference period. It provides the principal quantitative basis for the assessment and is cited without adjustment to its reported values. No supplementary source is required for the two measures presented below.\n\nTwo figures are reproduced directly from the Monthly Remote Sensing Census. The recorded total is 31 households, while the corresponding coverage is 68 sites. These values are retained in their reported form and provide the central evidence for the results presented in this note.\n\nThe figures support a concise assessment of observed coverage during the reference period. The difference between the two counts is not interpreted as a change over time, since the available evidence consists of a single census reference. The reported totals are therefore used descriptively, with no additional estimation or projection. Their inclusion is intended to preserve traceability between the narrative findings and the underlying source.\n\nMention: \"Monthly Remote Sensing Census\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage reports 31 households and 68 sites drawn directly from the Monthly Remote Sensing Census for 2024.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:5fb712f5d42f035a50c5b3f7", "occurrence_id": "84f960f1290cbec8a89b815e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Biennial Socioeconomic Survey (BSS) is cited as the survey covering Lusaka in 2018. It provides a locally relevant reference for assessing household conditions and service access. The source is used in its reported form, without adjustment to the underlying measures or substitution of administrative estimates.\n\nThe discussion indicates that access remained uneven across households and that observed differences were material for interpretation of welfare outcomes. Two figures are reproduced directly from the BSS to support this assessment: 62.4 percent of surveyed households are reported as having access to the specified service, while 18.7 percent are reported as lacking it. These values are presented in support of the broader finding rather than as a basis for a new calculation.\n\nThe figures are therefore treated as descriptive observations within the assessment. Their use supports comparison of conditions across the covered population while preserving the scope and terminology of the source. No inference is drawn beyond the reported survey results, and the cited evidence is confined to the conditions documented for Lusaka.\n\nMention: \"Biennial Socioeconomic Survey (BSS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the BSS figures of 62.4 percent with access and 18.7 percent without access to support the finding of uneven household conditions in Lusaka in 2018.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:d5be6d7dc102feb7c35ffee2", "occurrence_id": "557430cf7c1d1562014ebefe", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment relies on the High-Frequency Diagnostic Capacity Registry (HDCR) database for 2016 as a supporting source for the reported distribution of diagnostic capacity. The records provide a consistent basis for comparing categories of service readiness and for identifying gaps in the underlying reporting coverage.\n\nFor analytical use, the available entries were collected from the registry, harmonised against a common classification, and archived with the accompanying metadata. Inconsistent labels were reconciled where the substantive meaning was sufficiently clear, while unresolved cases were retained as limitations rather than adjusted through unverified assumptions. This treatment supports the claim that observed differences reflect both variation in capacity and variation in reporting completeness.\n\nThe source is used within the broader assessment to substantiate the interpretation of the capacity profile, not as a stand-alone measure of performance. Results derived from the database are therefore presented with the relevant qualification that coverage and comparability depend on the completeness of the recorded entries. The resulting evidence provides a documented reference point for subsequent review and for the maintenance of comparable records in future implementation reporting.\n\nMention: \"High-Frequency Diagnostic Capacity Registry (HDCR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The passage uses the HDCR database for 2016 and describes collecting, harmonising, and archiving its records for the capacity assessment.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:f2612a66343e30a59775e48c", "occurrence_id": "0e7c307ad366930080d90de2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Biennial Land Cover Dataset is used as the principal geospatial source for the assessment. It was produced by Lagos State Government for 2015 and is treated as the authoritative basis for the land-cover observations presented below. The source is retained in its reported form to preserve consistency between the underlying classification and the figures cited in the analysis.\n\nTwo figures are reproduced directly from the dataset. Built-up surfaces account for 43.7 percent of the mapped area, while vegetated cover is reported at 12,860 hectares. These values provide the central quantitative evidence for the assessment and are not recalculated from secondary summaries. Their inclusion allows the discussion to distinguish the relative extent of developed surfaces from the absolute area associated with vegetation.\n\nThe figures should be interpreted as reported measurements within the source framework. They support comparison across the land-cover categories used in the assessment, while avoiding assumptions about changes that are not documented in the dataset. No adjustment has been applied to either value, and the reported units have been retained to support traceability in subsequent review.\n\nMention: \"Biennial Land Cover Dataset\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported values of 43.7 percent for built-up surfaces and 12,860 hectares for vegetated cover from the Biennial Land Cover Dataset.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:c95b61838f4a4a389293cce3", "occurrence_id": "67247274f407834688330ce4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis draws on the Longitudinal Value Chain Dataset (LVCD), a geospatial source covering Gabon for 2018. The dataset is used to establish a consistent descriptive basis for examining the distribution of productive activity and related service access. Its coverage is considered suitable for the spatial interpretation required in this assessment, while the reported observations remain subject to the limitations of the underlying records.\n\nTwo figures are taken directly from the LVCD. It records 31 local facilities within the assessed network and indicates that 64 per cent of observed transactions are associated with the principal activity group. These values are reproduced without recalculation or adjustment. They provide a limited but direct indication of the scale and relative weight of activity represented in the source.\n\nTaken together, the figures support the assessment that value-chain activity is unevenly distributed across Gabon, with a substantial share concentrated in a defined segment of the network. The finding is used to inform prioritisation of follow-up analysis and does not constitute an estimate of unobserved activity. Interpretation should therefore remain confined to the coverage and definitions applied in the LVCD.\n\nMention: \"Longitudinal Value Chain Dataset (LVCD)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the LVCD figures of 31 facilities and 64 per cent to support an assessment of concentrated activity in Gabon.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:4b5c0956ce39e600ffab8bbe", "occurrence_id": "0230bfae4240c5f31dfcc058", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment finds that disparities remain material across the population groups covered by the analysis, with implications for the targeting of compensatory measures and the monitoring of distributional outcomes. The evidence is used to support the broader finding that average improvements have not eliminated unequal conditions.\n\nFor this purpose, the High-Resolution Inequality Index (HII) indicator for 2020 provides a consistent reference for the reported comparison. The assessment takes two figures directly from this source: 31 grades are recorded across the measured distribution, while 18.4 percent is reported for the corresponding share. These figures are cited as observed values and are not recalculated in the present assessment.\n\nThe results should therefore be read as a concise statistical basis for the claim regarding persistent inequality rather than as a separate estimate of change over time. Their use is limited to establishing the scale and dispersion of the recorded outcome, while interpretation remains aligned with the wider discussion of implementation performance and equity.\n\nMention: \"High-Resolution Inequality Index (HII)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the High-Resolution Inequality Index (HII) indicator for 2020 and reports 31 grades and 18.4 percent directly from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:5a78d9e837062f2f53a66451", "occurrence_id": "12a1fc72517979c6cd168822", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Municipal Settlement and Elevation Dataset is a geospatial source for 2019 and supplies the principal evidence for the settlement and elevation assessment. It is treated as an analytical base rather than a catalogue of figures. The available observations are reviewed for consistency, aligned to a common unit of analysis, and retained for aggregation.\n\nFrom these observations, the assessment calculates an average elevation and derives a settlement trend across the reporting units. The resulting measures summarise the distribution and direction of change without presenting any individual record as a stand-alone finding. This approach reduces sensitivity to isolated observations and permits comparison within a single analytical framework.\n\nThe computed average is used to describe the overall physical setting represented in the source, while the derived trend indicates whether settlement conditions are broadly increasing, stable, or declining over the observed sequence. These outputs support the interpretation of exposure and service requirements, but they are not treated as direct quotations from the dataset.\n\nResults should therefore be read as estimates generated through aggregation of the 2019 observations. Any apparent movement reflects the calculation procedure and the underlying coverage of the Municipal Settlement and Elevation Dataset, rather than a reported statistic reproduced without modification.\n\nMention: \"Municipal Settlement and Elevation Dataset\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage derives an average elevation and a settlement trend from the 2019 geospatial source rather than quoting individual observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:59afc08fabea737ca6364ff8", "occurrence_id": "d1a4a84f827492ad05b47d72", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Monthly Reproductive Health Survey is used as the principal survey for the assessment of reproductive health conditions in Johor during 2015. Its coverage is treated as sufficiently aligned with the analytical boundary of the assessment, while its limitations are retained in the interpretation of results. No supplementary institutional attribution is required for the use reported here.\n\nThe records were collected from the available survey materials and reviewed for internal consistency. Entries with differing terminology, presentation, or measurement conventions were harmonised to support a consistent reading across the reporting period. This process established a common basis for tabulation and reduced avoidable variation arising from transcription and arrangement.\n\nThe harmonised material was then archived with the supporting documentation and a clear record of the transformations applied. The archive serves as the central evidence base for the assessment and permits subsequent verification of the reported patterns without altering the underlying observations. In this form, the source supports controlled analysis while preserving its original geographic and temporal scope.\n\nUse of the survey is therefore limited to evidence directly contained in the archived material. Interpretive statements are kept proportionate to the coverage of Johor in 2015, and no inference is made beyond the information retained through collection, harmonisation, and archiving.\n\nMention: \"Monthly Reproductive Health Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The author collected, harmonised, and archived the Monthly Reproductive Health Survey for Johor in 2015 as the central evidence base.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:909f5f9ade1de6c39e438297", "occurrence_id": "ba7a86f8c61c1f723579b849", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Quality-Assured Public Works System (QPWS) is treated as an administrative source for 2024 in the assessment. Its records provide the basic observation set for work entries, completion statuses, and reported labour inputs. Coverage is interpreted at the record level, with duplicate entries screened through internal identifiers and incomplete observations retained only where the relevant measure remains usable.\n\nThe assessment does not reproduce individual QPWS figures as reported. Instead, the usable records are grouped by reporting interval and indicator, after which arithmetic means are calculated for each interval. These means are then compared sequentially to obtain a direction and rate of change. The resulting trend is used to describe movement in implementation performance while limiting the influence of unusually large or small observations.\n\nThis procedure gives greater weight to the distribution of records than to any single reported value. It also permits comparison of average completion status, labour input, and work-entry activity within one consistent analytical treatment. Where the number of usable records differs across indicators, the calculation uses the available denominator and records the resulting limitation in the assessment notes.\n\nThe derived averages and trend measures constitute the principal empirical basis for the findings. They are presented as analytical outputs from QPWS rather than as direct quotations, and interpretation remains restricted to patterns supported by the administrative record.\n\nMention: \"Quality-Assured Public Works System (QPWS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses QPWS records to calculate interval averages and derive a trend in implementation indicators.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:44b71db264db66bfba2ffe4c", "occurrence_id": "4ccf3d5e808bcad6284a2a31", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Cross-Sectoral Wage Structure Index is an indicator used in the assessment of relative wage positioning across the observed structure. It provides a standardized basis for describing dispersion between the lower and higher portions of the distribution, while avoiding interpretation beyond the reported measure. The indicator is treated as the primary quantitative reference for this paragraph.\n\nTwo figures are reproduced directly from the Cross-Sectoral Wage Structure Index. The lower recorded value is 42.7 index points, while the higher recorded value is 58.3 index points. Read together, these figures show that the measured position differs materially across the wage structure. The comparison is presented as reported evidence rather than as a recalculated estimate, and no additional adjustment has been applied.\n\nThe figures support a limited assessment of internal wage differentiation. The distance between the two reported positions is consistent with variation in the structure captured by the indicator, but it does not by itself establish the causes of that variation. Accordingly, the results are used to characterize the observed pattern and to inform subsequent review of distributional conditions. The Cross-Sectoral Wage Structure Index remains the sole quantitative source cited for this comparison.\n\nMention: \"Cross-Sectoral Wage Structure Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Cross-Sectoral Wage Structure Index to report 42.7 and 58.3 as directly reproduced figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:116f69663047980a3b5e9207", "occurrence_id": "49947b661eb28a31e7cb2828", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis draws on the Quality-Assured Informal Trade Survey (QITS), a survey produced by Fundação_ _Getúlio_ _Vargas. The QITS material is treated as an input to the statistical assessment rather than as a set of figures to be reproduced individually. Its scope and construction provide a consistent basis for examining movements in informal trade indicators.\n\nFor the indicator considered in this section, reported observations are converted into a comparable series and used to calculate an average level and an underlying trend. This procedure limits the influence of isolated observations and supports interpretation of the general direction of change. The resulting measures are presented as analytical estimates derived from the source, not as direct quotations from it.\n\nThe evidence supports the broader assessment that participation in informal trade has evolved unevenly across the observation period, with the calculated average providing a more stable reference than any single reported observation. The QITS-derived trend is therefore used to substantiate this finding while other elements of the assessment address coverage, measurement limitations, and the interpretation of changes in the indicator. The source is retained in the documentation to permit review of the calculation and to distinguish derived results from source-reported content.\n\nMention: \"Quality-Assured Informal Trade Survey (QITS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses QITS to calculate an average and trend that support a broader finding about uneven changes in informal trade participation.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:89fe04f28c67088b02587289", "occurrence_id": "6a189a3bea1978c64a866a72", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Coordinated Diagnostic Capacity System as an administrative source covering Timor-Leste. The system provides the central evidence for the description of diagnostic capacity and is treated as a registry of reported service points rather than as an analytical estimate. Its entries are used without adjustment, aggregation, or extrapolation.\n\nTwo figures are reproduced directly from the system. It records 31 diagnostic facilities within the reporting network and identifies 8 laboratories with designated testing functions. These values are presented as recorded, with no attempt to infer coverage beyond the entries contained in the source. The distinction is relevant because facility presence and laboratory function are not interchangeable measures.\n\nThe reported distribution indicates a service network with a limited number of facilities undertaking the specified functions. It also provides a fixed reference for subsequent review of operational readiness, staffing, and equipment, although those dimensions are not quantified here. The evidence should therefore be read as a description of registered capacity at the time represented in the system, rather than as a measure of effective service utilisation or population access.\n\nFor monitoring purposes, the two reported figures establish a concise baseline for verification against supporting records. Any later revision should preserve the distinction between the 31 diagnostic facilities and the 8 laboratories, since combining them would obscure the structure of the reported network.\n\nMention: \"Coordinated Diagnostic Capacity System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the figures of 31 diagnostic facilities and 8 laboratories directly from the Coordinated Diagnostic Capacity System.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:6de55c2bdf697c766fee82b8", "occurrence_id": "e47ced2f160e852b2775ff1e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Record-Based Informal Sector Registry, an ActivityInfo database maintained for reference in the analysis. The source is used as a documentary basis for describing the scale and composition of activity recorded in the informal sector. Its inclusion supports a measured interpretation of the available administrative evidence.\n\nThe registry records 31 registered units. This figure is used directly to indicate the observed number of units represented in the source and to support the finding that activity is distributed across a substantial set of small operations. No adjustment is applied to this count in the present assessment.\n\nA second figure taken directly from the same source is 18 active workers. Together, the two figures provide a limited but relevant indication of the relationship between recorded units and reported participation. They are presented as source values rather than as estimates generated through further calculation.\n\nThe evidence should nevertheless be read within the limits of registry coverage and reporting practice. The figures support the paragraph’s central claim regarding the documented scale of informal activity, while they do not by themselves establish completeness, persistence, or change over time. Further interpretation would require additional observations and corroborating records.\n\nMention: \"Record-Based Informal Sector Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The assessment takes the figures 31 registered units and 18 active workers directly from the Record-Based Informal Sector Registry.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:ee65e6be60dcc4d33891dd77", "occurrence_id": "4a3661025a30eba61fab096b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Continuous Living Standards Index as an indicator produced by NCHS/CDC/WHO for Kenya and referring to 2024. The source is treated as the relevant statistical basis for a single claim concerning reported living conditions. Its stated coverage and reference period are retained without adjustment in the analysis.\n\nFor this purpose, the text reproduces two figures directly from the source. The first records 31 percent, while the second records 47 percent. These values are presented as reported measurements rather than as estimates generated through an independent calculation. Their inclusion provides quantitative support for the assessment of the distribution described in the surrounding discussion.\n\nThe figures are used to substantiate the broader claim that reported conditions are not uniform across the population covered by the indicator. No interpolation, averaging, or conversion is applied to either value. The source citation therefore functions as evidentiary support for the stated comparison, while the interpretation remains limited to the information conveyed by the two reported figures. This treatment is consistent with an implementation report that distinguishes directly cited observations from analytical conclusions.\n\nMention: \"Continuous Living Standards Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage reports the two figures of 31 percent and 47 percent taken directly from the source for Kenya in 2024.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:62432942278a6c8104e34850", "occurrence_id": "08bee52200156bfeae627bb6", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Audited Ecosystem Carbon Collection (AECC) as an other source for the 2021 reference period. The collection is cited in the supporting evidence for the reported carbon estimates, with no adjustment applied to the figures reproduced in this note.\n\nThe note reproduces two numerical entries directly from the collection, namely 12.4 and 8.7. These values are retained in their reported form to support traceability between the assessment text and the underlying evidence. They are presented as quoted observations rather than as results generated through a separate calculation.\n\nThe figures are used only to document the values recorded for the reference period. No additional interpretation is assigned to their difference, and no extrapolation is made beyond the information contained in the cited source. This treatment is consistent with the requirement to distinguish reported observations from estimates produced within the assessment process.\n\nFor audit purposes, the source name and acronym are stated in full at first mention, while the two reproduced entries remain linked to the same citation. The resulting presentation permits the figures to be checked against the source without introducing an alternative classification or a separate reporting basis.\n\nMention: \"Audited Ecosystem Carbon Collection (AECC)\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"source\", \"summary\": \"The author takes the two directly reported figures, 12.4 and 8.7, from the Audited Ecosystem Carbon Collection (AECC) for 2021.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:d11f33b36b7ed0151209d5e9", "occurrence_id": "af3519c6d9e334d06562f996", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Nationwide Supply Chain Survey (NSCS), a survey conducted in 2020, provides the primary empirical basis for the assessment of supply continuity. Its observations are used to examine the regularity of deliveries, the persistence of interruptions, and the extent to which reported conditions vary across successive reporting periods. The source is treated as an underlying record for calculation rather than as a set of figures to be reproduced individually.\n\nThe assessment derives an average level of continuity from the observations and calculates a trend across the reporting periods. This approach reduces the influence of isolated results and provides a more stable indication of operational performance. The resulting measures are used to distinguish sustained movement from short-term fluctuation, while retaining the variation recorded in the survey.\n\nThe calculated average indicates the central level observed over the period covered by the source, whereas the trend indicates the direction and pace of change. Together, these measures support interpretation of whether service conditions improved, deteriorated, or remained broadly stable. The findings are therefore presented as analytical results generated from the Nationwide Supply Chain Survey (NSCS), not as direct quotations of individual survey entries.\n\nMention: \"Nationwide Supply Chain Survey (NSCS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the Nationwide Supply Chain Survey to calculate an average and a trend from its observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:92f274605969a9746cb71887", "occurrence_id": "3df9228be6757b5e8ecfbd77", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Monthly Vocational Training Survey is treated as a survey referring to 2015 and is used as a supplementary benchmark for the analysis. Its reported participation levels provide a contemporaneous basis for reviewing the magnitude of the estimate presented in the main text. The source is not used to replace the primary calculation, but to assess whether the result falls within a plausible range.\n\nFor this purpose, the survey evidence is compared with an estimate obtained from a separate source. The comparison supports the claim that participation was substantial among the population covered by the analysis, while indicating that the estimated level should be interpreted with moderate caution. Differences may reflect coverage, timing, or measurement practice, and they do not by themselves establish an inconsistency. The cross-check is therefore considered confirmatory at the level of direction and approximate magnitude rather than as an exact reconciliation.\n\nThe finding is retained as one supporting element within the broader assessment. No adjustment is applied solely on the basis of the survey, since the available information does not justify recalibration of the estimate. The source nonetheless strengthens the evidentiary basis for the reported conclusion and provides a transparent reference point for subsequent review. In interpreting the result, attention should remain on the consistency of the overall pattern rather than on an exact match between the two sources.\n\nMention: \"Monthly Vocational Training Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the Monthly Vocational Training Survey for 2015 to cross-check a separate estimate and support its direction and approximate magnitude.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:5ff0127a1f8d57ff5f952ec9", "occurrence_id": "58d9c085adf359bcb44a7cd5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws its central evidence from the Monthly Multidimensional Poverty Survey (MMPS), which is treated as a survey of household conditions and service access. The material provides a consistent basis for comparing the incidence and intensity of multidimensional poverty across the observation period. The source is retained as the principal statistical input, while interpretation remains limited to measures that can be reproduced from its reported structure.\n\nRather than reproducing isolated observations, the analysis aggregates the available records into period-specific rates and calculates an average for the observation window. It then compares these derived measures to identify the direction and pace of change. This procedure gives greater weight to overall movement than to any single reported figure and reduces sensitivity to short-term variation. The resulting trend is therefore an analytical estimate based on MMPS material, not a direct quotation of a headline statistic.\n\nCalculations are performed using consistent definitions for deprivation, eligibility, and aggregation throughout. Missing or non-comparable entries are excluded from the relevant denominator, and the resulting coverage is documented in the working tables. The analysis reports the average level, the change between successive periods, and the broad trend indicated by those calculations. These outputs support monitoring and prioritisation while preserving a clear distinction between source observations and measures derived for this assessment.\n\nMention: \"Monthly Multidimensional Poverty Survey (MMPS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives period-specific poverty rates, an average level, and a directional trend from the MMPS survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:219ccb2c6d75e2a31d9c0f61", "occurrence_id": "90b0cf5442c73634e5093c6e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Independent Enterprise and Jobs System (IEJS) is used as an administrative data source for the assessment of enterprise activity and employment records. Its entries provide a consistent basis for reviewing the recorded scale of participation and for checking the internal coherence of reported results. The source is cited in this section because the relevant observations are drawn directly from its records.\n\nThe IEJS records 31 registered enterprises and 14 associated jobs. These two figures are reported as recorded in the source and are not recalculated for the purposes of this assessment. They provide the numerical support for the finding that enterprise activity remained limited in scope while generating a measurable, though modest, level of employment.\n\nThe evidence should be interpreted with attention to the coverage and recording practices applied to the administrative system. The figures describe entries present in the source and should not be treated as an estimate of activity outside those records. Nevertheless, they provide a clear reference point for monitoring implementation, comparing reported outputs with documented cases, and identifying areas requiring further verification. On this basis, the assessment considers the recorded enterprise base and associated employment to be material indicators of implementation progress, subject to the usual limitations of administrative reporting.\n\nMention: \"Independent Enterprise and Jobs System (IEJS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the IEJS figures of 31 registered enterprises and 14 associated jobs to support the assessment of recorded enterprise activity and employment.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:1f98b4cfbe2ef8b771252295", "occurrence_id": "81da367c21cb770e07c8319a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Quality-Assured Informal Trade System, an administrative source produced by the World Inequality Lab for 2023, provides the principal empirical basis for this assessment. Its records support a consistent review of reported activity, quantities, and values across the observation period. The source is used as an underlying series rather than as a collection of isolated figures.\n\nThe analysis applies uniform checks to the available entries and aggregates observations into comparable intervals. This treatment reduces the influence of individual reports and permits the calculation of an average level for the period under review. The resulting measure is used to establish the central tendency against which subsequent changes are assessed.\n\nA trend is then calculated from the ordered observations, with attention to the direction and relative persistence of change. The computed average and trend, rather than any single reported entry, constitute the central evidence for the assessment. Small movements are therefore interpreted cautiously where they do not alter the broader pattern.\n\nThe findings remain dependent on the completeness and internal consistency of the source records. They should be read as an analytical summary derived from the Quality-Assured Informal Trade System and not as a direct reproduction of its individual observations. This approach provides a transparent basis for monitoring changes while limiting undue emphasis on short-term variation.\n\nMention: \"Quality-Assured Informal Trade System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives a period average and trend from the Quality-Assured Informal Trade System rather than reproducing individual observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:f2f8ff27b4866cf2c399c5ea", "occurrence_id": "c571af34dd4e86c4f8af513a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The enrolment and retention position was reviewed using the Panel Enrolment and Retention System, an administrative source referring to 2021. The record provides a consistent basis for examining reported participation and continuation across the observation period, subject to the coverage and reporting limitations normally associated with administrative records.\n\nThe estimate presented in the main analysis was derived from a different reporting stream and was cross-checked against the corresponding measures in the Panel Enrolment and Retention System. This comparison supports the claim that the reported level of participation was broadly consistent with the underlying administrative record. Differences between the two measures remain within a range that does not alter the assessment.\n\nThe source was therefore used as a corroborating reference rather than as the sole basis for the estimate. Its enrolment and retention entries also assist in identifying possible discrepancies in aggregation, delayed reporting, or the treatment of continuing participants. These checks strengthen the interpretation of the indicator while recognising that administrative coverage may not fully capture all relevant movements.\n\nMention: \"Panel Enrolment and Retention System\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The author uses the 2021 administrative enrolment and retention records to cross-check a separate estimate and support its consistency.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:20eaab4d02371d983198f616", "occurrence_id": "4a9ae3e631999d0faf36c00f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Joint Working Conditions Survey (JWCS), a 2018 survey used as the principal empirical reference for the analysis. The source is treated as a documentary basis for describing reported working conditions, with no adjustment to the values reproduced below.\n\nThe report reproduces two figures directly from the survey. It records that 62 percent of respondents reported access to written information on working conditions, while 38 percent reported that such information was not available to them. These figures are presented as reported observations rather than as estimates generated through further calculation.\n\nThe results indicate a substantial difference in access to basic workplace information. The higher proportion suggests that written communication was present for a majority of respondents, but the remaining share is sufficiently large to warrant attention in implementation planning. The evidence therefore supports continued monitoring of information practices and the inclusion of clear reporting requirements in subsequent assessments. Interpretation is limited to the two reported figures and does not imply a broader trend beyond the evidence contained in the cited source.\n\nMention: \"Joint Working Conditions Survey (JWCS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Joint Working Conditions Survey (JWCS) 2018 survey figures of 62 percent and 38 percent to assess access to written information on working conditions.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:8edafa88de563d865b0f968e", "occurrence_id": "e2472854a084644b5fa7fb42", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Baseline Temperature Records Collection (BTRC) is cited as an other source associated with 2015. Its inclusion provides documentary background for the discussion of baseline conditions and indicates that historical records were considered in framing the assessment. No additional identifying detail is required for the present annex.\n\nThe collection is mentioned to establish continuity with earlier temperature documentation and to support the completeness of the source review. It does not provide the evidentiary basis for any statement in this section, and no numerical observation, estimate, or classification from it enters the analysis.\n\nAccordingly, the findings reported here are based on the evidence described in the accompanying assessment materials. The BTRC reference is retained for traceability and orientation only. It does not determine the reported results, substantiate a comparison, or supply a measurement used in the calculations.\n\nMention: \"Baseline Temperature Records Collection (BTRC)\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"No values or derived findings were taken from the Baseline Temperature Records Collection (BTRC); it was retained only as background for source traceability.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:5373217b3be233ecaa0d32a1", "occurrence_id": "6e46506f7da10084ef09bb0a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Integrated Apprenticeship Survey is used as the principal survey for this assessment of participation and completion patterns. Its tabulations provide a consistent basis for interpreting the distribution of apprentices across the reporting categories and for checking the internal coherence of the results. No adjustment has been applied to the figures cited below.\n\nThe survey reports that 68 percent of apprentices completed the required period of training. This figure is reproduced directly from the source and is treated as the principal measure of observed completion. The same source reports that 42 percent of apprentices received a formal assessment at the end of training. This second figure is also taken directly from the survey and is presented without recalculation.\n\nTogether, the two reported proportions indicate a material difference between completion and documented assessment. The result should therefore not be interpreted as evidence that all completed training was formally assessed. It instead identifies a reporting gap relevant to implementation monitoring and to the design of subsequent verification procedures.\n\nThe figures are retained in their source form to preserve comparability with the underlying tabulations. Further analysis may examine the causes of the difference, but such analysis is outside the scope of this note and is not required for the use of the cited evidence.\n\nMention: \"Integrated Apprenticeship Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Integrated Apprenticeship Survey for two directly reported figures, 68 percent completing training and 42 percent receiving a formal assessment.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:957e2c218c537e3b192443bd", "occurrence_id": "545e9dc1b3679fd4c12027fd", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Analytical Targeting and Payments Survey (ATPS) as the principal empirical reference for the discussion of household reach and payment coverage in Tunisia. This survey refers to 2017 and provides the relevant basis for interpreting observed differences across beneficiary groups. The source is treated as a statistical record for the purposes of this annex, with its reported values retained without recalculation or adjustment.\n\nTwo figures are taken directly from the ATPS and provide the central evidence for the assessment. The survey reports that 62.4 percent of surveyed households received the relevant payment, while 18.7 percent of surveyed households were recorded as excluded from that payment. These values are presented as reported, rather than as estimates derived from a comparison or as results generated through a separate computation. Their inclusion permits the discussion to remain anchored in the source record.\n\nRead together, the figures indicate a material difference between observed receipt and recorded exclusion within the population covered by the survey. The distinction is used only to clarify the scale of coverage documented in the source and does not imply a causal explanation. No additional adjustment is applied to either value, and the interpretation remains limited to the evidence contained in the ATPS. Accordingly, the two reported percentages support the assessment of implementation performance while preserving the source’s original numerical form.\n\nMention: \"Analytical Targeting and Payments Survey (ATPS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the ATPS figures of 62.4 percent and 18.7 percent as direct evidence on payment coverage and exclusion.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:3319308f3883d97a3c2c9208", "occurrence_id": "fe2c4de87b7b6c7164683b48", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Continuous Student Transport Index is used as an indicator for reviewing continuity in student transport provision. Produced by ALCS, it provides a compact reference for assessing whether recorded services are sufficiently regular to support attendance and access objectives. The source is treated as a statistical input to the assessment rather than as an independent measure of learning outcomes.\n\nThe assessment reproduces two figures reported in the indicator, recording 31 routes and 84 percent of services as meeting the specified continuity threshold. These figures support the finding that transport arrangements are broadly established, while also indicating that a measurable share of services remains below the threshold. No extrapolation is applied to either figure, and the values are retained in their reported form.\n\nThe resulting interpretation is limited to the transport dimension. The figures inform the assessment of operational coverage and continuity, but they do not establish causation between transport provision and attendance. Their principal use is therefore to substantiate the paragraph’s assessment of implementation status and to provide a consistent basis for subsequent monitoring.\n\nMention: \"Continuous Student Transport Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Continuous Student Transport Index figures of 31 routes and 84 percent of services to support the assessment of transport continuity.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:dcdfaf98842e08505a358c8b", "occurrence_id": "2839b8596034b5c1efe413f2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Nationwide Primary Care Survey (NPCS) is treated in this assessment as the principal survey covering Ecuador for 2019. Its observations provide the empirical basis for the comparison of primary care conditions across the measured service domains. The analysis retains the source’s unit definitions and applies a common calculation procedure so that differences in scale do not drive the resulting summary. No individual observation is presented as a standalone national estimate.\n\nRather than reproducing selected entries, the assessment calculates an average from the NPCS observations and uses the ordered results to estimate a trend in service performance. The average is used as a compact measure of the distribution captured by the survey, while the trend indicates whether performance is relatively stronger or weaker across the assessed domains. These calculations are reported as analytical outputs and should not be read as additional observations collected outside the NPCS.\n\nResults are interpreted with attention to coverage, continuity, and basic readiness, with equal treatment of the observations that meet the stated inclusion criteria. The derived measures support the implementation discussion by identifying broad patterns and by limiting reliance on isolated values. The NPCS therefore functions as the central evidence for the assessment, while the reported average and trend provide the basis for comparing conditions within Ecuador in 2019.\n\nMention: \"Nationwide Primary Care Survey (NPCS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The assessment uses the NPCS to calculate an average and derive a service-performance trend for Ecuador in 2019.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:10dc88bff6ec7fab79e7ffe6", "occurrence_id": "9857b4468db9328baf974542", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Independent Informal Sector System as an administrative source covering Italy for 2024. It provides the central empirical basis for examining changes in informal activity across the observation period. The material is treated as a structured administrative record, with attention to consistency across reporting units and to the comparability of observations within the covered period.\n\nThe analysis derives an average level from the available observations and uses successive values to estimate the direction and pace of change. These calculations are used in preference to reproducing individual entries, since the assessment is concerned with the overall movement of activity rather than isolated measurements. The resulting trend is interpreted cautiously where coverage or reporting intensity may affect the observed pattern.\n\nThe derived average supports the assessment of the prevailing level of informal activity, while the calculated trend provides an indication of whether conditions are expanding, stable, or contracting. Together, these measures inform the discussion of implementation performance and the identification of areas requiring closer monitoring. No separate institutional estimate is introduced, and the results remain anchored in the cited source and the calculations performed for this assessment.\n\nMention: \"Independent Informal Sector System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the Independent Informal Sector System for Italy in 2024 to calculate an average and trend from administrative observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:6a4579c42dfb7234183ef92a", "occurrence_id": "0adc91c76759436f67f6f1a5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Clustered Classroom Practice Survey is a 2020 survey produced by WTO and is used as the principal empirical reference for the classroom practice assessment. Its definitions and tabulations are treated as the applicable basis for interpreting the reported indicators, without additional adjustment in this section.\n\nThe analysis reproduces two figures directly from the survey. It records that 62 percent of observed classrooms applied the specified practice consistently, while 47 percent reported access to the associated instructional materials. These values are presented as reported in the source and are not recalculated from subsidiary observations.\n\nThe difference between the two figures indicates a substantial implementation gap between classroom practice and material availability. The result supports attention to operational constraints affecting routine application, while avoiding inferences beyond the measurement recorded by the survey. The figures should therefore be read as descriptive evidence for implementation monitoring rather than as an estimate of causal effects.\n\nFor reporting purposes, the source provides a common reference point for subsequent assessment of classroom conditions and delivery arrangements. Any future comparison should retain the survey’s definitions and measurement conventions so that changes in the reported indicators are not attributed to altered interpretation.\n\nMention: \"Clustered Classroom Practice Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the 62 percent and 47 percent figures directly from the 2020 WTO survey and uses their difference to identify an implementation gap.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:afff28bc7ac016ddf930735b", "occurrence_id": "253e8eaf6455a751323714de", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Household Host Community Registry (HHCR) is the principal database used for the analysis of household conditions and access to basic services. It provides a consistent basis for examining changes across successive reporting points and for assessing differences in the distribution of recorded households. No individual entry is treated as a reported result in isolation.\n\nThe assessment derives average values from the HHCR records and uses grouped observations to identify the direction and stability of observed trends. This approach reduces the influence of unusually large or small entries and supports comparison across the indicators retained for monitoring. The resulting measures are therefore analytical estimates produced from the database rather than figures reproduced directly from a registry display.\n\nThe analysis indicates gradual movement in household composition and service access, with variation in the pace of change across indicators. These findings are used to inform implementation monitoring and to identify areas requiring closer review. Interpretation remains subject to the coverage and recording practices reflected in the HHCR database, but the source is sufficiently structured to support repeatable calculations and a transparent assessment of emerging patterns.\n\nMention: \"Household Host Community Registry (HHCR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The assessment derives average values and trends in household composition and service access from the Household Host Community Registry (HHCR) database.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:b4e4da7047219603b6c6253b", "occurrence_id": "9fbbfff6cc8fe4e2e541b514", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Civil Child Wellbeing Survey (CCWS), a survey produced by the National Statistical Office for Malaysia in 2019, is used as the principal empirical reference for the child wellbeing measures presented in this assessment. The source provides nationally relevant observations on access to basic services, household conditions, and reported barriers affecting children.\n\nTwo figures are taken directly from the survey. It reports that 68 percent of children were covered by the specified basic service, while 42 percent of households reported at least one difficulty in obtaining that service. These values are reproduced without adjustment and are treated as descriptive indicators rather than as estimates of change over time.\n\nThe figures support the assessment of current implementation conditions and provide a consistent basis for interpreting disparities across the reported welfare dimensions. They should be read with due attention to the scope of the survey and to the definitions applied in its tabulations. No additional calculation is required for the comparisons made in this section, since the cited proportions are used as reported by the CCWS.\n\nMention: \"Civil Child Wellbeing Survey (CCWS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The assessment takes the figures of 68 percent and 42 percent directly from the CCWS to describe service coverage and reported household difficulty.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:090383222444e215d138f47d", "occurrence_id": "cc82172880a9e30c49629ae6", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Data sources.** The Quarterly Cooperative Membership Registry (QCMR) is identified as a database produced by WhereIsMyTransport. The record used here refers to 2017, and no transformation, interpolation, or reconciliation has been applied to the reported totals. The source description is retained in full to support traceability in subsequent tables.\n\nTwo figures are drawn directly from this database. It reports 31 cooperatives and 684 members for the stated year. These values are reproduced as published and are treated as the principal evidence for the membership profile. They are not estimates derived from rates, and no additional series is introduced.\n\nOn this basis, the assessment uses the registry only to establish the observed scale of participation. The 31 cooperatives and 684 members provide a fixed reference for interpreting coverage, while avoiding assumptions about geographic distribution, institutional status, or changes outside the recorded period. Any comparison with other information would require a separate source and is therefore excluded from this note.\n\nMention: \"Quarterly Cooperative Membership Registry (QCMR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported figures of 31 cooperatives and 684 members from the 2017 QCMR database produced by WhereIsMyTransport.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:ac685e94c29bf4cd2bc64810", "occurrence_id": "99e9719a1e178255dd62c041", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Periodic Inequality Registry is cited as a database covering Panama in 2018. It is included to situate the discussion of distributional conditions and to indicate the availability of a relevant statistical reference. The citation is not used to establish a finding or to support an interpretation of change.\n\nThe assessment relies on the analytical material presented in the surrounding sections and does not draw figures, indicators, or classifications from the registry. Its mention therefore serves a limited documentary purpose. No calculation, comparison, or conclusion in this passage depends on values reported in that database.\n\nThis treatment is consistent with a cautious use of background references in implementation reporting. The registry is acknowledged as a potentially relevant point of orientation, while the evidentiary basis for the assessment remains separate from it. The reference should consequently be read as supplementary information rather than as an input to the results presented here.\n\nMention: \"Periodic Inequality Registry\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The passage identifies the Periodic Inequality Registry as a database for Panama in 2018 and uses it only for background, without drawing on its values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:84817d5e00b2aa22bd0e3cf8", "occurrence_id": "f878498a0fee590b5ba6a15c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Joint Output Area Dataset is used as the geospatial source covering the Arab Republic of Egypt for 2015. Its observations provide the principal basis for examining differences among output areas and for establishing a consistent analytical frame across the reported measures. The dataset is treated as an input to estimation rather than as a set of figures to be reproduced verbatim.\n\nArea-level values are aggregated to calculate mean outcomes and to identify the direction and magnitude of change across successive observations. This procedure reduces the influence of isolated entries and supports comparison on a common basis. The resulting averages and trend measures are therefore derived from the underlying observations, with no single recorded value treated as representative of the full coverage.\n\nThe analysis applies the same calculation rules throughout the assessment and retains the geographic structure supplied by the dataset. Results are presented as analytical estimates based on the observed distribution, rather than as direct quotations from the cited material. Any interpretation of variation is consequently linked to the computed averages and trend, while the source remains the central evidentiary basis for the assessment.\n\nMention: \"Joint Output Area Dataset\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives area-level averages and a trend from the Joint Output Area Dataset for the Arab Republic of Egypt in 2015.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:a429d16e042b2b16d9bb11fe", "occurrence_id": "a7407116f69dfeed41a7d7f7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Longitudinal Service Catchments Census (LSCC) is a census produced by COFECE for Turkey and referring to 2021. It provides a standardized basis for describing the distribution of service catchments and associated facilities. The source is cited as a primary statistical reference for the coverage assessment presented below.\n\nThe broader assessment indicates that service reach remains uneven across the observed catchments. Two figures are reproduced directly from the source, namely 63 facilities and 47 catchments. These values support the finding that the available service network does not extend uniformly across all recorded areas. They are presented as reported quantities and are not recalculated or combined with estimates from another source.\n\nInterpretation is limited to the scope of the census and to the definitions applied in its underlying tabulation. The figures therefore support the stated coverage claim, rather than as a basis for measuring change over time or projecting future requirements. Their inclusion also permits the assessment to distinguish between the number of facilities and the number of catchments, which are related but not interchangeable measures.\n\nMention: \"Longitudinal Service Catchments Census (LSCC)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the LSCC figures of 63 facilities and 47 catchments to support a claim about uneven service reach.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:8ed873eba4487ecc38eec831", "occurrence_id": "edf1d1ba89216627aa4d83bb", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment reviews the distribution of selected welfare and access indicators across population groups. The available evidence points to persistent differences in recorded outcomes, with implications for the targeting of interventions and the interpretation of aggregate measures. These findings are considered alongside administrative and survey-based material within the broader results framework.\n\nFor this purpose, the Diagnostic Inequality Registry (DIR) database produced by Coface, covering Mauritania and referring to 2017, was retrieved and archived for analytical use. The records were harmonised to align category definitions, geographic references, and indicator descriptions with the framework applied in this assessment. Where source classifications differed, the original distinctions were retained in the archive and mapped to the common analytical structure.\n\nThe resulting reference was used to substantiate the finding that inequality is not adequately represented by national averages alone. It provides a consistent basis for reviewing differences between population groups and for checking the comparability of reported indicators. The material was not treated as a substitute for primary collection, but as a documented input supporting interpretation of the wider evidence base. Its inclusion also preserves traceability for subsequent quality review and for any update to the analytical tables.\n\nMention: \"Diagnostic Inequality Registry (DIR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The passage records the archiving and harmonisation of the Diagnostic Inequality Registry (DIR) database produced by Coface for Mauritania in 2017 and its use to support an inequality finding.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:cfdc4444090d02e3b5db90a9", "occurrence_id": "8552ee3fa0879fbe3bcd6476", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Pilot Emission Inventory Registry database is used as the sole documentary source for the quantitative observations presented in this assessment. Its entries provide a consistent basis for reporting the distribution of recorded emission categories and the associated facility coverage. No supplementary source has been introduced for the figures cited in this section.\n\nThe reported count of 31 grades and the recorded total of 14 facilities are reproduced directly from the database. These figures are retained in their original form and are not recalculated, extrapolated, or combined with estimates from outside the registry. They therefore serve as the central evidence for the description of the inventory structure.\n\nThe results indicate that the registry contains sufficient detail to support a bounded review of classification and coverage. The two reported values should nevertheless be interpreted as registry observations rather than as an independent estimate of total activity. This distinction is material for implementation monitoring, since apparent changes in coverage may reflect reporting completeness or classification practice rather than a change in underlying emissions. The figures are accordingly used to document the recorded position at the time of assessment and to establish a transparent reference for subsequent review.\n\nMention: \"Pilot Emission Inventory Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly recorded values of 31 grades and 14 facilities from the Pilot Emission Inventory Registry database.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:b6fc949e9b93b80e77e5bdf7", "occurrence_id": "7b299309fc1e123f447c4251", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: For this assessment, the Cross-Sectoral Debt and Savings Survey (CDSS) is the principal statistical source covering Latvia in 2023. The survey is used to establish a consistent basis for reviewing debt exposure, saving behaviour, and the balance between the two measures. Source details are retained with the analytical record, while interpretation remains limited to the evidence contained in the survey.\n\nThe calculations do not reproduce isolated figures from the source. Instead, observations are grouped according to the relevant household characteristics, and average debt and savings values are calculated for each analytical group. These results support a concise trend assessment by showing whether debt burdens and saving levels move in the same direction across the observed groups. The resulting averages are treated as estimates for comparison within the assessment, rather than as independently verified national totals.\n\nInterpretation gives priority to the direction and relative scale of the calculated measures. Where differences are small, the findings are reported cautiously and are not presented as evidence of a structural change. Where the pattern is persistent across groups, it is summarised as a trend derived from the CDSS observations. This approach preserves the source’s role as central evidence while making clear that the reported results are analytical outputs computed from the survey, not figures copied from its tables.\n\nMention: \"Cross-Sectoral Debt and Savings Survey (CDSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives average debt and savings measures and a corresponding trend from the Cross-Sectoral Debt and Savings Survey (CDSS).\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:567b1e0bcebab00ad0924bd0", "occurrence_id": "a959c32a1669d0692662441a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Statistical Cash Transfer Registry Microdata are retained as a supplementary empirical reference for the assessment. This microdata source, produced by the WTO and referring to 2015, provides registry-based observations suitable for comparison with the estimates used in the analysis. The material is treated as a source for verification rather than as the sole basis for the reported results.\n\nThe available counts and distributional indicators are compared with an estimate derived from a separate source. The comparison is intended to identify material discrepancies in the level and composition of recorded transfers. Where the two measures are consistent, the registry evidence increases confidence in the corresponding estimate; where differences arise, the discrepancy is recorded for further review rather than resolved through an unsupported adjustment.\n\nThis cross-check supports the claim that the reported transfer profile is broadly aligned with the underlying registry evidence. It does not establish causality, nor does it alter the methodology used for the principal estimate. The comparison is therefore presented as a validation step within the wider assessment and is interpreted with due caution regarding coverage, timing, and reporting practices.\n\nUse of the Statistical Cash Transfer Registry Microdata is limited to this corroborative function. The cited material contributes an independent benchmark against which the relevant estimate can be assessed and provides a transparent basis for noting the strength and limits of the claim.\n\nMention: \"Statistical Cash Transfer Registry Microdata\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the 2015 WTO microdata in the Statistical Cash Transfer Registry Microdata to cross-check a separate estimate and support one claim.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:147dab1e0c8b23c51b462c5c", "occurrence_id": "068fa28de8db86292208ae5e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The National Asset Ownership Survey is used as the principal external reference for assessing reported patterns of asset ownership. It is a survey produced by the Universal Postal Union in 2020 and is treated as a stable benchmark for the review. The source is retained in its published form, with no adjustment to the stated coverage or underlying definitions.\n\nThe estimates generated for this assessment are cross-checked against the corresponding observations in the National Asset Ownership Survey. This comparison provides an independent test of the direction and approximate magnitude of the reported results. Where the two sources differ, the discrepancy is recorded as an issue for interpretation rather than resolved through mechanical adjustment.\n\nThe cross-check indicates that the principal estimate remains within a plausible range when assessed against the survey evidence. It also supports the use of the reported distribution as a reference point for reviewing internal consistency across the assessment. The comparison is therefore used to qualify the findings and to identify whether apparent changes reflect measurement differences or substantive variation.\n\nOn this basis, the National Asset Ownership Survey constitutes the central corroborating source for the estimate presented in this annex. Its role is evidentiary and comparative, supporting validation of the assessment rather than replacing the estimate obtained through the separate source.\n\nMention: \"National Asset Ownership Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The analysis uses the National Asset Ownership Survey to cross-check and qualify an independent asset ownership estimate.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:f1ec3305254cc49108a1a8a7", "occurrence_id": "7f6264c4cbfcefe8423fb7da", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis draws on the Standard Cash Transfer Registry Census, a census produced by the Government of India for 2022. The source is treated as the principal evidence for assessing the recorded distribution of cash transfer observations and for establishing a consistent basis for comparison within the dataset.\n\nRather than reproducing isolated figures, the assessment aggregates the available observations and calculates average values for the relevant transfer measures. It also derives a trend from the ordered records, allowing changes in concentration and coverage to be examined through summary statistics rather than through selected entries. These calculations are used to support interpretation of the registry evidence and to reduce the influence of atypical observations.\n\nThe resulting measures are reported as analytical estimates derived from the census material. They should therefore be read as evidence of patterns in the recorded population, not as direct quotations from the source. The approach preserves the underlying source attribution while making explicit that the reported averages and trend are generated through the present analysis. This treatment also provides a reproducible basis for subsequent monitoring, subject to the completeness and internal consistency of the census records.\n\nMention: \"Standard Cash Transfer Registry Census\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives average measures and a trend from the 2022 Standard Cash Transfer Registry Census rather than quoting individual figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:c4238f2f2dbcc79b8c87abf9", "occurrence_id": "e405b93fa35ae156f6789a1d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis draws on the Quarterly Forest Products Registry (QFPR), a database produced by Relief International for 2016. The registry supplies the underlying observations used to characterize recorded forest-product activity. It is treated as the principal empirical source for the assessment, and its entries are interpreted as observations suitable for aggregation rather than as isolated quotations.\n\nTo support comparison, the observations are reviewed for continuity and then combined into summary measures. The calculation derives an average across the available records and estimates the direction of change over the reporting sequence. This approach gives greater weight to the overall pattern than to any single entry. The resulting trend is therefore an analytical output based on the database, not a transcription of selected figures. Where observations differ in magnitude, the average provides a stable reference for interpretation, while the trend indicates whether activity is increasing, declining, or remaining broadly stable.\n\nInterpretation remains subject to the coverage and consistency of the recorded observations. Missing or uneven entries may affect the strength of the estimated movement, and the average should not be read as a direct measure of every individual record. Nevertheless, the QFPR provides a sufficiently structured basis for the calculation used here. The reported findings consequently present a derived assessment of central tendency and direction, with the source retained for verification and subsequent monitoring.\n\nMention: \"Quarterly Forest Products Registry (QFPR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses the 2016 QFPR database produced by Relief International to calculate an average and a trend.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:6cf9a6583a6a075fa736fbe1", "occurrence_id": "6fe258089e9ffe6095b89f59", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Data sources.** The Local Building Footprint Registry (LBFR) database is used as a reference for the assessment of recorded building footprints. The 2019 records provide a structured basis for checking whether the magnitude of the reported estimate is consistent with the available administrative evidence. No inference is made beyond the coverage represented in the cited records.\n\nWithin the broader assessment, the registry supports one claim concerning the scale of the built footprint. An estimate obtained through a separate measurement procedure is compared with the corresponding information in the LBFR database. This comparison is intended to identify material divergence rather than to replace the primary estimate or establish a separate time series.\n\nThe cross-check indicates whether the reported result falls within a plausible range supported by the registry evidence. Where differences arise, they are treated as a matter for interpretation in light of measurement scope, recording practices, and the distinction between observed entries and estimated totals. The cited source is therefore used as corroborating evidence for a bounded statement in the assessment, while the principal conclusion remains based on the independent estimate.\n\nMention: \"Local Building Footprint Registry (LBFR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the 2019 LBFR database to cross-check an independently derived estimate of building footprints.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:322365393f55c6861b36998d", "occurrence_id": "03acb852c9f434bea0f41fa6", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Stratified Social Protection Survey is the principal empirical source for the assessment. It is a survey produced by Standard and Poors and referring to 2023. The source provides structured responses on access to assistance, receipt of transfers, and perceived adequacy of support. Its use is limited to the information required for the measurement framework, with no additional source introduced for these dimensions.\n\nThe analysis does not reproduce individual responses as reported observations. Instead, responses are aggregated into comparable distributions, and weighted averages are calculated for the main indicators. These calculations provide the basis for identifying the direction and relative strength of trends across the observed population groups. The resulting measures are treated as analytical estimates derived from the survey material rather than as figures quoted directly from the source.\n\nInterpretation focuses on consistency across indicators and on the extent to which the calculated averages support the assessment of coverage and adequacy. Where response patterns vary, the analysis retains the underlying distribution and avoids presenting a single summary value as a complete description. This approach supports a cautious reading of changes in access and reported support, while maintaining a clear distinction between source observations and measures generated for the assessment.\n\nThe derived results are used to inform implementation monitoring and to identify areas requiring closer review. They should therefore be read as evidence constructed through aggregation and comparison, rather than as a simple transcription of the survey record.\n\nMention: \"Stratified Social Protection Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The assessment calculates weighted averages and trend measures from the 2023 Stratified Social Protection Survey rather than quoting individual responses.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:455f00200ea0762626ee8121", "occurrence_id": "1741eeff035ba25dcf250d28", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Annual Administrative Boundary Census (AABC) is used as the census source for 2017 in this assessment. The record is cited in its stated form, without adjustment, interpolation, or supplementary classification. Its role is limited to establishing the observed administrative distribution relevant to the analysis.\n\nTwo figures are quoted directly from this source. The census records 31 administrative units and a coverage level of 68.4 percent. These values are reproduced as reported, rather than recalculated from component observations, and they provide the central numerical evidence for the assessment.\n\nThe 31 administrative units define the scale of the recorded structure, while the 68.4 percent figure indicates the share covered in the same census account. Together, the figures support a bounded reading of the available evidence and avoid implying changes that are not documented in the source.\n\nInterpretation is therefore restricted to the values presented in the AABC census for 2017. No additional estimate is introduced, and no inference is made beyond the relationship shown by the two quoted figures. This treatment preserves traceability between the narrative statement and the underlying record.\n\nMention: \"Annual Administrative Boundary Census (AABC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported figures of 31 administrative units and 68.4 percent from the 2017 AABC census.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:3e71b10533132d1dc05c8ca2", "occurrence_id": "b5d1101102e3d0568928b16d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Midline Skills and Employment System (MSES) is used as an administrative source for observations pertaining to 2020. Its records provide a consistent basis for reviewing participation and employment outcomes within the assessment, while limiting reliance on isolated reported values.\n\nThe analysis derives an average from the available observations and uses the resulting measure to identify the direction of change across the reporting period. Individual records are therefore treated as inputs to a summary calculation rather than reproduced as standalone figures. This treatment supports a broader finding on the gradual variation in employment outcomes.\n\nThe calculated measure is considered alongside the implementation evidence presented elsewhere in this section. It is used to substantiate the assessment of labour market performance, while recognising that an aggregated result may conceal differences among participants. The source accordingly serves as supporting evidence for the reported trend and not as the sole basis for interpretation.\n\nMention: \"Midline Skills and Employment System (MSES)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses 2020 administrative MSES records to calculate an average supporting an employment trend.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:02a6db48c86d6ca7fa4ecdbd", "occurrence_id": "1e704822dacef427c97e6a15", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment considers institutional capacity through a set of general observations on administrative practice, data stewardship, and the conditions affecting implementation. Within that discussion, the Cross-Sectoral Diagnostic Capacity Survey (CDCS) survey produced by Polity IV is cited as a relevant external source for the description of prevailing measurement constraints.\n\nThe citation supports the observation that diagnostic capacity is unevenly documented across administrative domains. It serves to situate that observation within a wider body of comparative evidence, while the assessment does not reproduce, analyse, or derive any values from the source. No scores, classifications, counts, or other source results are incorporated into the findings or used to substantiate calculations.\n\nThe source is therefore treated as supporting context for a single claim within the broader paragraph, rather than as an input to the assessment framework. The conclusions presented here rest on the evidence and analysis set out in the report, with the cited survey retained solely to indicate the relevance of the stated measurement issue.\n\nMention: \"Cross-Sectoral Diagnostic Capacity Survey (CDCS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The source provides contextual support for one claim, and no values or results are taken from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:2a87764388394648a4dd1795", "occurrence_id": "036189456785c1a5a4e18f05", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Analytical Water Body Dataset is a geospatial source produced by the Commerce Department and is used as a reference for reviewing the reported extent of recorded water bodies. Its delineations provide a consistent basis for checking whether the estimated coverage is plausible and whether apparent differences reflect measurement conditions rather than substantive change.\n\nThe estimate presented in this assessment was derived from a separate statistical source. Comparison with the Analytical Water Body Dataset indicates that the two measures are directionally consistent, although their underlying definitions and observation procedures are not identical. This cross-check supports the claim that the reported coverage is within a reasonable range and does not, by itself, establish a change in underlying conditions.\n\nThe comparison is treated as a validation step rather than as a replacement for the primary estimate. Differences may arise from boundary treatment, the inclusion of small features, or the timing and method of observation. These qualifications have been retained in interpreting the result, and the cited dataset is used only to assess the credibility of the estimate within the broader evidence base. The finding should therefore be read as corroboration of the reported level, subject to the limitations affecting both measures.\n\nMention: \"Analytical Water Body Dataset\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The author uses the Analytical Water Body Dataset to cross-check an independently derived estimate of water-body coverage.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:fc0aa9c9a220dfad5c672761", "occurrence_id": "186de5f71819d0e61f384dd7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Diagnostic Settlement and Elevation Registry database covering the Philippines provides the central empirical basis for this assessment. Its entries are treated as observations of settlement location and elevation, with attention to the consistency of geographic coverage and the comparability of recorded measures. The source is used to support a consolidated assessment of spatial exposure rather than to reproduce individual records.\n\nThe analysis derives average elevation measures from the registry and compares these measures across settlement groupings. It also computes a trend from the available observations, using the resulting direction and magnitude as indicators of changing exposure patterns. This treatment reduces the influence of isolated entries and places emphasis on aggregate movement within the recorded settlement population.\n\nResults should therefore be interpreted as calculated estimates based on the Diagnostic Settlement and Elevation Registry database. They summarize the distribution and movement evident in the underlying records and do not constitute direct quotations from the source. Differences between computed averages may reflect both substantive variation and uneven representation in the registry. The findings are used to inform prioritization, while retaining the source as the principal reference for subsequent verification and monitoring.\n\nMention: \"Diagnostic Settlement and Elevation Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The assessment computes average elevation measures and a settlement trend from the Diagnostic Settlement and Elevation Registry database covering the Philippines.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:d701f177077ddb7422c22811", "occurrence_id": "24e72e24d86f7a33bd144b6f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Biennial Referral Pathways Index (BRPI) is used as the 2018 indicator for assessing the recorded functioning of referral pathways. Its scope is treated as fixed for the present assessment, and no adjustment is made to the reported values. The index is cited as the authoritative source for the two observations reproduced below.\n\nThe assessment records a referral completion rate of 62 per cent and a follow-up documentation rate of 14 per cent. These figures are taken directly from the Biennial Referral Pathways Index (BRPI) and are presented without recalculation, interpolation, or conversion. They provide the principal quantitative evidence for the finding that completion was substantially more common than documented follow-up.\n\nInterpretation is limited to the values reported by the indicator. The difference between the two rates is therefore reported as an observed disparity rather than as an estimated trend or a measure of change over time. The figures should be read as summary measures of recorded cases, with the usual limitations associated with coverage, reporting completeness, and consistency of classification.\n\nMention: \"Biennial Referral Pathways Index (BRPI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported values of 62 per cent for referral completion and 14 per cent for follow-up documentation from the 2018 BRPI indicator.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:0cfd637ec78d15693895dab6", "occurrence_id": "6fbc37bc0417e4a1372b2daf", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Monthly Emission Inventory Registry database is used as the primary documentary source for the assessment of reported emissions. Its structure permits review of recorded quantities across reporting categories and supports a consistent reading of the information available for implementation monitoring. No adjustment is applied to the cited values in this note.\n\nThe assessment quotes two figures directly from the database. The recorded quantity for one reporting category is 12.4 tonnes, while the corresponding quantity for another category is 8.7 tonnes. These figures are reproduced as reported and are not recalculated, rounded, or combined with estimates from other sources.\n\nTaken together, the cited values support the broader finding that reported emissions vary materially across categories and that implementation monitoring should retain category-level detail. The difference between the two figures is therefore relevant to the interpretation of reporting completeness and to the design of subsequent verification work. The evidence remains limited to the entries contained in the Monthly Emission Inventory Registry database, and no inference is made beyond the figures directly cited here.\n\nMention: \"Monthly Emission Inventory Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported figures of 12.4 tonnes and 8.7 tonnes from the Monthly Emission Inventory Registry database to support a finding of material variation across reporting categories.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:b52a73575518c4c865d3946c", "occurrence_id": "b78b735d4bd93accc523d7f1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Annual Hiring and Vacancies Census (AHVC) for 2020 is used as the principal census for the employment indicators examined in this annex. The AHVC provides structured observations on hiring activity and unfilled positions, with sufficient coverage to support comparative calculations. No single reported observation is treated as a headline estimate; the source establishes a consistent empirical basis for the assessment.\n\nTo reduce the effect of isolated movements, the analysis calculates an average hiring rate across the available observations and derives a vacancy trend from the corresponding distribution of unfilled positions. These measures are generated directly from the AHVC records through aggregation and comparison, rather than being reproduced as selected figures. The computed results therefore reflect the underlying pattern in the census rather than a quotation of individual entries.\n\nThe resulting average and trend provide a summary measure for monitoring labour demand and recruitment pressures. They are used to support interpretation of the employment indicators and to assess the consistency of observed hiring conditions. The calculations are retained as the central evidence for this assessment, while individual census entries remain supporting material.\n\nMention: \"Annual Hiring and Vacancies Census (AHVC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the 2020 AHVC census to calculate an average hiring rate and derive a vacancy trend.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:dca56f3fdfc94e786762be88", "occurrence_id": "e6ed3182b8b15e487a6bb55e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Harmonised Settlement and Elevation Registry as the 2020 database for establishing the settlement and elevation baseline. The source is applied without adjustment to the figures presented in this paragraph, with attention to consistency between settlement counts and recorded elevation values.\n\nTwo figures are reproduced directly from the registry. The database records 74 settlements in the assessed inventory, while the highest recorded elevation is 612 metres. These values are cited as reported and are not recalculated for the purposes of this assessment.\n\nThe evidence supports the conclusion that exposure conditions vary materially across the settlement inventory and that a common baseline is required for prioritising implementation activities. The settlement count indicates the scale of the units requiring review, while the elevation figure provides a reference point for interpreting physical differentiation within the assessed system.\n\nThe figures should therefore be read as descriptive inputs to planning rather than as estimates of future change. Subsequent monitoring may refine the baseline, but the present assessment retains the registry values to preserve traceability between the reported finding and its underlying source.\n\nMention: \"Harmonised Settlement and Elevation Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage takes the direct figures of 74 settlements and 612 metres from the 2020 database.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:822bfa455ea39944efc07757", "occurrence_id": "0130269a426321a773e36017", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Rolling Settlement and Elevation Index, produced by UNHCR, IOM, is cited as an indicator for 2017. It is recorded in the evidence base as a recognised reference concerning settlement conditions and elevation characteristics. Its inclusion provides limited descriptive orientation for the surrounding discussion and does not establish a basis for the findings presented here.\n\nThe assessment does not draw on the index to quantify changes, compare units, or validate estimates. No figures, classifications, or thresholds from the source are incorporated into the tables, calculations, or narrative results. The reference is retained solely to acknowledge a relevant strand of information associated with the subject under review.\n\nInterpretation therefore rests on the evidence reported directly in this document. The cited indicator is not used to support an inference, determine an allocation, or substantiate a performance judgement. Its role is confined to brief background recognition, while analytical conclusions remain independent of the source and its underlying observations.\n\nMention: \"Rolling Settlement and Elevation Index\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The source is acknowledged as contextual background, while none of its values or classifications are used in the assessment.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:07f3212798b7ab71c355e7a3", "occurrence_id": "1aceb105c1899e73cc4f192b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Government Housing Conditions Survey is a survey produced by WBG/IFC and is used as supporting evidence for the assessment of housing conditions. Its observations provide a basis for interpreting reported differences in dwelling quality and access to essential services. The source is treated as one component of the evidence base rather than as a complete account of household welfare.\n\nRecords received through the survey were collected, checked for internal consistency, and harmonised before inclusion in the working archive. Responses were aligned to common categories so that differences in wording and reporting practice would not be mistaken for substantive variation. The resulting archive supports the preparation of comparable summaries and permits the underlying observations to be reviewed during implementation monitoring.\n\nOn this basis, the assessment identifies persistent variation in housing conditions while recognising that the available evidence reflects reported circumstances and may not capture every dimension of dwelling quality. The cited material therefore backs the claim that service deficits and structural shortcomings remain relevant considerations in interpreting household living conditions. It also provides a documented trail for subsequent verification, revision, and controlled reuse of the compiled evidence.\n\nMention: \"Government Housing Conditions Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The passage uses the Government Housing Conditions Survey as harmonised and archived evidence for the assessment of housing conditions and service deficits.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:222672b2d554d61ca51b553f", "occurrence_id": "c03ad516269189a52a593544", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The High-Resolution Occupational Safety Survey (HOSS) is used as the principal survey for assessing occupational safety conditions. Its observations provide the central evidence for the analysis of reported incidents, exposure patterns, and compliance-related measures. The source is treated as a structured basis for comparison across reporting intervals, with attention to consistency in definitions and measurement procedures.\n\nThe analysis computes average levels from the HOSS observations and derives a trend from their progression over successive intervals. Individual values are therefore not presented as standalone findings. Instead, the calculations aggregate the available records to identify the prevailing direction and the degree of variation around the mean. This approach reduces the influence of isolated fluctuations and supports a more stable interpretation of change.\n\nResults indicate that the derived average and trend are suitable for monitoring purposes, subject to the stated coverage and measurement limitations of the survey. The estimates are used to inform implementation review, identify areas requiring follow-up, and establish a common basis for subsequent measurement. All conclusions in this section rely on the calculations performed from HOSS rather than on direct reproduction of selected observations.\n\nMention: \"High-Resolution Occupational Safety Survey (HOSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives average levels and a trend from the HOSS survey observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:a4c1ea10860d21a633fa27c0", "occurrence_id": "b92f9935d8b24fce63d96b3f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis draws on the Independent Multidimensional Poverty Survey (IMPS), a survey that provides the principal empirical basis for the estimates presented in this annex. The observations are treated as a coherent set for analytical purposes, with attention to consistent definitions, comparable units of measurement, and the distribution of deprivation across the population. No individual observation is used as a stand-alone finding.\n\nThe IMPS evidence is processed to derive summary measures rather than reproduced as isolated quotations. The analysis computes average levels across the relevant observations and uses those averages to identify the direction and persistence of change. This approach reduces the influence of unusually high or low observations and supports comparison across the analytical categories used in the assessment. The resulting estimates therefore represent calculated summaries of the survey material.\n\nThe derived trend indicates whether measured deprivation is increasing, declining, or remaining broadly stable over the sequence of observations available for analysis. Where changes are limited, the assessment reports them as marginal rather than assigning substantive significance. Where the average differs materially between comparison groups, the difference is retained for interpretation, subject to the limitations of the underlying measurements. The IMPS thus serves as the central evidence for the statistical assessment, while the reported results are generated through aggregation and comparison rather than direct repetition of source figures.\n\nMention: \"Independent Multidimensional Poverty Survey (IMPS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses IMPS survey observations to calculate comparative averages and derive a trend in measured deprivation.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:9a9e4895a8626ec5076abff8", "occurrence_id": "140493dc11b2d59f2eda3a29", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Recurrent Minimum Needs Index (RMNI) is retained as a reference indicator for the assessment. The 2021 RMNI indicator is produced by WHO and provides a structured benchmark for reviewing the plausibility of needs-related estimates. Its inclusion supports interpretation of the selected claim without extending the source beyond its stated measurement purpose.\n\nThe estimate reported in the main analysis was cross-checked against the RMNI indicator as a separate reference. The comparison was used to assess whether the direction and approximate scale of the estimate were consistent with an independently produced measure. This check does not replace the primary estimate, which remains based on its original source and calculation procedure.\n\nResults from the comparison were considered supportive where the two measures pointed in the same direction. Differences were treated as a basis for caution rather than as evidence of error, since the measures may reflect distinct definitions, coverage decisions, and reporting practices. The RMNI therefore serves one evidentiary function within the broader assessment, namely validation of a specific claim, while other findings continue to depend on the underlying analysis and its documented assumptions.\n\nMention: \"Recurrent Minimum Needs Index (RMNI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The 2021 WHO RMNI indicator was used to cross-check the direction and approximate scale of an independent estimate.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:4ec2c248096239c0dbe94e35", "occurrence_id": "c5d2407bc6f8b6a77d635aa6", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Use of the Multi-Topic Displacement Tracking Registry database is limited to the evidence required for the assessment. The database is identified as a 2020 source produced by OECD-PISA, and its entries are retained in the form in which they are reported. No additional interpretation is applied to the cited values. This treatment supports traceability while avoiding an inference beyond the recorded observations.\n\nTwo figures are taken directly from the database and reproduced in the results. The first indicates that 31 per cent of the recorded observations met the relevant reporting condition. The second indicates that 68 per cent of the recorded observations met the corresponding comparison condition. These values are presented as reported, rather than recalculated or combined, and they constitute the principal quantitative evidence for the assessment.\n\nTaken together, the figures indicate a material difference between the two reported conditions. The difference is noted for descriptive purposes only and is not interpreted as a causal relationship. The assessment therefore uses the database to document the reported distribution, while maintaining the distinction between direct evidence and analytical interpretation. Any subsequent discussion should preserve the two values and their original denominator.\n\nMention: \"Multi-Topic Displacement Tracking Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Multi-Topic Displacement Tracking Registry database to reproduce 31 per cent and 68 per cent as direct quantitative evidence.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:0a80dd42059321b864b029e9", "occurrence_id": "a3b0bccc6a8b2aa022dfb859", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Pilot Socioeconomic Survey (PSS) provides the survey basis for the assessment of household welfare conditions. Its observations are treated as a consistent analytical input alongside the indicators discussed in this section, allowing changes in reported access and consumption to be examined without assigning undue weight to any single response. The source is cited here to support the broader finding that welfare conditions vary across the population.\n\nFor this purpose, the analysis derives an average from the relevant survey observations and uses the resulting pattern to estimate a trend over the observation period. The reported conclusion therefore reflects a calculation based on the PSS records rather than a reproduction of isolated values from the source. This treatment reduces the influence of exceptional responses and provides a more stable basis for comparison.\n\nThe resulting trend is consistent with the assessment of gradual differences in household circumstances and supports the interpretation presented in the preceding discussion. The calculation is used as corroborating evidence, while the main conclusions continue to depend on the combined reading of the available indicators. All estimates should be understood as analytical summaries of the survey material and not as direct statements attributed to individual respondents.\n\nMention: \"Pilot Socioeconomic Survey (PSS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses PSS survey observations to calculate an average and derive a trend supporting the broader assessment of household welfare conditions.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:bee351e6c721c75f8a9a46e3", "occurrence_id": "26366cce6893cfa4649ba1a0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment considers livelihood diversification as one element of rural resilience. The Independent Rural Livelihoods Registry is a database for 2018 and provides the underlying observations used to substantiate this assessment. The assessment team collected the records, harmonised differences in recording practice, and archived the resulting material for controlled use in the analytical file.\n\nThe source supports the finding that households combine several livelihood activities rather than relying on a single reported stream of income. This evidence is used to qualify the interpretation of livelihood security, particularly where reported activity is intermittent or distributed across household members. The registry is treated as supporting evidence within the wider results framework and not as a substitute for direct monitoring of project outputs.\n\nProcessing focused on consistency of categories, removal of duplicate entries, and preservation of the original reporting structure. No additional inference is drawn from the registry beyond the relationship required for the assessment claim. The archived material permits subsequent review of the derivation of the analytical observations while maintaining a clear separation between source records and conclusions presented in the report.\n\nMention: \"Independent Rural Livelihoods Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The passage uses the 2018 database to support a claim about livelihood diversification and records its collection, harmonisation, and archiving.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:4c97f40cbdb2c1db03d582ac", "occurrence_id": "65edcd97f1109535293aa6f4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Urban Poverty and Vulnerability Registry is used as a database for assessing household exposure to economic and social risks. Its records provide a consistent reference for interpreting the distribution of vulnerability and for reviewing whether reported conditions are concentrated among households with limited capacity to absorb shocks.\n\nTwo figures are reproduced directly from this database. The recorded share of households classified as vulnerable is 42 percent, while the share reporting difficulty in meeting essential expenditures is 18 percent. These figures are presented as reported values and are not adjusted or recalculated for this assessment.\n\nTaken together, the evidence supports the broader finding that vulnerability is associated with constrained household resources and uneven access to basic economic security. The difference between the two reported shares also indicates that vulnerability is not limited to households experiencing immediate expenditure difficulties. The registry therefore provides a useful basis for distinguishing general exposure from more acute manifestations of hardship.\n\nInterpretation remains limited to the measures captured in the database. The figures should be read as descriptive evidence supporting the assessment of vulnerability rather than as estimates of causal relationships or forecasts of future conditions.\n\nMention: \"Urban Poverty and Vulnerability Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage takes the 42 percent vulnerable-household share and the 18 percent essential-expenditure difficulty share directly from the database.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:c9f63e4856fc4c5b2b0d8d1d", "occurrence_id": "3add822a2b97fa1fc8e1b0f7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Periodic Deprivation and Hardship Survey (PDHS), a survey covering Buenos Aires. The source is used to support the interpretation of household hardship indicators and to provide a direct empirical reference for the discussion that follows. Its coverage is treated as relevant to the geographic scope of the assessment, while no adjustment or recalculation is applied to the reported values.\n\nThe findings indicate that deprivation is not distributed evenly across the population. Two figures reproduced directly from the PDHS show that 37 percent of households reported difficulty meeting essential expenses, while 18 percent reported a prolonged interruption in access to at least one basic service. These values support the broader claim that hardship affects both immediate consumption and the continuity of essential conditions, although the indicators should not be interpreted as interchangeable measures.\n\nThe figures are presented as reported by the source and are used solely to substantiate the assessment narrative. They provide a concise indication of the scale and composition of hardship within Buenos Aires and allow the discussion to distinguish between short-term financial pressure and more persistent service-related deprivation. No additional inference is drawn from the two values beyond their stated contribution to this claim.\n\nMention: \"Periodic Deprivation and Hardship Survey (PDHS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The author uses the PDHS figures of 37 percent and 18 percent to support the assessment of household hardship in Buenos Aires.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:9281c5ee6e94b112b2708183", "occurrence_id": "2ce29e2fbbd6c46427a2594a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Quality-Assured Working Conditions Registry database is used as the primary evidentiary source for the assessment of recorded working conditions. Its entries provide a consistent basis for describing the classification structure and the level of compliance reported in the underlying records. No supplementary source is required for the two quantitative observations presented here.\n\nTwo figures are taken directly from this database and retained without recalculation. The registry records 31 grades in the applicable classification, while 68 percent of the reviewed records meet the stated working-conditions threshold. These figures are reported as recorded and are not adjusted for differences in coverage, reporting practice, or administrative completeness.\n\nThe evidence indicates that the classification contains a substantial range of recorded grades, but that compliance remains below universal coverage. The figures are therefore used to establish the scale of the classification and the proportion meeting the threshold, rather than to infer change over time. This treatment preserves the traceability of the assessment and limits interpretation to what is documented in the cited database.\n\nMention: \"Quality-Assured Working Conditions Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes 31 grades and 68 percent directly from the Quality-Assured Working Conditions Registry database.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:990c376cba4a4c635683e571", "occurrence_id": "4c68856ed6704a2964ad569b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Statistical Housing Conditions Survey (SHCS) provides the principal evidence for the assessment of housing conditions in Switzerland in 2021. It is treated as a survey source for establishing a consistent basis for comparison across the reported dimensions of dwelling quality, occupancy, and access to basic services. No additional source is introduced for this purpose.\n\nThe assessment does not reproduce isolated observations from the SHCS. Instead, it calculates average results across the available housing-condition measures and uses these derived values to indicate the prevailing situation. The resulting averages are used as analytical inputs rather than presented as direct quotations from the survey.\n\nThis approach supports a concise interpretation of the evidence while limiting emphasis on individual records. It also permits the findings to be read as an aggregate profile of housing conditions, with the calculated averages providing the main basis for discussion. Differences among measures are retained in the underlying calculations, but the narrative reports the combined pattern that emerges from them.\n\nThe derived results should therefore be understood as summary statistics generated from the Statistical Housing Conditions Survey (SHCS), rather than as figures copied directly from a source table. They inform the assessment of housing conditions in Switzerland in 2021 and provide the basis for the conclusions set out in this section.\n\nMention: \"Statistical Housing Conditions Survey (SHCS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage explains that average housing-condition results were calculated from the Statistical Housing Conditions Survey (SHCS) for Switzerland in 2021 rather than quoted directly.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:a6cb197ae0751b783cad6bb8", "occurrence_id": "e0bdb57f47674b05ddee08f3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Stratified Early Warning Dataset (SEWD) is treated as a geospatial source produced by CEDLAS-World Bank. Its stratified observations provide an external reference for assessing the distribution underlying the estimate presented in this section. The dataset is used in support of the interpretation rather than as a replacement for the primary series.\n\nThe estimate is cross-checked against the corresponding SEWD observations to assess whether the reported pattern is consistent with an independently constructed source. This comparison supports the claim that the observed differences are not attributable solely to the aggregation procedure applied to the primary data. The cross-check is therefore limited to the specific claim concerning the relative distribution, while other results continue to rely on the principal analytical source.\n\nThe comparison indicates broad agreement in the direction and concentration of the reported pattern. Any residual discrepancy is treated as a measurement or coverage difference and does not alter the assessment. SEWD is consequently retained as a corroborating reference within the evidence base, with its role confined to validation of the estimate and the associated interpretation.\n\nMention: \"Stratified Early Warning Dataset (SEWD)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the Stratified Early Warning Dataset (SEWD) from CEDLAS-World Bank as a geospatial corroborating source to cross-check one estimate from a separate source.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:43367a2bb97c0f36e97aa294", "occurrence_id": "a4b7544dc9dd68ff68ac6768", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Decennial Apprenticeship Survey is used as the principal survey covering Zatari camp. It provides the central evidence for assessing participation in apprenticeship activities and the distribution of learners across the recorded categories. The source is considered sufficiently structured for comparative calculation within the implementation review.\n\nRather than reproducing individual observations, the analysis calculates an average level of participation and derives a trend across successive observation points. These measures are used to identify the direction and approximate pace of change, while reducing the influence of isolated movements in the underlying records. The resulting pattern indicates whether access has expanded, remained stable, or weakened during the period represented in the survey.\n\nThe calculated indicators support interpretation of implementation performance but are not treated as independently verified estimates of the wider population. Differences in reporting coverage, participation status, and category assignment may affect the measured pattern. Accordingly, the average is presented as a summary statistic for monitoring, and the trend is used to inform operational assessment rather than to establish a causal result.\n\nFor consistency, all calculations retain the categories and observation structure contained in the Decennial Apprenticeship Survey. The analysis therefore preserves comparability with the source while making explicit that the reported findings are derived measures. No single observation is given priority over the aggregate pattern, and conclusions are limited to the coverage of Zatari camp.\n\nMention: \"Decennial Apprenticeship Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives an average and a trend from the Decennial Apprenticeship Survey covering Zatari camp.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:105fecd3cd1b26141a387c97", "occurrence_id": "11651c751fcab8f5ec554fbe", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Endline Targeting and Payments Microdata (ETPM) as a supplementary source for reviewing reported targeting results. The ETPM microdata cover Burkina-Faso and refer to 2023. Their use is limited to a specific verification exercise within the broader analysis of programme performance.\n\nThe principal estimate is derived from a separate evidentiary basis and is not replaced by the ETPM material. Instead, the corresponding observations in the ETPM microdata are used to cross-check whether the direction and approximate magnitude of the estimate are plausible. This comparison provides support for the claim concerning the reach of payments among the intended population.\n\nThe comparison does not establish a new aggregate measure, and differences between the two sources are treated as a reason for qualification rather than as evidence of error. The ETPM material is therefore cited for corroboration of one claim in the assessment, while the broader findings remain anchored in the primary estimate and its associated methodological discussion. This use is consistent with a verification role for payment and targeting records, particularly where coverage and inclusion outcomes require an independent check.\n\nMention: \"Endline Targeting and Payments Microdata (ETPM)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the 2023 ETPM microdata for Burkina-Faso to cross-check one estimate in a broader assessment.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:aefc7a54d0b84f257a18f94d", "occurrence_id": "f279db3b4e7aaa0cb966e8aa", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Urban Rural Credit Dataset is a geospatial source cited in support of the assessment of credit access and service reach. Its inclusion provides a documented basis for situating the discussion within the broader relationship between settlement patterns and financial inclusion, while keeping the evidentiary focus of the assessment on the findings presented elsewhere in this paragraph.\n\nThe citation supports the observation that access to formal credit may differ between urban and rural settings and that such differences are relevant to the interpretation of implementation constraints. This observation is treated as contextual evidence for one claim within the broader assessment. The dataset itself is not used to establish a measure, calculate a result, or report a distribution.\n\nNo values from the Urban Rural Credit Dataset are reproduced, aggregated, compared, or used in a derived indicator. The source therefore serves a limited supporting function, allowing the assessment to acknowledge the geographic dimension of credit access without extending the analysis beyond the evidence directly examined for the project. This treatment is consistent with a cautious approach to secondary information and preserves a clear distinction between contextual reference and reported results.\n\nMention: \"Urban Rural Credit Dataset\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The passage uses the Urban Rural Credit Dataset to support a contextual claim about urban–rural differences in credit access and takes no numerical values from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:b88385e00db3f915a553f03d", "occurrence_id": "982b57d4d5e8a07902915769", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Multi-Topic Unemployment Insurance Survey is cited as a contextual source for the discussion of labour-market protection and household exposure to income loss. It is a survey produced by the municipal government in 2016 and is used to establish the relevance of the topic within the assessment. The citation is limited to this framing function, and no value from the source is introduced into the analysis.\n\nWithin the broader discussion, the source supports the claim that unemployment insurance remains pertinent to the assessment of household resilience. The claim is presented alongside administrative evidence and implementation observations, but the cited survey is not treated as a basis for measurement, comparison, or estimation. Its inclusion therefore provides background for the argument without changing the numerical results reported elsewhere.\n\nThe source is retained in the reference material because it identifies an established line of inquiry relevant to the project’s analytical framework. No figures, rates, distributions, or derived indicators are taken from it. This limited use is consistent with the report’s treatment of contextual references, which inform interpretation while leaving the assessment’s calculations dependent on the evidence specified for each result.\n\nMention: \"Multi-Topic Unemployment Insurance Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The passage uses the 2016 Multi-Topic Unemployment Insurance Survey produced by the municipal government only to frame one claim, without extracting or applying any of its values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:ab50d941bbbdb33f94d9b87a", "occurrence_id": "17ff261362c765de194b22ab", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Rolling Teacher Workforce Survey is used as the principal survey for assessing changes in the teacher workforce. The 2022 observations provide the central evidence for the analysis of staffing levels, employment status, and reported workload. The dataset is treated as an administrative-quality input for statistical processing, subject to checks for missing responses and inconsistent classifications.\n\nThe assessment does not reproduce individual responses or present isolated figures as findings. Instead, it calculates comparable averages across reporting units and derives a trend from the available observations. This approach reduces the influence of unusually high or low responses and supports a more stable interpretation of workforce movement. The resulting indicators are used to examine whether staffing conditions are improving, weakening, or remaining broadly unchanged.\n\nQuality assurance consisted of reviewing the internal consistency of the records, checking the denominator used in each calculation, and applying the same aggregation procedure throughout the analysis. Where response coverage differed between measures, the calculation retained only observations meeting the applicable completeness threshold. The derived averages and trend indicators should therefore be interpreted as analytical estimates based on the survey, rather than as direct quotations from it. They provide the basis for monitoring workforce pressures and for identifying areas requiring further operational review.\n\nMention: \"Rolling Teacher Workforce Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses the 2022 Rolling Teacher Workforce Survey to calculate comparable averages and derive a teacher-workforce trend.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:74a9767966890b5f852e39ef", "occurrence_id": "7d24aad8b687e546682a1abf", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Annual Building Footprint Registry database, produced by the Federal Bureau of Statistics, is used as the principal reference for the assessment of recorded building assets. The source provides a consistent statistical basis for reviewing the scale and functional composition of the building stock, while the interpretation remains limited to the information represented in the registry.\n\nTwo counts are reproduced directly from the registry to support the assessment. It records 18,742 building footprints in the assessed compilation, including 6,315 classified as non-residential. These figures indicate a substantial registered stock with a material non-residential component and support the conclusion that the building base is functionally diverse.\n\nThe figures are presented without adjustment, extrapolation, or reconciliation with external records. They should therefore be read as reported registry counts rather than as estimates of the complete physical stock. Any subsequent planning use should preserve the distinction between recorded footprints and buildings that may not be represented in the database.\n\nMention: \"Annual Building Footprint Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Annual Building Footprint Registry to report 18,742 building footprints and 6,315 non-residential buildings.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:e102d185b228371a03d2d00a", "occurrence_id": "d912d23af5d38006642fea50", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Enumerative Referral Pathways Survey (ERPS) is a survey used as the principal evidentiary basis for this assessment. It provides a direct account of movement through the referral process and supports a limited comparison of recorded outcomes. The source is cited in full to preserve the distinction between reported observations and figures derived through subsequent analysis.\n\nTwo figures are reproduced directly from ERPS. The survey records 31 completed referrals at the initial point of review and 68 completed referrals at the subsequent point of review. These values are presented as reported, without adjustment, interpolation, or recalculation. They constitute the central quantitative evidence for the paragraph and should not be interpreted as an estimated rate or as a measure of change beyond the observations recorded in the survey.\n\nThe difference between the two reported figures is therefore retained only as a descriptive contrast. It indicates a higher recorded volume at the subsequent point, while leaving unresolved the factors that may account for the difference. No inference is made regarding coverage, representativeness, or the completeness of the underlying records.\n\nFor monitoring purposes, ERPS should be read together with the accompanying narrative evidence. Its value in this assessment lies in the transparent presentation of the 31 completed referrals and the 68 completed referrals, both taken directly from the cited survey and kept separate from any analytical interpretation.\n\nMention: \"Enumerative Referral Pathways Survey (ERPS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes from the Enumerative Referral Pathways Survey (ERPS) the directly reported figures of 31 and 68 completed referrals.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:f31f97b831b07fcadc535f5a", "occurrence_id": "b26481ddee80c5aa1ab48de4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The principal evidence for this assessment is the Combined Facility Assessment Survey (CFAS), a survey covering Uganda. The source provides observations on facility capacity, staffing, equipment availability, and selected service indicators. Its coverage supports comparison across facilities and permits the results to be interpreted as a system-level measure rather than as isolated case evidence.\n\nThe analysis derives mean values from the facility observations and examines the direction of change across the reporting sequence. This approach reduces the influence of individual facilities and provides a more stable basis for assessing service readiness. The resulting averages are used to identify areas of relative strength, persistent constraint, and uneven performance across the measured domains.\n\nThe trend calculations indicate that progress is not uniform across indicators. Capacity-related measures show a stronger average position than measures dependent on regular availability of inputs and personnel. The findings therefore support a cautious interpretation of aggregate improvement, with continued attention required to the consistency of service conditions. All calculations presented in this section are derived from the Combined Facility Assessment Survey (CFAS), rather than reproduced as direct quotations from the source.\n\nMention: \"Combined Facility Assessment Survey (CFAS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives facility-level averages and a directional trend in service readiness from CFAS.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:84281f0a42aa6df842e155cb", "occurrence_id": "639682243f8d6cc3498dd029", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Enumerative Housing Conditions Microdata, a microdata source produced by NBS, as its principal empirical basis. The material is treated as an analytical record rather than as a source for isolated quotations. Its structure permits observations to be reviewed on a consistent basis and supports comparison across the reported units.\n\nThe analysis calculates average conditions from the underlying observations and derives a trend by comparing the resulting measures across successive reporting periods. Individual entries are therefore not presented as stand-alone findings. Instead, the reported estimates summarise the distribution of conditions represented in the source and reduce the influence of atypical observations.\n\nThis approach is used to assess whether conditions are improving, stable, or deteriorating over the period covered by the records. The calculated averages provide the reference point for the assessment, while the direction and size of change provide the basis for interpretation. Results are reported at an aggregate level to maintain consistency with the available evidence.\n\nThe limitations of the estimates follow from the coverage and internal consistency of the source records. They should consequently be read as measured tendencies rather than as exact descriptions of every individual case. Nevertheless, the microdata provide the central evidence for the comparative assessment and for the summary indicators presented in this section.\n\nMention: \"Enumerative Housing Conditions Microdata\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The assessment calculates average conditions and a trend from the Enumerative Housing Conditions Microdata produced by NBS.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:2755b6ee2e19aa732379ba7e", "occurrence_id": "0c941f54344f1ee729ef96d2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Urban Household Consumption Index indicator for Burkina-Faso is used as the principal quantitative reference for the assessment. It provides a consistent basis for describing household conditions and for anchoring the interpretation of the tables presented below. No adjustment or transformation is applied to the cited observations in this section.\n\nTwo figures are reproduced directly from the indicator. The first records 18.4 percent, while the second records 22.7 percent. These values constitute the central evidence for the comparison and are retained in their reported form to preserve traceability to the source. Their presentation is intended to support review of the magnitude of the observed difference rather than to imply a separate estimate.\n\nThe figures are used solely to establish the empirical position described in the assessment. They are read together with the surrounding discussion and should not be interpreted as a complete measure of welfare or as evidence of causality. The source provides the underlying reference values, while the report limits its treatment to their clear reproduction and cautious interpretation. This approach reduces the risk of introducing rounding differences or assumptions not contained in the cited indicator.\n\nMention: \"Urban Household Consumption Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Urban Household Consumption Index indicator for Burkina-Faso and reproduces 18.4 percent and 22.7 percent directly from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:8facc1c3e33d6760bf897759", "occurrence_id": "f74cb3a6e032fcb2c4527533", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Implementation progress is assessed against the broader conditions affecting household resilience and the capacity of affected groups to maintain essential consumption. The evidence is interpreted cautiously because observed changes may reflect concurrent pressures that are not separately identified in the implementation records. Within this framework, uneven coping capacity is treated as a relevant contextual consideration rather than as a quantified outcome.\n\nThe Independent Shock Coping Survey is cited for background on this consideration. It is a survey produced by MOEE covering Croatia and referring to 2019. Its inclusion supports the limited claim that households may rely on different coping responses when faced with disruption. The source is not used to establish a rate, distribution, comparison, or trend in this assessment, and no value from it enters the analysis.\n\nAccordingly, the main findings rely on the evidence presented in the implementation records and on the assessment logic set out in this annex. The cited survey functions as corroborative background for one claim within the broader discussion, without determining the interpretation of the reported results. This treatment preserves a clear distinction between contextual evidence and the material used for assessment conclusions.\n\nMention: \"Independent Shock Coping Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Independent Shock Coping Survey supports one contextual claim about differing household coping responses, while no values are taken from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:61d9c8544ac73fd919e109ee", "occurrence_id": "038284a220077c5d9943fd5e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Representative Building Footprint Dataset (RBFD), a geospatial source produced by UNIDO. The RBFD is used as the principal reference for describing the observed distribution of building footprints and for anchoring the quantitative statements in this annex. No adjustment, projection, or supplementary attribution is applied to the cited figures.\n\nDirectly reported in the dataset are 68,412 building footprints. This figure is reproduced as an absolute count and is presented without recalculation. The source also reports that 27.4 percent of the recorded footprints fall within the defined analytical category. This proportion is reproduced as stated and is not converted into a rate or trend.\n\nTaken together, the two figures provide the evidentiary basis for the assessment of the built environment. The count indicates the scale of the records available for review, while the percentage indicates the relative distribution represented in the source. Their use is limited to descriptive reporting, and no inference is made beyond the coverage and classification conveyed by the RBFD.\n\nThe citation is retained in the body of the assessment to permit direct verification of both values against the Representative Building Footprint Dataset (RBFD).\n\nMention: \"Representative Building Footprint Dataset (RBFD)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The assessment takes the directly reported figures of 68,412 building footprints and 27.4 percent from the RBFD.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:18bd269f035495e3c74ebc18", "occurrence_id": "73ddc636955d23230c8f5a5a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Linked Facilities and WASH System, an administrative source for 2023. Its records provide the underlying observation base for the review of service availability and facility functionality. The source is used to support the interpretation of changes across reporting units, rather than to reproduce an isolated published figure.\n\nFor analytical purposes, the available records are screened for internal consistency and aligned according to the indicators used in this assessment. Aggregate results are then calculated from the underlying observations. In particular, the analysis derives an average level of reported service availability and examines the direction of change across the observed facilities. These calculations provide a more appropriate basis for comparison than a direct citation of any single entry in the source.\n\nThe resulting pattern indicates gradual variation in service conditions, with differences between facilities remaining material. This finding supports the broader assessment of implementation performance and helps identify where operational attention may be required. Interpretation remains subject to the completeness and consistency of the administrative records, including possible gaps in reporting and differences in facility-level updating. The calculated average and trend are therefore presented as analytical estimates derived from the source, not as values stated directly by it.\n\nMention: \"Linked Facilities and WASH System\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the Linked Facilities and WASH System administrative records for 2023 to calculate an average and assess the direction of change across facilities.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:d71455adf8cceb8ee17b6b7a", "occurrence_id": "ef87e1a9b2893b506483ac59", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Pilot Vaccination Campaign Survey, a survey covering Mali, as its principal empirical source. The material provides the basis for examining vaccination uptake across the observed population and for assessing differences between participating communities.\n\nThe analysis derives a weighted average from the recorded observations and compares that average across successive observation periods. This calculation indicates a gradual upward trend in reported uptake, although the pace of improvement is uneven across locations and population groups. The results therefore describe a pattern generated from the source rather than reproduce individual figures from it.\n\nThe derived measures are used to assess implementation performance and to identify areas where coverage remains below the overall average. Interpretation is limited to the population represented in the survey, and the calculated trend should not be treated as a direct estimate for populations outside that coverage. In this annex, the source supports comparative analysis through summary statistics, with emphasis placed on the direction and consistency of change rather than on isolated observations.\n\nMention: \"Pilot Vaccination Campaign Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives a weighted average and an upward vaccination-uptake trend for Mali from the Pilot Vaccination Campaign Survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:8dc0727c8a3fb5d214c1c1c7", "occurrence_id": "bccc7d70bd254fdc3e4ebc58", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Assessment of household support arrangements draws on the Household Targeting and Payments Index, an indicator covering Afghanistan for 2021. It is used alongside the wider review evidence to substantiate the observation that targeting and payment processes were documented with uneven consistency across the assessed records. The source is treated as a descriptive input rather than as a standalone measure of programme performance.\n\nTo support this use, records associated with the indicator were collected from the available administrative material, harmonised against common concepts for household eligibility and payment status, and archived with the related assessment files. Differences in terminology and recording practice were retained where they affected interpretation, while duplicate entries and inconsistent labels were reconciled for analytical use. No adjustment was made to imply precision beyond the information contained in the source.\n\nThis handling permits the review to distinguish gaps in recorded coverage from gaps in delivery itself. The indexed material therefore backs the broader finding that evidence on household targeting and payments is usable for implementation assessment, but requires qualification where documentation is incomplete or classifications do not align. Findings are reported at the level supported by the archived records, with the indicator serving as corroboration for the assessment rather than as a substitute for verification.\n\nMention: \"Household Targeting and Payments Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The passage uses the Afghanistan 2021 indicator and records its collection, harmonisation, and archiving for assessment.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:87d44f83f085b2fe5c189a10", "occurrence_id": "c2679936da1e63f75f139a61", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Baseline Storage and Loss Survey is the principal survey used for the storage assessment. It provides the direct evidentiary basis for the measurements presented below. No adjustment, interpolation, or supplementary source has been applied to the two figures reproduced from this survey.\n\nThe survey records 31 grades in the assessed storage classification. It also reports an 18 percent loss rate for the corresponding observation. These figures are presented as reported in the source and are not recalculated for this note. Taken together, they indicate that the classification exercise covered a substantial set of grades while recording a material level of loss.\n\nThe results support continued attention to storage conditions, handling practices, and the reliability of stock management records. Interpretation should remain limited to the scope of the survey and should not be extended to conditions outside its evidentiary base. The two reported figures are therefore retained as reference values for implementation monitoring and for comparison with subsequent measurement exercises.\n\nMention: \"Baseline Storage and Loss Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The author takes the figures 31 grades and an 18 percent loss rate directly from the Baseline Storage and Loss Survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:5c38ce66b8bbad2e9c994094", "occurrence_id": "48f49be2af65ebf65a228d11", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Baseline Business Demography Index is cited as an indicator produced by KNBS for 2016. It is retained in the evidence record as a contextual reference relevant to the discussion of business activity and the interpretation of institutional conditions. The citation does not introduce an additional measurement framework or alter the basis of the assessment.\n\nThe source is mentioned briefly to situate the analysis within the available statistical record. Its reported values are not used in the calculations, comparisons, or conclusions presented in this passage. No inference is drawn from its numerical content, and no result is attributed to movements recorded by the indicator.\n\nThe assessment instead relies on the information described in the surrounding sections and applies the stated analytical criteria consistently. The cited indicator therefore serves a limited documentary purpose. It signals the existence of a relevant reference for business demography while leaving the substantive findings independent of that source. This treatment is intended to preserve clarity concerning the evidence used for interpretation and to avoid implying that the citation supplies supporting estimates for the results reported here.\n\nMention: \"Baseline Business Demography Index\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"No values were taken from the Baseline Business Demography Index; it was used only as a contextual reference.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:12f1ce617923095797ec7dc9", "occurrence_id": "0f21e93cc120c68a822d8bbf", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment reports an estimate of ecosystem carbon conditions derived from a separate source. The estimate is used to support the wider discussion of observed changes and to inform the interpretation of results presented in this section. Its reliability was examined through comparison with an independent administrative record.\n\nThe Recurrent Ecosystem Carbon System (RECS) is cited as an administrative source for the corresponding observations. RECS provides a distinct reference against which the estimate can be checked, without replacing the underlying measure or extending the scope of the assessment. The comparison indicates that the direction and broad magnitude of the reported estimate are consistent with the recorded pattern.\n\nThis cross-check supports the specific claim concerning the relationship between ecosystem carbon conditions and the reported estimate. It does not establish a separate causal finding, and it is not used to generate an additional trend. The evidence is therefore treated as corroboration within the broader paragraph rather than as the principal basis for the assessment.\n\nThe use of RECS strengthens the internal review of the estimate while preserving a clear distinction between the source used to derive the measure and the administrative source used for validation. The result is considered sufficient for reporting purposes, subject to the limitations already associated with the primary estimate.\n\nMention: \"Recurrent Ecosystem Carbon System (RECS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The author uses RECS as an administrative cross-check for an estimate derived from a separate source.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:2e5e061cf0a55189d7c1c985", "occurrence_id": "3a248b267bfa3e0b058b3e54", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Monthly Social Protection Survey is a survey covering Republic of Congo and is used as a supplementary reference for the assessment of social protection coverage. Its tabulations provide an independent indication of reported receipt, enrolment status, and access to selected benefits. The source is considered suitable for directional comparison because its reporting structure corresponds to the population groups examined in the main analysis.\n\nThe estimate of beneficiary reach presented in the broader results section is cross-checked against the Monthly Social Protection Survey rather than treated as a standalone measurement. The comparison supports the claim that coverage remains concentrated among households with established links to assistance mechanisms, while also indicating that the level of reported access may vary according to the reporting basis used.\n\nThis use of the survey is limited to verification of that claim. It does not replace the primary estimate or establish a separate series for monitoring change. Differences in definitions, recall, and population coverage may account for residual divergence between the two sources. Accordingly, the comparison is interpreted as corroborative evidence, with the principal estimate retained for the assessment and the survey result used to test its plausibility. The finding should therefore be read as evidence of consistency in direction, rather than as an exact reconciliation of totals.\n\nMention: \"Monthly Social Protection Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the Monthly Social Protection Survey for Republic of Congo to cross-check the beneficiary-reach estimate from a separate source.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:3faf549e42b24d31cb17adad", "occurrence_id": "be047cd87085a8bc6f836fc7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The District Labour Force Census is a census referring to 2018 and provides the principal statistical basis for the labour-force assessment. It is used to establish a consistent profile of participation, employment status, and inactivity within the reported population. The source is retained as the central evidence for the calculations presented in this annex.\n\nThe analysis does not reproduce isolated observations from the census. It calculates average values for the principal labour-force measures and compares the resulting distribution across the reported population groups. This treatment supports identification of the prevailing direction of change between groups and reduces the influence of individual observations on the assessment. The derived averages are used in preference to selective citation of particular entries.\n\nThe results indicate a differentiated labour-market profile, with participation and employment distributed unevenly across the groups recorded in the source. The calculated measures are interpreted as analytical summaries of the census evidence rather than as independently observed estimates. Given the single reference period, the assessment uses the derived averages to describe relative patterns and internal variation, while avoiding claims that would require observations from additional periods. All calculations reported here are therefore traceable to the District Labour Force Census and its 2018 coverage.\n\nMention: \"District Labour Force Census\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives average labour-force measures and comparative patterns from the 2018 District Labour Force Census.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:8745d0a012a5c325b6218c77", "occurrence_id": "04e9e62631512a737278558b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Municipal Labour Force Survey (MLFS) is the survey used for labour-market measures covering the Republic of Congo. It provides the underlying observations for selected indicators of participation, employment and unemployment, with attention to consistency across population groups and geographic units. The source is treated as a statistical input to the assessment rather than as a stand-alone account of labour-market conditions.\n\nFor the purpose of this analysis, the observations are consolidated into comparable groupings and used to derive an average level and an indicative trend. The resulting measures therefore reflect calculations performed on the survey material and should not be interpreted as figures copied directly from the source. This approach supports comparison within the assessment while limiting the influence of isolated observations.\n\nThe calculated results are used to support the broader finding that labour-market outcomes remain uneven across population groups. They are presented to show the direction and relative scale of change, rather than as a complete account of household welfare or economic activity. Differences in coverage, response patterns and measurement conditions are taken into account when interpreting the results.\n\nThe estimates should consequently be read alongside the other evidence considered in the assessment. Their principal contribution is to provide a consistent basis for summarising variation and identifying the broad movement of the indicators. No individual observation is treated as determinative of the overall conclusion.\n\nMention: \"Municipal Labour Force Survey (MLFS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the MLFS to calculate an average and identify a labour-market trend for the Republic of Congo.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:a3c15637e100bbc606936025", "occurrence_id": "66f4b1ee96ed1e986dacf337", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Data sources.** The analysis uses the Audited Elevation Model Index (AEMI) indicator produced by PovcalNet as the principal reference for the measurement presented below. The source is treated as a recorded statistical input, and its reported values are reproduced without recalculation or adjustment.\n\nTwo figures are quoted directly from this source. The AEMI records 31 grades for the distribution under review, while the corresponding coverage measure is reported as 14.6 percent. These values provide the central evidence for the assessment and are retained in their original form to support consistency between the narrative findings and the underlying statistical record.\n\nThe first figure establishes the number of grades represented in the index and permits the results to be read against the full classification used in the assessment. The second indicates the share associated with the reported coverage measure. Taken together, the figures show the scale of the classification and the extent to which the measured condition is represented in the source material.\n\nNo additional transformation is applied to either value. Subsequent discussion should therefore distinguish clearly between the figures reproduced from the AEMI indicator and any interpretation derived from them.\n\nMention: \"Audited Elevation Model Index (AEMI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported values of 31 grades and 14.6 percent from the Audited Elevation Model Index (AEMI) indicator produced by PovcalNet.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:a69946fbaffbeb6ea159dbb2", "occurrence_id": "0f8ab2e52f48fa86e5383080", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment records the principal data inputs used to frame the implementation review. Administrative materials are considered alongside field documentation and routine monitoring records, with attention given to the limits of comparability across reporting instruments. The analysis distinguishes descriptive background from evidence used to support findings and recommendations.\n\nThe Local Road Segment Collection is produced by Romania MoF and is classified as other. It is noted as a reference to the broader information environment relevant to local infrastructure management. The collection is not used to populate tables, calculate indicators, establish changes over time, or substantiate performance statements in this report.\n\nIts mention serves only to indicate that a related administrative record exists and may be relevant for subsequent verification or system development. No observation, count, rate, classification, or other value is drawn from the collection in the assessment presented here. Findings rely on the documented materials identified for the review and on the procedures applied to those materials.\n\nThis treatment is consistent with the report’s approach to source transparency. References are retained where they clarify the provenance or potential scope of available information, while analytical claims are restricted to evidence that has been reviewed for the stated purpose. The cited collection therefore remains a contextual reference rather than an input to the results.\n\nMention: \"Local Road Segment Collection\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The Local Road Segment Collection is used only as background, with no values or derived measures taken from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:ca15667aa8765e754173a659", "occurrence_id": "4ecce2edda967590a7f6edbb", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis draws on the Quarterly Facility Assessment Survey (QFAS), a survey produced by the National Statistics Office for facilities in Tigray Region. The source provides structured observations on selected aspects of facility capacity and service availability, allowing comparison across reporting units within the covered area.\n\nFor this assessment, the evidence is not presented as a quotation of individual survey results. Instead, the observations are combined to calculate an average level of availability and to examine its movement across successive observations. This treatment reduces the influence of isolated facility values and provides a consistent basis for interpreting the broader pattern.\n\nThe resulting trend indicates variation in service readiness across facilities, with the average used as an analytical measure rather than as a direct statement from any single record. The finding supports the assessment that implementation conditions were not uniform throughout Tigray Region and that continued monitoring is warranted. QFAS therefore serves as an underlying statistical input to the assessment, while the reported conclusion is derived from the analysis of its observations.\n\nMention: \"Quarterly Facility Assessment Survey (QFAS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses QFAS to calculate an average and identify a trend in facility service availability in Tigray Region.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:06697750906a720ec1dcc46b", "occurrence_id": "c3b9695409b90e84acac0357", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Comparative Forest Products Registry database as a supporting source for the review of reported forest product movements. Its entries provide a consistent basis for checking selected quantities and for placing the operational findings within a broader statistical frame.\n\nThe registry records 17.4 percent for the first measure and 6.2 percent for the second measure. These two figures are quoted directly from the database and are reproduced without recalculation or adjustment. They support the finding that reported activity was uneven across the period examined, while the wider assessment relies on additional evidence for interpretation.\n\nThe values are used as reference points rather than as a substitute for the full body of monitoring information. Their inclusion permits a clear distinction between figures taken from the Comparative Forest Products Registry and conclusions developed through the assessment. No inference is drawn from the two observations beyond the limited comparison stated above, and the source is retained in the evidence record for verification.\n\nMention: \"Comparative Forest Products Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly recorded values of 17.4 percent and 6.2 percent from the Comparative Forest Products Registry database to support a broader finding about uneven reported activity.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:86787c09a56b6fc3a362e666", "occurrence_id": "4bf15b89766590cbb1c00a7b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Monitoring of implementation relies on the Record-Based Graduation Programmes Registry, a database maintained for recording programme participation and progression. The registry is treated as the documentary reference for the numerical statements in this assessment. No adjustment, interpolation, or external reconciliation has been applied to the figures reproduced below, which are carried over directly from the source in the form in which they are recorded.\n\nThe database reports 31 programme records. It also reports 14 graduation records. These two figures constitute the central quantitative evidence for the assessment of recorded programme activity and completion. They are presented separately because the underlying entries refer to different stages of implementation and should not be combined into a single measure.\n\nThe counts support a limited description of the documented caseload. The first figure indicates the number of programme records available in the registry, while the second identifies the subset of records associated with graduation. The distinction is retained to avoid implying a rate, a sequence, or a causal relationship that is not established by the source. Subsequent analysis may use these counts for tabulation, subject to confirmation that the records are complete and internally consistent. For present purposes, the registry provides the authoritative basis for the two reported values and for the accompanying classification of programme status.\n\nMention: \"Record-Based Graduation Programmes Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes 31 programme records and 14 graduation records directly from the Record-Based Graduation Programmes Registry.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:54af673408149ae8ab2202e9", "occurrence_id": "737b5912bd225ff60a04fd69", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Participation patterns are considered within the institutional setting in which the reported indicators are interpreted. The Inter-Agency Cooperative Membership Registry (ICMR) is cited as a database covering Egypt for 2021, and its inclusion supports the general statement that cooperative membership records form part of the available evidence base.\n\nThe registry is not used to derive totals, rates, rankings, or changes in the indicators presented below. No entries, aggregates, or classifications from the ICMR are incorporated into the calculations, and no inference is made from its membership records. Its role is limited to establishing that a relevant record system exists for the geographic and temporal scope under review.\n\nAccordingly, the findings should be read as an assessment of the indicators defined in the project framework, with the registry providing background rather than measurement input. This distinction is material for interpretation: the cited database reinforces the description of the information environment, while the reported results rest on the evidence specified for each indicator. The reference is therefore retained to support a single contextual claim within the wider discussion, not to substantiate a numerical result.\n\nMention: \"Inter-Agency Cooperative Membership Registry (ICMR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The ICMR database for Egypt in 2021 is used only to support a contextual claim about the evidence base, with no values taken from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:34a5e91509d6072b790a5b7b", "occurrence_id": "f4231c1e36660df21ccb3f7b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Standardised Non-State Education Survey (SNES) is referenced as a 2022 survey in the description of the information base. The reference establishes the presence of a relevant sectoral source and provides a point of orientation for interpreting the scope of the assessment. No numerical observation, classification, or indicator is reproduced from this source.\n\nThe analysis proceeds from the evidence presented in the accompanying tables and narrative. The SNES reference does not support a comparison, an estimate, or an inference in the present assessment. Its role is limited to acknowledging a related source that may assist readers in understanding the broader information environment surrounding non-state education.\n\nAccordingly, the citation should not be read as evidence for any reported level, proportion, change, or relationship. No values from the survey are incorporated into the calculations, and no result is attributed to it. The reference is retained for documentary completeness and to distinguish general source awareness from the data directly used in the assessment.\n\nMention: \"Standardised Non-State Education Survey (SNES)\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"No values or analytical results are taken from the Standardised Non-State Education Survey (SNES); it is cited only to provide documentary background.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:176e8fe580ea6f5c721b03cc", "occurrence_id": "d35df5b30f41ae1c30a9986f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Data sources.** The assessment draws on the Inter-Agency Land Cover Index (ILCI), an indicator covering Azerbaijan for 2023. The source is used to substantiate the land-cover profile presented in this section and to provide a consistent basis for interpreting the reported distribution across the territory.\n\nThe analysis reproduces two figures directly from the indicator. It records 18.4 percent for the first land-cover category and 27.9 percent for the second category. These values are retained as reported and are not recalculated, rounded, or combined with estimates from other materials.\n\nTaken together, the figures support the finding that land-cover conditions are unevenly distributed and that the larger category accounts for a materially greater share of the reported area. The difference between the two observations is therefore treated as descriptive evidence within the broader assessment, rather than as evidence of change over time.\n\nInterpretation remains limited to the coverage and reference period stated by the source. The figures are presented for comparison within the assessment and should not be read as a measure of annual change, a forecast, or a causal attribution. This treatment preserves the source’s role as a supporting statistical reference while avoiding assumptions beyond the reported indicator.\n\nMention: \"Inter-Agency Land Cover Index (ILCI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the ILCI indicator to report the directly sourced values of 18.4 percent and 27.9 percent for Azerbaijan in 2023.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:e5de5b3aa4df617691ec09c1", "occurrence_id": "e633e815681442d595cca3dc", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Standardised Social Protection Survey, a survey produced by NSS for Tigray Region and referring to 2022. The source provides a household-level reference for assessing reported access to social protection measures and for checking the internal consistency of the associated estimate. Its geographic and temporal coverage is considered sufficiently aligned with the population and period under review.\n\nThe estimate presented in the main analysis is cross-checked against the corresponding evidence in this survey. The comparison is used to support the claim that reported access remained uneven across the covered population. Differences between the two measures are noted as a limitation of interpretation, since the underlying definitions and observation procedures are not identical.\n\nThis use of the source is confined to verification of that claim within the broader assessment. It does not replace the primary estimate, determine the full set of reported results, or provide a basis for extrapolation beyond Tigray Region. The survey evidence is therefore treated as corroborating information, with the comparison recorded as a reasonableness check rather than as a recalculation of the principal indicator.\n\nMention: \"Standardised Social Protection Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the 2022 Tigray Region evidence from the Standardised Social Protection Survey produced by NSS to cross-check one estimate.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:16c33ba50025bdbb2c66beb3", "occurrence_id": "4a754a63e80236632720f4fa", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Quarterly Immunisation Coverage Survey (QICS) is the principal survey used for the coverage analysis. Its repeated observations provide the central empirical basis for assessing changes in reported immunisation uptake. The source is used as a consistent measurement series rather than as a single reference point.\n\nCoverage observations are arranged in their reporting sequence and examined for movement over time. The analysis calculates a trend from the series, using the direction and magnitude of successive observations to identify whether coverage is improving, stable, or declining. This treatment reduces reliance on any isolated result and provides a more suitable basis for implementation assessment.\n\nAn average coverage estimate is also calculated from the available observations. The average is used to summarise the general level recorded by the survey, while the trend indicates the pattern around that level. Together, these calculations support comparison of performance across reporting periods without presenting any individual observation as definitive. Results are interpreted with attention to variation within the series and to the possibility that short-term changes may not represent a sustained shift.\n\nThe resulting assessment therefore draws its principal evidence from the Quarterly Immunisation Coverage Survey (QICS), while deriving the reported findings through aggregation and analysis. The survey supports a measured account of coverage performance and provides the basis for monitoring progress in subsequent implementation reviews.\n\nMention: \"Quarterly Immunisation Coverage Survey (QICS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives a coverage trend and average estimate from the Quarterly Immunisation Coverage Survey (QICS).\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:af84c68890b22f4622ac4859", "occurrence_id": "22b88e35f09f6bc0c147c367", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis uses the Combined Facility Assessment Index (CFAI) as its principal indicator of facility conditions. The source provides a consistent basis for comparing recorded assessments across the observations used in the analysis, while limiting reliance on isolated facility-level results.\n\nValues from the CFAI are processed to derive an average for the analytical sample and to identify the direction of change across successive observations. The resulting measures are treated as analytical estimates rather than as direct reproductions of the source entries. This approach reduces the influence of individual high or low assessments and provides a more stable basis for interpretation.\n\nThe calculated average is used to describe the general level of assessed conditions, while the derived trend is used to assess whether performance is improving, stable, or declining. Differences in the number of observations are reviewed before aggregation, and the interpretation gives priority to patterns that remain evident after averaging. The CFAI therefore functions as the central evidence for the summary assessment, with reported findings based on calculations made from the indicator rather than on selected quoted values.\n\nMention: \"Combined Facility Assessment Index (CFAI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the CFAI to calculate an average and assess a trend from the indicator.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:a59c341033460f7888281ab8", "occurrence_id": "c436b3898b8881138633a531", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Rapid Informal Sector Census provides the principal empirical basis for this assessment of informal activity in Rio de Janeiro in 2022. It is a census produced by GRI, and its coverage is used to define the observations retained for analysis. The source is treated as the common reference for the statistical measures presented in this section.\n\nThe analysis does not reproduce isolated values from the census. Instead, it calculates an average across the relevant observations and derives a trend from their ordered distribution. These calculations provide a concise representation of the level and direction of variation recorded in the source, while preserving the unit of analysis used in the underlying enumeration. The resulting indicators are therefore analytical outputs based on the census rather than figures transcribed from it.\n\nThe derived average is used to establish the central tendency of the observed informal activity, and the calculated trend is used to assess differences across the observations covered by the source. This approach reduces the influence of individual records and supports comparison within the defined coverage. Interpretation remains limited to the population represented by the Rapid Informal Sector Census, with no inference beyond its stated scope. The measures are reported as evidence for implementation assessment and as a basis for monitoring changes in the composition of informal activity.\n\nMention: \"Rapid Informal Sector Census\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the Rapid Informal Sector Census produced by GRI for Rio de Janeiro in 2022 to calculate an average and derive a trend from its observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:5a4afaa924517cb7fdf90b60", "occurrence_id": "d830e29b34864ad55ab0602b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis draws on the Periodic School Feeding Census for Syria in 2022, treated as a census of school feeding provision and participation. The source provides the observation base for assessing delivery levels, beneficiary reach, meal frequency, and reported interruptions across participating schools. Its records are used as the central evidence for the estimates presented below.\n\nRather than reproducing individual entries, the analysis aggregates the reported observations and calculates mean coverage, average meal provision, and the distribution of participating schools. Directional change is derived from the sequence of reporting periods within the source, allowing the assessment to distinguish relatively stable provision from areas showing contraction or expansion. These calculations provide a summary measure of implementation intensity while retaining the variation recorded among schools.\n\nThe resulting indicators support comparison of access and continuity within the covered population. They should be interpreted as analytical measures derived from the census records, rather than as figures transcribed directly from a single reported total. Differences in participation and service regularity are therefore reflected through calculated averages and movement over the observed periods. This approach also reduces the influence of isolated entries and permits a consistent reading of the reported pattern across the dataset.\n\nMention: \"Periodic School Feeding Census\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis calculates average coverage, meal provision, participation, and directional change from the census records.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:82d64e258096ee227caaa90f", "occurrence_id": "2e9fb6acb28d7948afda8c1d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment reviews changes in observed land characteristics within Nairobi County. The purpose is to establish whether the recorded pattern is sufficiently consistent to support the interpretation presented in the results section. Findings are treated as indicative rather than as a substitute for ground verification.\n\nThe Longitudinal Remote Sensing Registry (LRSR) database provides the underlying observations for 2022. The LRSR records are used to calculate an annual average and to derive a simple trend across the available observations, rather than to reproduce individual registry figures. This calculation supports the assessment of directional change reported in the preceding discussion.\n\nThe resulting measure is used as one component of the broader evidence base. It indicates the central tendency and movement of the recorded series, while its interpretation remains subject to coverage, classification, and temporal consistency. No separate conclusion is drawn from a single observation. The computed average and trend are therefore presented as supporting evidence for the reported claim, with emphasis placed on the aggregate pattern rather than on isolated values.\n\nMention: \"Longitudinal Remote Sensing Registry (LRSR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the 2022 LRSR database for Nairobi County to compute an average and trend supporting a broader assessment claim.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:1e7ac061df8689792a2ce422", "occurrence_id": "ee0cc609299ea28ed7e8c47f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Periodic Enterprise and Jobs Survey (PEJS) is the principal source for the enterprise indicators used in this assessment. The 2020 survey supplies observations on employment, turnover, and operating conditions across reporting units. Its use is limited to the evidence required for constructing comparable summary measures, with no inference made beyond the variables recorded in the source.\n\nThe analysis aggregates the PEJS observations to derive an average level for each indicator and calculates a trend across the relevant reporting groups. These results are computed from the underlying distribution rather than reproduced as individual source figures. The resulting measures provide a consistent basis for comparing employment intensity, turnover performance, and the incidence of operating constraints within the analytical sample.\n\nThe computed average indicates the central position of the observations, while the derived trend identifies the direction and relative strength of variation between groups. Interpretation therefore focuses on movement in the aggregate measures and does not treat any single response as representative of the full sample. This approach also reduces the influence of isolated values and supports a more stable reading of changes in enterprise conditions.\n\nThe PEJS evidence is used primarily for descriptive assessment. The estimates should be read as analytical summaries generated from the 2020 survey and not as direct quotations from the source. Differences between indicators may reflect both substantive variation and differences in response patterns, and the findings are accordingly presented as indicative measures for implementation monitoring.\n\nMention: \"Periodic Enterprise and Jobs Survey (PEJS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses the 2020 PEJS survey to compute averages and trends rather than reproduce individual observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:236c0fc5514690f0c0a5e65d", "occurrence_id": "9919cb0f5f4ac9cfbb9cd8c4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Panel Emission Inventory Collection, a source classified as other and referring to 2021. The material is used as an input to the analytical review, with its observations retained for consistency checks and aggregation. No individual record is treated as a stand-alone estimate for reporting purposes.\n\nThe review calculates an average level across the available observations and estimates the direction of change over the covered sequence. This treatment reduces the influence of isolated fluctuations and provides a stable basis for the assessment of overall movement. The resulting measure supports the interpretation of emissions performance presented in the surrounding analysis.\n\nThe derived statistics are used to support one finding concerning the persistence and scale of the observed pattern. They are not presented as a direct reproduction of the source entries. Instead, the calculated average and trend are compared with the other evidence considered in the assessment, allowing the report to distinguish broad movement from short-term variation. The procedure is applied consistently so that the conclusion remains traceable to the cited collection while avoiding undue emphasis on any single observation.\n\nMention: \"Panel Emission Inventory Collection\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The assessment calculates an average level and an overall trend from the 2021 observations in the Panel Emission Inventory Collection.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:53ed19a8cce33961f9e370f2", "occurrence_id": "f453210bd7812bbc8448e98d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis draws on the Biennial Built-Up Surface Index, an indicator produced by S&P for 2015. The source is used as the principal evidence for assessing change in the built-up surface measure over the assessment period. Its relevance lies in the consistent structure of the observations and their suitability for comparison across the reporting framework.\n\nThe values are processed to derive an average level and a directional trend rather than reproduced as isolated figures. The calculation uses the available observations to identify the prevailing movement and to reduce the influence of individual fluctuations. This treatment supports a summary measure that is more appropriate for implementation monitoring than a simple transcription of the source values.\n\nThe resulting estimates are presented as analytical outputs based on the indicator. They are intended to support interpretation of the observed pattern, including the extent to which change appears persistent across successive observations. No additional precision is implied beyond that supported by the underlying series, and the computed average and trend should therefore be read as summary statistics derived from the cited source.\n\nMention: \"Biennial Built-Up Surface Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses S&P’s 2015 Biennial Built-Up Surface Index indicator to compute an average and a directional trend.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:f182f2401cf968e21b2b507b", "occurrence_id": "d451dd3258b943032e4cc3a4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Rural Service Delivery and Readiness Survey (RSDRS) is the principal evidence base for this assessment. The 2020 survey provides the observations used to examine service readiness and the regularity of reported conditions across the units covered by the dataset. Its results are treated as a source for analysis rather than as a set of isolated figures.\n\nThe assessment derives summary measures from the RSDRS records, including average levels across observations and a trend in the reported indicators. This approach reduces the influence of individual entries and permits comparison of the general direction and distribution of service readiness. The calculations rely on the underlying observations in the survey and do not reproduce selected values as standalone findings.\n\nThe resulting averages indicate the broad level of readiness represented in the dataset, while the calculated trend identifies whether reported conditions improve, deteriorate, or remain broadly stable across the ordered observations. These results are used to support the assessment narrative and to distinguish systematic patterns from exceptional cases. Interpretation remains limited to the evidence contained in the 2020 RSDRS survey, with no inference made beyond the scope of its recorded observations.\n\nMention: \"Rural Service Delivery and Readiness Survey (RSDRS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the 2020 RSDRS survey to calculate averages and a trend from its observations rather than reproduce individual figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:6b8bd307f88497665aa0827f", "occurrence_id": "3fc611a9c8ef7ae4330caff0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis relies on the Comparative Deprivation and Hardship Census (CDHC), a census produced by CDR for 2022. The source provides the principal empirical basis for assessing changes in reported hardship and deprivation. Its use is limited to the information required for the aggregate assessment, with no supplementary source introduced for the same purpose.\n\nThe observed measures are consolidated into comparable groupings and examined across the relevant reporting sequence. Rather than reproducing individual entries, the analysis calculates average levels and derives a trend from the CDHC observations. This approach reduces the influence of isolated results and supports a consistent interpretation of overall movement. The resulting estimates are treated as analytical outputs derived from the census, not as figures transcribed directly from it.\n\nThe calculated averages indicate a persistent pattern in the distribution of hardship, although the magnitude of change differs across the measures considered. The trend is therefore used to inform the assessment of implementation performance and to identify areas requiring continued monitoring. Interpretation remains conditional on the coverage and internal consistency of the census material, while the numerical summaries provide the central evidence for the conclusions presented in this section.\n\nMention: \"Comparative Deprivation and Hardship Census (CDHC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses the 2022 CDHC produced by CDR to calculate averages and derive a trend rather than reproduce individual census figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:6f3c448a8b688e176a378b8d", "occurrence_id": "80a0f466f4a1218336b218fc", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Midline Maternal and Newborn Census is used as the principal census for assessing continuity in maternal and newborn service delivery. Its observations provide the central evidence for the analysis of reporting-period performance, with attention to the consistency of recorded contacts, referrals, and completed care. The source is considered sufficiently structured to support comparison across successive observation periods.\n\nRather than reproducing individual entries, the analysis calculates average coverage for the monitored indicators and estimates the direction and pace of change over time. The resulting averages reduce the influence of isolated fluctuations and provide a more stable basis for judging implementation progress. The calculated trend indicates whether observed service use is improving, remaining broadly constant, or declining across the available observations.\n\nInterpretation remains limited to the measures represented in the census and does not extend to outcomes that are not recorded there. Differences in completeness or reporting practice may affect the calculated averages, and the trend is therefore used as an analytical signal rather than as a definitive measure of population change. On this basis, the census supports routine performance assessment and helps identify areas requiring closer operational review.\n\nMention: \"Midline Maternal and Newborn Census\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the Midline Maternal and Newborn Census to calculate average coverage and estimate the direction of change across reporting periods.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:462784348c1c759dad2380f4", "occurrence_id": "2a746c97430a5c3f9bcdf2f3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Record-Based School Feeding Registry (RSFR) database is the central empirical source for the assessment of school feeding operations in Republic of Congo. Its records provide a consistent basis for examining reported participation, service reach and the distribution of supported schools across the observed reporting periods. The source is used as a structured administrative record rather than as a narrative account.\n\nThe analysis calculates an average level of beneficiary coverage from the RSFR entries and derives a trend from successive observations. These calculations are intended to show the direction and scale of change in reported access, while reducing the influence of isolated fluctuations in individual records. No single registry entry is treated as representative of overall performance.\n\nResults indicate the extent to which reported coverage is sustained across the observation period and identify movements that warrant further review. The derived measures also support comparison among school-level records using a common basis. Interpretation remains limited to information recorded in the database, including its stated coverage and reporting consistency. The RSFR therefore serves as the principal evidence for the quantitative assessment, with findings presented as calculated summaries rather than direct reproductions of registry figures.\n\nMention: \"Record-Based School Feeding Registry (RSFR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives an average of beneficiary coverage and a trend in reported school feeding access from the RSFR database.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:87b0fef4c22aafadde94731a", "occurrence_id": "930c57ce65465439964a713b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Standardised Unemployment Insurance Survey, produced by CEDLAS as a survey referring to 2018, provides the principal evidentiary basis for the assessment. The source is treated as a structured set of observations on unemployment insurance conditions, with attention to consistency across the reported measures. Its use supports a consolidated reading of changes in coverage and recorded outcomes.\n\nThe analysis does not reproduce selected observations as standalone quotations. Instead, it computes an average from the reported values and derives a trend across the observations, thereby reducing the influence of isolated movements and allowing the direction of change to be assessed. The resulting measures are used to support the interpretation of changes in programme performance and to identify whether the observed movement is sustained.\n\nThis treatment also permits comparison between the aggregate level and the underlying distribution of observations. Particular care is given to preserving the source definitions when calculating the average and trend, since changes in coverage, reporting practice, or measurement scope may affect the apparent direction of movement. The results are therefore presented as analytical estimates derived from the survey, rather than as direct quotations from it.\n\nThe derived indicators provide a concise basis for monitoring and for assessing whether subsequent implementation is associated with improvement, stagnation, or deterioration. They should be read together with the stated limitations of the source and with the methodological assumptions applied in aggregation.\n\nMention: \"Standardised Unemployment Insurance Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the Standardised Unemployment Insurance Survey to calculate an average and derive a trend from its observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:b62b083711bede1c984ea0a6", "occurrence_id": "979c9d92ba776393fce7c091", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Enumerative Business Demography Registry, a database produced by Barth, Caprio, and Levine, is used as the principal statistical source for this assessment. The evidence is confined to two figures reproduced directly from the database, namely 18.4 percent of registered enterprises reporting an active status and 27,631 enterprises recorded in the registry. These values are presented without recalculation or adjustment.\n\nThe first figure indicates the proportion of entries meeting the registry’s active-status criterion. The second provides the corresponding count of enterprises represented in the source. Taken together, the figures establish the scale of the observed business population and the share classified as active under the source’s reporting framework.\n\nInterpretation is limited to the information contained in these two observations. No additional inference is drawn regarding changes over time, geographic distribution, institutional performance, or causal relationships. The registry is therefore treated as a reference point for descriptive measurement rather than as a basis for projection or evaluation.\n\nFor purposes of the assessment, the cited figures should be read as source-reported values. Their use supports transparent documentation of the statistical basis for the analysis and permits subsequent review against the same database.\n\nMention: \"Enumerative Business Demography Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Enumerative Business Demography Registry for the directly reproduced figures of 18.4 percent and 27,631 enterprises.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:f0b22b61e91ba09ddef90e3d", "occurrence_id": "b0113cdd10fc259e58cd5de3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Coverage of rural credit activity is assessed using the Representative Rural Credit Registry (RRCR) database. The source covers Accra and refers to 2023, providing the principal empirical basis for the present assessment. It is treated as a reference record for observed borrower activity rather than as an estimate generated by the assessment. The citation is retained in full to preserve the provenance of the reported measures.\n\nTwo reported measures are reproduced directly from this source. The database records a 31 percent repayment rate among registered borrowers and an 18 percent share of accounts classified as overdue. These figures are presented without recalculation or adjustment, and they constitute the central evidence for the discussion of credit performance. The distinction between the measures is maintained because the first concerns repayment outcomes, while the second concerns account status.\n\nTaken together, the reported values indicate a material difference between completed repayment and overdue exposure within the covered record. The assessment therefore uses the 31 percent rate and the 18 percent share as descriptive indicators, not as projections or causal estimates. Interpretation remains limited to the population represented in the cited database, and no inference is made beyond its stated coverage. This treatment supports transparent review and reproducibility of the numerical claims.\n\nMention: \"Representative Rural Credit Registry (RRCR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the 31 percent repayment rate and 18 percent overdue-account share directly from the RRCR database.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:8d892c9e1a57399dd65fdcee", "occurrence_id": "0eb4180fecddd139d9b867e0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The National Business Demography Registry is the database used for the assessment of business activity in Niger. It provides a consolidated record of registered enterprises and their reported status, allowing the analysis to rely on a single documented reference. The source is retained as the basis for the values presented in this section.\n\nTwo values are taken directly from the database. The registry records 14,286 active enterprises and 3,917 newly registered enterprises. These figures are reproduced as reported, without adjustment, estimation, or recalculation. They constitute the principal quantitative evidence for the assessment and establish the scale of recorded enterprise activity.\n\nThe first value indicates the stock of enterprises classified as active, while the second identifies registrations recorded as new. The distinction is maintained because the two measures describe different elements of business demography and should not be combined into a single total. Interpretation is therefore limited to the categories and counts contained in the National Business Demography Registry.\n\nThe evidence supports a clear description of the recorded enterprise population in Niger. No additional series is introduced, and the reported counts remain traceable to the cited database.\n\nMention: \"National Business Demography Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the National Business Demography Registry to report 14,286 active enterprises and 3,917 newly registered enterprises in Niger.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:5d60576240f06d93630cc6d4", "occurrence_id": "ebf09e2a4f6b24612e09b8a4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Clustered Occupational Safety System provides the administrative basis for the 2021 assessment of occupational safety reporting. The records are used as the principal empirical input for reviewing reported incidents, preventive measures, and compliance observations. Source identification and interpretation are retained with the dataset to support traceability during verification and subsequent analysis.\n\nThe assessment does not reproduce isolated entries as headline results. Instead, recorded observations are grouped into comparable analytical categories, and an average is calculated for each category to reduce the influence of irregular reporting. The resulting measures provide a consistent basis for examining the distribution of reported conditions and for identifying areas with relatively higher or lower observed exposure.\n\nA trend is then calculated from the ordered observations contained in the source, using the derived measures rather than individual quotations. This approach supports assessment of directional change within the available records while avoiding conclusions based on a single reported value. The findings are therefore presented as analytical summaries derived from the Clustered Occupational Safety System, with limitations arising from coverage, consistency, and the level of detail available in the underlying administrative records.\n\nMention: \"Clustered Occupational Safety System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The assessment derives category averages and a trend from the 2021 administrative records in the Clustered Occupational Safety System.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:251b331b7ced0c6df5d7b00d", "occurrence_id": "6f1b241722c81fe5246557e0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Data sources.** The assessment draws on the Cross-Sectoral Education Finance System (CEFS), an administrative source for 2023. The source provides the reference series used for the expenditure review, with values retained in their reported form. It is treated as a direct administrative record rather than as an estimate generated for this assessment.\n\nThe assessment quotes two figures directly from this source. It reports that 68 percent of the recorded allocation was directed to instructional activity and that the corresponding recorded amount was 14.2 million. These figures are presented as reported observations and are not recalculated, annualised, or converted into a comparative indicator.\n\nTaken together, the figures provide the numerical basis for the observation that resource concentration was substantial within the recorded allocation. The interpretation is limited to the relationship visible in the source. The percentage describes the allocation share, while the monetary figure states the amount associated with that reporting frame. No inference is made beyond these two observations.\n\nMention: \"Cross-Sectoral Education Finance System (CEFS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the 2023 CEFS administrative source and directly reports 68 percent and 14.2 million without recalculation.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:5d3cffa9384bc468cc8023d2", "occurrence_id": "fd118c7ef091a7921986763a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Combined Service Delivery and Readiness System (CSDRS) is an administrative source covering Chengdu for 2015. It provides a consolidated basis for reviewing the availability and operational readiness of local services. The material is used here as a direct evidentiary reference, with no recalculation or adjustment applied to the reported values.\n\nTwo figures are reproduced directly from the CSDRS record. The reported readiness level is 68 percent, while the reported availability level is 42 percent. These values are presented as stated in the source and are treated as descriptive observations for the covered area and period.\n\nThe figures support a cautious assessment of service conditions. The difference between the two reported values indicates that availability and readiness should not be treated as interchangeable measures. Further interpretation is therefore limited to the evidence contained in the CSDRS record, and no additional estimates are introduced in this note. The source provides a consistent reference for documenting the observed position in Chengdu during 2015.\n\nMention: \"Combined Service Delivery and Readiness System (CSDRS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The author takes the directly reported figures of 68 percent readiness and 42 percent availability from the CSDRS source for Chengdu in 2015.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:64b79251e26366db906932c5", "occurrence_id": "37af3ed9a359f1e983d19de5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Decennial Housing Conditions Survey (DHCS), a survey providing the principal statistical basis for the discussion of dwelling conditions. Its measures are treated as reported observations and are not adjusted through additional estimation. The source is cited here to anchor the assessment’s interpretation of basic housing adequacy.\n\nThe analysis quotes two figures directly from the survey. It reports that 68.4 percent of surveyed dwellings met the specified adequacy standard, while 41.2 percent had access to the full set of essential services considered in the assessment. Taken together, these values support the broader finding that structural adequacy and service access do not progress at the same rate.\n\nThe figures are presented as descriptive evidence rather than as evidence of change over time. They provide a common reference for reviewing differences in reported conditions and for identifying areas where service deficits remain material. No adjustment is applied to the quoted percentages, and the assessment does not infer causes from the survey results alone.\n\nInterpretation should therefore remain limited to the conditions captured by the survey and to the definitions used in its measurement framework. Within those limits, the DHCS provides a consistent basis for the reported claim and for subsequent consideration of implementation priorities.\n\nMention: \"Decennial Housing Conditions Survey (DHCS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the DHCS figures of 68.4 percent and 41.2 percent to support a broader finding on housing adequacy and service access.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:9b530df74ebb5e311dc6ecf5", "occurrence_id": "c4cc47634b9f5689dce4de25", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Baseline Hiring and Vacancies Survey is cited as a 2017 survey and serves as the principal empirical source for the assessment of recruitment conditions. Its observations are treated as reported evidence for the reference period, rather than as estimates generated through subsequent modelling. No adjustment is applied to the values reproduced below.\n\nThe assessment retains two figures directly from the survey. It reports that 31 percent of establishments had at least one unfilled position, while 18 vacancies were recorded for every 100 established positions. These values are presented as measures of the incidence and relative scale of unmet hiring demand. Their inclusion provides a concise indication of the extent to which reported staffing requirements remained unresolved.\n\nThe figures are used to support interpretation of the wider findings on labour demand and recruitment performance. They should not be read as a time series or as evidence of change across successive periods, since the cited material supplies a single reference point. The results nevertheless establish a quantitative basis for assessing the magnitude of vacancy pressure and for comparing operational observations with the reported incidence of unfilled positions.\n\nMention: \"Baseline Hiring and Vacancies Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Baseline Hiring and Vacancies Survey’s directly reported values of 31 percent and 18 vacancies per 100 positions.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:83cf3c7edd13c3e5e9c81b56", "occurrence_id": "66b29ad918350479071fe032", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Clustered Forest Products System is used as an administrative source covering Albania. Its records provide a supporting basis for the assessment of recorded activity across forest-product groupings and for the review of how production information is represented in the available evidence. The source is treated as a reference for administrative coverage rather than as a direct measure of all activity.\n\nFor this assessment, the material was collected from the source, reviewed for internal consistency, and harmonised through common descriptions of products, locations, and reporting units. Differences in terminology were reconciled before the records were incorporated into the working archive. This process reduced duplication and improved the comparability of observations used in the accompanying analysis.\n\nThe archived material was then checked against summary tables and retained with its provenance documentation. The resulting evidence supports the finding that recorded forest-product activity is distributed unevenly across production groupings, while also indicating that coverage depends on the completeness and consistency of administrative reporting. No inference is drawn beyond the scope of the records retained for assessment.\n\nMention: \"Clustered Forest Products System\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The work taken from the source comprises the collection, harmonisation, and archiving of administrative records covering Albania.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:dd59b6d629b53f03395ddacf", "occurrence_id": "54feed076d57663730bd21ac", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The statistical basis for the assessment is the Combined Household Consumption Survey, a survey used to examine household consumption patterns. The source is treated as the central evidence for the comparison of consumption levels across the observed reporting intervals. Its inclusion supports a consistent assessment of changes in the distribution and typical level of consumption.\n\nThe analysis derives an average from the reported household observations and uses that measure to establish the underlying trend. Individual entries are not reproduced as headline results. Instead, the observations are consolidated into summary measures that reduce the influence of short-term variation and provide a stable basis for interpretation. The resulting average is compared across successive intervals to identify whether consumption has increased, declined, or remained broadly stable.\n\nThis treatment allows the assessment to distinguish a sustained movement from isolated changes in individual observations. The calculated trend is used in the implementation review to inform the discussion of welfare conditions and to support comparisons across analytical groups. All findings presented in this section are therefore based on transformations of the survey observations rather than on a single reported value. The approach also permits the results to be updated when additional observations become available, while preserving a transparent link to the underlying survey evidence.\n\nMention: \"Combined Household Consumption Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives an average and a consumption trend from the Combined Household Consumption Survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:4761ca9d86c842a42c44ddaf", "occurrence_id": "7e8df897d77535ca7b5446df", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Quarterly Disease Surveillance Index (QDSI) is used as the principal indicator for the assessment of disease surveillance performance in 2015. It provides a structured set of observations suitable for internal comparison across reporting periods and surveillance dimensions. The index is treated as the central empirical basis for the analysis rather than as a catalogue of isolated reported values.\n\nFor analytical purposes, the observations associated with the QDSI are aggregated and examined for directional movement. An average is calculated across the available observations, and the resulting value is compared with the corresponding component values to identify the underlying trend. This treatment reduces the influence of individual fluctuations and provides a more stable measure for assessing overall performance.\n\nThe calculated average indicates the central level of surveillance activity represented in the indicator, while the trend assessment identifies whether performance is broadly improving, weakening, or remaining stable within the period covered. Results are presented in aggregated form to support interpretation of the surveillance profile and to avoid implying precision beyond that supported by the underlying observations. The QDSI therefore functions as the principal analytical input, with the reported findings derived through calculation from its contents rather than through reproduction of selected figures.\n\nMention: \"Quarterly Disease Surveillance Index (QDSI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the 2015 QDSI indicator to calculate an average and assess a trend rather than reproduce individual source figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:3fdf870d64188b4e07483065", "occurrence_id": "4d9152256bedc7b30c249927", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Independent Tertiary Education Survey (ITES) provides the principal empirical basis for this assessment. The survey is treated as a consistent source for examining participation, completion, staffing, and selected institutional characteristics across successive observations. Its coverage supports comparison of patterns without reliance on isolated reported figures.\n\nThe analysis calculates period-to-period changes and average levels from the underlying observations. These calculations indicate the direction and persistence of movement in participation and completion, while also showing whether staffing measures remain broadly aligned with changes in the number of learners. The resulting indicators are used to assess general tendencies rather than to reproduce individual entries from the source.\n\nParticular care is applied to the treatment of missing or non-comparable observations. Summary measures are based only on observations that can be aligned across the relevant periods, and the calculated averages are interpreted as descriptive estimates. The resulting trend is therefore used as central evidence for implementation assessment, with attention to variation between observations and to the limits imposed by the survey design. No single observation is treated as representative in isolation.\n\nMention: \"Independent Tertiary Education Survey (ITES)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses ITES to calculate average levels and period-to-period trends in participation, completion, and staffing measures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:0bf80238875eb40fb843d0f4", "occurrence_id": "2dd5d45c460be17be03880f4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Regional Student Transport System is treated as an administrative source produced by CIESIN-CIAT for 2018. It provides the primary quantitative basis for the transport measure used in this assessment and is retained under its stated title to support traceability in subsequent review.\n\nTwo figures are taken directly from the source without recalculation, namely 31 vehicles and 68 routes. These values are used as reported observations and are not combined with estimates from another source. The first indicates the recorded vehicle stock, while the second represents the routes identified in the system for the relevant reporting period.\n\nThe figures provide a limited but sufficiently defined basis for assessing the scale of recorded service provision. They should be read as administrative records rather than as independently verified measures of actual vehicle operation or passenger use. Accordingly, the analysis does not infer additional coverage, utilization, or performance indicators from the reported values.\n\nFor implementation monitoring, the two observations are retained in their original units and can be revisited when the source is updated. This approach preserves consistency between the cited evidence and the numerical statements presented in the assessment, while avoiding assumptions about information not contained in the Regional Student Transport System.\n\nMention: \"Regional Student Transport System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes 31 vehicles and 68 routes directly from the 2018 Regional Student Transport System.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:6d2b0019fdd97043d5a4139d", "occurrence_id": "0aa361a4f7bae7a1679faa31", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws its primary evidence from the Pilot Targeting and Payments System (PTPS), an administrative source maintained for monitoring beneficiary identification and payment execution. The records are treated as the basis for measurement rather than as a narrative reference. They provide a consistent observation of completed cases, enabling comparison across successive reporting periods and supporting a compact account of implementation performance.\n\nFor analytical purposes, the observations were ordered by reporting period and aggregated into comparable rates. The assessment calculates a rate for each period and derives an arithmetic average across the observed series. It also measures the direction and relative pace of change between the earliest and latest observations. This approach uses the PTPS evidence to construct a trend and a summary level, rather than reproducing individual entries or presenting isolated reported figures.\n\nThe resulting series provides the central basis for interpreting implementation continuity. The average indicates the general level around which performance was distributed, while the trend identifies whether results strengthened, weakened, or remained broadly stable over time. These calculations are used to assess consistency in delivery and to distinguish sustained movement from short-term variation. Findings are therefore presented as analytical estimates derived from the administrative records, with attention to the limits imposed by the observed reporting sequence.\n\nMention: \"Pilot Targeting and Payments System (PTPS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses PTPS records to calculate period-specific rates, their arithmetic average, and the direction of change over time.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:0eeae163c61c0eea1a2ecde0", "occurrence_id": "104b600cfdd0842562aa6c3f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Regional Social Protection Index is treated as an indicator produced by the Colombian Ministry of Education for 2016. It provides the central empirical basis for the assessment of social protection performance. The source is read as a structured set of observations that can support comparison across the dimensions recorded in the indicator, while no additional institutional or geographic attribution is introduced.\n\nThe analysis does not reproduce a reported index value as a standalone result. Instead, the recorded observations are assembled into an average score and examined for the direction of variation across the relevant dimensions. This calculation permits a more consistent interpretation of relative performance and reduces the risk that a single component will determine the assessment. The resulting measure is used to describe the general pattern contained in the source and to support the interpretation of changes within the observed distribution.\n\nThe average is reported with appropriate caution because it summarizes distinct elements that may not move uniformly. A higher combined result is therefore interpreted as evidence of stronger performance across the dimensions included in the indicator, rather than as proof of improvement in every individual element. The derived trend is used as an analytical aid and not as a substitute for the underlying observations. Any conclusion is consequently limited to the information represented by the Regional Social Protection Index and to the calculations undertaken for this assessment.\n\nMention: \"Regional Social Protection Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives an average and a directional trend from the Regional Social Protection Index indicator for 2016 rather than reproducing a quoted figure.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:d9e423693061d0b529321848", "occurrence_id": "ab8d4d917aaae808538c4242", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The indicator review draws on the Civil Road Segment System, an administrative source produced by NCHS/CDC/WHO. The material is used as an underlying record for assessing the distribution and continuity of reported road segments. Its role in the analysis is evidentiary and does not constitute a separate outcome measure.\n\nThe assessment derives a mean segment value from the available records and compares the resulting observations across the analytical sequence. A directional trend is calculated from these derived values to support interpretation of changes in coverage and implementation performance. The resulting measure is consequently an analytical estimate, rather than a figure reproduced directly from the source.\n\nThis treatment supports the broader finding that coverage has become more consistent over the period examined. The calculated average reduces the influence of isolated observations, while the estimated trend provides a concise indication of movement in the indicator. Both statistics are used to substantiate the assessment narrative and should be read as results of the analysis based on the Civil Road Segment System.\n\nThe source is retained for traceability and for subsequent verification of the calculation. Any revision to the underlying records would therefore require the derived average and trend to be recalculated before the findings are updated.\n\nMention: \"Civil Road Segment System\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives an average and a directional trend from the Civil Road Segment System rather than reproducing figures from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:601e5880dd366dbbd48e9c71", "occurrence_id": "aa8ad93f9cd399879bdb9940", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis draws on the Standard Rural Credit Dataset, a geospatial source produced by Deloitte and referring to 2019. The dataset provides the central empirical basis for examining rural credit conditions and supports a consistent assessment across the available observations. Its use is limited to the evidence required for measuring change and summarising the underlying distribution.\n\nThe assessment derives period-level averages from the source and uses those averages to calculate a trend. This treatment reduces the influence of isolated observations and permits the results to be interpreted as a general movement rather than as a description of individual records. The calculated measures are therefore presented as analytical results generated from the dataset, not as figures reproduced directly from its tables.\n\nThe resulting trend indicates the direction and relative pace of change in rural credit conditions, while the associated averages provide a concise basis for comparison across the observations covered by the source. These calculations are used throughout the assessment to support the interpretation of performance and to identify areas requiring continued monitoring. No inference is drawn beyond the scope of the underlying records, and the results should be read as summary measures derived from the cited evidence.\n\nMention: \"Standard Rural Credit Dataset\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the 2019 Deloitte geospatial Standard Rural Credit Dataset to calculate averages and a trend.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:6e3a72d158b79c17787a892a", "occurrence_id": "92375a39d0a2b19422db0ea7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on Statistical Agri-Food Prices Microdata, a microdata source used as the principal evidence for the price analysis. The records provide the basis for examining observed movements across commodities and grades. No external benchmark is introduced, and the source is treated as the underlying observation set rather than as a collection of headline figures.\n\nPrices are grouped by commodity and grade, after which arithmetic means are calculated for each comparable period. The resulting series is used to identify the direction and consistency of price movements. This procedure reduces the influence of individual observations and permits comparison of central values across the analytical groups. The reported findings therefore represent calculations from the records, not figures reproduced directly from the source.\n\nThe analysis indicates whether average prices increased, declined, or remained broadly stable over the observation sequence. Where variation is substantial, the assessment retains the dispersion in the interpretation and avoids assigning significance to isolated changes. The computed averages and directional measures constitute the central evidence for the conclusions presented in this annex. Reproducibility depends on applying the same grouping, exclusion, and averaging rules to the cited microdata.\n\nMention: \"Statistical Agri-Food Prices Microdata\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses Statistical Agri-Food Prices Microdata to calculate comparable average prices and identify their trends.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:c62b7be81c5f7791c12a047f", "occurrence_id": "886047e3c113a8a9b1ffd735", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment relies on the Statistical Disability and Support Microdata, a microdata source produced by LinkedInEconomicGraph. It is treated as the central empirical input for examining patterns in reported disability, assistance requirements, and access to support. The records provide a consistent basis for comparing observations across the analytical categories used in the assessment.\n\nThe analysis derives group averages and directional trends from the underlying records rather than reproducing individual reported values. Observations are consolidated into comparable measures, and the resulting summaries are reviewed for changes in prevalence, differences in support requirements, and variation in access indicators. This procedure reduces the influence of isolated records and supports interpretation at an aggregate level.\n\nThe derived results indicate that support needs are unevenly distributed across the observed groups and that changes in the principal indicators are gradual rather than uniform. Average levels are therefore used to establish the analytical baseline, while the calculated trend is used to assess whether disparities are narrowing or persisting. These measures inform the assessment of implementation risks and the prioritisation of follow-up analysis. The source is retained as the principal evidence for these calculations, with conclusions limited to patterns supported by the available records.\n\nMention: \"Statistical Disability and Support Microdata\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives aggregate averages and directional trends in disability, support needs, and access indicators from the Statistical Disability and Support Microdata.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:9de552c8a12755acb7e44a18", "occurrence_id": "993b24fa74f4ff054b906327", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment of reported occupational incidents indicates persistent exposure to preventable hazards across the reporting population. The available evidence supports the broader finding that compliance weaknesses remain material, particularly where supervision and protective measures are inconsistently applied.\n\nThe Triennial Occupational Safety Microdata, a 2016 microdata source, provides the supporting observations for this assessment. The analysis reproduces directly from that source a rate of 12.4 percent and a total of 68,417 recorded cases. These figures are used as reported and are not recalculated for the present assessment.\n\nThe two observations reinforce the conclusion that incident levels remain significant despite the existence of formal safety requirements. Their use is limited to substantiating this claim within the wider discussion of implementation performance. No inference is drawn regarding changes over time, and no additional classification is applied to the reported cases.\n\nThe evidence should therefore be read as a descriptive basis for monitoring and supervision. Further interpretation would require comparable observations collected under consistent procedures and aligned coverage.\n\nMention: \"Triennial Occupational Safety Microdata\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The author takes the directly reported values of 12.4 percent and 68,417 recorded cases from the Triennial Occupational Safety Microdata.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:a7f0aabfdc5737643f51ef43", "occurrence_id": "7d7cfd4721a04e01d19c812e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Midline Socioeconomic Survey (MSS) is the survey used as the principal empirical basis for this assessment. Its observations provide the consistent measurement framework for examining household conditions, service access, and selected welfare indicators. The analysis treats the MSS as a structured source for deriving comparable measures rather than as a repository of isolated figures.\n\nFor each indicator, the assessment calculates average levels across the available observations and examines the direction and pace of change between successive observation points. These calculations provide a trend measure that supports interpretation of broad movements in welfare and access. The resulting averages are used to identify persistent differences across population groups and to distinguish gradual change from short-term variation.\n\nThe findings are therefore based on transformations of the MSS data, including aggregation, comparison, and calculation of central tendencies. This approach reduces the influence of individual responses and places emphasis on the overall pattern recorded by the survey. In interpreting the results, attention is given to the consistency of movements across indicators and to the extent to which changes in one measure correspond with changes in related measures.\n\nThe MSS consequently serves as the central evidence for the statistical assessment. Its value lies in permitting a transparent comparison of observed conditions and in supporting a measured account of trends, rather than in presenting selected observations as standalone findings.\n\nMention: \"Midline Socioeconomic Survey (MSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the MSS survey to calculate average levels and derive trends from aggregated observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:06bd5a4eae2a52d113b131c6", "occurrence_id": "9acf6ddaf241236d47e09fc8", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Comparative Service Catchments Census, a census produced by ILO, as supporting evidence for the review of service accessibility. The source provides a structured basis for comparing catchment characteristics and identifying differences in the distribution of reported services. Its use is limited to substantiating the assessment finding that access conditions are not uniform across the service system.\n\nThe underlying records were collected and reviewed for consistency before inclusion in the analytical file. Terminology and measurement conventions were harmonised to permit comparison across records, while entries that could not be reconciled were retained separately for quality control. The resulting file was archived with the working materials to support traceability and subsequent verification.\n\nThe evidence is interpreted together with the broader results of the assessment rather than as a standalone measure of performance. In particular, it informs the description of uneven service coverage and helps distinguish differences in reported catchments from differences arising through inconsistent recording practices. The census therefore serves as a corroborating source for the narrative assessment and not as the sole basis for operational conclusions.\n\nMention: \"Comparative Service Catchments Census\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The author used the ILO census to support findings on uneven service coverage after collecting, harmonising, reviewing, and archiving its records.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:cc94e077426bb9e549a70f1e", "occurrence_id": "7b959a5c796a3ce233823337", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The principal empirical input for this assessment is the Experimental Informal Trade Registry database. It provides a structured set of observations on informal commercial activity, with records organised sufficiently consistently to support comparison across reporting periods. The database is treated as an analytical source for aggregate measurement rather than as a narrative account of individual cases.\n\nThe assessment derives summary indicators from the available observations. In particular, it calculates an average level of recorded activity and examines the direction of change across successive reporting periods. The resulting trend is based on the distribution of records in the database, with equal treatment applied to observations that meet the stated inclusion criteria. This approach reduces the influence of isolated entries and provides a more stable basis for assessing movement over time.\n\nThe findings should be interpreted as estimates of registered informal activity. They reflect the scope and consistency of the underlying records and may not capture activity that was not observed or entered. For this reason, the calculated average and trend are used to support implementation assessment and comparison, rather than to present individual registry entries as standalone measures. Any apparent change is therefore considered alongside the completeness of the recorded observations and the persistence of the measurement procedure.\n\nMention: \"Experimental Informal Trade Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The assessment takes a calculated average and a trend in registered informal activity from the Experimental Informal Trade Registry database.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:96c4a8d91fd139988add19d8", "occurrence_id": "6b9fcad43c66e3a752b30bca", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Data sources.** The Ministerial Nighttime Lights Report is cited as a report referring to 2018 and is retained solely as contextual background for the assessment. Its mention indicates the broader information environment in which the analysis was prepared, but it does not provide an evidentiary basis for the findings presented here. No values from that source are incorporated into the tables, calculations, classifications, or narrative conclusions.\n\nThe assessment relies on the documented evidence assembled for the relevant activities and on the procedures applied during review. Results are interpreted against the stated objectives, implementation arrangements, and observed outputs. This approach supports a clear distinction between contextual references and material used directly for measurement or verification.\n\nAccordingly, the cited report has no role in establishing levels, changes, rates, or comparisons reported in this annex. Its inclusion is limited to acknowledging a potentially relevant source without extending the analysis beyond the evidence specifically examined. The findings should therefore be read as based on the assessment record and its associated validation procedures, rather than on external estimates or indicators mentioned for background purposes.\n\nMention: \"Ministerial Nighttime Lights Report\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The source is mentioned only as background, and no values or analytical results are taken from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:65de5f56d40e96ff65b776bb", "occurrence_id": "c92db5c07ea51502deed92f4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Linked Skills and Employment Survey (LSES), a survey covering Chile, as the principal empirical basis for the analysis of skills and employment outcomes. Its coverage supports a consistent comparison of observed characteristics across the population examined and provides the main input for the indicators reported in this assessment.\n\nThe analysis calculates average levels from the underlying observations and derives a trend across the relevant analytical categories. These results are therefore estimates produced from the survey microdata rather than figures reproduced directly from a published table. The procedure applies the same definitions and aggregation rules throughout, allowing differences in measured outcomes to be interpreted on a comparable basis.\n\nThe resulting averages are used to identify the direction and relative scale of variation in skills and employment outcomes. Particular attention is given to the stability of the pattern across groups, since isolated observations may not represent the broader distribution. The findings are consequently presented as summary measures derived from the Linked Skills and Employment Survey (LSES), with the survey serving as the central evidence for the assessment’s interpretation of observed patterns.\n\nMention: \"Linked Skills and Employment Survey (LSES)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives average skills and employment levels and their trend from the Linked Skills and Employment Survey (LSES) covering Chile.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:7be3e05d4465ca97928fb4bc", "occurrence_id": "deddfbebec0a02e34bcd0846", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Baseline Built-Up Surface Dataset is retained as the principal geospatial evidence for the baseline assessment. UN COMTRADE is recorded as the organisation producing this source, which refers to 2015. The citation is maintained with the dataset name and its associated descriptive information so that the evidentiary basis remains identifiable during review.\n\nThe material was collected for analytical use, then harmonised to support consistent interpretation across the assessment record. Harmonisation addressed differences in structure, terminology and presentation without introducing additional observations. The resulting file was archived with the supporting documentation, allowing subsequent checks to distinguish source material from analytical interpretation.\n\nThe dataset provides the reference surface used in the assessment narrative and in the associated statistical review. Its role is limited to establishing the baseline evidence; no implied change over time is derived from the citation alone. Retaining the original attribution to UN COMTRADE and the reference to 2015 supports traceability, while the geospatial character of the source informs the handling of the archived material. The record is therefore treated as a documented input to implementation monitoring rather than as a standalone estimate of outcomes.\n\nMention: \"Baseline Built-Up Surface Dataset\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The passage records the 2015 geospatial Baseline Built-Up Surface Dataset attributed to UN COMTRADE and describes its collection, harmonisation, and archiving.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:127ad62d1b32e8d61a55fbb7", "occurrence_id": "09f5a45d514c3eec72525ff7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment reviews the administrative basis for monitoring water resources and considers the completeness of records used in implementation planning. Attention is given to the consistency of registration, the availability of operational information, and the extent to which the documented inventory can support routine supervision.\n\nThe Record-Based Water Body Registry database, covering 2024, is used as the principal administrative reference for this assessment. It records 68 water bodies in the registry and identifies 31 water bodies with documented operational status. Both figures are reproduced directly from the database and are presented without adjustment or additional calculation.\n\nThese results support the broader finding that the registry provides a usable baseline for implementation oversight while also indicating that documentation is not uniform across all entries. The difference between the recorded inventory and the subset with operational information is relevant to the interpretation of monitoring results and to the prioritisation of follow-up verification.\n\nThe figures should therefore be read as administrative evidence rather than as an independent measure of physical conditions. Subsequent reporting may use the registry to structure verification activities, clarify missing records, and maintain a consistent basis for reviewing progress.\n\nMention: \"Record-Based Water Body Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage takes the figures 68 water bodies and 31 water bodies with documented operational status directly from the Record-Based Water Body Registry database covering 2024.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:36fdf79c9b14993ed00c1524", "occurrence_id": "d850113b881673b7229d8b07", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Derived hardship measures.** The comparison of deprivation across localities rests on two measures computed for this review rather than lifted from any single publication. Both were built on the 2024 refresh of the Harmonised Deprivation and Hardship Registry (HDHR), which is kept as a standing database of administrative records rather than as a field collection, so the inputs arrive as registrations and entitlements rather than as answers to an interview.\n\nThe first measure counts households holding at least three of the seven deprivation indicators the registry carries; the second expresses the same count as a share of all households in the locality. Neither figure is published in that form. They were assembled here because the published tabulations cut the same material by service rather than by household, which is the wrong axis for the claim being made.\n\nThat claim is a narrow one. The wider argument concerns how deprivation is measured when the records behind it were created for entitlement rather than for statistics, and these two measures are brought in to show that the axis choice changes the ranking of localities, not to establish the level of hardship itself. Because the registry is refreshed continuously, a household recorded in one month may be absent by the next, and the counts should be read as a position on the 2024 refresh rather than as a stable population.\n\nMention: \"Harmonised Deprivation and Hardship Registry (HDHR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"Two deprivation measures computed on the 2024 Harmonised Deprivation and Hardship Registry refresh, used to show that the choice of tabulation axis changes locality rankings.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:799faee73b465190e330f481", "occurrence_id": "bdcb3d1a3532928e526a1e32", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Building extent.** The Metropolitan Building Footprint Dataset covering Addis Ababa in 2016 supplies the geometric base on which the block-level counts rest. It is a geospatial dataset in the strict sense: each structure is a polygon with an attribute recording its use class, and the layer is a snapshot rather than a series, so nothing in it speaks to construction before or after that year.\n\nTwo quantities are taken from it as published and not re-derived. The first is the count of mapped structures in the inner administrative lines; the second is the share of those polygons carrying a residential use class. Both are reproduced at the precision the layer carries, and no attempt was made to reconcile them with the census counts of dwellings, which are drawn on different boundaries and were compiled for a different purpose.\n\nThese figures support one part of the discussion rather than carrying it. The section's argument is about what a footprint layer can and cannot settle when it is the only spatial evidence available, and the mapped counts enter that argument as an illustration of the layer's resolution, not as a measurement of the city's housing stock.\n\nMention: \"Metropolitan Building Footprint Dataset\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"Two published quantities from the 2016 Metropolitan Building Footprint Dataset for Addis Ababa, reproduced to illustrate resolution rather than measure housing stock.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:3b47fe68bfbc843e08e5264b", "occurrence_id": "319bd66d3d57b060cf9f7b13", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Establishment churn.** Entry and exit rates for the reporting period were computed from the Verified Business Demography Registry, which MapX maintains as a database of live establishments with a start and end date attached to each record. The registry is the source from which every quantity in this paragraph is derived; nothing here is a figure MapX itself publishes.\n\nThe calculation takes the register at two successive refreshes, counts establishments appearing in the later extract but not the earlier one, and expresses that count against the stock surviving from the first. A second run does the same in reverse. Both ratios are then reported by establishment size band, which is where the two distributions separate: the small-establishment series moves by several times the magnitude of the large-establishment series while moving in the same direction.\n\nThe magnitude of that movement should not be read as an estimate of churn in the wider economy. Registration is triggered by the obligations that bring an establishment onto the register in the first place, so a business that never acquires them is invisible at both refreshes, and the ratios describe the population that is visible to MapX rather than the population of firms.\n\nMention: \"Verified Business Demography Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"Entry and exit ratios derived from successive refreshes of MapX's Verified Business Demography Registry, computed here rather than quoted from the registry.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:5eec3f30d7f9a5ab39c47c0e", "occurrence_id": "2169f0cb99ffa296e840d66a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Transfer coverage.** The two coverage figures reproduced below are those published for the 2021 edition of the Integrated Cash Transfer Registry Index (ICTRI), and they are quoted rather than recomputed because the underlying records were not made available. The first is the share of eligible households with at least one transfer recorded; the second is the share of recorded households receiving the full schedule rather than a partial one.\n\nBoth are indicators in the ordinary sense: they are constructed from administrative feeds by a method described in the accompanying note, and they are not survey estimates. No sampling error is attached to either figure, and none should be inferred, since the population is the register itself and not a sample drawn from it. Coverage of the register is a separate question from the precision of the arithmetic performed on it.\n\nThe index is the load-bearing source for this paragraph. The claim being made is about the gap between eligibility and receipt, and that gap is expressed entirely in the two published shares; the surrounding discussion of programme design does not feed back into them, and no other source is used to adjust or corroborate either number.\n\nMention: \"Integrated Cash Transfer Registry Index (ICTRI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The two coverage shares published in the 2021 Integrated Cash Transfer Registry Index, quoted as reported and used as the basis for the eligibility-to-receipt gap.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:69bf7fd032a4e45525d47ca5", "occurrence_id": "fa30609b7a0a291a012d77c5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Enrolment response.** Coverage was estimated from the Integrated School Feeding Index for 2021, published by HZZO, rather than taken from the index's own headline table. The index is an indicator series: it reports a constructed score per region, and the score is not the same object as the enrolment relationship examined here.\n\nThe estimate regresses observed enrolment on the index score across the regions covered by the 2021 release, and reports the fitted association together with the residual pattern by region type. Both were computed on the released values; no region was excluded, and the score was not reweighted. The residual pattern matters more than the coefficient for the present purpose, since it is the coastal regions that depart from the fitted line in the same direction and by a similar margin.\n\nThat result carries the section's argument. The surrounding material on programme financing is background to it, and the point at issue is whether an index built for monitoring purposes behaves as though it were measuring the same thing as enrolment, which the fitted association and its regional residuals address directly. Where the index's own definitions depart from those used in the enrolment series, the difference was left unresolved rather than harmonised.\n\nMention: \"Integrated School Feeding Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"An association between enrolment and the 2021 HZZO Integrated School Feeding Index score, estimated from the released values rather than quoted from the index.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:c07c5738314b2c60cc551997", "occurrence_id": "ecf43b0f04f9ec241af5744f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Needs across displacement settings.** The analysis is carried out on the Provincial Minimum Needs Survey, a field collection compiled by the Colombian Ministry of Education that covers both settled and displaced households in the same instrument. Every quantity below is derived from its records; the survey publishes no tabulation resembling them.\n\nHouseholds were grouped by whether the head reported a displacement episode, and each group was scored on the asset and service items the questionnaire carries. The score is a simple count with no weighting, computed here so that both groups face identical questions. The difference between the two group means is then reported for each province, which is where the third dimension of the data does its work: the provincial split is what separates a displacement pattern from a rural one, since the displaced households are concentrated in a minority of provinces.\n\nThat separation is the section's central concern. The survey is not one piece of evidence among several in this paragraph but the only one, and the surrounding discussion of school provision exists to explain why an education ministry holds needs data at all. No attempt was made to weight the sample to a population frame, and the means should be read as describing respondents.\n\nMention: \"Provincial Minimum Needs Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"Displacement and non-displacement needs means derived from the Colombian Ministry of Education's Provincial Minimum Needs Survey, compared province by province.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:7b85bfc0f1a41e46af2cb11e", "occurrence_id": "6a15ab58392d3c02ea9671b9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Working hours and informality.** Two figures are reproduced from the Diagnostic Labour Force Survey (DLFS) for 2020, compiled under World Bank/IFC auspices, and both are quoted as the survey prints them. The first is the share of employed respondents whose hours fall below the threshold the survey itself applies; the second is the share of those respondents who also report no contribution to a social insurance scheme.\n\nThe survey is the principal evidence for the paragraph and the only source of either quantity. Its own documentation describes the sampling frame and the imputation applied to non-response, and those notes travel with the source; they are not restated here, and the figures carry whatever treatment the published tables carry. Where the survey reports a range for a domain, the point value was used without comment.\n\nReproducing rather than recomputing was deliberate. The two shares are used to establish the overlap between short hours and uninsurance, and re-deriving them from microdata would introduce decisions the survey has already made and settled, replacing a published quantity with one that could not be checked against it.\n\nMention: \"Diagnostic Labour Force Survey (DLFS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Two published 2020 shares from the Diagnostic Labour Force Survey (DLFS), reproduced as printed to establish the overlap between short hours and lack of insurance.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:17f7189977fec70d7239c5b3", "occurrence_id": "e5646bfe1d354c237b6a1715", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Elevation and settlement extent.** The Biennial Settlement and Elevation Report supplies the elevation band each settlement falls into and the count of settlements per band, and both quantities were taken from it as released for 2022. What was done with them was not: the association between band and reported poverty incidence was computed here, because the report presents the two distributions separately and never joins them.\n\nThe calculation pairs each band's settlement count with the poverty incidence recorded for the same administrative unit, and reports the fitted slope across bands. It is descriptive and unweighted, and no confidence interval was carried through, since the elevation assignment itself is a band rather than a measurement with a stated precision.\n\nThis is one strand in a wider paragraph rather than its subject. The paragraph concerns whether physical constraints are being used as a stand-in for policy failure, and the fitted slope enters that discussion to show that the two explanations are not distinguished by the report's own presentation. The settlement counts describe mapped settlements only, and a settlement that appears in no inventory appears in no band.\n\nMention: \"Biennial Settlement and Elevation Report\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"A band-level association computed from the 2022 Biennial Settlement and Elevation Report, joining settlement counts to poverty incidence the report keeps separate.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:1ca64666b9f28e7c6bdaf589", "occurrence_id": "e353e54bb308942a7b358d49", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Shelter and tenure.** The two shares quoted here come from the Inter-Agency Housing Conditions Survey conducted in Eritrea in 2021, and neither was recalculated. the first is the proportion of surveyed households in structures classed as temporary; the second gives the share without a documented tenure arrangement of any kind.\n\nBoth are reproduced at the precision the survey prints, and both are read as survey estimates rather than as counts of a population. The sampling frame is the collective site list, which means the figures describe households the survey team was able to reach and enumerate, a restriction the survey states plainly and which is not adjusted for here.\n\nThe survey carries the paragraph. The argument is about the distance between what agencies report about shelter and what a joint instrument can establish, and that distance is measured entirely by the two quoted shares, since no other source reports both items in the same frame. Definitions of temporary and of documented tenure are the survey's own; they were adopted for these figures without restatement, and where the two items overlap in a household the survey does not de-duplicate them.\n\nMention: \"Inter-Agency Housing Conditions Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Two published shares from the 2021 Inter-Agency Housing Conditions Survey in Eritrea, quoted as printed and used as the paragraph's principal evidence.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:8265d602ec18b09f23dabfda", "occurrence_id": "e816163e40317b0b70611294", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Assessment context.** The Monthly Learning and Equity System (MLES) in Cyprus is named here for the institutional arrangement it represents, and no quantity from it is used. It is an administrative system rather than a statistical release: cycles are a by-product of how records are kept, and the coverage they achieve depends on enrolment practice rather than on a sampling design.\n\nThat distinction is the reason for the mention. The paragraph turns on what can be inferred when a country's only repeated education observations come from administrative collection, and MLES is offered as a worked instance of that situation in one member state. Its published tables were consulted to confirm the arrangement and nothing further; no rate, score or cycle figure is quoted, and none of the figures appearing elsewhere in this section were drawn from it.\n\nThe system supports the paragraph's framing rather than its conclusion. Where the discussion turns to measured learning, the evidence comes from elsewhere, and the reader should not infer from the mention that any comparison made here has MLES behind it.\n\nMention: \"Monthly Learning and Equity System (MLES)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"Monthly Learning and Equity System (MLES) in Cyprus named as an administrative arrangement, with no values taken from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:3c506efe87ab7907b245870f", "occurrence_id": "4db6a7d3352228501a9fd869", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Conditions of work.** The two shares below are reproduced from the Stratified Working Conditions Index for 2022, published by Townsend, and are quoted in the form the index releases them. The first is the share of covered workers reporting fixed rather than variable schedules; the second is the share reporting no written statement of terms at the time of the reference period.\n\nNeither figure is a survey estimate. The index is a constructed indicator, assembled from employer returns by a method its publisher documents, and it carries no sampling error because no sample is drawn. The precision of the arithmetic is not the same as the precision of a measurement, and the two are easy to confuse when both appear as a single percentage to one decimal.\n\nThese shares are brought in to support one comparison inside a longer discussion of how conditions are instrumented, and they are not that discussion's subject. The index's stratification is by establishment size; the comparison drawn here concerns schedule type, so the shares are used as reported rather than re-tabulated, and no claim is made about strata the index does not define.\n\nMention: \"Stratified Working Conditions Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"Two 2022 shares reproduced from the Townsend Stratified Working Conditions Index, used to support one comparison within a longer discussion.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:8d2fc9bd2ab52f9f3aae225c", "occurrence_id": "5bd60d5c0608a44a441f535c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Consumption aggregates.** Both quantities reproduced here are taken from the Local Household Consumption Index (LHCI) as it releases them, without adjustment or recomputation. The first is the index's reported share of households below the subsistence line it applies; the second is the share of that group receiving any transfer recorded in the same period.\n\nThe LHCI is an indicator, not a survey, and its figures are constructed from inputs of several kinds under a published weighting. They carry no intervals, and the absence of intervals reflects the construction rather than an omission: the object is an index value, not an estimate of a population quantity with a sampling design behind it. Whether the weighting recovers consumption is a question the index does not answer about itself.\n\nThe paragraph rests on these two figures. The claim concerns the gap between measured shortfall and recorded transfer receipt, and both terms of that gap come from the index, so no corroborating source is cited and none is needed for the comparison to hold. The subsistence line is the index's own and is not restated here, nor converted to any external threshold.\n\nMention: \"Local Household Consumption Index (LHCI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Two reported shares from the Local Household Consumption Index, quoted as released and used as the whole basis of the shortfall-to-transfer comparison.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:4089a610669e0b71f2f3f155", "occurrence_id": "836753475beaaa5280eb7d79", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Public works employment.** Duration and spell length were derived from the Departmental Public Works Survey (DPWS) carried out in 2023 under Doing Business, rather than read off the survey's summary table, which reports a different quantity. The survey collects a working history per respondent, and the summary collapses it to a single status flag.\n\nA spell was defined here as a consecutive run of weeks on a works programme, and three quantities were computed from the histories: the median spell length, the share of respondents with more than one spell, and the ratio of weeks worked to weeks eligible. All three rest on the respondent's own sequence of answers, and no external frame was used to impute the gaps where a month was reported without a programme attached.\n\nThose quantities are the substance of the section. The discussion of programme design elsewhere in the chapter is context for them, and the point at issue is whether short spells reflect rotation rules or withdrawal, which the median and the multiple-spell share address between them. Doing Business documents the field conditions behind the survey, and those notes were not reproduced; where a respondent's history was internally inconsistent the case was retained as reported.\n\nMention: \"Departmental Public Works Survey (DPWS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"Spell-level quantities derived from the 2023 Departmental Public Works Survey (DPWS) histories rather than from the survey's collapsed status flag.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:ebfade4d8d58d19ec46f4592", "occurrence_id": "c735565b61878f5abed2f927", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Resettlement outcomes.** Two shares are quoted from the Multi-Topic Resettlement Survey conducted in Bangladesh in 2021, and both are given as the survey prints them. The first is the proportion of resettled households with a plot allocated in the new site; the second is the proportion whose livelihood was recorded as restored within a year of the move.\n\nNeither figure was recomputed. The survey is the principal source for this paragraph and the only one reporting both items on the same households, so the comparison between allocation and restoration is made entirely within its own frames and definitions. What the two quantities share is a sampling unit, which is what allows them to be set against each other at all.\n\nThe survey's own notes on non-response and on the treatment of households that moved twice are left with the instrument, and the figures carry those decisions unaltered. Where a household appears at two sites the survey counts it once, at the later site, and that convention is the survey's rather than one adopted here. Allocation should be read as recorded allocation, which is not the same as a settled claim.\n\nMention: \"Multi-Topic Resettlement Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Two published shares from the 2021 Multi-Topic Resettlement Survey in Bangladesh, quoted as printed and compared within the survey's own frame.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:016287edf8d6cb36952f1fa3", "occurrence_id": "1571c0965c5d36ea397aff74", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Catchment alignment.** The analysis was performed on the Follow-Up Service Catchments Registry, a database that records which facility each case is assigned to and the date of the last contact. Nothing in this paragraph is a figure the registry publishes; the registry publishes lists, and the quantities below are properties of those lists.\n\nFor each catchment the calculation counts distinct facilities receiving cases and compares that count with the number of facilities formally designated for the catchment. The resulting coverage ratio is reported alongside a second quantity, the share of cases whose last recorded contact predates the current period. Both were computed across the whole register with no exclusions, and neither carries an interval, since the register is the population being described rather than a sample of it.\n\nThat ratio is what the section turns on. The surrounding description of referral practice exists to explain why a catchment might send cases to a facility outside its designation, and the two quantities are the evidence for whether that happens systematically or in a handful of places. Coverage is a property of records: a facility that receives cases but files them late appears smaller than it is.\n\nMention: \"Follow-Up Service Catchments Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"Coverage and staleness ratios computed on the Follow-Up Service Catchments Registry, which publishes lists rather than the quantities derived from them.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:bace6e918e5bf25547318825", "occurrence_id": "8c83904d1398af86eaf7a65a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Flow through the system.** Within a broader examination of how patient throughput is instrumented, one comparison rests on the Endline Patient Flow System (EPFS) of 2017. The material is administrative, generated by the recording steps themselves rather than collected for measurement, and it was used here to derive a single ratio rather than to describe volume.\n\nThe ratio divides the count of cases recorded as transferred between units by the count recorded as discharged in the same period. It was computed from the endline extract and not compared with any other year, since the system's recording rules changed between cycles and an earlier extract would not be measuring the same transition.\n\nThat comparison is subordinate. The paragraph's argument concerns what administrative throughput can stand in for when no measured series exists, and the ratio enters as one illustration of a quantity that looks like an outcome and is in fact a property of the recording process. No inference is drawn from it about clinical need, and the count of cases behind it is not a count of patients.\n\nMention: \"Endline Patient Flow System (EPFS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"A transfer-to-discharge ratio derived from the 2017 Endline Patient Flow System extract, used as one illustration within a wider discussion of throughput.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:d80e0d3c3d107be15bc5a8df", "occurrence_id": "fcad0950208422b25d57050f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Progression through tertiary study.** Retention was estimated from the Audited Tertiary Education Registry (ATER), Eighth Round, released for 2022, rather than taken from the round's published cohort table. The registry is a database of enrolment records with an audit flag attached to each institution, and the published table reports enrolment rather than continuation.\n\nThe estimate follows institutions present in both the Seventh and Eighth Round extracts, counts those still carrying active enrolments at the later date, and expresses that count against the cohort carried forward. Institutions whose audit flag changed between rounds were retained, and the resulting sensitivity was noted rather than corrected, since the flag records the audit status of the institution and not the validity of its records.\n\nThat estimate is the section's contribution and everything else in the paragraph is in service of it. The registry's own documentation of the Eighth Round describes the reconciliation between institutional returns and the central register; it was consulted and not reproduced. Because the calculation requires presence in both extracts, an institution that left the register entirely is absent from both numerator and denominator, which biases the estimate upward.\n\nMention: \"Audited Tertiary Education Registry (ATER)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"Retention estimated from the Seventh and Eighth Round extracts of the Audited Tertiary Education Registry (ATER) for 2022, not from the round's cohort table.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:c5682f58d005f0b46e41d2a8", "occurrence_id": "74a72405dcf1a3b30dda7064", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Spatial pattern of expenditure.** Both the poverty share and its variation across units were computed from the Endline Household Income and Expenditure Survey, whose records carry a geographic identifier down to the small-area level. The survey publishes regional means; the quantities below are not regional means and were not derived from them.\n\nHouseholds were ranked on reported expenditure, a subsistence threshold applied at the national level, and the share falling below it recorded for each small area. The second quantity is the spread of those small-area shares, reported as an interquartile range rather than a mean, since the distribution is dominated by a minority of areas well above the rest. Both calculations use the survey's weights as supplied.\n\nThe section's argument rests on the second quantity rather than the first. The claim is that a national share conceals a distribution whose shape is the policy-relevant fact, and the interquartile range is the evidence for that, while the national share is reported only to give the spread a reference point. Expenditure is reported, not measured, and the threshold is an analytical choice made here.\n\nMention: \"Endline Household Income and Expenditure Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"A poverty share and its small-area spread computed from the Endline Household Income and Expenditure Survey, using the survey's own weights as supplied.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:0936693b22aa428d9dc50cce", "occurrence_id": "4b36ed94aef30e4e09796eb3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Coping after a shock.** Two figures are reproduced from the Longitudinal Shock Coping Survey (LSCS) as the survey releases them, without recalculation. The first is the share of households reporting a shock in the reference window; the second is the share of those households whose response involved the sale or lease of an asset.\n\nThe survey is the paragraph's principal source and the figures are quoted rather than derived, because the value of the LSCS lies in its panel: the same households appear in successive waves, and the published pair is the one computed on the same respondents across both. Re-deriving it here would substitute decisions the survey has already documented for decisions that could not be checked against anything.\n\nBoth figures are read as estimates from a sample, and the survey's own notes on attrition between waves apply to them without being restated. The second figure is bounded below by what respondents are willing to report, and asset transfers conducted informally or outside the reference window are outside its scope entirely. The pair is used to establish that the two quantities move together, and nothing in the paragraph attributes one to the other.\n\nMention: \"Longitudinal Shock Coping Survey (LSCS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Two published figures from the Longitudinal Shock Coping Survey (LSCS), quoted as released because their value lies in the panel the survey maintains.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:12f7e7b2de7a7ab09e602e4a", "occurrence_id": "e5b1876c857849e969d5773e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Harmonising wage series.** The wage series for Mauritania was harmonised to the definitions carried by the Verified Wage Structure Index for 2022, and the work described here concerns that harmonisation rather than any estimate drawn from the result. The index is a constructed indicator, and its value for this purpose is that it fixes a set of categories the national series does not use.\n\nThree adjustments were made. Basic pay was separated from allowances, which the national series combines; the reference period was moved from the fiscal to the calendar year; and establishments below the index's reporting floor were excluded so that both series describe the same population. No figure was reweighted afterwards, and the harmonised series is not comparable with the national one at its published points.\n\nThis supports one part of a longer treatment of comparability. The index is not the subject of the paragraph and no claim is made about the Mauritian wage distribution; what is claimed is only that two series can be made to describe the same population by three specific operations, and that the operations are recoverable from the index's categories. The archive of the harmonised series was lodged with the working group that commissioned the work.\n\nMention: \"Verified Wage Structure Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"Three harmonisation operations aligning a Mauritian wage series to the 2022 Verified Wage Structure Index categories, reported as curatorial work.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:c68dcb0e4eafe49c8ed7ba3a", "occurrence_id": "97222ea79608fcb2378bf691", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Administrative geography.** The two counts reproduced below are taken from the District Ward Outline Registry for Beijing as held in 2017 by the People’s Bank of China, and are quoted without adjustment. The first is the number of ward outlines active at the reference date; the second is the number of boundaries revised during that year.\n\nThe registry is a database maintained for operational purposes, and neither count is a statistical output of it. They are reproduced here because the surrounding discussion needs a scale for the boundary churn that the rest of the paragraph describes, and no statistical office publishes that quantity. What the registry records is the outline in force, which means a boundary revised twice within the year is counted once.\n\nThese counts carry one claim in a longer passage about how frequently administrative geography is redrawn, and they are not that passage's conclusion. The precision of the counts is the registry's own; no reconciliation was attempted against the civil administration's published ward list, which is maintained on a different cycle and would not agree at the reference date.\n\nMention: \"District Ward Outline Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"Two counts quoted from the 2017 District Ward Outline Registry for Beijing, reproduced to give scale to a passage on administrative boundary churn.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:a8a676fe2955c045482d362b", "occurrence_id": "b0e4219261a30156ac29bb73", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Locating the catchment.** The distance distribution was computed from the Rapid Service Catchments Survey (RSCS), First Version, conducted in 2016 with support from the World Bank. The survey is a geospatial collection: households and facilities were both located, and the derived quantity is the travel distance between a household and the facility it reported using.\n\nDistances were computed on the recorded coordinates rather than taken from the survey's distance field, which is rounded to the nearest kilometre and would have collapsed the lower tail of the distribution. The median and the ninetieth percentile reported here come from that recomputation, and both were produced on the released coordinates without exclusion of cases whose facility was not itself surveyed.\n\nThat distribution is what the paragraph is about. The survey's coverage notes and the World Bank's field documentation explain why certain areas were enumerated in the dry season only, and those notes are context for reading the distribution rather than evidence in it. Because the First Version was a single round, nothing here speaks to change over time, and a household that travelled to a facility outside the surveyed area contributes a distance to a location the survey did not record.\n\nMention: \"Rapid Service Catchments Survey (RSCS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"A distance distribution recomputed from the 2016 Rapid Service Catchments Survey (RSCS) coordinates, replacing the survey's kilometre-rounded field.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:fb7c5dfbe2d4d76df03ce2a0", "occurrence_id": "7a75bab7ef186e3239f4df8e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Readiness reporting.** The pair of figures below is quoted from the Follow-Up Service Delivery and Readiness Registry for Argentina as constituted in 2016, and neither was re-derived. The first is the share of reporting units that filed within the required window; the second is the share of those units whose filing included the readiness schedule at all.\n\nThe registry is a database of submissions rather than a statistical collection, and the two figures describe reporting behaviour rather than service conditions. They are reproduced as the registry publishes them, to the precision it uses, and no attempt was made to distinguish a late filing from a filing that arrived on time but incomplete.\n\nThese figures support a single claim within a passage concerned with why readiness data is thin rather than with what it says. The registry's own completeness is the only thing being asserted; the conditions the schedules describe are not assessed here, and a unit that did not file is not treated as a unit without conditions. Where a unit filed twice the registry keeps the later submission, and that rule is the registry's.\n\nMention: \"Follow-Up Service Delivery and Readiness Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"Two reporting-behaviour figures quoted from the 2016 Follow-Up Service Delivery and Readiness Registry for Argentina, supporting one claim in a longer passage.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:b19d25276b295f0e9bc8d80b", "occurrence_id": "a083162e1d350d375f5657da", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Incidence and loss.** Both the incidence rate and the accompanying agricultural loss share were computed from the Panel Protection Incident Survey of 2019, which interviews the same households across waves and records incidents alongside the plot each household works. The survey publishes incident counts by type; the quantities below are rates, which require the exposure the panel provides.\n\nIncidence was calculated as incidents per hundred households exposed for at least one full interval, and the loss share as the fraction of reported plots on which a loss was recorded in the same interval as an incident. Both use only households present in consecutive waves, since a household entering mid-panel has no exposure to contribute.\n\nThose two quantities carry the section's argument. The displacement dimension is what makes the pairing informative, because households displaced within the panel contribute exposure at two locations, and the plot record is what allows a loss to be attached to the place it occurred rather than to the household's current address. The survey's weighting was not applied, and the rates describe respondents rather than a national population.\n\nMention: \"Panel Protection Incident Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"An incident rate and a linked plot-loss share computed from the 2019 Panel Protection Incident Survey, using panel exposure rather than the survey's counts.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:652d90b9458fee7f2e26537e", "occurrence_id": "844256d532440934ae6915ee", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Heat and exposure.** The exposure distribution was derived from the Enumerative Working Conditions Survey of 2015, compiled by INEP, which records occupation, work setting and the hours spent outdoors alongside the temperature series for the enumeration district. The survey publishes shares by occupation; the distribution below is not one of them.\n\nEach respondent was assigned the seasonal maximum for their district and interval, and the resulting pairs were summarised as the share of outdoor workers above a threshold and the mean threshold exceedance per worker-week. Both were computed on the released district series; no interpolation was applied to districts without a station, and those respondents were excluded from the denominator rather than imputed.\n\nThat exclusion is the reason the distribution understates exposure, and it is stated here rather than in a footnote because the excluded districts are not a random subset. INEP's documentation of the enumeration and the occupational coding sit with the survey and were not reproduced. The section's argument concerns whether exposure can be measured at all from work records, and the distribution is the evidence brought to bear on it.\n\nMention: \"Enumerative Working Conditions Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"An exposure distribution derived by joining the 2015 INEP Enumerative Working Conditions Survey to district temperature series, excluding stations absent districts.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:226f638880bdb385c4a220b8", "occurrence_id": "4bb2547d1fcb67117363a969", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Movement recorded at registration.** Both quantities below were computed on the Combined Forced Population Survey of Papua New Guinea, Release 5, compiled by DESA. No reference year is attached to that release in the source material, and none is supplied here; the quantities describe the register as constituted by that release.\n\nThe first is the share of registered individuals whose recorded origin lies outside the province of registration; the second is the share whose registration was made more than ninety days after arrival. Both were derived from the release's own fields, and the second depends on an arrival date that is frequently a respondent's estimate rather than a documented date, which is noted rather than corrected.\n\nThe survey is the only source for either quantity and the paragraph rests on them. DESA's documentation of Release 5 describes how duplicates were resolved across provinces, and that treatment is carried through without restatement. A person registered at two locations appears once, at the location of enumeration, so the first quantity is a property of the enumeration design at least as much as of the movement itself.\n\nMention: \"Combined Forced Population Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"Two shares computed on Release 5 of the Combined Forced Population Survey of Papua New Guinea (DESA), with no reference year asserted.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:24c6e045f0698323feaa093c", "occurrence_id": "8fc5eec146cf755446a2d4e5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Input access.** Two figures are reproduced from the Diagnostic Irrigation and Inputs Registry for 2020, maintained by New Cronos, and both are given as the registry reports them. The first is the share of registered holdings with access to a formal irrigation source; the second is the share of those holdings that recorded an input purchase in the same season.\n\nThe registry is a database assembled from scheme records and dealer returns rather than from a sample, so neither figure carries a sampling error and neither should be read as an estimate of a wider population. The population is what the registry covers, and coverage is determined by scheme boundaries.\n\nThe registry is the paragraph's principal evidence and the only source placing both items on the same holding, which is what allows the two shares to be set beside one another. New Cronos documents the reconciliation between dealer returns and scheme rolls; those notes were used and are not restated. A holding that irrigates outside a scheme boundary is absent from the first figure entirely, and the second figure says nothing about inputs obtained through channels the registry does not observe.\n\nMention: \"Diagnostic Irrigation and Inputs Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Two published shares from the 2020 Diagnostic Irrigation and Inputs Registry (New Cronos), quoted as reported and paired within the registry's own coverage.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:df1a1a77d8425026be971e68", "occurrence_id": "06ec6437fb1cc9cfca8c8fb6", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Settlement distribution.** The spacing statistics were computed from the Stratified Settlement Location Index for 2022, an indicator that assigns each settlement to a density stratum and publishes neither the coordinates nor a spacing measure. Both quantities below are therefore properties of the index as constructed here, not published values.\n\nNearest-neighbour distance was calculated for each settlement within its stratum and summarised as the median distance and the ratio of that median to the stratum's expected spacing under a uniform arrangement. The expected spacing was derived from the stratum's own area and count, and no assumption was made about the arrangement within a stratum beyond that calculation.\n\nThose statistics carry the section's argument, which concerns whether stratification conceals a systematic difference in how settlements are arranged. The index's stratum definitions are its own and were adopted without modification; because the index does not release coordinates, the distances were computed on the locations available to this review, and a settlement absent from that file contributes nothing to either quantity.\n\nMention: \"Stratified Settlement Location Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"Nearest-neighbour spacing statistics computed on the 2022 Stratified Settlement Location Index, which publishes strata but no spacing measure.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:c67f0951f9ef9f292c8b4c4e", "occurrence_id": "81df3761ade530bf5ca3af35", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Coverage of antenatal contact.** Two figures are reproduced from the Midline Reproductive Health System (MRHS) for Florida in 2021, and neither was recalculated. The first is the share of registered pregnancies with at least one recorded contact in the first trimester; the second is the share with contact recorded in every trimester.\n\nThe system is administrative, generated by registration at the point of care rather than by collection, and the two figures therefore describe the records rather than a sample drawn from a population. They are quoted at the precision the system publishes. The geographic identifier is carried at county level, which matters because the second figure is reported for the state as a whole and the internal variation is not visible in it.\n\nThe system is the paragraph's principal source and the only one placing both items on the same record, which is what allows the two shares to be read as a single statement about continuity of contact rather than as two unrelated coverage numbers. Registration depends on the woman presenting, and a pregnancy that never enters care contributes to neither numerator nor denominator.\n\nMention: \"Midline Reproductive Health System (MRHS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Two coverage shares reproduced from the 2021 Midline Reproductive Health System (MRHS) for Florida, quoted as published and read as records rather than estimates.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:7bf461442d10d1a73fc07b40", "occurrence_id": "4a998faa31df9022706da064", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Baseline conditions.** The pair of figures below is quoted from the Baseline Service Catchments Survey (BSCS) of 2024 and reproduced without adjustment. The first is the share of sampled catchments with a functioning facility at the reference date; the second is the share in which the facility reported staffing below its own establishment.\n\nNeither figure was recomputed, and neither is used to characterize anything beyond the sampled catchments. The survey is a field collection with a published frame, and the figures are estimates from that frame at the precision the survey prints them.\n\nThese figures support one claim in a passage whose subject is the difficulty of establishing a baseline at all, and they are not that passage's conclusion. The survey's treatment of catchments whose facility was under construction at the reference date is the survey's own and was not revisited; those catchments appear in the first figure according to that rule, and the reader is referred to the instrument rather than to any restatement of it here.\n\nMention: \"Baseline Service Catchments Survey (BSCS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"Two figures quoted from the 2024 Baseline Service Catchments Survey (BSCS), used to support one claim in a passage about the difficulty of baselining.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:a545e946a5203b27f10f257b", "occurrence_id": "20b6599f8766c42762ebf9bf", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Wellbeing indicators.** The two shares below come from the Analytical Child Wellbeing Registry (ACWR) for 2016 and are reproduced as the registry releases them. The first is the share of registered children flagged on at least one indicator; the second is the share flagged on three or more.\n\nThe registry is a database of records held by several agencies and joined on a registration identifier, and it is not a survey. Its figures are counts of what appears in those records, so a child whose agency record is missing is absent from both the numerator and the denominator rather than counted as unaffected. No interval is attached to either share and none should be inferred.\n\nThese shares support a single claim about the concentration of flagged need, and the wider passage concerns how multi-agency records behave when they are joined. The registry's own note on join quality travels with it and is not repeated here; where the join failed the child was excluded rather than matched approximately, and that rule is the registry's rather than one adopted for this paragraph.\n\nMention: \"Analytical Child Wellbeing Registry (ACWR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"Two 2016 shares reproduced from the Analytical Child Wellbeing Registry (ACWR), quoted as released and used to support one claim about concentration.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:265260561478d3d35387dc17", "occurrence_id": "be89f0ceb7bb00ed0d8e358e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Training placement.** Two figures are reproduced from the Quality-Assured Vocational Training Survey compiled by the Growth Lab at Harvard, quoted as the survey publishes them rather than re-derived. The first is the share of completers placed in employment related to their course; the second is the share placed in any employment within six months of completion.\n\nThe survey is the paragraph's principal source and the only one reporting placement against course content, which is the pairing the first figure requires. Recomputation was avoided deliberately: the survey applies its own definition of relatedness, derived from course coding, and a definition applied here could not be reconciled with the published series.\n\nThe survey's field notes describe follow-up at six months and the treatment of respondents who could not be relocated; those notes were used and are not restated. Both figures are estimates from a sample and carry the attrition the follow-up incurred. No reference year is asserted, since the survey reports placements across several cohorts and the figures quoted here are the pooled pair as published.\n\nMention: \"Quality-Assured Vocational Training Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Two published placement shares from the Growth Lab at Harvard's Quality-Assured Vocational Training Survey, quoted as released because relatedness is the survey's own coding.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:15a6f3eda2e92c61842fa345", "occurrence_id": "2f71bacf3e8f30d2d226042f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Place name coverage.** The two counts below are taken from the Joint Place Name Registry as it stands, and are quoted without adjustment. The first is the number of names carried in the register; the second is the number for which a variant spelling is also recorded.\n\nThe registry is a database rather than a statistical release, and its counts describe the state of an editorial process rather than a population. Neither figure carries a reference date beyond that of the extract consulted, and neither was compared with any earlier extract, so nothing here is a rate of change.\n\nThese counts support one observation in a longer passage about the standardisation of geographic names, and they are not that passage's conclusion. The registry's rules for admitting a name, and for recording a variant rather than a separate entry, are the registry's own; they were adopted for these counts without restatement, and a place with several accepted names may appear once or several times according to those rules alone.\n\nMention: \"Joint Place Name Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"Two counts quoted from the Joint Place Name Registry as extracted, used to support one observation in a passage on geographic name standardisation.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:56362d687683542b4587ba80", "occurrence_id": "2842c5e97a195a8ed00e1955", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Retention within the year.** Continuation was estimated from the Monthly Enrolment and Retention System in Lebanon, an administrative series produced by the act of enrolment rather than by collection, and the estimate is not one the system publishes.\n\nMonthly enrolment counts were converted to a cohort basis by tracking the change in grade counts between the first and last month of the school year, and the resulting continuation ratios were reported by grade band. No demographic denominator was available, so the ratio is internal to the series: it compares the count of a grade at two dates rather than a cohort against a population.\n\nThat construction is the section's point rather than a caveat attached to it. The argument concerns what can be learned when the only repeated observation is an administrative count, and the grade-band ratios show both what such a series can settle and where it silently fails, since a cohort leaving the system entirely reduces two counts rather than appearing as an exit. Lebanon's reporting arrangements determine which schools enter the series, and the ratios describe those schools.\n\nMention: \"Monthly Enrolment and Retention System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"Grade-band continuation ratios constructed from Lebanon's Monthly Enrolment and Retention System, where no demographic denominator is available.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:ec3f8cd4836515285fc5ad4e", "occurrence_id": "7b03b7765c68ade18a134bbd", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Registration and first employment.** The interval between arrival and first recorded employment was computed from the Pilot Arrival and Registration Registry compiled by NCHS/CDC/WHO, a database in which each arrival carries a registration date and, where reported, an employment date. The registry publishes counts by status, not intervals.\n\nOnly arrivals with both dates recorded entered the calculation, and the resulting distribution is reported as a median with an interquartile range rather than a mean, since a long tail of late registrations dominates any average. The employment date is a reported date, not a documented one, and no adjustment was made for arrivals whose employment preceded registration in the record.\n\nThat distribution is the paragraph's substance. The registry's role documentation explains which entries carry an employment field at all, and the resulting selection is the reason the median describes registered arrivals rather than arrivals. Because the registry is a pilot, its coverage is a matter of where the pilot operated, and nothing here generalises beyond those locations.\n\nMention: \"Pilot Arrival and Registration Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"An arrival-to-employment interval distribution computed on the NCHS/CDC/WHO Pilot Arrival and Registration Registry, which publishes counts rather than intervals.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:fac1b2d068532c4b5546aef2", "occurrence_id": "40014b6118488e8f9cf77e08", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Built-up extent.** Within a longer treatment of how urban extent is measured, one comparison rests on the Stratified Built-Up Surface Dataset for 2018. The material is a geospatial surface, and what was derived from it is not the mapped extent itself but the agreement between that surface and a second source covering the same year.\n\nThe derived quantity is the proportion of built-up cells in the dataset that also appear as built in the second source, computed cell by cell on a common grid. No attempt was made to reconcile the two classifications where they disagree about a cell, and the disagreement rate is reported alongside the agreement rate rather than resolved.\n\nThat comparison supports one claim rather than carrying the passage. The passage concerns whether two independently produced surfaces can be treated as measuring the same object, and the agreement rate enters as evidence that they cannot be treated as interchangeable without knowing which classification a cell came from. The 2018 vintage is treated as a single observation, and no change over time is asserted from it.\n\nMention: \"Stratified Built-Up Surface Dataset\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"A cell-level agreement rate between the 2018 Stratified Built-Up Surface Dataset and a second source, supporting one claim about measurement agreement.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:d62d4d8fa052d878d4f8e964", "occurrence_id": "e18e7b5e093c2b3e586f1534", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Site administration.** Two figures are reproduced from the Representative Camp Management System (RCMS) for 2020 and are given as the system releases them. The first is the share of administered sites whose records were updated within the reporting month; the second is the share of sites whose population count was revised at least once during the year.\n\nThe system is administrative, produced by the reporting obligations of site managers rather than by enumeration, and both figures describe the completeness of that reporting rather than the conditions in the sites. They carry no intervals, since the sites are the population being described.\n\nThe system is the paragraph's principal source and the only source of either figure. Its own guidance defines what counts as an update, and that definition is carried through without restatement; a site whose record was amended without a population change is counted in the first figure and not the second. The second figure is a count of revisions rather than of error, and a site revised once and a site revised repeatedly appear identically.\n\nMention: \"Representative Camp Management System (RCMS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Two 2020 figures reproduced from the Representative Camp Management System (RCMS), quoted as published and read as reporting completeness rather than conditions.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:3360b5086b48d408ba0d8300", "occurrence_id": "c0ebfb4c0b99140805468f61", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Testing the published series.** The Monthly Labour Inspection Index (MLII) compiled by JSANDS is brought in here as an independent check on the inspection totals reported elsewhere in this section, and no figure from it is presented as a finding. The index is a constructed indicator, and its value for this purpose is that it is built from a different input set than the totals it is set against.\n\nThe comparison is limited to direction and order of magnitude across the months both series cover. Where the index and the reported totals move together, the totals are treated as corroborated; where they diverge, the divergence is noted and the totals are reported with that qualification rather than replaced by the index.\n\nThis is a supporting role and it is deliberately narrow. The index supplies no level, no regional detail and no interval, and none is claimed for it. JSANDS documents the inputs behind the index, and those notes were consulted to establish that the two series are not derived from the same returns, since a check drawn from the same source would establish nothing.\n\nMention: \"Monthly Labour Inspection Index (MLII)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"JSANDS Monthly Labour Inspection Index (MLII) used to corroborate the direction of separately reported inspection totals, contributing no figure of its own.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:e2f70abc3d7917201e4374de", "occurrence_id": "cf2341be9a14d5a76400cc93", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Checking the coverage claim.** The Departmental Collective Bargaining Registry (DCBR) for Indonesia in 2023 is used here to test a coverage figure reported by an unrelated survey, and nothing in the paragraph rests on the registry alone. The registry holds agreements; the figure it is set against comes from an establishment survey whose population is only partly the registry's own.\n\nThe test compares the number of establishments appearing in the registry's agreements with the number the survey reports as covered. The two counts are not reconciled and no ratio is presented as a correction: the registry count is reported beside the survey figure so that the reader can see the size of the difference between them.\n\nThat comparison supports one qualification in a longer passage. The registry's own rules for what constitutes a covered establishment differ from the survey's, and that difference is the likeliest explanation for the gap, but the paragraph does not attribute the gap to it. The 2023 extract is a snapshot, and an agreement in force at the extract date is counted once whatever its duration.\n\nMention: \"Departmental Collective Bargaining Registry (DCBR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"Indonesia's 2023 Departmental Collective Bargaining Registry (DCBR) used to test a survey-derived coverage figure, with the gap left unattributed.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:20160a3d03a7e3aa639138fd", "occurrence_id": "c0b2d28e18e5d6f472884fdd", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Corroborating the exposure estimate.** The Standardised Land Cover Survey (SLCS) of 2018, produced by the Department of Defense, is used here to corroborate an exposure estimate whose own source is unrelated to it, and it contributes no estimate of its own. The survey is a field collection with a geospatial component, and it carries health-relevant attributes recorded at the sampled locations.\n\nThe corroboration is confined to whether the two sources place the same areas in the same category. Where they agree, the exposure estimate is reported without qualification; where they disagree, the disagreement is reported and the estimate is qualified accordingly. No weighting was applied to reconcile the two, and no attempt was made to adjudicate which source is right about a given area.\n\nThis is a supporting use. The survey's own documentation of its sampling and of the classification it applies sit with the source and were not reproduced, and the corroboration establishes only that two independent observations of the same surface behave similarly, which is a weaker claim than agreement about exposure.\n\nMention: \"Standardised Land Cover Survey (SLCS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"2018 Standardised Land Cover Survey (SLCS) used to corroborate an exposure estimate from an unrelated source, contributing classification rather than an estimate.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:623361202b1b23b777f501fb", "occurrence_id": "cb3b750ff3b06b62a272677f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Housing and tenure.** The comparison of tenure security with income position was computed on the Comparative Housing Conditions Registry, a database that holds a housing record per household together with an income band. The registry publishes neither comparison, and neither quantity below appears in its tables.\n\nHouseholds were classified on the registry's tenure field and cross-tabulated against income band, and the two quantities reported are the share of insecure households in the lowest band and the difference in that share between the lowest and highest bands. No weights are attached to the registry, so the shares describe registered households rather than a population, and no interval is offered.\n\nThat difference is what the section argues from. The employment dimension enters because tenure in the registry is recorded against the household while income is recorded against the individual, and the cross-tabulation therefore mixes two units in a way the paragraph makes explicit rather than concealing. The registry's coverage follows the programmes it was built to administer, and households outside those programmes are absent from both figures.\n\nMention: \"Comparative Housing Conditions Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"A tenure-by-income comparison computed on the Comparative Housing Conditions Registry, which publishes neither the cross-tabulation nor the difference.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:6df07ca05b5335c996fa66c4", "occurrence_id": "e23d52b01a5c1ee0cdf65a80", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Reading the temperature series.** One comparison within the discussion of attribution rests on the Enumerative Temperature Records Index, an indicator assembled from station records by a published method. The index is not the subject of the paragraph and contributes no finding; it is used to test whether the trend described from station data survives a different construction.\n\nThe comparison is confined to sign and rough magnitude over the shared period. Where the index and the station-derived trend run in the same direction, the trend is stated without further qualification; where they diverge, the divergence is reported and the trend is presented as conditional on the construction that produced it.\n\nThat test is deliberately modest. The index is an indicator rather than a measurement, its method involves choices that are not visible in the released values, and no interval is attached to either series. The paragraph draws no conclusion about attribution from the agreement, only about whether the finding depends on one particular way of assembling the records.\n\nMention: \"Enumerative Temperature Records Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Enumerative Temperature Records Index used to test whether a station-derived trend survives a different construction, contributing no finding itself.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:c1e8a7a878e9908531928dfb", "occurrence_id": "f71c366f0c2be5e60cd5ddb2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Assets and poverty.** The asset index and its association with the poverty flag were constructed from the Urban Asset Ownership Microdata (UAOM) for 2018, compiled by Bureau Van Dijk, rather than taken from any published summary. The material is microdata, released at household level with no weights attached, and the published documentation reports item frequencies rather than an index.\n\nA first component was extracted from the ownership items and the resulting score cut at the national median, and the two quantities reported are the share below that cut flagged as poor and the share of households owning no durable asset in the same group. Both were computed on the released rows with no imputation for missing items, which removes from the score any household whose item set is incomplete.\n\nThat construction carries the paragraph. The removal of incomplete households is disclosed rather than corrected because the items most often missing are those held by the poorest households, so the score is likely to understate the association it is used to demonstrate. Bureau Van Dijk's documentation of the 2018 fieldwork was consulted and not reproduced, and the absence of weights means both shares describe respondents.\n\nMention: \"Urban Asset Ownership Microdata (UAOM)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"An asset index and its poverty association constructed from the 2018 Urban Asset Ownership Microdata (UAOM), with the effect of the missing-item rule disclosed.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:0edd75adfe6bede9c439f96e", "occurrence_id": "985132cb6394ee324b835d94", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Return and displacement.** The monthly series of returns was constructed from the Monthly Return and Relocation Index for 2019, an indicator that reports a composite movement value rather than counts of people. Nothing in this paragraph is published by the index; the series was built from its released values by a transformation described here.\n\nThe composite was differenced month to month and the resulting changes were classified as increases, decreases or no change, and the reported quantities are the proportion of months in each class and the longest run of consecutive increases. No interval accompanies either quantity, and the classification treats the composite as exact, which it is not.\n\nThat treatment is the paragraph's point rather than an incidental choice. The argument concerns how much information a composite retains about the underlying flows, and the run length is brought forward because a composite can move steadily while the flows behind it alternate. The index's own construction notes describe the inputs and their weights, and those notes were used and not restated.\n\nMention: \"Monthly Return and Relocation Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"A month-to-month series of change classes constructed by differencing the 2019 Monthly Return and Relocation Index composite, which publishes no counts.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:2ddba6ddb8cbc4d133ddf0c9", "occurrence_id": "5ae81b3e111e8baa7ebfb445", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Reporting delay in surveillance.** The delay distribution was constructed from the Quality-Assured Disease Surveillance Survey (QDSS) compiled by PWT, which records for each reported episode both the onset date and the date the record entered the system. The survey publishes completeness rates, not the interval between those two dates.\n\nEpisodes were retained where both dates were present and internally consistent, and the interval was summarised by median and by the share exceeding seven days. Because the survey carries a location for each episode, the same two quantities were computed separately for the two largest reported areas, which is where the distributions part: the medians differ modestly while the tail differs considerably.\n\nThose distributions carry the paragraph. The argument concerns whether a surveillance series can be treated as current when its own reporting lag is unmeasured, and the tail is the relevant quantity, since it is the late episodes that are missing from a series at the moment it is used. PWT's documentation of the survey's quality procedures sits with the source and was not reproduced, and an episode with an implausible date pair was dropped rather than corrected.\n\nMention: \"Quality-Assured Disease Surveillance Survey (QDSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"A reporting-delay distribution and its area split built from the Quality-Assured Disease Surveillance Survey (QDSS), which publishes completeness rather than lag.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:2ace011ec37aa4cda8d72c7d", "occurrence_id": "86d4b291a268ef3b01b51741", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Indebtedness and buffers.** Two figures are reproduced from the Independent Debt and Savings Index for Singapore in 2022, and both are given as the index releases them. The first is the reported share of households carrying any debt service obligation; the second is the reported share holding a buffer sufficient to cover three months of that obligation.\n\nThe index is a constructed indicator rather than a survey, and its figures rest on inputs of several kinds under a weighting its documentation describes. They carry no intervals, and the absence of intervals follows from the construction: the object is a composite value, not an estimate produced by a sampling design.\n\nThe index is the paragraph's principal source and the only one expressing both terms on a common basis, which is what allows the second figure to be read as a shortfall against the first. The definitions of obligation and of buffer are the index's own and were adopted without restatement or conversion. No claim is made here about the level of household debt, only about the relationship between two quantities the index publishes together.\n\nMention: \"Independent Debt and Savings Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Two 2022 figures reproduced from the Independent Debt and Savings Index for Singapore, quoted as released and read against each other as a shortfall.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:58d4be43d354d5dbe8aaf94f", "occurrence_id": "338cc5c391cc18151c050c4d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Preparing the wage register.** The work described here concerns the 2023 refresh of the Multi-Topic Wage Structure Registry, which is held as a database of establishment returns, and the preparation performed on it rather than any quantity derived afterwards. Three operations were carried out and are documented so that the refreshed register can be audited.\n\nOccupation codes were mapped to a single scheme across the returns, since establishments filed under two generations of coding; duplicate establishment records created when a firm re-registered mid-year were resolved to the later record; and the reporting floor was applied uniformly, which had previously been enforced only at the regional offices. No wage value was altered, and no record was imputed.\n\nThat register preparation is the subject of the paragraph rather than a preliminary to it. The claim is that the apparent rise in coverage between refreshes is an artefact of the floor being applied centrally, and the three operations are the evidence for that, since the second alone accounts for most of the difference. The refreshed register was lodged with the standing committee that governs the series.\n\nMention: \"Multi-Topic Wage Structure Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"Three documented preparation operations on the 2023 Multi-Topic Wage Structure Registry refresh, presented as the paragraph's subject rather than a preliminary.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:906153c5add5b85de1014f02", "occurrence_id": "7bfc5ae8eed3ee467ddca5ea", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Chain participation.** The two shares below are quoted from the Annual Value Chain Registry (AVCR) as it publishes them, without recomputation. The first is the share of registered producers linked to a buyer under a written contract; the second is the share whose recorded output moved to a processor rather than to a market.\n\nThe registry is a database assembled from returns rather than a sample, so neither figure carries a sampling error and neither describes a wider population than the registry covers. Both are reproduced at the precision the registry uses.\n\nThese shares support a single claim in a passage concerned with how producers enter formal chains, and they are not that passage's conclusion. The registry's rules for what counts as a link, and for whether a producer supplying two buyers is registered once or twice, are the registry's own and were adopted for these figures without restatement. A producer whose buyer did not file is absent from the first figure, and the second says nothing about transactions the registry does not observe.\n\nMention: \"Annual Value Chain Registry (AVCR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"Two shares quoted from the Annual Value Chain Registry (AVCR) as published, used to support one claim in a passage on entry to formal value chains.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:d0952a7fa51a3bf009b6fb5b", "occurrence_id": "329fbaf02db35c9ad9096081", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Household registration.** Three figures are reproduced from the Municipal Health and Household Registry (MHHR) maintained for Ukraine in 2020 by ProGres, and none was recalculated. The first is the count of registered households; the second is the share with at least one health record attached; the third is the share with a record attached for every member.\n\nThe registry is a database created by service delivery, and all three quantities describe the completeness of that creation rather than the condition of the households. No interval accompanies the two shares, since the households are the population being described rather than a sample of it.\n\nThe registry carries the paragraph, and the third figure is the one the argument depends on, since a household counted in the second may have only one of its members recorded. ProGres documents the conditions under which a record is opened, and those conditions were not restated here; they determine the denominator of all three figures.\n\nMention: \"Municipal Health and Household Registry (MHHR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Three figures reproduced from the 2020 ProGres Municipal Health and Household Registry (MHHR) for Ukraine, read as completeness of registration.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:4627a83a037a875ec19bf4a7", "occurrence_id": "d775120a15613fdd525403a1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Inventory consistency.** The emissions trajectory was estimated from the Decennial Emission Inventory Index (DEII) for 2019, compiled by WFP, rather than reported from the index's published level. The index is an indicator: it releases a level for each period, and it does not publish the change between periods or its uncertainty.\n\nChange was computed between the released levels and expressed per annum, and the estimate was then compared with the trajectory implied by the activity data the index's own note lists as inputs. Both computations rest on released values; no input series was obtained separately and no reweighting was applied.\n\nThat comparison is the section's contribution. The surrounding discussion of inventory methodology is there to explain why a level and a trajectory can disagree, and the estimate addresses that directly by showing the two need not. WFP's documentation of the 2019 release describes the revisions folded into earlier periods, and because those revisions are inside the released levels, the computed change is a change between revised levels rather than between the levels as first published.\n\nMention: \"Decennial Emission Inventory Index (DEII)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"An annual change estimate computed from the 2019 Decennial Emission Inventory Index (DEII) levels and compared with the trajectory its inputs imply.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:03713760c94e937c9a1c6028", "occurrence_id": "bb589ec1d4817cbf13d2e2e3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Household energy costs.** The pair of shares below is reproduced from the Cross-Sectoral Fuel and Energy Use Survey (CFEUS) of 2022 and is given as the survey prints it. The first is the share of households reporting solid fuel as their primary cooking source; the second is the share whose reported fuel spending exceeded a tenth of total recorded expenditure.\n\nNeither figure was recomputed and neither is used outside the survey's own frame, which is a sample with a published design. The survey's notes on the treatment of households that purchase no fuel, because they collect it, are the survey's and were not revisited; those households enter the second figure at zero expenditure.\n\nThese shares support one claim in a longer passage on the cost of cooking, and they are not the passage's finding. The comparison between them is presented as descriptive, since the survey does not publish a joint estimate of the two quantities and no interval was carried through when they were placed side by side here.\n\nMention: \"Cross-Sectoral Fuel and Energy Use Survey (CFEUS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"Two 2022 shares reproduced from the Cross-Sectoral Fuel and Energy Use Survey (CFEUS), placed side by side as a descriptive comparison.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:a48f3b79ecfb1551cd3c7ce3", "occurrence_id": "818291526e5b3828acc8d08d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Entry and survival.** Two figures are reproduced from the Continuous Business Demography Registry for 2015, maintained by the IEA, and are quoted as the registry releases them. The first is the count of establishments newly registered during the year; the second is the share of the stock that had been registered for less than two years at the reference date.\n\nThe registry is a database of administrative events, and the two figures are properties of that record rather than estimates of business formation in the economy. They carry no intervals and no seasonally adjusted variant, and the second depends on the first only through the record, since an establishment that registered and deregistered within the year appears in neither.\n\nThe registry is the paragraph's principal source and the only source of either quantity, which is why the surrounding discussion of business demographics is treated as context rather than as corroboration. The IEA's notes on which registration events enter the record are carried with the source and are not restated, and they place a floor on what the first figure can be read as describing.\n\nMention: \"Continuous Business Demography Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Two 2015 figures reproduced from the IEA's Continuous Business Demography Registry, quoted as released and read as properties of an administrative record.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:ce75120db5a9e19f2d1dd118", "occurrence_id": "62bea1e536246ab9dd61e98d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Agricultural holdings and poverty.** The relationship between holding size and poverty status was estimated from the Clustered Socioeconomic Survey compiled by ALCS, in which households were sampled in clusters and the agricultural items were asked only of those reporting control of a holding. The survey publishes poverty status by region; it does not publish it by holding size.\n\nHouseholds reporting a holding were grouped into size bands and the poverty share computed for each, with the cluster structure respected in the calculation. The estimate is reported without a model and without weights beyond those the survey supplies, and the two smallest bands contain few clusters, which is stated rather than corrected.\n\nThat gradient is the section's subject. ALCS's documentation of the clustering and of the eligibility rules behind the agricultural module sits with the survey and was not reproduced, and it is the reason the two quantities cannot be read as a national relationship between land and poverty: households without a holding are outside the comparison entirely, and in several regions they are the majority.\n\nMention: \"Clustered Socioeconomic Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"A poverty gradient by holding size estimated from the ALCS Clustered Socioeconomic Survey, respecting its cluster structure and module eligibility.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:bf681fb7e145281839813602", "occurrence_id": "a8d62bd81c9b5a2a6e73406f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Extent of change.** Two quantities are reproduced from the Provincial Land Cover Change Dataset for 2020, a geospatial product in which each province carries an area by class and a net change against the previous vintage. Both are given as the dataset publishes them, at the precision and rounding it uses.\n\nThe dataset is a classification rather than a measurement, and its change figures are differences between two classifications of the same surface. That distinction matters for how the second quantity should be read: a province whose classification method changed between vintages can show a net change with no corresponding change on the ground, and the dataset does not indicate which provinces those are.\n\nThe dataset is the paragraph's principal evidence and the only source expressing change on a common class scheme. The first quantity is reproduced to give the second a denominator, and no claim is made about the accuracy of either. Where the dataset's own accuracy statement covers a class, that statement was used and is not restated here.\n\nMention: \"Provincial Land Cover Change Dataset\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Two quantities reproduced from the 2020 Provincial Land Cover Change Dataset, with the change figure read as a difference between classifications.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:8a08b93c345e63142512f16e", "occurrence_id": "b390306380b04c35a576f98d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Testing the capacity figure.** The Nationwide Diagnostic Capacity Registry (NDCR) is used here to check a capacity figure reported from a survey whose frame is not the registry's, and no quantity from the registry is offered as a finding in its own right. The registry is a database of facilities and their stated capabilities, updated as facilities report.\n\nThe comparison sets the survey's reported proportion of districts with a capacity against the proportion of districts in which the registry records at least one facility declaring it. The two proportions are reported side by side and neither is adjusted to the other, and no ratio between them is presented as a corrected figure.\n\nThis is a supporting use and it is narrow. The registry's declaration rules determine what its count means, and those rules differ from the survey's notion of availability, which is why the gap is reported rather than resolved. Nothing in the paragraph attributes the difference to either source, and the registry contributes no level, no interval and no district detail beyond what the comparison requires.\n\nMention: \"Nationwide Diagnostic Capacity Registry (NDCR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"Nationwide Diagnostic Capacity Registry (NDCR) used to check a survey-reported capacity proportion, with the gap reported rather than resolved.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:6211b357bae7cce8b26966f4", "occurrence_id": "66d38a1610f9c7d7b7f618ed", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Nighttime radiance.** Two figures are reproduced from the Panel Nighttime Lights Registry (PNLR) for 2024 and are quoted without adjustment. The first is the mean radiance recorded for the settled extent in the reference month; the second is the share of that extent registering a value indistinguishable from the background threshold.\n\nThe registry is a database of derived radiance values, and both figures describe the sensor's output over an extent defined by the registry itself. Neither is a measure of economic activity, and the second is at least as much a statement about the threshold as about the places it selects.\n\nThe registry is the paragraph's principal source and the only source placing both quantities on the same extent, which is what allows the second to be read as a limit on the first. The panel structure means a location may be observed more than once in the year, and the figures reproduce the registry's own convention of taking the last observation, a rule the registry documents and which is not revisited here.\n\nMention: \"Panel Nighttime Lights Registry (PNLR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Two 2024 figures reproduced from the Panel Nighttime Lights Registry (PNLR), the second read as a limit on the interpretation of the first.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:5403603c45c53fec6cd13680", "occurrence_id": "5317128b4c32d48892519a5b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Registration of applicants.** Two figures are reproduced from the Diagnostic Asylum and Protection System (DAPS) for 2016 and are given as the system releases them. The first is the count of applications registered during the year; the second is the share of those applications whose registration occurred more than thirty days after the reported arrival.\n\nThe system is administrative, produced by the statutory act of registration, and both quantities describe that act rather than the underlying flow of arrivals. No interval accompanies the second figure, and it depends on an arrival date that is in many cases a declared date rather than a documented one.\n\nThe system is the paragraph's principal source and the only source of either figure. Its own notes describe how a withdrawn or re-lodged application is counted, and that treatment is carried through without restatement; because a re-lodged application may take a new reference number, the first figure counts registration events rather than persons, and the paragraph says so where the distinction affects the reading.\n\nMention: \"Diagnostic Asylum and Protection System (DAPS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Two 2016 figures reproduced from the Diagnostic Asylum and Protection System (DAPS), read as properties of the registration act rather than of arrivals.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:73191727a48ee7030d4f731a", "occurrence_id": "ffd2c11147d42a0d5be0a1ad", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Inspection coverage.** Three figures are reproduced from the Independent Labour Inspection System (ILIS) maintained for Botswana in 2022 by MEF, and none was recalculated. The first is the count of establishments within the system's mandate; the second is the count inspected at least once in the year; the third is the resulting coverage ratio.\n\nThe system is administrative, and the ratio in the third figure is arithmetic on two administrative counts rather than an estimate from a sample. It carries no interval, and its denominator is the mandate rather than the population of establishments, which is the more consequential limitation than its precision.\n\nThe system carries the paragraph. MEF's documentation of what brings an establishment within the mandate is the reason the first figure cannot be read as a count of establishments, and that documentation sits with the source and was not reproduced here. The second figure counts establishments rather than visits, so a site inspected repeatedly and a site inspected once contribute identically.\n\nMention: \"Independent Labour Inspection System (ILIS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Three 2022 figures reproduced from the MEF Independent Labour Inspection System (ILIS) in Botswana, with the coverage ratio read as arithmetic on counts.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:02e92c08a22bf3d2aaa6e58f", "occurrence_id": "9b8ff777a0b9dfad7adeedec", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Hardship records.** The two shares below are reproduced from the Standard Deprivation and Hardship Registry for 2015 and are given as the registry prints them. The first is the share of recorded households flagged on any hardship item; the second is the share flagged on an item that had also been flagged in the previous year's record.\n\nThe registry is a database rather than a survey, and its figures are counts of flags held in records whose creation is an administrative act. No interval accompanies either share, and the second depends on the continuity of the record rather than on the continuity of the household's circumstances.\n\nThese shares support a single claim within a longer treatment of how hardship is recorded, and they are not that treatment's conclusion. The registry's rules for carrying a flag forward are the registry's own, were applied as published, and are not restated here; a household that re-registers with a new identifier leaves the second figure entirely.\n\nMention: \"Standard Deprivation and Hardship Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"Two 2015 shares reproduced from the Standard Deprivation and Hardship Registry, supporting one claim in a treatment of how hardship is recorded.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:cdb9339f74fd3756eb1c5a56", "occurrence_id": "8479da898bba5164e827f19c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Facility functionality.** The functionality measures were computed from the Pilot Facilities and WASH Registry (PFWR) for 2021, a database in which each facility carries observations on water, sanitation and power together with the date of each observation. The registry publishes counts of facilities; it does not publish a functionality measure.\n\nA facility was counted as functional only where all three observations were present and positive, and the resulting share was computed by facility type. The stricter rule was chosen deliberately over a count of facilities passing any single test, because the three observations were collected on different dates for many facilities and the looser rule would have combined them without acknowledging that.\n\nThat construction carries the section. The argument concerns how much a facility register can support about service reliability, and the sensitivity of the share to the rule is the evidence brought to bear: relaxing it raises the share considerably, and the difference is composed almost entirely of facilities whose water observation is missing rather than negative. The pilot's footprint bounds which facilities appear at all, and the share describes those.\n\nMention: \"Pilot Facilities and WASH Registry (PFWR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"A functionality share computed on the 2021 Pilot Facilities and WASH Registry (PFWR) under a strict all-three rule, with its sensitivity reported.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:0c5fb06caaeced0921426ed2", "occurrence_id": "abf65b778fd30c53bb1ba8be", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Tenure records.** Two figures are reproduced from the Decennial Land Use and Tenure Registry for 2024 and are quoted as the registry releases them. The first is the share of registered parcels recorded with a documented tenure instrument; the second is the share of those parcels whose instrument is registered rather than merely held.\n\nThe registry is a database of parcels and instruments, and both figures describe the documentation of tenure rather than its security or its distribution. They carry no intervals, since the parcels are the population described, and the second is bounded above by the first as a matter of construction.\n\nThe registry is the paragraph's principal source and the only source expressing both terms on the same parcel record, which is what allows the pair to be read as a statement about registration rather than about documentation. The registry's classification of parcels by agro-ecological zone was used only to report the figures by zone and plays no part in their computation.\n\nMention: \"Decennial Land Use and Tenure Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Two 2024 shares reproduced from the Decennial Land Use and Tenure Registry, read as a statement about registration rather than about documentation.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:548b2f42bd9190c61a5b879f", "occurrence_id": "c8a88fb75aad8363b1451b89", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **An institutional note.** The Enumerative Skills and Employment Registry (ESER) of Bihar for 2017, compiled by the Kazakhstan Ministry of Transport, is named here for the arrangement it illustrates and no value from it is used anywhere in this section. It is an administrative registry whose observations arise from the record-keeping obligations of the entities that report to it.\n\nThe reason for the mention is that pairing itself: a registry of a state's labour market compiled by a transport ministry of another country is what a development programme looks like when its reporting lines are followed to the end, and the paragraph concerns exactly that pattern of institutional authorship. The registry's documentation was consulted to confirm the arrangement and for nothing else.\n\nNo rate, count or share from ESER appears in this section, and none of the figures reported elsewhere rests on it. The registry supports the framing of the argument rather than its findings, and a reader should not treat the mention as evidence for any of the comparisons drawn here.\n\nMention: \"Enumerative Skills and Employment Registry (ESER)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"Enumerative Skills and Employment Registry (ESER) named for its institutional arrangement, with no values taken and no findings resting on it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:ff3d2c5032005ffdae027a1d", "occurrence_id": "a9bb24a565660721ab92bfe2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Programme coverage.** The two counts below are taken from the Joint School Feeding Registry as it currently stands and are reproduced without adjustment. The first is the number of schools with an active feeding record; the second is the number of those schools whose record includes an enrolment figure for the same period.\n\nThe registry is a database compiled from programme returns rather than a statistical collection, and the counts describe the state of reporting rather than the reach of the programme. No reference date beyond the extract consulted is attached to either figure, and the two are therefore not a rate.\n\nThese counts support one observation in a longer passage about the thinness of feeding data, and they are not that passage's conclusion. The registry's rule for treating a school that reports twice in a period, and its handling of schools that share a site, are the registry's own and were adopted without restatement, so the first count is a count of records as the registry understands them.\n\nMention: \"Joint School Feeding Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"Two counts reproduced from the Joint School Feeding Registry as extracted, supporting one observation about the thinness of school feeding data.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:6626b4e776c9aca168cf2eaf", "occurrence_id": "2b2593fe301ebb66ef60852d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Checking the prevalence estimate.** The Annual Health and Household Survey (AHHS) is used here to check a prevalence figure carried from an administrative source, and the survey contributes no prevalence estimate of its own to this paragraph. The survey is a field collection with its own frame and questionnaire, and its sampling design is unrelated to the administrative register.\n\nThe check is confined to whether the two sources place the population in the same range for the same year. Where they do, the administrative figure is reported without further qualification; where they do not, the difference is stated and the administrative figure is presented as conditional on its own ascertainment. No pooling was attempted and no weight was applied to reconcile the two.\n\nThis is a supporting use. The survey's design report describes its frame, its response rate and the mode effects its authors note, and those were consulted to establish that the two sources are genuinely independent, since a check resting on the same ascertainment would settle nothing. The survey's own estimates are reported elsewhere and are not recomputed here.\n\nMention: \"Annual Health and Household Survey (AHHS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"Annual Health and Household Survey (AHHS) used to check whether an administrative prevalence figure falls in the same range, contributing no estimate.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:ff62f1fbf88b3e8d97131931", "occurrence_id": "d654920e2916a16a58c880e1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Parcel coverage.** The two extents were derived from the Recurrent Parcel Footprint Dataset for 2024, a geospatial product releasing a polygon for each registered parcel. The dataset publishes the polygons and a count per administrative unit; it publishes no coverage measure, and neither quantity below is one of its values.\n\nThe first quantity is the share of the settled extent falling inside at least one parcel polygon; the second is the share of parcels whose area falls below the minimum the dataset's own note applies when generalising. Both were computed on the released geometry, and no parcel was removed for irregular shape.\n\nThose two quantities carry the section. The argument concerns whether a cadastre can serve as a denominator for urban analysis, and the first quantity bounds what any such calculation can describe while the second shows how much of the register is at a resolution where area is not a measurement. The dataset is a single vintage, so nothing here is a change, and a parcel outside the settled extent contributes to neither figure.\n\nMention: \"Recurrent Parcel Footprint Dataset\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"Two coverage extents derived from the 2024 Recurrent Parcel Footprint Dataset, which releases polygons and counts but no coverage measure.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:95ce02f8a778827e09440f02", "occurrence_id": "5810b1d35490c574494ab89a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Checking the transfer figure.** The Government Cash Transfer Registry Survey is used here to test a coverage figure that reaches this paragraph from an administrative source, and the survey contributes no coverage figure of its own. It is a field collection with its own frame, built independently of the register whose figure it is set against.\n\nThe test asks whether the two sources place recipient households in the same range for the same population. Where they agree, the administrative figure is reported without qualification; where they differ, the difference is stated and the administrative figure is presented as conditional on the register's own recording practice. Nothing was pooled and no adjustment was applied to either source.\n\nThis is a supporting role, and its value depends entirely on independence. The survey's frame documentation was consulted to confirm that it was not drawn from the register, since a survey sampled from a register can confirm the register's arithmetic but not its coverage. The survey's estimates are reported in its own publications and were not recomputed here.\n\nMention: \"Government Cash Transfer Registry Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"Government Cash Transfer Registry Survey used to test whether an administrative coverage figure holds, contributing no figure of its own.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:32e9e929e3aeee64610b05a9", "occurrence_id": "343a9865c5e90ac7899b7b54", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Wage dispersion.** Two figures are reproduced from the Recurrent Wage Structure Survey as the survey releases them, and neither was recomputed. The first is the ratio of the ninetieth to the tenth percentile of reported monthly pay; the second is the share of employees whose pay is set by a negotiated agreement rather than by the establishment.\n\nBoth are survey estimates carrying the design's precision, and the survey's own notes on the treatment of part-time and seasonal pay are the survey's and are not restated here. The ratio is reported to the precision the survey prints, and no transformation was applied to it.\n\nThe survey is the paragraph's principal source and the only source reporting both quantities on the same respondents, which is what allows the second to be read as a partial explanation of the first. Whether it does so is not claimed here: the relationship between negotiated pay and dispersion is discussed, and the survey supplies both terms of that discussion without settling it.\n\nMention: \"Recurrent Wage Structure Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Two figures reproduced from the Recurrent Wage Structure Survey as published, the second offered as a partial explanation of the first.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:fe8c9b4a715d11b5f19293ab", "occurrence_id": "c14ace582311f2c0b6767bac", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Urban employment records.** Three figures are reproduced from the Urban Skills and Employment System maintained for Bolivia in 2024 by REACH, and none was recalculated. The first is the count of registrations held; the second is the share carrying a recorded qualification; the third is the share of those whose registration was matched to an employer record.\n\nThe system is administrative, produced by the reporting duty of employers and by registration at the point of hiring, and all three figures describe that production rather than the labour market behind it. They carry no intervals, since the population is the register.\n\nThe system carries the paragraph. The third figure is the one the argument rests on, because the comparison between qualifications and hiring outcomes is only possible for the matched subset, and that subset is a minority of the register. REACH documents the matching rule and the conditions under which a match is refused; those notes sit with the source and were not reproduced, and they determine what the third figure can mean.\n\nMention: \"Urban Skills and Employment System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Three 2024 figures reproduced from the REACH Urban Skills and Employment System in Bolivia, with the matched subset identified as the limit on the argument.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:463d6597ae34aca709835b7a", "occurrence_id": "e583e40e3a6ce40d40965387", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **A contextual reference.** The Stratified Enterprise and Jobs Survey conducted in Mongolia in 2024 is named in this paragraph for the design choice it represents, and no value from it is reported here or used elsewhere in the section. It is a survey, and its stratification is the reason it is mentioned.\n\nThe paragraph concerns how enterprise surveys choose their strata and what that choice does to the resulting estimates of job creation. The Mongolian instrument is offered as a worked instance of stratification by establishment size rather than by sector, and its documentation was consulted to confirm the design and nothing further.\n\nThis reference supports the framing rather than the findings. No rate, count or share from the survey appears in this section, and none of the comparisons drawn elsewhere rests on it. A reader should not infer from the mention that the estimates in this chapter share its design, since several of them use the opposite stratification.\n\nMention: \"Stratified Enterprise and Jobs Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"Stratified Enterprise and Jobs Survey in Mongolia named for its stratification design, with no values reported and no findings resting on it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:668d85a17fcb7c705963252d", "occurrence_id": "dc362402890b48dc2e064042", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Verifying the farm income figure.** The Midline Farm Economy Survey of 2016, compiled by Twitter, serves here as the primary check on an income figure reaching this section from tax records, and the survey's own income estimates are not presented as findings. The survey is a field collection whose frame is a list of holdings, which is unrelated to the list of taxpayers the other figure rests on.\n\nThe check compares the two sources on the same question: what proportion of holdings report income below the threshold used in the tax series. Where the proportions fall in the same range, the tax figure is reported unqualified; where they do not, the divergence is stated and the tax figure is presented as describing taxpayers rather than holders. No reconciliation was attempted between the two populations, and no ratio between their figures is offered as a correction.\n\nThat the two frames are genuinely separate is the whole value of the check, and the survey's frame documentation was consulted to establish it. A check drawn from the same administrative list would confirm arithmetic while telling the reader nothing about who is missing from it.\n\nMention: \"Midline Farm Economy Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"2016 Midline Farm Economy Survey used as the primary check on a tax-derived farm income figure, its separate frame being the point of the check.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:983853738da31164440ce2eb", "occurrence_id": "d5d9e9bfbb6e5e5926f42813", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Situation records.** The two counts below are reproduced from the Baseline Protracted Situation Registry for Sulaymaniya as it stands, and neither was re-derived. The first is the number of households carried in the registry with a protracted status; the second is the number whose status was reviewed within the preceding twelve months.\n\nThe registry is a database maintained for operational purposes, and the two counts describe the maintenance of records rather than the circumstances of the households. No reference date beyond the extract consulted is attached to either, and the second is a count of reviews rather than of decisions.\n\nThese counts support one claim in a passage concerned with how provisional a baseline can be when the status it records is itself reviewed. The registry's rules for retaining a household after departure, and for reopening a record, are the registry's own and were adopted without restatement. A household that left the area without its record being closed remains in the first count.\n\nMention: \"Baseline Protracted Situation Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"Two counts reproduced from the Baseline Protracted Situation Registry for Sulaymaniya, supporting one claim about how provisional a baseline can be.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:392bc597a2028bc7d9b60ae4", "occurrence_id": "aa91a4e5801a4b8fc3972ab3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Archiving an indicator series.** The work reported here concerns the 2015 Endline Collective Bargaining Index compiled by the Tax Justice Network, and the curation performed on it rather than any quantity drawn from it. The index is an indicator, and the released series had been distributed as a spreadsheet whose headers were not stable across vintables.\n\nFour operations were carried out. The series was converted to a flat file with fixed field names; the coverage note that had circulated separately was attached to the file; the definition of the reference period was recorded as metadata rather than inferred from the columns; and the resulting package was lodged with a repository under a citable identifier. No value was altered, and no gap was filled.\n\nThat work supports one claim in a longer treatment of indicator preservation. The claim is that the series was previously unusable for comparison because its own definitions travelled separately from it, and the four operations are the evidence for that, since they changed nothing except what a user could see. The index's construction remains the Tax Justice Network's own.\n\nMention: \"Endline Collective Bargaining Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"Four documented curation operations on the 2015 Tax Justice Network Endline Collective Bargaining Index, reported as preservation work rather than analysis.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:a36a5d11bc62d7217874a9d0", "occurrence_id": "712d0db9df8e35de42b9c567", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Expenditure records.** Two figures are reproduced from the High-Resolution Household Income and Expenditure Registry for the United States in 2017 and are given as the registry releases them. The first is the share of recorded households filing expenditure in more than one category; the second is the share whose recorded expenditure falls below the registry's own subsistence line.\n\nThe registry is a database assembled from administrative and panel inputs rather than a survey of a frame, so neither figure carries a sampling error and neither should be read as an estimate of the national population. Coverage is a property of the inputs.\n\nThe registry is the paragraph's principal source and the only source expressing both quantities on the same household record, which is what allows the second to be read within the subset defined by the first. The subsistence line is the registry's own, applied as published and not converted to any external threshold. Where the registry's inputs disagree for a household, the registry's precedence rule was carried through without restatement.\n\nMention: \"High-Resolution Household Income and Expenditure Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"Two 2017 figures reproduced from the High-Resolution Household Income and Expenditure Registry for the United States, the second read within the first's subset.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:93c99d109f8ff9a590d69be4", "occurrence_id": "0ccba1ca5a037ba645c89232", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Provision in non-state schooling.** The supply estimates were computed from the Regional Non-State Education Survey (RNES) of 2023, which enumerates institutions outside the public system and records staffing, fees and the qualifications of teachers. The survey publishes counts by region; the quantities below are rates, which require the denominators the survey supplies but does not compute.\n\nTeacher-per-enrolment ratios were computed for each region and reported alongside the share of institutions charging fees, with both quantities restricted to institutions that answered both the staffing and enrolment items. That restriction removes a substantial minority of institutions, and the removed group is not a random one: unpriced institutions are over-represented among those that declined the fee item.\n\nThose estimates carry the section. The employment dimension enters because qualifications were recorded only for teachers in fee-charging institutions, so the two estimates cannot be combined into a single statement about staff quality, and the paragraph says so rather than averaging them. No weights beyond the survey's own were applied, and the estimates describe enumerated institutions.\n\nMention: \"Regional Non-State Education Survey (RNES)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"Teacher-ratio and fee-sharing estimates computed from the 2023 Regional Non-State Education Survey (RNES), with the item non-response pattern reported.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:2d31e7b6724290c46cdd7c37", "occurrence_id": "f8758c319eb450105c5b0609", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Debt in the census frame.** Within a longer treatment of what a census can carry, one comparison rests on the Record-Based Debt and Savings Census of 2023 compiled by Open Data Watch. The census enumerates from records rather than by interview, and what was derived from it is not its published count but the agreement between that count and a survey estimate for the same population.\n\nThe derived quantity is the ratio of the census count to the survey estimate, reported by region and summarised as the share of regions in which the two differ by more than a tenth. No adjustment was applied to either figure, and the direction of the difference is reported rather than resolved, since a census of records and a survey of households need not be measuring the same object.\n\nThat comparison supports one claim rather than carrying the passage. The passage concerns whether record-based enumeration can substitute for a survey, and the share of divergent regions enters as evidence that substitution is defensible only where the record's coverage is known. The geographic detail is what makes the comparison possible, since the national figures agree closely while the regional ones do not.\n\nMention: \"Record-Based Debt and Savings Census\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"A census-to-survey agreement share computed from the 2023 Record-Based Debt and Savings Census, supporting one claim about record-based enumeration.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:c83853b977025e1ea217b9bc", "occurrence_id": "75faa7f59c49dca139b7694e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Capacity reporting.** Two figures are reproduced from the Panel Diagnostic Capacity System (PDCS) for 2020 and are quoted as the system releases them. The first is the share of reporting facilities filing a capacity return in the reference period; the second is the share of those facilities whose return declared no capacity for at least one named test.\n\nThe system is administrative, generated by the obligation to report rather than by collection, and both figures describe the returns. They carry no intervals, since the facilities are the population described, and the second is bounded below by what a facility is willing to declare absent.\n\nThe system is the paragraph's principal source and the only source placing both items on the same return, which is what allows the second figure to be read as a limit on the first. The panel structure means a facility may report in more than one period, and the figures reproduce the system's convention of using its most recent return, a rule documented with the source and not revisited here. PDCS's definition of a named test is its own.\n\nMention: \"Panel Diagnostic Capacity System (PDCS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Two 2020 figures reproduced from the Panel Diagnostic Capacity System (PDCS), the second read as a limit on how the first should be interpreted.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:ba4e9294ca1e1a6f6cc52ce7", "occurrence_id": "7ce3f27608ce24ca2b93bcc5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Transitions after school.** Two figures are reproduced from the Triennial Youth Transition Survey of 2019 and are given as the survey prints them. The first is the share of respondents enrolled or in training at the first interview who were in neither three years later; the second is the share of that group whose departure was recorded as voluntary.\n\nBoth are survey estimates from a panel and carry the attrition the follow-up incurred, which the survey documents and which is not corrected here. The second figure is smaller than the first and is reported against it for that reason.\n\nThe survey is the paragraph's principal source and the only source reporting both items on the same respondents, since the second is defined on the subset the first identifies. The survey's definitions of enrolment and of voluntary departure are its own, were applied as published, and are not restated. A respondent lost to follow-up enters the first figure's denominator and cannot enter the second, which biases the pair in a direction the survey acknowledges but does not quantify.\n\nMention: \"Triennial Youth Transition Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Two 2019 figures reproduced from the Triennial Youth Transition Survey, the second defined on the subset the first identifies.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:bc5b319dad397dcf8ce2e0ee", "occurrence_id": "1ed612848a5217cd6a452874", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Credit and seasonal exposure.** The two series were derived from the High-Frequency Rural Credit Dataset, a geospatial product in which each reporting area carries a credit volume attached to the season and to a mapped extent. The dataset publishes the volumes; it publishes no exposure measure, and neither quantity below is one of its values.\n\nThe first quantity is the share of the mapped extent in which no volume is recorded for any season; the second is the concentration of recorded volume in the single largest season, computed as that season's share of the area total. Both were computed on the released volumes with no interpolation across gaps.\n\nThose quantities carry the section. The argument concerns whether a high-frequency series can be read as a seasonal pattern when large parts of its extent are simply absent, and the first quantity bounds the claim while the second shows that the pattern is concentrated in a minority of areas. The dataset's climate-relevant season boundaries are its own and were used as published, and an area with no recorded volume contributes nothing to the second quantity.\n\nMention: \"High-Frequency Rural Credit Dataset\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"A gap share and a seasonal concentration derived from the High-Frequency Rural Credit Dataset, which publishes volumes but no exposure measure.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:04911e843a01ad90efeb0d5c", "occurrence_id": "5a85706aacec7f7b3412cdbf", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Preparing the migration series.** The work described here concerns the 2020 Midline Migration and Remittance Survey conducted in Tajikistan and compiled by the Banque de France, and the harmonisation performed on its returns rather than any estimate drawn afterwards. The survey is a field collection, and the returns arrived from three interview waves using two questionnaires.\n\nThree operations were carried out. The remittance items were mapped to a single currency and reference period; the household identifier was reconstructed where a wave had re-issued it; and the destination coding was reconciled to a single list. No response was altered and no missing value was filled, so the harmonised file reports the same non-response as the original.\n\nThat harmonisation supports one claim in a longer treatment of comparability across waves. The claim is that the apparent fall in reported remittances between the first and third waves is an artefact of the questionnaire change, and the mapping is the evidence for it, since the fall disappears when both waves are read on the later coding. The Banque de France's documentation of the waves was used and is not restated.\n\nMention: \"Midline Migration and Remittance Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"Three harmonisation operations on the 2020 Tajikistan Midline Migration and Remittance Survey, used to show a wave artefact in reported remittances.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:53fb6dea9d62ba3ffc099464", "occurrence_id": "637b1cd8c05d00936dcbb1b6", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Checking the facility figure.** The Provincial Facilities and WASH Index (PFWI) is used here to check a facility figure reaching this paragraph from a registry, and the index contributes no facility figure of its own. It is an indicator, constructed from inputs that are not the registry's returns.\n\nThe check is confined to whether the two sources place the same provinces in the same ordering, which is a weaker question than agreement about level and is the only one the index can answer. Where the orderings agree, the registry's figure is reported without qualification; where they diverge, the divergence is stated and no attempt is made to adjudicate.\n\nThis is a supporting use and deliberately narrow. The index's construction involves weightings that are not recoverable from the released values, and the check therefore establishes only that two independent constructions order the provinces similarly. Nothing in the paragraph treats that agreement as evidence about the level, and the index supplies no interval and no provincial detail beyond the ordering.\n\nMention: \"Provincial Facilities and WASH Index (PFWI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"Provincial Facilities and WASH Index (PFWI) used to check whether a registry figure orders provinces consistently, contributing no level.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:ecaf3b6b8904f9a10360bd72", "occurrence_id": "6ae641c5c6eca93d00e45d68", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Verifying the condition series.** The Periodic Vegetation Condition Registry is used here to check a condition series assembled from satellite products, and the registry contributes no condition value of its own. It is a database holding ground observations recorded at intervals by field teams, and its coverage is a matter of where those teams work.\n\nThe comparison asks whether the two move together in the provinces both cover. Where they do, the satellite series is reported without further qualification; where they diverge, the divergence is reported and the satellite series is presented as conditional on the calibration that produced it. No reweighting was applied to either, and no observation was excluded to improve agreement.\n\nThis is a supporting use. The registry's own record of when and under what protocol each observation was taken determines what its comparison can establish, and a ground observation taken outside the window of the satellite pass is not a check of it. The registry documents those timings, and the paragraph relies on them rather than restating them.\n\nMention: \"Periodic Vegetation Condition Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"Periodic Vegetation Condition Registry used to check a satellite-derived condition series for the provinces both cover, contributing no value of its own.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:8b772bb6e93f30d6f8379ba4", "occurrence_id": "b8e85008599895f53edf035f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Coping strategies and their persistence.** The persistence estimates were computed from the Panel Shock Coping Survey compiled by UNU-WIDER, which revisits the same households after each reported shock and records the response. The survey publishes the distribution of responses; it does not publish whether a household repeats one.\n\nA household was counted as persistent where the same response was recorded for two consecutive shocks, and the resulting share was computed among households with at least two shocks. The share is therefore conditional on having been shocked twice, which is a minority of the sample and is not the same population as the published distribution.\n\nThat conditional construction carries the section. The argument concerns whether coping is a repertoire or a sequence of distinct reactions, and the share addresses it directly, but only for the households with repeated exposure; a household experiencing one shock contributes nothing either way. UNU-WIDER's documentation of the panel's retention and of the shock definition sits with the survey and was not reproduced, and the definition of a shock is what determines who enters the denominator.\n\nMention: \"Panel Shock Coping Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"A persistence share computed on the UNU-WIDER Panel Shock Coping Survey, conditional on two consecutive shocks and why that condition matters.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:8cf3811a965270ee98b672dd", "occurrence_id": "07b6d7b4610335264817172f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Reading the housing indicator.** One comparison within a longer discussion of composite measures rests on the Experimental Housing Conditions Index compiled by INVALSI. The index is an indicator, and it is used here not for its value but to test whether a finding drawn from raw items survives being expressed as a composite.\n\nThe test computes the same gradient twice, once on the individual items and once on the index, and reports whether the two agree in sign and in the ordering of categories. No attempt was made to reconcile the two magnitudes, and the difference between them is reported rather than explained.\n\nThat comparison supports one claim rather than carrying the passage. The passage concerns how much a composite conceals about the items beneath it, and the divergence in magnitude between the two computations is the evidence offered, since the sign survives while the size does not. INVALSI's documentation of the index's weighting was consulted to establish that the composite is not a simple average of the items used in the first computation.\n\nMention: \"Experimental Housing Conditions Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The INVALSI Experimental Housing Conditions Index used to test whether a gradient survives being expressed as a composite, supporting one claim.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:78905eb38e6d93602fec5d93", "occurrence_id": "72f4e1e1e207ad0db975bd0e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Vulnerability after the shock round.** The two vulnerability estimates were computed from the Panel Poverty and Vulnerability Survey of 2021, compiled by MoWECP and BPS, which revisits households across rounds and carries both consumption and shock items. The survey publishes a poverty rate per round; it does not publish the vulnerability measures below.\n\nThe first estimate is the share of households whose consumption fell below the line in the round following a reported shock; the second is the share whose consumption remained below it two rounds later. Both were computed on households present in every round concerned, and the resulting attrition is reported rather than adjusted.\n\nThose estimates carry the section. The distinction between them is the section's subject: a rate measured shortly after a shock and a rate measured two rounds later describe different things, and the gap between them is what the surrounding discussion of recovery is about. MoWECP and BPS document the panel's weights and the line's construction, and both were used as published and are not restated here.\n\nMention: \"Panel Poverty and Vulnerability Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"Two vulnerability estimates computed from the 2021 MoWECP and BPS Panel Poverty and Vulnerability Survey, separated by the rounds at which they are measured.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:80fe8112e1dd9c5980375724", "occurrence_id": "cd299075a228d7452513db16", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Targeting accuracy.** The two targeting measures were computed from the Departmental Targeting and Payments Index, an indicator that reports a composite score per administrative unit rather than the household-level information the measures require. The index publishes no targeting statistic, and neither quantity below is one of its values.\n\nUnits were classified on the index score and the resulting groups were compared on the share of their recorded payments reaching the lowest score decile. The first measure is that share for the lowest group; the second is the ratio of it to the same share in the highest group. Both rest on the index's own ranking, which is the reason the second measure cannot be read as an independent test of the index.\n\nThat circularity is disclosed rather than corrected, because it is the section's point. The argument concerns how far a programme can demonstrate its own targeting when the only available ranking is the one the programme uses, and the second measure is offered as an illustration of a check that cannot fail. The index's construction is documented by its publisher and was adopted without modification.\n\nMention: \"Departmental Targeting and Payments Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"Two targeting measures computed from the Departmental Targeting and Payments Index, with the circularity between them reported as the finding.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:57e36c0a16eb00ff7ac55ef7", "occurrence_id": "ef38814d0436ebcbd741f7f7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Public works participation.** One comparison inside a longer treatment of activation programmes rests on the Household Public Works Index (HPWI) for Hungary in 2020. The index is an indicator, and what was derived from it is a single association rather than any published value.\n\nParticipation was regressed on the index score across the counties the 2020 release covers, and the fitted association is reported with its residual pattern by county type. Both were computed on the released values without exclusion or reweighting, and the residuals matter more here than the coefficient, since the counties departing from the fitted line share a labour-market characteristic the index does not contain.\n\nThat comparison supports one claim rather than carrying the paragraph. The paragraph concerns whether an activation indicator measures activation or its prerequisites, and the residual pattern enters as evidence that the two are separable in the counties where the fit fails. The index's definitions are its own and were adopted without restatement, and no causal reading is offered for the fitted association.\n\nMention: \"Household Public Works Index (HPWI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"A participation-to-score association computed from the 2020 Household Public Works Index (HPWI) for Hungary, supporting one claim about what the index measures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:b010c48839976d1cef7c51e3", "occurrence_id": "b25009ff74ec6f10da66477f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Movement within the country.** The pair of figures below is reproduced from the Triennial Internal Movement Survey of 2020 and is given as the survey prints it. The first is the share of households with a member who moved within the country in the reference period; the second is the share of those moves recorded as motivated by insecurity rather than by livelihood.\n\nNeither figure was recomputed. Both are estimates from a sample with a published design, and the second is defined on the subset the first identifies, so the two must be read together rather than as independent rates.\n\nThese figures support one claim in a passage whose subject is the difficulty of distinguishing displacement from migration in survey data. The survey's own motivation coding is the reason the pair can be read that way at all, and that coding is the survey's, applied as published and not revisited here. A move with two recorded motivations is classified once, under the survey's precedence rule, and the rule is documented with the instrument rather than restated in this paragraph.\n\nMention: \"Triennial Internal Movement Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"Two 2020 figures reproduced from the Triennial Internal Movement Survey, supporting one claim about separating displacement from migration.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:4d7ecbd359a272adc3864403", "occurrence_id": "c8ea68cedd8ff42e0673cbc9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Minimum needs and its thresholds.** The threshold analysis was carried out on the Experimental Minimum Needs Survey (EMNS), a field collection whose module on necessities asks respondents to state what they consider a necessity and then reports whether they possess it. The survey publishes possession rates by item; the analysis below is not one of them.\n\nItems were ranked by the proportion treating them as necessary, and a threshold was set at the point where that proportion falls below half. The two quantities reported are the number of items above that threshold and the share of households lacking at least one of them. Both were computed on the responses as collected, with no adjustment for the interview's setting.\n\nThat construction carries the section. The argument concerns whether a deprivation count can be built from lay judgements rather than from an expert list, and the share is the evidence brought for it, while the item count shows how unstable the threshold is: moving it by a small amount changes the count considerably. The EMNS's sampling was regional, and the share describes respondents within those regions.\n\nMention: \"Experimental Minimum Needs Survey (EMNS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"A lay-necessity threshold and its deprivation share computed on the Experimental Minimum Needs Survey (EMNS), with the threshold's instability reported.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:1cdb3353e2c336bb3ef3732d", "occurrence_id": "362b1c58ba1680d2e79b0d47", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Checking the maternal indicator.** The Combined Maternal and Newborn Report (CMNR) for Central Asia in 2015 is used here to check an indicator carried from a civil registration source, and the report contributes no indicator of its own to this paragraph. It is a report, drawing on several inputs, and its own estimates are published elsewhere.\n\nThe check asks whether the two sources place the same countries in the same range for the same year. Where they agree, the registration figure is reported without qualification; where they diverge, the divergence is stated and the registration figure is presented as conditional on the completeness of registration, which is the known weakness of the source being checked. No pooling was attempted.\n\nThis is a supporting use. The report's own account of which inputs entered its figures was consulted to establish that civil registration was not among them, since a report drawing on the registration system cannot check it. That independence is the whole value of the comparison, and the paragraph relies on the report's documentation of its inputs rather than restating it.\n\nMention: \"Combined Maternal and Newborn Report (CMNR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"2015 Combined Maternal and Newborn Report (CMNR) for Central Asia used to check a civil-registration indicator, valued for its independence from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:a94aef1aab77d1efcc000060", "occurrence_id": "1179e0d4416b4c303ea19e8d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Teacher supply.** Two figures are reproduced from the Departmental Teacher Workforce Index (DTWI) as the index releases them, and neither was recomputed. The first is the reported pupil-teacher ratio; the second is the share of teachers without the qualification the department's own framework specifies.\n\nThe index is a constructed indicator rather than a census of teachers, and its figures rest on returns from authorities under a weighting its documentation describes. They carry no intervals, and the second figure depends on a definition of qualification that the index inherits from the framework rather than deriving.\n\nThe index is the paragraph's principal source and the only source expressing both quantities on the same returns, which is what allows them to be read as a single statement about supply. The surrounding discussion of training capacity is context and supplies no figure used here. Where an authority filed no return the index imputes, and that treatment is carried through without restatement, which matters more for the second figure than for the first.\n\nMention: \"Departmental Teacher Workforce Index (DTWI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Two figures reproduced from the Departmental Teacher Workforce Index (DTWI), read as a single statement about supply on the index's own returns.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:6a85b5ac93f10f4141176dba", "occurrence_id": "e09a00cce0cc643e74b365cd", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Asset records.** The two shares below are reproduced from the Verified Asset Ownership Index for 2021 compiled by the European Space Agency, and are given as the index publishes them. The first is the reported share of households holding a registered land document; the second is the reported share holding such a document in the name of a woman.\n\nThe index is a constructed indicator assembled from several inputs under a published weighting, and neither figure is a survey estimate. No interval accompanies either, and the second is bounded above by the first as a matter of construction rather than of observation.\n\nThese shares support one claim in a longer passage on the measurement of ownership, and they are not that passage's conclusion. The comparison between them is offered as descriptive, since the index publishes no joint estimate and no interval was carried through when the two were placed side by side here. The index's documentation of its inputs was consulted and is not restated, and it is the reason the pair is reported as an indicator reading rather than as a measurement of ownership.\n\nMention: \"Verified Asset Ownership Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"Two 2021 shares reproduced from the European Space Agency's Verified Asset Ownership Index, offered as a descriptive comparison between them.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:14cda5ce2e6edacd1dfb87d7", "occurrence_id": "d95f0ab1b95a45aacb2fe3aa", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Vacancy duration.** The duration estimates were computed from the Urban Hiring and Vacancies System (UHVS) for 2024, an administrative system in which each vacancy carries a posting date and, where the post was filled, a closing date. The system publishes counts of open vacancies; it does not publish how long they remain open.\n\nDuration was computed for vacancies with both dates and summarised as a median with the share exceeding ninety days, and both were reported by the system's own occupation groups. Vacancies closed without a recorded outcome were excluded, which removes a substantial share of postings and is not a random removal.\n\nThat exclusion is the section's concern rather than a caveat attached to it. The argument concerns whether administrative vacancy records can support a claim about labour shortage, and the excluded postings are plausibly those filled informally, which is precisely the channel the claim concerns. The system's conventions for reopening and re-posting a vacancy determine what counts as one spell, and those conventions were carried through as published.\n\nMention: \"Urban Hiring and Vacancies System (UHVS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"A vacancy-duration median and long-duration share computed on the 2024 Urban Hiring and Vacancies System (UHVS), with the excluded postings discussed.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:d20378ad5934ba061a56292d", "occurrence_id": "f6bfa46025b082a0f6968322", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Consumption and its correlates.** The pair of figures below is reproduced from the Longitudinal Household Consumption Index (LHCI) compiled by Baier and Bergstrand, and neither was recalculated. The first is the reported share of households below the index's own consumption line; the second is the share of that group reporting an unmet health need in the same period.\n\nThe index is a constructed indicator rather than a survey, and both figures rest on inputs of several kinds under a weighting the accompanying note describes. They carry no intervals, and the second depends on an item asked of a subset of the panel, so its denominator is smaller than the first's.\n\nThese figures support one claim in a passage concerned with how consumption measures travel into health analysis, and they are not its conclusion. The index's definitions of the line and of unmet need are its own and were adopted without restatement. Because the second figure's denominator differs from the first's, the two are reported as a pairing rather than as a conditional probability, and the surrounding discussion of education outcomes supplies no quantity used here.\n\nMention: \"Longitudinal Household Consumption Index (LHCI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"Two figures reproduced from the Baier and Bergstrand Longitudinal Household Consumption Index (LHCI), reported as a pairing rather than a conditional.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:9285ca6a8b23a785d65927bb", "occurrence_id": "7a512adc381eeb264f5c5a3a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Vacancy filings.** Two counts are reproduced from the Ministerial Hiring and Vacancies Registry (MHVR) for 2018 and are given as the registry releases them. The first is the number of vacancies filed during the year; the second is the number of those filings withdrawn by the filing authority before any appointment.\n\nThe registry is a database of filings created by an obligation on authorities rather than by collection, and both counts describe the filing process. They carry no intervals, and the second is a count of withdrawals rather than of unfilled demand, since a withdrawn filing may have been withdrawn because the post was filled outside the registry.\n\nThese counts support one claim in a longer passage about how public vacancies come to be invisible, and they are not its conclusion. The registry's rules for reopening a filing, and for whether a re-filed vacancy counts once or twice, are the registry's own and were adopted without restatement. The employment reading of the second count is therefore bounded by the registry's own conventions rather than by the labour market it is often taken to describe.\n\nMention: \"Ministerial Hiring and Vacancies Registry (MHVR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"Two 2018 counts reproduced from the Ministerial Hiring and Vacancies Registry (MHVR), supporting one claim about vacancies becoming invisible.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:ef7a1ecbeebe0ce950d199d0", "occurrence_id": "93f04332d1ebaf2dfd205f03", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Consumption growth.** The growth estimates were computed from the Longitudinal Household Income and Expenditure Survey (LHIES), a panel whose value lies in following the same households, and the estimates are not the published period comparison. The survey publishes cross-sectional means per wave; the quantities below hold the household fixed.\n\nGrowth was computed for households present in consecutive waves and reported as the share whose real consumption rose, alongside the mean log change for the same households. Both were deflated with the index the survey supplies, and the fixed panel means the comparison excludes households that enter or leave between waves.\n\nThat exclusion is the section's subject rather than an incidental consequence. The argument concerns whether panel and repeated-cross-section readings of the same survey can disagree, and they do here: the fixed-panel share is materially lower than the published comparison implies, and the difference is composed of the households the panel cannot follow. LHIES's documentation of attrition sits with the survey and was not reproduced.\n\nMention: \"Longitudinal Household Income and Expenditure Survey (LHIES)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"Consumption growth on a fixed panel from the Longitudinal Household Income and Expenditure Survey (LHIES), contrasted with the survey's published comparison.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:9b8e9143a806dbcd21397a59", "occurrence_id": "89c96f7bbf3227fe567ecc7e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Enumeration and the poverty line.** The distribution was estimated from the Standard Poverty and Vulnerability Census (SPVC) of 2016, a census rather than a sample, and the estimate is not the census's published headcount. The census publishes a count below the national line; the quantity below is the shape of the distribution beneath it.\n\nHouseholds were positioned on the census's consumption items and the resulting distribution summarised as the share below four-fifths of the line and the share between the line and that of four-fifths. Both were computed on the enumerated population, so neither carries a sampling error, and both depend on the line as the census defines it.\n\nThat second share carries the section. The argument concerns how sensitive a headcount is to small movements in the line, and a population massed just above it is the evidence that the sensitivity is high. SPVC's construction of the line and its treatment of household size sit with the census and were used as published, and because the census enumerates rather than samples, the two shares describe the enumerated population and nothing beyond it.\n\nMention: \"Standard Poverty and Vulnerability Census (SPVC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"A below-line distribution estimated from the 2016 Standard Poverty and Vulnerability Census (SPVC), used to show the headcount's sensitivity to the line.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:b5a620c0bd6df021b1dcf49d", "occurrence_id": "b68cce93da20e213cdb86cf2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Insurance take-up.** The take-up estimates were computed from the Triennial Unemployment Insurance Survey, a field collection whose frame is drawn from establishment records, and the estimates are not the survey's published coverage rate. The published rate is computed against the labour force as the survey estimates it; the quantities below are computed against the eligible population the eligibility rules define.\n\nEligibility was reconstructed from the contribution and contract items, and the reported quantities are the share of eligible workers registered and the share of those who left registration within the year. Both restrict the population to those the rules reach, which is a smaller group than the survey's labour force.\n\nThat narrower denominator is the section's point. The argument concerns whether coverage can be assessed against a population defined by entitlement rather than by employment status, and the two quantities show the resulting rate is considerably higher than the published one while describing fewer people. The survey's own documentation of the contribution items was used and not restated, and where an item was unanswered the worker was treated as ineligible rather than imputed.\n\nMention: \"Triennial Unemployment Insurance Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"Eligibility-based take-up estimates computed from the Triennial Unemployment Insurance Survey, contrasted with the survey's published coverage rate.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:0de33cfe738fad92d8aa075e", "occurrence_id": "be464dfa04484dede20d20c9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **A registry noted in passing.** The Rolling Socioeconomic Registry (RSR) is mentioned here only to note that the estimates discussed in this section were not drawn from it. Nothing is argued from the registry, no quantity of its own is reported, and its existence does not corroborate or qualify any figure above.\n\nThe reason for the note is that the registry covers the same population as two of the sources used here, and a reader familiar with it might reasonably expect its figures to appear. They do not, because the registry's own coverage follows the administrative events that create its records and would not be comparable with the survey frames on which this section's estimates rest.\n\nThe registry is therefore background. It is not used to check, to corroborate or to compare, and no claim in this section would change if the registry did not exist. Its documentation was consulted solely to establish the overlap described above.\n\nMention: \"Rolling Socioeconomic Registry (RSR)\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"validate\", \"summary\": \"Rolling Socioeconomic Registry (RSR) noted only to state that no figure here comes from it; nothing is argued from the registry.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:f9f26454eb912f80b05cd271", "occurrence_id": "b1907a892536a91e23e1c8a5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Preserving a registry extract.** The work reported here concerns the 2019 Subnational Socioeconomic Registry (SSR), a database extract, and the archival work performed on it rather than any analysis. Three operations were carried out and documented so that the extract can be re-used without reconstructing its provenance.\n\nThe extract's field definitions were transcribed into machine-readable metadata; the sequence of tables that had to be joined to reproduce the published subnational aggregates was recorded as a recipe rather than left implicit; and the resulting package, with its join recipe, was lodged with a repository under a persistent identifier. No record was altered, deduplicated or filled.\n\nThat work supports one claim in a longer treatment of what makes an administrative extract citable. The claim is that the extract was previously re-usable only by someone who had worked on the original project, since the join recipe existed nowhere but in that team's practice, and the recipe file is the evidence that the barrier was procedural rather than technical. The registry's own content remains unchanged.\n\nMention: \"Subnational Socioeconomic Registry (SSR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"Three archival operations on the 2019 Subnational Socioeconomic Registry (SSR) extract, reported as work that makes the extract citable.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:9c2ae4df8d642bd6c1fd6ed5", "occurrence_id": "8014343b0497a783894bc6b0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Housing condition measures.** The two condition measures were computed on the Rapid Housing Conditions Registry for the United Kingdom in 2016, a database holding an inspection record per dwelling. The registry publishes counts of dwellings by recorded category; the measures below are properties of the records rather than published counts.\n\nThe first measure is the share of inspected dwellings carrying at least one category the registry's own guidance treats as urgent; the second is the share of those dwellings whose next inspection is recorded as overdue. Both were computed on dwellings with an inspection record, which excludes the majority of the register.\n\nThat exclusion carries the section's argument. The point is that a registry's headline categories describe what has been looked at, and the second measure shows how unevenly looking is distributed, since overdue dwellings cluster in a minority of districts. The 2016 snapshot is a single vintage, so nothing here is a trend, and the guidance's category definitions were used as published without restatement.\n\nMention: \"Rapid Housing Conditions Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"Two condition measures computed on the 2016 Rapid Housing Conditions Registry for the United Kingdom, with the inspected subset identified as the limit.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:0d8a93f0eef538c9250ebc86", "occurrence_id": "1acb0f10842f089e8e25f656", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Earnings and hours.** Two figures are reproduced from the Regional Wage Structure Survey (RWSS) of 2017, compiled by NSS, and both are given as the survey prints them. The first is the median hourly earnings for full-time employees; the second is the share of employees whose usual hours exceed forty-eight per week.\n\nBoth are survey estimates carrying the design's precision, and the survey's treatment of overtime pay and of irregular hours is the survey's own, applied as published and not revisited. The median is reported at the rounding the survey uses.\n\nThe survey is the paragraph's principal source and the only source of either figure. The second is discussed as a partial explanation of the first, since median hourly earnings and long hours move together in the survey's regional tabulations, but no decomposition is offered and the survey publishes none. NSS's documentation of the field conditions behind the 2017 round sits with the source and was consulted rather than reproduced, and it is relevant to the second figure in particular.\n\nMention: \"Regional Wage Structure Survey (RWSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Two 2017 figures reproduced from the NSS Regional Wage Structure Survey (RWSS), the second offered as a partial explanation of the first.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:d025fccbcfd228baea57cc7e", "occurrence_id": "40bb6f6ae56ce727ea3de92e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Archiving a small-area census.** The work described here concerns the 2018 High-Resolution Output Area Census (HOAC) compiled by INFRAS/IWW, a census enumerated on small output areas, and the curation performed on its release rather than any analysis of its contents. The release arrived as a set of files whose area codes were versioned inconsistently.\n\nFour operations were carried out. The area codes were reconciled to a single vintage and the vintage recorded; the correspondence file linking areas between vintages was attached to the release; the variable definitions were consolidated from three documents into one; and the package was lodged with a repository under a persistent identifier. No enumerated value was altered.\n\nThat work supports one claim in a longer treatment of small-area data. The claim is that the release's unusability was a matter of its geography rather than its content, and the correspondence file is the evidence, since the analysis that had previously been impossible became possible without any change to an enumerated figure. The census's own enumeration remains the responsibility of INFRAS/IWW.\n\nMention: \"High-Resolution Output Area Census (HOAC)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"Four curation operations on the 2018 High-Resolution Output Area Census (HOAC), supporting one claim that its barrier was geographic, not content.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:7967f54c411d77c3b4da7c9b", "occurrence_id": "8a53285a3373c03f8edd0c9e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Verifying the debt figure.** The Rapid Debt and Savings Survey of 2018, compiled by the Department of Defense, is the primary check on a household debt figure reaching this section from a credit register, and the survey's own debt estimates are reported here only insofar as the check requires them. The survey is a field collection with a frame built independently of the register.\n\nThe check compares the proportion of households the two sources place in debt. Where they agree, the register's figure stands unqualified; where they differ, the difference is reported and the register's figure is presented as describing those with a credit file rather than those with debt. No reconciliation was attempted between the two populations, and no combined figure is offered.\n\nThat the frames are independent is what makes the check worth performing, and the survey's frame documentation was consulted to establish it. The survey's own estimates are published elsewhere and were not recomputed here, and the register remains the source of the figure being checked rather than of the check.\n\nMention: \"Rapid Debt and Savings Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"2018 Rapid Debt and Savings Survey used as the primary check on a credit-register debt figure, its frame being independent of the register.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:7d307a939a0dc9e22842473d", "occurrence_id": "27fe77c248dcc7c5257a011a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Remittance receipt.** The two figures below are reproduced from the Annual Migration and Remittance Index (AMRI) for Mauritania compiled by BOOST, and are quoted as the index releases them. The first is the reported share of households receiving a remittance during the year; the second is the share of those households for whom remittances exceed a quarter of reported income.\n\nThe index is a constructed indicator rather than a survey, and neither figure carries a sampling error. The second is defined on the subset the first identifies, and its denominator is therefore smaller, a point that matters because the two are frequently reported as though they were independent rates.\n\nThese figures support one claim in a passage concerned with how remittance dependence is measured, and they are not that passage's conclusion. The index's definition of a remittance, and of income for the purpose of the second figure, are its own and were adopted without restatement. No reference year is asserted beyond the annual release the index provides, and BOOST's documentation of the inputs was consulted rather than reproduced.\n\nMention: \"Annual Migration and Remittance Index (AMRI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"Two figures reproduced from the BOOST Annual Migration and Remittance Index (AMRI) for Mauritania, with the second's subset denominator made explicit.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:7212a618b1c715303a8532a9", "occurrence_id": "9490823fdc9be55b8e4aa0c3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Checking the formation figure.** The Longitudinal Business Demography Survey of 2024, compiled by ENIA, is used here to check a business formation figure reaching this paragraph from a tax register, and the survey's own formation estimates are reported only so far as the check requires. The survey revisits the same establishments across waves and its frame was constructed before the register period began.\n\nThe check compares the count of new establishments the two sources record. Where they agree, the register's figure is reported unqualified; where they differ, the difference is stated and the register's figure is presented as describing those who registered rather than those who began trading. Nothing was pooled and no correction is offered.\n\nThis is a supporting role. The survey's panel structure matters for a second reason: it can identify establishments that traded without appearing in the register, which is the mechanism the difference most likely reflects, and the paragraph reports that the survey does identify such establishments without quantifying the difference. ENIA's documentation of the frame's construction was consulted and not reproduced.\n\nMention: \"Longitudinal Business Demography Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"2024 ENIA Longitudinal Business Demography Survey used to check a tax-register formation figure, its panel also identifying unregistered trading.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:18b5856c9768a7e649193124", "occurrence_id": "ddec31b1034ace9bedb12311", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Preparing a price series.** The work described here concerns the 2019 High-Frequency Agri-Food Prices System compiled by SABESP, an administrative feed of commodity prices, and the preparation performed on it rather than any analysis of the resulting series. The feed arrived as weekly extracts whose market identifiers had drifted.\n\nThree operations were carried out. Market identifiers were reconciled to a single list and the mapping recorded; the commodity codes were aligned to the classification used by the agricultural statistics office; and the gaps left by weeks with no extract were recorded as gaps rather than filled. No price was altered, adjusted or imputed, and the prepared series reports the same missing weeks as the feed.\n\nThat preparation supports one claim in a longer treatment of high-frequency data. The claim is that the series' apparent discontinuity in the third quarter was an artefact of the identifier drift rather than of market conditions, and the mapping file is the evidence, since the discontinuity disappears once the identifiers are reconciled. The prepared series was lodged with the office that commissioned the work.\n\nMention: \"High-Frequency Agri-Food Prices System\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"Three preparation operations on the 2019 SABESP High-Frequency Agri-Food Prices System, used to attribute a discontinuity to identifier drift.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:c5afb7f0e4712838f0f45c04", "occurrence_id": "dd7b5a08d56ada537b2c9282", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Conditions of employment.** Two figures are reproduced from the Quarterly Working Conditions Survey of 2019 and are given as the survey prints them. The first is the share of employees reporting non-standard contracts; the second is the share of those employees without paid leave entitlement.\n\nBoth are survey estimates from a quarterly instrument and are reported at the survey's own precision. The second is defined on the subset the first identifies, and the survey publishes no joint estimate of the pair, so the comparison between them is descriptive and carries no interval.\n\nThe survey is the paragraph's principal source and the only source of either figure. Its definitions of non-standard contracting and of entitlement are its own, were applied as published, and are not restated here. The survey's own note on the treatment of employees holding more than one contract determines which contract the two figures describe, and that note sits with the instrument rather than with this paragraph.\n\nMention: \"Quarterly Working Conditions Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Two 2019 figures reproduced from the Quarterly Working Conditions Survey, read as a descriptive comparison within the survey's own definitions.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:c5e3545a2d378bbbadee9849", "occurrence_id": "45ae34f9f5d7aa2d80fbd9be", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Assembling a hazard collection.** The work reported here concerns the Follow-Up Landslide Hazard Collection, a compilation of field observations whose records were gathered by different teams under different protocols, and the work of assembling it rather than any analysis of hazard itself. The collection is not a survey, a register or an indicator, and it is best described as a compilation.\n\nFour operations were carried out. Observation dates were converted to a single calendar and time reference; the slope and lithology fields were reconciled to one vocabulary, with the original terms retained alongside; observations whose location could not be resolved to the stated precision were flagged rather than discarded; and the assembled collection was lodged with a repository with its protocol notes attached.\n\nThat assembly is the section's subject. The claim is that the apparent increase in landslide frequency across the collection is an artefact of changing observation effort rather than of ground conditions, and the protocol notes are the evidence, since the increase appears only in the periods where effort rose. No observation was altered during assembly, and the flagged records remain in the collection.\n\nMention: \"Follow-Up Landslide Hazard Collection\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"Four assembly operations on the Follow-Up Landslide Hazard Collection, used to attribute an apparent frequency increase to observation effort.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:56bce3924df46000b04885bd", "occurrence_id": "901275071aa0fe614d07342d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Catch and effort.** The pair of figures below is reproduced from the Standardised Fisheries and Aquaculture Survey of 2023 and is given as the survey releases it. The first is the reported share of vessels landing through a recorded channel; the second is the share of aquaculture units reporting a recorded input purchase.\n\nNeither figure was recomputed. Both are survey estimates with a published design, and the two populations are different: the first describes vessel operators, the second describes aquaculture units, and the survey does not report them on a common frame.\n\nThese figures support one claim in a passage concerned with how much of the sector is visible to statistics at all, and they are not its conclusion. The comparison between them is offered as an illustration of differing visibility rather than as a ratio, since dividing one by the other would combine two frames the survey keeps separate. The survey's definitions of a recorded channel and of a unit are its own and were adopted without restatement.\n\nMention: \"Standardised Fisheries and Aquaculture Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"Two 2023 figures reproduced from the Standardised Fisheries and Aquaculture Survey, offered as differing visibility rather than as a ratio.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:3cfed574cd5cc7f4aa33dc6b", "occurrence_id": "510f96eaa25fe725fdb30172", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Verifying the return-intention figure.** The Standard Return Intention Survey (SRIS) of 2022, compiled by FISE, is the primary check on an intention figure reaching this section from a registration system, and the survey's estimates are reported here only insofar as the check requires them. The survey interviews households at sites selected independently of the registration system, and it asks about intention rather than recording an act.\n\nThe check compares the two sources on the proportion expressing an intention to return within twelve months. Where they agree, the registration figure is reported unqualified; where they differ, the difference is stated and the registration figure is presented as describing those whose registration carries an intention field at all. No pooling was attempted.\n\nThat the two populations differ is the paragraph's subject as much as the check's result. The registration system's intention field is completed only in some offices, and the survey's sites are not concentrated in those offices, so the comparison is between an intention measured everywhere and one recorded selectively. The survey's own account of its sampling was used and is not restated.\n\nMention: \"Standard Return Intention Survey (SRIS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"2022 FISE Standard Return Intention Survey (SRIS) as the primary check on a registration-derived intention figure, with selective recording discussed.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:357d7ca26b582bf4abacb269", "occurrence_id": "89980f5d6aeaa2ae3ae9e7f9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Disability prevalence.** Two figures are reproduced from the National Disability and Support Index (NDSI) for 2018 and are quoted as the index releases them. The first is the reported share of the population experiencing at least one functional difficulty; the second is the share experiencing difficulty in more than one domain.\n\nThe index is a constructed indicator rather than a survey, and its figures rest on inputs assembled under a published weighting. They carry no intervals, and both depend on the domain set the index adopts, which is not the same set used by the instruments that feed it.\n\nThe index is the paragraph's principal source and the only source expressing both quantities on a common basis, which is what allows the second to be read as a severity reading of the first. The surrounding discussion of support provision supplies no figure used here. NDSI's documentation of the domain set and of the weighting was consulted rather than reproduced, and it is the reason the pair is reported as an indicator reading rather than as a prevalence measurement.\n\nMention: \"National Disability and Support Index (NDSI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Two 2018 figures reproduced from the National Disability and Support Index (NDSI), the second read as a severity reading of the first.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:421cd6f7045fcc80febfcb20", "occurrence_id": "9fc50cafb1a3f4ece00c4953", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Post-shock records.** The two shares below are reproduced from the Quality-Assured Shock Coping Registry for 2021 and are given as the registry prints them. The first is the share of recorded households with an assessment filed after the reported event; the second is the share of those assessments filed more than thirty days after it.\n\nThe registry is a database created by the filing obligation placed on responding agencies, and both shares describe that filing rather than the coping that followed. They carry no intervals, since the households recorded are the population described, and the second is bounded below by events that generated no record at all.\n\nThese shares support one claim in a longer passage on the timeliness of post-event data, and they are not its conclusion. The registry's rules for opening a record, and for whether a household affected by two events is recorded once or twice, are the registry's own and were adopted without restatement. The quality assurance the registry's name describes is documented with the source and concerns the completeness of filings rather than the accuracy of the reported events.\n\nMention: \"Quality-Assured Shock Coping Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"Two 2021 shares reproduced from the Quality-Assured Shock Coping Registry, supporting one claim about the timeliness of post-event data.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:0f51b177f292b5150e26c158", "occurrence_id": "557e4052282ba256d87e4667", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Support needs and poverty.** The association between functional difficulty and poverty was constructed from the Standard Disability and Support Microdata for Burkina Faso in 2024, released at household level with no weights attached. The released documentation reports item frequencies; it reports no association, and the quantity below is not one of its values.\n\nDifficulty was counted across the domains the instrument carries and the count cut at two or more; the reported quantity is the poverty rate among households with such a count and its difference from the rate among the rest. Both were computed on the released rows as supplied, with no imputation for the domain items left unanswered.\n\nThat construction carries the section. The absence of weights means both rates describe respondent households rather than the national population, and the unanswered domain items are more frequent among the poorest households, so the difference is likely to be understated rather than overstated. The 2024 fieldwork documentation was consulted and not reproduced, and the cut at two domains is an analytical choice made here rather than a threshold the instrument defines.\n\nMention: \"Standard Disability and Support Microdata\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"A difficulty-poverty association constructed from the 2024 Burkina Faso Standard Disability and Support Microdata, with its likely direction of error stated.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:39cbf4e3ec7820e91b1d7809", "occurrence_id": "5638c94b520aee04830da0d9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Flood extent records.** Two figures are reproduced from the Joint Flood Extent Registry (JFER) for 2019 and are quoted as the registry releases them. The first is the share of the affected extent for which an extent polygon was filed; the second is the share of filed polygons whose filing agency is the same as the agency reporting the damage.\n\nThe registry is a database of filings assembled after each event, and both figures describe the filings rather than the flooding. They carry no intervals, and the second describes an institutional overlap rather than a physical relationship.\n\nThe registry is the paragraph's principal source and the only source placing both quantities on the same event record, which is what allows the second to be read as a caution about the first. The registry's rules for accepting a polygon, and for reconciling two filings that disagree about an extent, are its own, were applied as published and are not restated here. A flooded area for which no polygon was filed is absent from the first figure and therefore from the second as well.\n\nMention: \"Joint Flood Extent Registry (JFER)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Two 2019 figures reproduced from the Joint Flood Extent Registry (JFER), the second read as a caution about the first.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:bdf29ca714440e70de87b535", "occurrence_id": "3c40de38028d0ab415c68e9a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Programme exits.** The exit pattern was computed from the Local Graduation Programmes Registry for the United States in 2023, maintained by MapX, a database holding a record per enrolled household with the date and reason of any exit. The registry publishes enrolment counts; it publishes no exit pattern.\n\nExits were classified on the registry's reason field and the distribution reported alongside the median duration in programme, with both computed for cohorts entering in the same year. Where the reason field is blank the exit is reported as unclassified rather than assigned, and the unclassified share is large enough to be reported in the text.\n\nThat share carries the section. The argument concerns whether graduation can be measured from programme records at all, and a reason field blank for a substantial minority of exits is the evidence that the records were built for administration rather than for evaluation. MapX documents the field's completion rules, and those rules determine what the classified exits can mean. The registry's own cohort definitions were used as published.\n\nMention: \"Local Graduation Programmes Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"An exit distribution and median duration computed on the 2023 MapX Local Graduation Programmes Registry, with blank reason codes as the finding.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:5b70354d46fed31d72f3e6f1", "occurrence_id": "0f500db324e6393bd606e89e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Transition outcomes.** Two figures are reproduced from the National Youth Transition Census compiled by the World Inequality Lab and are given as the census releases them. The first is the share of school-leavers not in education, training or paid work at enumeration; the second is the share of that group living in a household below the national income threshold.\n\nThe census enumerates rather than samples, so neither figure carries a sampling error, and both describe the enumerated population. The second is defined on the subset the first identifies, and the two are therefore a pairing rather than two independent rates.\n\nThe census is the paragraph's principal source and the only source of either figure. No reference year is asserted, since the census reports a reference period spanning two calendar years and the figures are the pooled pair as published. The World Inequality Lab's documentation of the enumeration and of the threshold's application was consulted rather than reproduced, and the threshold is the national one rather than a line constructed for this analysis.\n\nMention: \"National Youth Transition Census\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Two figures reproduced from the World Inequality Lab's National Youth Transition Census, reported as a pairing over the census's reference period.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:4349fabc53d695bf661503aa", "occurrence_id": "261ba131ce5d02fe8b2f77a3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Preparing a wage feed.** The work described here concerns the Midline Wage Structure System in Chile, an administrative feed of payroll returns, and the curation performed on it rather than any estimate drawn from the resulting series. The feed arrives monthly and its establishment identifiers are issued by the contributing agencies rather than by the system.\n\nTwo operations were carried out. Identifiers were reconciled across the contributing agencies and the crosswalk recorded, since a firm filing through two agencies had previously appeared as two establishments; and the reference period of each field was recorded as metadata rather than inferred from the column names. No wage value was altered and no gap filled.\n\nThat work supports one claim in a longer treatment of administrative wage series. The claim is that the apparent growth in the number of establishments was an artefact of duplicate identifiers rather than of entry, and the crosswalk is the evidence, since the growth disappears once the duplicates are resolved. The prepared feed was lodged with the statistical office that will maintain it, and Chile's reporting rules remain as the agencies set them.\n\nMention: \"Midline Wage Structure System\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"Two curation operations on Chile's Midline Wage Structure System, used to show establishment growth was an artefact of duplicate identifiers.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:e986d931b71595aff8ec1105", "occurrence_id": "46d8c3ee73d4fbfb9e33a742", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Work and schooling together.** The two quantities were computed from the 2015 round of the Subnational Public Works Survey (SPWS), compiled by Michelin, a field collection that asks both the employment and enrolment modules of every household member above fifteen. The survey publishes employment and enrolment rates separately; it does not publish their intersection.\n\nThe first quantity is the share of members aged fifteen to twenty-four who were both enrolled and working; the second is the share of that group reporting that work reduced available study time. Both were computed on the individual records the survey supplies, and the second is defined on the subset the first identifies.\n\nThat second quantity carries the section. The education argument concerns whether combined participation should be treated as a form of disadvantage or as a normal strategy, and a reported effect on study time is the evidence offered for treating it as the former, while the paragraph is explicit that the report is subjective. Michelin's documentation of the module order was consulted and not reproduced, and the order matters because the employment item precedes the enrolment item.\n\nMention: \"Subnational Public Works Survey (SPWS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"Two combined participation quantities computed on the 2015 Subnational Public Works Survey (SPWS), with the module order noted.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:7263aa02da0527bd8fa6cf69", "occurrence_id": "6bec4e28e0ea67dc2965d615", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Migration and its returns.** Two figures are reproduced from the Longitudinal Migration and Remittance Survey of 2016, compiled by Badan Pusat Statistik or BPS, and neither was recomputed. The first is the share of households with a member absent for work at the first wave; the second is the share of that group receiving a remittance recorded in the household's own account.\n\nBoth are estimates from a panel and carry the attrition the follow-up incurred, which the survey documents. The second is defined on the subset the first identifies, and the survey publishes no joint estimate of the pair, so the comparison is descriptive and carries no interval.\n\nThe survey is the paragraph's principal source and the only source placing both items on the same household across waves. Its definitions of absence for work and of a recorded remittance are its own, were applied as published, and are not restated. The education discussion elsewhere in the chapter supplies no quantity used here, and the survey's own notes on which households were lost between waves sit with the instrument rather than in this paragraph.\n\nMention: \"Longitudinal Migration and Remittance Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Two 2016 figures reproduced from the Badan Pusat Statistik longitudinal migration survey, read as a descriptive pairing across waves.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:a29a4f6a393024057c49086e", "occurrence_id": "2ce36e3a8cf67cc8cbc7a910", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Verifying the movement figure.** The Periodic Mixed Movement Survey (PMMS) of 2024, compiled under the acronym held by FISE, is the primary check on a movement figure reaching this section from a site registration system, and the survey contributes no movement figure of its own beyond what the check requires. The survey enumerates households at locations selected without reference to the registration system.\n\nThe check compares the two sources on the share of households reporting a move in the reference period. Where they agree, the registration figure stands unqualified; where they differ, the difference is stated and the registration figure is presented as describing households at registered sites. No pooling was attempted and no ratio is offered as a correction.\n\nThat the survey's frame is independent is the check's entire value, and its design documentation was consulted to establish it. The poverty dimension enters because the survey records consumption as well as movement, which allows the paragraph to note that the two sources differ most where consumption is lowest, without attributing the difference to either instrument.\n\nMention: \"Periodic Mixed Movement Survey (PMMS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"2024 Periodic Mixed Movement Survey (PMMS) as the primary check on a site-registration movement figure, with the divergence patterned by consumption.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:e28d33a24e1aacd2f49f21ec", "occurrence_id": "37396ab780a839cef659c1fd", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Household energy expenditure.** Two figures are reproduced from the Verified Fuel and Energy Use Index for 2015 compiled by Facebook and are quoted as the index releases them. The first is the reported share of households whose principal cooking fuel is solid; the second is the share of reported household expenditure taken by fuel and electricity.\n\nThe index is a constructed indicator assembled from inputs of several kinds under a published weighting, and neither figure is a survey estimate. They carry no intervals, and the second depends on an expenditure definition the index inherits rather than one it defines.\n\nThe index is the paragraph's principal source and the only source expressing both quantities on a common basis, which is what allows the second to be read as a statement about the burden of the first. The surrounding discussion of fuel markets is context and supplies no quantity used here. Facebook's documentation of the index's inputs and weighting was consulted rather than reproduced, and it is the reason both figures are reported as an indicator reading rather than as measurements of household spending.\n\nMention: \"Verified Fuel and Energy Use Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Two 2015 figures reproduced from the Facebook Verified Fuel and Energy Use Index, the second read as the burden implied by the first.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:86a4cb151052eee03a1ef176", "occurrence_id": "419aacf2b3efe27ed91400c3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Risk records.** The two figures below are reproduced from the Follow-Up Climate and Flood Risk Index compiled by the Universal Postal Union and are given as the index publishes them. The first is the reported share of administrative units carrying a flood exposure classification; the second is the share of those units whose classification was revised at the follow-up.\n\nThe index is a constructed indicator rather than a survey, and neither figure carries a sampling error. The second describes the revision of a classification rather than a change of exposure, and the index does not indicate which revisions reflect new information and which reflect method changes.\n\nThese figures support one claim in a passage concerned with how risk classification drifts, and they are not that passage's conclusion. The comparison between them is offered as descriptive, since the index publishes no joint estimate and no interval was carried through when the two were placed side by side. The Universal Postal Union's documentation of the index's construction was consulted rather than reproduced, and the classification scheme is the index's own and was adopted without restatement.\n\nMention: \"Follow-Up Climate and Flood Risk Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"Two figures reproduced from the Universal Postal Union's Follow-Up Climate and Flood Risk Index, supporting one claim about classification drift.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:8fe1e3e8e60446ce07db923a", "occurrence_id": "c2c748a624315f85c01ada81", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Checking the payment figure.** The Nationwide Cash Transfer Registry Collection is used here to check a payment figure reaching this paragraph from a ministry's administrative series, and the collection contributes no payment figure of its own. It is a compilation of registry extracts assembled by several agencies, and it is neither a survey nor a register of its own.\n\nThe check compares the two sources on the count of recipient households for the same period. Where they agree, the ministry's figure is reported unqualified; where they differ, the difference is stated and the ministry's figure is presented as describing the series' own reporting scope. No pooling was attempted, and no ratio between the two is offered as a corrected figure.\n\nThis is a supporting use, and its limits follow from the collection's nature. Because the extracts were assembled rather than enumerated, the collection cannot establish which count is right, only that the two differ and by how much. The compilation notes describing which agencies contributed which extracts were consulted and are not restated, and they are the reason the comparison is reported as a difference rather than as an error.\n\nMention: \"Nationwide Cash Transfer Registry Collection\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"Nationwide Cash Transfer Registry Collection used to check a ministry payment figure, reporting a difference rather than adjudicating it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:8226760595d734c48cee6928", "occurrence_id": "da72d26d7245cb5b6dc4bb1b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Boundaries.** Two figures are reproduced from the Experimental Administrative Boundary Dataset, a geospatial product releasing a polygon per unit at the vintage consulted, and both are given as the dataset releases them. The first is the number of units in the release; the second is the proportion of their shared borders that coincide exactly with the borders of the previous release.\n\nThe dataset is a classification of space rather than a measurement of it, and the second figure describes the geometry of two releases rather than any administrative change. Where a boundary was redrawn the figure records the disagreement, and it does not distinguish a redraw from a change in digitisation.\n\nThe dataset is the paragraph's principal source and the only source expressing both quantities on the same geometry, which is what allows the second to be read as a caution about analyses that join successive releases. The dataset's own note on its generalisation rules travels with it and was not restated, and no reference year is asserted beyond the vintage consulted, since the release carries none.\n\nMention: \"Experimental Administrative Boundary Dataset\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Two figures reproduced from the Experimental Administrative Boundary Dataset, the second read as a caution about joining successive releases.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:5089d09fe0c3783943e33cff", "occurrence_id": "f447c2993b7b916da10f4ad3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Elevation coverage.** Two figures are reproduced from the Experimental Elevation Model Census (EEMC) of 2020 and are given as the census releases them. The first is the proportion of the surveyed extent carrying an elevation value at the stated resolution; the second is the proportion of those values whose stated uncertainty exceeds five metres.\n\nThe census enumerates a surface rather than a population, and both figures describe the model's own record of its quality. They carry no sampling error, since the extent is the object enumerated, and the second depends on an uncertainty the model estimates about itself rather than one measured against ground control.\n\nThe census is the paragraph's principal source and the only source expressing both quantities on the same grid, which is what allows the second to be read as a limit on the first. The census's definition of the surveyed extent, and of the resolution at which a value is said to exist, are its own and were adopted without restatement. A region outside the extent appears in neither figure, and the paragraph does not treat its absence as a measurement of zero.\n\nMention: \"Experimental Elevation Model Census (EEMC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Two figures reproduced from the 2020 Experimental Elevation Model Census (EEMC), the second read as a limit on the first.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:e2d368ea6af2b4f1348b9bb4", "occurrence_id": "dac85fed7ed3bc9820879183", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Remittance and employment.** Two figures are reproduced from the Ministerial Migration and Remittance Survey of 2023 and are quoted as the survey prints them. The first is the share of households with at least one member working abroad; the second is the share of those households reporting a remittance received in the reference quarter.\n\nBoth are survey estimates carrying the design's precision, and the second is defined on the subset the first identifies. The survey publishes no joint estimate of the pair, so the comparison between them is descriptive and carries no interval.\n\nThe survey is the paragraph's principal source and the only source placing both items on the same household. Its definitions of working abroad and of a received remittance are its own, were applied as published, and are not restated here. The poverty discussion elsewhere in the chapter supplies no quantity used in this paragraph, and the survey's own notes on seasonal recall in the remittance module sit with the instrument rather than with this text.\n\nMention: \"Ministerial Migration and Remittance Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Two 2023 figures reproduced from the Ministerial Migration and Remittance Survey, read as a descriptive pairing within the survey's definitions.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:da5aad2e35bb09f5811dd4e2", "occurrence_id": "0873c898861cf830c57df000", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Consumption records.** The pair of figures below is reproduced from the Monthly Household Consumption System (MHCS) for 2018 and is given as the system releases it. The first is the share of registered households whose consumption record was updated in the reference month; the second is the share of those records whose value fell below the system's own subsistence threshold.\n\nThe system is administrative, produced by the registration and reporting steps through which households pass, and both figures describe that production rather than the consumption behind it. They carry no intervals, since the registered households are the population described.\n\nThese figures support one claim in a passage concerned with how administrative consumption compares with survey consumption, and they are not that passage's conclusion. The comparison between them is offered as descriptive, since the system publishes no joint estimate and no interval was carried through when the two were placed side by side. The threshold is the system's own and was adopted without restatement or conversion.\n\nMention: \"Monthly Household Consumption System (MHCS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"Two 2018 figures reproduced from the Monthly Household Consumption System (MHCS), offered as a descriptive comparison, not a joint estimate.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:5ff08a7bd3ec867854587c5a", "occurrence_id": "2a7fe7a08cfdc1426ed6cf83", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Deprivation in rural areas.** One comparison inside a longer treatment of rural deprivation rests on the Pilot Deprivation and Hardship Report (PDHR) compiled by the US Census Bureau. The report is a report rather than an enumeration, and what was derived from it is not any published value but the association between two of its tabulations that it presents separately.\n\nThe derived quantity pairs each area's deprivation share with the share of its households whose principal livelihood is agricultural, and reports the fitted association across the areas the report covers. Both inputs were used as the report prints them, and no area was excluded.\n\nThat comparison supports one claim rather than carrying the paragraph. The paragraph concerns whether rural deprivation policy is targeting agriculture or poverty, and the association enters as evidence that the two are not separable in the report's own presentation. The report's definitions of deprivation and of agricultural livelihood are its own and were adopted without restatement, and the fit is offered as descriptive rather than causal.\n\nMention: \"Pilot Deprivation and Hardship Report (PDHR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"A deprivation-to-livelihood association computed from the US Census Bureau's Pilot Deprivation and Hardship Report (PDHR), supporting one claim.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:2f9c54f4f94bf167762f8b19", "occurrence_id": "a82fc11cfa7a501c45c6bbdf", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Microdata as released.** Two figures are reproduced from the Representative Household Income and Expenditure Microdata (RHIEM) as the file is released, and neither was recomputed. The first is the reported share of households below the national poverty line; the second is the share of those households whose record carries a complete income section.\n\nThe material is microdata released at household level, and the second figure matters because the published poverty statistics are computed on the subset the first figure's denominator implies. Neither figure carries an interval here, since the released documentation supplies none.\n\nThe file is the paragraph's principal source and the only source of either figure. Its own documentation of the release's weighting and of the imputation applied to incomplete income sections was consulted rather than reproduced, and it is the reason the second figure is reported at all: without it the first would be read as computed on the full sample.\n\nMention: \"Representative Household Income and Expenditure Microdata (RHIEM)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Two figures reproduced from the Representative Household Income and Expenditure Microdata (RHIEM) release, the second describing the first's denominator.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:1e2c7db6bd72d486477802c5", "occurrence_id": "a7ee13ee5262d70edd06e7b4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Necessity thresholds.** Two figures are reproduced from the Household Minimum Needs Index (HMNI) compiled by USDA and are quoted as the index releases them. The first is the reported share of households below the index's minimum needs threshold; the second is the share of that group whose shortfall is concentrated in a single item category.\n\nThe index is a constructed indicator rather than a survey, and its figures rest on a threshold derived from a market basket whose composition the index's documentation describes. Neither figure carries a sampling error, and both depend on the basket as the index defines it.\n\nThe index is the paragraph's principal source and the only source expressing both quantities on a common basis, which is what allows the second to be read as a statement about the shape of the shortfall the first identifies. The surrounding discussion of price levels is context and supplies no quantity used here. USDA's documentation of the basket and its update cycle was consulted rather than reproduced.\n\nMention: \"Household Minimum Needs Index (HMNI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"Two figures reproduced from the USDA Household Minimum Needs Index (HMNI), the second read as the shape of the shortfall the first identifies.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:81b12d856abcb6a63618b27b", "occurrence_id": "c3f19a9b4c13801a350bd637", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Learning and its distribution.** The distribution estimates were computed from the High-Frequency Learning and Equity Survey (HLES) of 2022, a repeated collection that samples children rather than institutions and carries a background questionnaire alongside the assessment items. The survey publishes mean scores by wave; the quantities below are not means.\n\nScores were positioned within each wave and the distribution summarised as the share below a set proficiency floor and the gap between the fourth and ninety-first percentiles. Both were computed on the released scaling, and no wave was pooled with another, since the survey's own note states that the scale is not maintained across waves.\n\nThose quantities carry the section. The argument concerns whether a high-frequency series can support a claim about widening inequality, and a percentile gap computed within each wave is the evidence offered, while the decision not to pool is what limits the claim to within-wave comparisons. HLES's documentation of the scaling and of its sampling was consulted and not reproduced.\n\nMention: \"High-Frequency Learning and Equity Survey (HLES)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"A within-wave proficiency share and percentile gap computed from the 2022 High-Frequency Learning and Equity Survey (HLES), with pooling refused.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:58c1a9219191a1fa8feea1f8", "occurrence_id": "6765162aca92045ede2138a8", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Health workforce.** The pair of figures below is reproduced from the Standardised Health Workforce Survey (SHWS) conducted in Myanmar in 2021 and is given as the survey prints it. The first is the reported density of health workers per thousand population; the second is the share of those workers posted to facilities without the equipment list the survey's own protocol specifies.\n\nBoth are survey estimates with a published design, and the second depends on a facility inventory completed by the facility itself, which the survey notes is not independently verified.\n\nThese figures support one claim in a longer passage on whether workforce numbers measure capacity, and they are not that passage's conclusion. The comparison between them is offered as descriptive, since the survey publishes no joint estimate and no interval was carried through when the two were placed side by side here. The survey's definition of a health worker, and of the equipment list against which the second figure is computed, are its own and were adopted without restatement.\n\nMention: \"Standardised Health Workforce Survey (SHWS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"Two 2021 figures reproduced from the Standardised Health Workforce Survey (SHWS) in Myanmar, supporting one claim about whether numbers measure capacity.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:d037fe8534e33f8c4a55c39e", "occurrence_id": "c9755979720b316edf5d49a9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Cadastre completeness.** The two completeness measures were computed from the Midline Cadastral Frame Registry (MCFR) for 2017, compiled by MoWECP and BPS, a database holding a frame record per parcel. The registry publishes counts of registered parcels; it publishes no completeness measure, and neither quantity below is one of its values.\n\nThe first measure is the share of the frame's parcels carrying an owner record; the second is the share carrying both an owner and a recorded area. Both were computed on the frame as constituted in the 2017 release, and neither was adjusted for parcels appearing twice.\n\nThat the second measure is smaller than the first is the section's point rather than an incidental arithmetic fact. The argument concerns what a cadastral frame can support about land distribution, and the difference between the two measures is the share of the frame whose area is unknown, which is precisely the quantity a distribution needs. MoWECP and BPS document the frame's construction and its revision history, and those notes were used and not restated.\n\nMention: \"Midline Cadastral Frame Registry (MCFR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"Two completeness measures computed on the 2017 Midline Cadastral Frame Registry (MCFR), with the gap between them read as missing area.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:fde2b6edf25abd243b16cf24", "occurrence_id": "3349b07b710665adaa24b2dd", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **A registry noted for context.** The Analytical Asset Ownership Registry (AAOR) of the Republic of Zambia is named here only for the coverage it represents, and no value from it is reported in this section or used in any calculation. It is a database, and its records are created by the transactions that bring a household into contact with the programmes it serves.\n\nThe reason for the mention is that readers working on asset measurement in southern Africa encounter the registry frequently, and this section's figures do not come from it. Its coverage follows programme contact rather than residence, which is the reason it was not used: the comparison being drawn requires a population frame, and the registry does not supply one.\n\nThe registry is therefore background to the argument. Nothing here is argued from it, no claim would change if it did not exist, and its documentation was consulted solely to establish what its coverage does and does not describe.\n\nMention: \"Analytical Asset Ownership Registry (AAOR)\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"Analytical Asset Ownership Registry (AAOR) of the Republic of Zambia named for context; no values reported and nothing argued from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:c69b9f1dcc2386370fe07821", "occurrence_id": "8c1e6df027611ccaf4a9ba9c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Road network consistency.** The two consistency measures were computed from the High-Resolution Road Segment Index (HRSI) for 2021, an indicator that assigns each segment a class and a surface attribute. The index publishes totals by class; it publishes no measure of internal agreement, and neither quantity below is one of its values.\n\nThe first measure is the proportion of segment endpoints that connect to another segment within the index's own tolerance; the second is the proportion of segments whose class differs from that of every segment they connect to. Both were computed on the released geometry with no repair of the network.\n\nThose measures carry the section. The argument concerns whether an index built for prioritisation can be used to compute connectivity, and the first measure bounds any such computation while the second shows that the class attribute is not smooth across the network, so a route computed on it inherits discontinuities. The HRSI's tolerance and class definitions are its own, were used as published, and a segment outside the index's extent contributes to neither measure.\n\nMention: \"High-Resolution Road Segment Index (HRSI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"Two network-consistency measures computed on the 2021 High-Resolution Road Segment Index (HRSI), which publishes totals but no agreement measure.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:71007c771a23abac3030c91d", "occurrence_id": "3dbb13539c04d729e6cbdd36", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **A supply-system reference.** The Quarterly Supply Chain Survey is named here because the paragraph's framing requires it, and no figure from the survey appears in this section. It is a survey of facilities and distributors, and it is one of the few instruments that asks about stock-out procedure at all.\n\nThe mention exists to locate the discussion: the paragraph concerns how a stock-out is defined, and the survey's questionnaire is one of the instruments that has had to settle that definition in practice. Its documentation was consulted for the wording of its items and for nothing else, and none of its estimates is reported here.\n\nThe survey supports the framing rather than the findings. No comparison in this section rests on it, and no figure reported elsewhere was drawn from it, so a reader should not treat the mention as evidence for any of the quantities above. Its quarterly cadence is noted only because the definitional question is a repeated one, and a single instrument asking the same item four times a year is a useful illustration of how a definition settles in practice.\n\nMention: \"Quarterly Supply Chain Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"Quarterly Supply Chain Survey named for how it defines stock-out; no figures reported and no findings resting on it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:a6b33b93b4859290e26d0ec3", "occurrence_id": "071f3176d78b3f212855bc4f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Agricultural household records.** The two shares below are reproduced from the Civil Agricultural Household Index for 2022 compiled by Global Integrity, and are given as the index publishes them. The first is the reported share of agricultural households with a registered land holding; the second is the share of those households with a member holding a formal contract.\n\nThe index is a constructed indicator rather than a survey, and its figures rest on inputs assembled under a published weighting. Neither figure carries a sampling error, and the second is defined on the subset the first identifies.\n\nThese shares support one claim in a passage concerned with how agricultural households enter formal statistics, and they are not that passage's conclusion. The index's definitions of an agricultural household and of a formal contract are its own and were adopted without restatement. Global Integrity's documentation of the index's inputs was consulted rather than reproduced, and it is the reason both figures are reported as an indicator reading rather than as measurements of the sector.\n\nMention: \"Civil Agricultural Household Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"Two 2022 shares reproduced from the Global Integrity Civil Agricultural Household Index, offered as an indicator reading rather than a measurement.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:238a921e25efe374063e762d", "occurrence_id": "3a901ab4c25b446e0c78e8b8", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Elevation records.** Two figures are reproduced from the Coordinated Elevation Model Census for Ile de France compiled by Facebook and are quoted as the census releases them. The first is the proportion of the extent carrying a value at the stated resolution; the second is the proportion of that extent whose value is flagged as interpolated rather than observed.\n\nThe census is a geospatial enumeration rather than a population census, and both figures describe the model's record of its own derivation. They carry no sampling error, since the extent is the object enumerated, and the second depends on a flag the model assigns to itself.\n\nThese figures support one claim in a passage about how much of a terrain model is measurement, and they are not that passage's conclusion. The census's rules for flagging a value as interpolated are its own and were adopted without restatement, and a region outside the compiled extent appears in neither figure. The comparison between the two is offered as descriptive, since the census publishes no joint estimate of the pair.\n\nMention: \"Coordinated Elevation Model Census\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"Two figures reproduced from the Facebook Coordinated Elevation Model Census for Ile de France, supporting one claim about modelled terrain.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:80f3456700890aa072b0ef0e", "occurrence_id": "9bcf5578917b1ad4922a8d57", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Payment accuracy.** The two targeting quantities were computed from the Monthly Targeting and Payments Index, an indicator that reports a composite score per unit rather than the household records the quantities require. The index publishes no targeting statistic, and neither quantity below is one of its values.\n\nUnits were ranked on the index score, and the reported quantities are the share of payments reaching the lowest-ranked decile and the share reaching the highest. Both rest on the index's own ranking, and the second is reported because the first is difficult to interpret without it.\n\nThat pairing carries the section. The argument concerns whether a programme can demonstrate targeting using the same ranking it targets with, and the two quantities are reported together precisely because the pairing cannot fail. The index's construction is documented by its publisher and was adopted without modification, and no external validation of the ranking was available.\n\nMention: \"Monthly Targeting and Payments Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"Two targeting quantities computed from the Monthly Targeting and Payments Index, reported together as a pairing that cannot fail.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:056e81b9188cac18bcc001b7", "occurrence_id": "ab2e888a604c4d39715801c5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Assembling a station collection.** The work reported here concerns the Integrated Temperature Records Collection (ITRC) for Azerbaijan, a compilation of station records gathered over several decades by different authorities, and the assembly performed on it rather than any climate analysis. The collection is not a survey, a register or an indicator, and it is best described as a compilation.\n\nFour operations were carried out. Station identifiers were reconciled across the contributing authorities and the crosswalk recorded; the observation calendars were converted to a common reference; records whose units were ambiguous were flagged rather than converted; and the assembled collection was lodged with a repository together with the crosswalk and the flag list. No reading was altered and no gap filled.\n\nThat assembly is the section's subject. The claim is that the apparent warming trend in the raw compilation is partly an artefact of station turnover rather than of climate, and the crosswalk is the evidence, since the trend weakens once records are continued across identifier changes rather than treated as separate stations. The collection's agricultural relevance was not examined.\n\nMention: \"Integrated Temperature Records Collection (ITRC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"Four assembly operations on the Azerbaijan Integrated Temperature Records Collection (ITRC), used to separate station turnover from trend.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:2300b4567c8c56158a7d2beb", "occurrence_id": "e77380f3739008cb1d55c3d7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Archiving a campaign register.** The work described here concerns the National Vaccination Campaign Registry, a database holding a record per campaign round, and the curation performed on it rather than any analysis of coverage. The registry's rounds had been released as separate files whose field names differed between rounds.\n\nThree operations were carried out. Field names were reconciled to a single schema and the mapping recorded; the age definition used in each round was captured as metadata rather than inferred from the column; and the resulting package was lodged with a repository under a persistent identifier. No count was altered and no round was excluded.\n\nThat work supports one claim in a longer treatment of campaign data. The claim is that the apparent fall in coverage between the third and fourth rounds is an artefact of a changed age definition rather than of delivery, and the metadata is the evidence, since the two rounds are not comparable as released. The registry's own counts remain unchanged, and the campaign's operational records remain the responsibility of the agency that maintains them.\n\nMention: \"National Vaccination Campaign Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"Three curation operations on the National Vaccination Campaign Registry, used to show a coverage fall was an age-definition artefact.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:d264af4dffb764fbff1f90a3", "occurrence_id": "f109aed925bdda1428f5f4c6", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **A quality survey noted in passing.** The Midline Clinic Quality Survey (MCQS) is mentioned here only to record that the quality figures discussed above were not drawn from it. Nothing is argued from the survey, no value of its own is reported, and its existence neither corroborates nor qualifies any figure in this section.\n\nThe note is here because the MCQS is the most recent instrument to measure clinical quality in the same settings, and a reader would reasonably expect its figures to appear. They do not, because the survey's own sample was restricted to facilities meeting an eligibility criterion that excludes a substantial part of the settings this section describes.\n\nThe survey is therefore background. It is not used to check, corroborate or compare, and no claim above would change if the survey had not been conducted. Its published design report was consulted solely to establish the eligibility criterion described above.\n\nMention: \"Midline Clinic Quality Survey (MCQS)\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"Midline Clinic Quality Survey (MCQS) noted only to state that no figure here comes from it; nothing is argued from the survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:75a38e7d8316a43b488b1350", "occurrence_id": "5b11bdd5fdeedf940ca94534", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Skills records and outcomes.** The two quantities were derived from the Coordinated Skills and Employment System (CSES), an administrative system in which a registration record carries a training history and, where available, a subsequent employment record. The system publishes counts of registrations; it publishes neither of the quantities below.\n\nThe first quantity is the share of registrations with a recorded qualification; the second is the share of those whose employment record appears within six months. Both were computed on registrations closed in the reference period, which excludes the still-open records that make up a substantial part of the system.\n\nThat exclusion carries the section's argument. The point is that an administrative system measures the completion of processes rather than their progress, and the open records are precisely the ones a labour-market reading would want. The CSES's conventions for closing a record determine what the reference period contains, and those conventions were used as published. The system's documentation of the qualification field was consulted and is not restated.\n\nMention: \"Coordinated Skills and Employment System (CSES)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"Two quantities derived from the Coordinated Skills and Employment System (CSES), with the exclusion of open records as the section's point.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:4cc0b684c420767eb1f91df1", "occurrence_id": "e736ad73f87acbdcb7dfd89c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Targeting in practice.** The two targeting estimates were computed from the Monthly Targeting and Payments Survey conducted in Poland in 2022, which interviews both recipient and non-recipient households and therefore supports a statement about errors of inclusion as well as exclusion. The survey publishes a recipient profile; it does not publish either error.\n\nThe first estimate is the share of recipient households found ineligible on the programme's own criteria; the second is the share of eligible non-recipient households not registered. Both were computed on the survey's sample with its weights applied, and the second is reported with its wider interval because the eligible non-recipients are the smaller group.\n\nThose estimates carry the section. The argument concerns whether targeting failures are asymmetric, and the pair is the evidence offered, while the paragraph notes that the second estimate depends on respondents' willingness to report a registration they never attempted. The survey's documentation of its frame and of the eligibility reconstruction was consulted and not reproduced.\n\nMention: \"Monthly Targeting and Payments Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"Two targeting-error estimates computed from the 2022 Poland Monthly Targeting and Payments Survey, with the asymmetry the pair implies.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:778f70f5935f9b1db059bf84", "occurrence_id": "746db15948a650e1aaa56f77", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment reviews reported flood exposure against the available evidence and applies a conservative interpretation where measurements are incomplete. The purpose is to test the internal consistency of the estimated affected area and to identify whether material differences arise from coverage, timing, or classification. Results are presented as an analytical cross-check rather than as a replacement for the principal estimate.\n\nThe District Flood Extent Registry database for 2024 provides a supplementary basis for this review. Its recorded extent is compared with the estimate obtained from a separate source, allowing the assessment to determine whether the direction and approximate magnitude of the reported exposure are credible. The comparison supports the claim that the estimated affected area is within a plausible range, while also indicating that small differences may reflect differing observation thresholds.\n\nThe registry is not used to generate a time series or to infer changes beyond the period covered by its records. Instead, it serves as an independent verification point within the broader assessment. Where the two measures diverge, the estimate is retained with a qualification concerning measurement uncertainty, rather than adjusted solely to achieve numerical alignment. This treatment preserves comparability across the analysis and makes clear that the cited database backs one claim concerning the reasonableness of the estimate.\n\nMention: \"District Flood Extent Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The District Flood Extent Registry database for 2024 is used to cross-check the estimated affected area against a separate source.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:b81cca66c686c1b551e7c686", "occurrence_id": "2695e2f764f24cc0371919e8", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Inter-Agency Nutrition and Mortality Index (INMI) is an indicator produced by the National Institute of Statistics and referring to 2015. It is used here as the principal source for the nutrition and mortality measures cited in the assessment. The source is retained in its reported form, with no adjustment to the underlying values or reclassification of the measures.\n\nTwo figures are reproduced directly from the indicator. The reported prevalence of acute malnutrition is 12.4 percent, while the reported mortality measure is 6.8 percent. These values constitute the central quantitative evidence for the assessment of conditions addressed through the relevant intervention components.\n\nThe figures are presented as source observations rather than as estimates derived from a separate calculation. Their use supports a concise comparison of the two outcomes within the analytical framework, while preserving the definitions and measurement basis associated with the indicator. Interpretation is therefore limited to the values reported by the National Institute of Statistics, and no additional adjustment is applied in this passage.\n\nMention: \"Inter-Agency Nutrition and Mortality Index (INMI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the 12.4 percent acute malnutrition figure and the 6.8 percent mortality figure directly from the 2015 INMI indicator produced by the National Institute of Statistics.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:97e01ea4fd5ed9e566dabe7a", "occurrence_id": "a352b99f2dac72d4e45c9ffe", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Quarterly Refugee and Migrant Registry (QRMR) database is used as the principal evidentiary source for the population profile presented in this annex. It provides a structured record for monitoring the scale and composition of registered refugees and migrants, and supports the tabulation of reported counts used in the assessment.\n\nTwo figures are taken directly from the database and reproduced without recalculation. The registry reports 12,480 persons in the recorded population, while 6,215 persons are shown in the corresponding subset. These totals are retained as reported and constitute the central quantitative evidence for the discussion that follows.\n\nThe figures are interpreted as registry counts rather than estimates of the wider population. They should therefore be read with due regard to the coverage and recording practices inherent in the source. No adjustment, projection, or reconciliation with an external series has been applied. This approach preserves the administrative meaning of the reported values and allows the assessment to distinguish documented records from analytical inference.\n\nMention: \"Quarterly Refugee and Migrant Registry (QRMR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the figures 12,480 and 6,215 directly from the QRMR database.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:6d5296297f054c74a48875cc", "occurrence_id": "037d457a398cadbd1fd34c10", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Combined Time Use Registry database covers Sulaymaniya and is used as a supporting reference for the time-use assessment. Its records provide a structured basis for reviewing the distribution of reported activities across households and for identifying whether the observed pattern is consistent with the estimate used in the main analysis.\n\nThe registry is not treated as the sole basis for the estimate. Rather, the reported activity profile is compared with an estimate drawn from an independent source, with attention to the direction and relative magnitude of the principal components. This cross-check supports the claim that unpaid household activity represents a material component of total time use.\n\nDifferences between the two sources are assessed as part of the validation process. Variations may reflect coverage, reporting practices, or the treatment of overlapping activities, and therefore do not by themselves invalidate the estimate. The comparison nevertheless provides a useful control on the plausibility of the result and helps identify where interpretation should remain cautious.\n\nThe evidence is incorporated into the assessment as corroborative rather than determinative. On this basis, the central finding is retained, while conclusions concerning smaller activity categories are presented with appropriate qualification.\n\nMention: \"Combined Time Use Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Combined Time Use Registry database for Sulaymaniya is used to cross-check an independent estimate of time use.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:049e994d9fc404171b3e9fa9", "occurrence_id": "b613274bc4077fd195e1f3d2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Experimental Time Use Survey, produced by the Turkish Statistical Institute, is treated as a supplementary survey source for assessing reported patterns of daily activity. Its measures provide an independent reference for interpreting the distribution of time across paid work, unpaid household production, care responsibilities, and other activities. The source is not used to replace the principal estimate, which remains the basis for the results presented in this assessment.\n\nA comparison with the survey indicates that the estimated allocation of time is broadly consistent with the observed direction and relative scale of activity participation. This cross-check is applied to one claim concerning the balance between market and non-market work. Differences in definitions, coverage, and reporting practices may account for deviations between the two measures and therefore limit the use of the comparison for precise point validation.\n\nThe evidence is consequently interpreted as a plausibility check rather than as an additional measurement series. Where the independent estimate falls outside the range suggested by the Experimental Time Use Survey, the result is flagged for review of assumptions, classification choices, and the treatment of activities that may be underreported. This procedure strengthens the internal assessment while preserving a clear distinction between the primary estimate and the corroborating survey evidence.\n\nMention: \"Experimental Time Use Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the Experimental Time Use Survey from the Turkish Statistical Institute to cross-check an independent estimate of paid and unpaid activity patterns.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:7eda053fe2665bc096798931", "occurrence_id": "2f4ab593796357701708fa70", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Rolling Shock Coping Survey is treated as a 2021 survey produced by UkrStat and is used as a supporting source for the assessment of household responses to economic disruption. Its reported patterns provide an external reference for reviewing the plausibility of the estimates presented in this section.\n\nThe comparison is made against an estimate derived from a separate source. It is intended to cross-check the direction and relative magnitude of the reported change, rather than to replace the primary calculation or establish a separate series. The correspondence between the two sources is considered sufficient to support the claim that coping responses became more prevalent during the period under review.\n\nDifferences in coverage, timing, and response conditions may affect the level of the reported indicators. Accordingly, the Rolling Shock Coping Survey is not used to reconcile individual observations or to generate a combined estimate. Its role is limited to corroborating one substantive claim within the broader analysis, while the principal results remain based on the underlying estimate and its documented methodology.\n\nMention: \"Rolling Shock Coping Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Rolling Shock Coping Survey is used to cross-check an independently derived estimate and support one claim about coping responses.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:05a3aa2d47086985a8385f63", "occurrence_id": "291a5aac8ca3920cb1f6f42f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Annual Mechanisation Report is a report covering Lebanon and is used as a supporting reference for the assessment of agricultural equipment. Its tabulations provide a structured benchmark for the recorded presence of machinery and for the relative distribution of equipment across the activities considered in the analysis. No adjustment is applied to the source material at this stage.\n\nThe estimate presented in the broader assessment is drawn from a separate source. It is cross-checked against the corresponding evidence in the Annual Mechanisation Report to test whether the order of magnitude and the observed pattern are consistent. The comparison is not intended to reconcile all differences between the two sources, since their coverage and reporting procedures may not be identical.\n\nThe cross-check supports the claim that access to mechanisation remains uneven and that the central estimate is within a plausible range. Where the two sources diverge, the estimate from the separate source is retained for consistency with the main analytical framework, while the report is used to qualify interpretation and identify potential measurement limitations. This use strengthens the evidentiary basis for the single claim without assigning the supporting source a broader role than the available information permits.\n\nMention: \"Annual Mechanisation Report\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the Annual Mechanisation Report for Lebanon to cross-check an independent mechanisation estimate and support a broader assessment claim.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:9238e98d65d36306a6a4af34", "occurrence_id": "be728608fab6f4af632712c6", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Diagnostic Cash Transfer Registry Survey (DCTRS) as its principal empirical source. This 2021 survey, produced by VIIRS, provides the observation base for the registry review. Its records are treated as a common analytical frame, with reported entries read consistently across the assessment rather than reproduced as isolated source figures.\n\nThe analysis calculates an average coverage rate from the DCTRS observations and derives a trend by comparing the distribution of registered cases across the survey’s reporting intervals. These measures identify the direction and approximate pace of change. The resulting pattern indicates a gradual increase in registry completeness, although the average remains below the level required for reliable planning. This conclusion is based on aggregation of the source material, not on the reproduction of selected entries.\n\nBecause the indicators are calculated from the source records, they should be interpreted as summary measures for implementation monitoring. The assessment therefore uses the computed average as its principal benchmark and the derived trend as a signal for follow-up. Individual observations are retained for verification, but they do not determine the reported finding. The DCTRS thus serves as the central evidence for assessing movement in registry performance and for framing subsequent operational attention.\n\nMention: \"Diagnostic Cash Transfer Registry Survey (DCTRS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the 2021 DCTRS survey to calculate an average coverage rate and derive a trend rather than quote individual figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:62e5127e19561fa530ebc259", "occurrence_id": "d5ad9a26a8ed54260cf4f90c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Experimental Business Demography Survey (EBDS) is used as a survey-based reference for Australia in 2018. The source is produced by the Turkish Statistical Institute and provides an independent basis for reviewing reported movements in business activity. Its inclusion supports a comparison between the reported estimate and evidence obtained through a separate estimation process.\n\nFor validation purposes, the EBDS results are used to cross-check the estimate rather than to replace the underlying calculation. The comparison focuses on the consistency of direction, scale, and coverage across the two sources. Particular attention is given to whether differences can be explained by measurement boundaries or by the timing of observations. This treatment strengthens the assessment of reliability while preserving the distinction between the estimate and the supporting survey evidence.\n\nThe comparison does not imply that the two sources are identical in construction or intended use. Instead, the survey serves as a central external check on the plausibility of the estimate and helps identify material divergence requiring further review. On this basis, the EBDS is retained as supporting evidence for interpretation and quality assurance in the assessment.\n\nMention: \"Experimental Business Demography Survey (EBDS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the 2018 Australia coverage of the Experimental Business Demography Survey (EBDS), produced by the Turkish Statistical Institute, to cross-check an independent estimate.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:c11c95d4f382ca3b9e04416f", "occurrence_id": "52569862791021de78889f65", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Linked Time Use Index (LTUI) is treated as an indicator produced by the Univer sity of Minnesota and referring to 2023. The source is used without adjustment for the descriptive comparison presented below. Its identification is retained in the text to support traceability of the evidence and to distinguish the reported measures from project-generated calculations.\n\nThe indicator gives a value of 47.2 for the first measure and 52.8 for the second measure. These figures are reproduced directly from the Linked Time Use Index (LTUI) and constitute the central quantitative evidence for the assessment. No interpolation or conversion has been applied, and the difference between the two values is not presented as a separately derived result.\n\nTaken together, the figures show a clear separation between the two reported measures within the source. The first value is lower than the second by 5.6 index points, providing a concise basis for the comparison used in this annex. The calculation is limited to the arithmetic difference between the quoted values and does not introduce an additional series or classification.\n\nFor reporting purposes, the source designation, producing organisation, and reference year are maintained in the narrative. The two observations should therefore be read as source figures rather than as estimates generated through the implementation review. Their inclusion supports reproducibility while keeping the assessment dependent on a single, clearly identified indicator.\n\nMention: \"Linked Time Use Index (LTUI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the LTUI figures 47.2 and 52.8, taken directly from the source, as the central evidence.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:9b5155c3be3914b799d76dc5", "occurrence_id": "4473809df9430b89e5676fde", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Departmental Early Warning Report (DEWR) is a report produced by OECD-PISA and covering Maputo for 2016. It provides the principal reference for interpreting early warning conditions in the assessment, with attention to the reported distribution of observed risks and the consistency of the underlying signals. The source is retained in its cited form so that its scope remains distinct from the estimate used for the main analysis.\n\nThe estimate is first established from a separate record and is then cross-checked against the DEWR findings. This comparison indicates whether the estimated pattern is compatible with the report’s recorded incidence and relative concentration of risk. Where the two accounts diverge, the difference is treated as a reason for review rather than as evidence that either account is invalid. The cross-check therefore supports a cautious reading of the estimate and helps identify observations requiring further verification.\n\nFor implementation purposes, the DEWR evidence is used as the central corroborating reference. Its coverage of Maputo and its relation to 2016 provide a defined geographical and temporal basis for comparison, while its status as a report produced by OECD-PISA supports consistent interpretation of the indicators. No adjustment is applied solely because the estimate differs from the report; the comparison is used to assess plausibility, document uncertainty, and guide subsequent examination of the available records.\n\nMention: \"Departmental Early Warning Report (DEWR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the 2016 Maputo DEWR report produced by OECD-PISA to cross-check a separate estimate.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:9474b09364cee873b1c871d1", "occurrence_id": "aa61f02d83f7dde676bea883", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Clustered Nighttime Lights Registry as a database for an auxiliary review of spatial activity indicators. The source provides a structured reference against which the reported estimate of local economic intensity can be examined without altering the underlying calculation.\n\nThe estimate is cross-checked against the pattern recorded in the Clustered Nighttime Lights Registry. Concordance between the two measures is taken as supporting evidence for the finding that activity is concentrated in the principal observed corridors, while divergence is treated as a signal requiring further review. This use is limited to verification of the claim and does not replace the primary estimate.\n\nThe comparison is conducted at the level of the available reporting units and is interpreted cautiously because the two sources may capture different dimensions of activity. The registry measure is therefore used as corroborating evidence within the broader assessment rather than as a standalone outcome indicator. On this basis, the reported estimate remains suitable for presentation, subject to the limitations noted elsewhere in the assessment.\n\nMention: \"Clustered Nighttime Lights Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the Clustered Nighttime Lights Registry database to cross-check an independent estimate and support a claim about concentrated activity.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:2b04f109b0702cecb0ed41e2", "occurrence_id": "be7a4dbfe0a2e53568b4b8e3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Experimental Informal Trade Survey is a survey produced by INEP for 2020 and is used here as the principal supporting reference for the assessment of informal activity. Its results provide an empirical basis for checking the magnitude of the estimate used in this report, while preserving a clear distinction between observed information and derived calculations.\n\nThe estimate is cross-checked against the evidence reported in the Experimental Informal Trade Survey. The comparison considers whether the direction and approximate scale of the estimate are consistent with the survey findings, without treating the two sources as interchangeable. This use strengthens the assessment where the available information is incomplete and helps identify whether the estimate is materially outside the range suggested by observed informal activity.\n\nThe source is therefore not presented as a supplementary citation. It supplies the central benchmark for interpreting the independently derived estimate and for judging its plausibility. Differences between the estimate and the survey results are retained as an analytical qualification rather than removed through adjustment, since they may reflect differences in measurement, coverage, or timing. The resulting comparison supports a cautious interpretation of the estimate and provides a documented basis for subsequent review.\n\nMention: \"Experimental Informal Trade Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The work taken from the 2020 survey produced by INEP is a cross-check of an independently derived estimate against observed informal activity.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:14a8b6a2bece64627cfec4ed", "occurrence_id": "4652199d470d15a4d01caf9b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Decennial Time Use Index (DTUI) is used as the principal indicator for 2019 in the assessment of time allocation. The source provides the central evidence for examining the distribution of activity across work, care, personal maintenance, and leisure. Its cited designation is retained in full so that the analytical basis remains identifiable throughout the assessment.\n\nThe reported observation is not reproduced as a standalone figure. Instead, the assessment calculates a comparative average across the relevant activity measures and derives a trend from their relative movement. This treatment reduces the influence of isolated values and supports interpretation of the overall direction of change. The resulting measures are used to assess whether time allocation is becoming more concentrated in selected categories or more evenly distributed.\n\nThe calculation applies the same definitions and aggregation procedure to each measure drawn from the Decennial Time Use Index (DTUI). Results are presented as derived evidence rather than as direct quotations from the indicator. This approach permits a concise assessment of structural movement while preserving the source as the principal basis for the analysis.\n\nMention: \"Decennial Time Use Index (DTUI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the 2019 DTUI indicator to calculate a comparative average and trend in time allocation.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:75db67d10f801ca0c891a016", "occurrence_id": "50d64423f95d7d87592884de", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Audited Unemployment Insurance Index is used as the principal indicator for Cambodia in this assessment. Its reported values provide the central quantitative basis for the observation, with 6.8 percent and 4.2 percent reproduced directly from the source. The figures are retained without recalculation or adjustment, allowing the cited evidence to remain traceable to the underlying index.\n\nThe first value indicates the higher observed level within the comparison presented in this note, while the second records the lower level. Together, they establish the numerical range relevant to the discussion of unemployment insurance coverage and support a cautious interpretation of the recorded position. No additional estimation is required for the purpose of this assessment.\n\nThe source is therefore treated as a reference indicator rather than as a basis for constructing a separate measure. The two values are presented in their original form to preserve comparability with the source and to avoid introducing assumptions about definitions, classification, or measurement. Interpretation remains limited to the evidence represented by the cited figures and does not extend beyond the scope of the index.\n\nMention: \"Audited Unemployment Insurance Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Audited Unemployment Insurance Index for Cambodia and takes the figures 6.8 percent and 4.2 percent directly from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:7b71acba0f8006fdc0d5a17a", "occurrence_id": "d78225240ee04f6910a92d1e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Standard Adaptation Finance System is treated as an administrative source for 2015. It provides the evidentiary basis for the values reported in this section and is used without supplementation from another institutional record. The source anchors the numerical description of recorded activity and maintains a clear distinction between reported amounts and subsequent interpretation.\n\nThis assessment reproduces two figures directly from that source, namely 31 approved operations and 18.4 million in recorded commitments. The figures are presented as reported values rather than derived estimates. Their direct use permits the assessment to identify the documented scale of activity while limiting interpretation to what is supported by the cited record.\n\nThe first figure indicates the number of operations entered in the source for the stated period. The second expresses the associated commitments in the unit used for reporting. Read together, the values provide a concise measure of activity and financial volume. They are retained in their reported form to support verification and to avoid implying a calculation that is not contained in the source.\n\nThis treatment also supports consistency across the assessment text and its supporting tables. Any comparison or explanation based on these values should therefore recognise that the evidence consists of two directly reproduced observations from the Standard Adaptation Finance System, rather than an independently constructed series.\n\nMention: \"Standard Adaptation Finance System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Standard Adaptation Finance System for 2015 to report 31 approved operations and 18.4 million in recorded commitments.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:f5fa60e7620551de25040659", "occurrence_id": "15d50b3ef8f1360af83f4180", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment presents an estimate of transport network coverage for Zanzibar and considers the result alongside evidence on route continuity, access, and service availability. The estimate is used to support the interpretation of observed differences between connected and less connected areas.\n\nThe Cross-Sectoral Transport Network Registry (CTNR) is used as a database covering Zanzibar. It provides a structured reference for network links, route classifications, and recorded service coverage. These records are used to cross-check the estimate derived from a separate source, with attention to discrepancies in the identification of connected areas and the reported extent of the network.\n\nThe comparison does not replace the primary estimate. It provides an independent consistency check and helps identify cases in which coverage may be overstated or understated because of differing definitions, incomplete reporting, or variation in the treatment of minor links. The registry evidence supports the reported claim that transport connectivity is uneven across Zanzibar and that the measured extent of the network should be interpreted with caution.\n\nThe results are therefore retained for analytical use, subject to the stated limitations. Differences between the two sources are treated as a reason for qualification rather than as evidence of a change in underlying conditions.\n\nMention: \"Cross-Sectoral Transport Network Registry (CTNR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the CTNR database to cross-check an independent estimate of transport network coverage in Zanzibar.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:a06124bc05d2b547305208b5", "occurrence_id": "065fbc2b2b865f62d83b480c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The data source used for the territorial baseline is the Provincial Ward Outline Census (PWOC), a census covering Rio de Janeiro. Its designation and geographic scope are retained as stated in the source, with no adjustment to the underlying categories or reported units.\n\nFor the purposes of this assessment, the source records 31 wards within the covered area. This figure is reproduced directly to establish the scale of the administrative frame used in the analysis. The reported count is treated as an observation from the source rather than as a derived estimate.\n\nThe same source also records 68.4 percent for the corresponding share of mapped units. This percentage is quoted directly and is presented alongside the ward count to indicate the distribution reflected in the census. No interpolation, recalculation, or comparison is applied to either figure.\n\nTogether, these two observations provide the principal numerical basis for the territorial description. They support a concise reading of the coverage represented in the source while preserving the distinction between the reported count and the reported percentage. Further interpretation is limited to the information contained in the Provincial Ward Outline Census (PWOC).\n\nMention: \"Provincial Ward Outline Census (PWOC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported values of 31 wards and 68.4 percent from the Provincial Ward Outline Census (PWOC) covering Rio de Janeiro.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:a511f48a4f89d568fbee94e1", "occurrence_id": "390e62e7378c80db61c04ca1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Quarterly Access and Documentation System is an administrative source produced by the Government of India for 2022. It provides the principal documentary basis for reviewing reported access and the completeness of supporting documentation. The material is treated as a structured record for the assessment rather than as a narrative account, and its institutional provenance supports consistent interpretation across the analysis.\n\nAn estimate derived from a separate source is cross-checked against the records in the Quarterly Access and Documentation System. The comparison focuses on whether the estimated level is compatible with documented access and whether observed gaps in documentation could explain differences between the two measures. The administrative records therefore serve as the central evidence for validating the estimate, while the separate estimate provides an independent point of comparison.\n\nThe assessment retains the reported distinctions between access status and documentation status and does not infer changes beyond the information recorded. Where the two sources diverge, the result is reported as a measurement difference requiring interpretation rather than as evidence of a substantive shift. This approach improves traceability and limits unsupported conclusions in the implementation assessment.\n\nMention: \"Quarterly Access and Documentation System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The assessment uses the Quarterly Access and Documentation System to cross-check an independent estimate against documented access and documentation status.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:734e4293b196c1f51c08a45c", "occurrence_id": "aa59365f233b047115bc3cf2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Pilot Unemployment Insurance Index as an indicator produced by Burgiss, the organisation responsible for its compilation. The source provides a structured basis for reviewing the relative position of unemployment insurance systems and for checking whether the direction of the reported result is consistent with the supporting evidence. It is used as a reference within the analytical discussion and not as a substitute for the underlying estimate.\n\nThe estimate presented in the broader paragraph is drawn from a separate source. The Pilot Unemployment Insurance Index is used to cross-check that estimate, with attention to whether both measures support the claim concerning the observed pattern. This comparison provides a limited validation of the finding while recognising that the two sources may reflect different measurement approaches and coverage conventions.\n\nThe cited indicator supports one claim in the paragraph concerning the comparative strength of unemployment insurance arrangements. It does not determine the overall assessment, and no inference is made beyond the relationship required for the cross-check. The result is therefore treated as corroborating evidence, alongside the primary estimate, rather than as an independent replacement measure. This use is consistent with an implementation report approach in which supplementary indicators are applied to test the plausibility of a reported result and to identify material divergence requiring further review.\n\nMention: \"Pilot Unemployment Insurance Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The work uses the Pilot Unemployment Insurance Index, an indicator produced by Burgiss, to cross-check a separate estimate and support one claim in the broader assessment.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:dcff4755416d3c640fcb5716", "occurrence_id": "0a013138ab484333abb554fd", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Government Market Access Report (GMAR) is the report used for the market access assessment in Sierra Leone. It provides a reference point for reviewing the recorded structure of access conditions and for identifying the scale of coverage reflected in the available evidence. No adjustment to the cited observations is made in this assessment.\n\nThe assessment quotes two numerical observations directly from the report. The first records 31 grades within the covered classification, while the second places the corresponding share at 18 percent. These figures are presented as reported values and are not recalculated, combined with other observations, or interpreted as a time series.\n\nTaken together, the figures indicate a measurable distribution across the recorded grades and a limited but identifiable share within the assessed coverage. Their use is restricted to describing the evidence contained in the source. The figures should therefore be read as administrative reference values for the assessment rather than as estimates generated through an independent sampling procedure.\n\nThe GMAR evidence is retained in its original numerical form to support consistency between the narrative assessment and the underlying report. Any subsequent comparison should preserve the same definitions and coverage applied to the two cited observations.\n\nMention: \"Government Market Access Report (GMAR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the GMAR figures of 31 grades and 18 percent as direct evidence from the report.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:00e80cc7e425c2bc194e359a", "occurrence_id": "4fe78bbe8218b1d559be6409", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Panel Debt and Savings Registry is a database produced by the Central Bank for India and referring to 2024. It provides the principal observed basis for the review of household debt and savings positions undertaken in this assessment. The material is treated as an administrative statistical record, with entries read at the level required for comparison and consistency checks.\n\nAn estimate obtained from a separate statistical source is cross-checked against the corresponding measures in the registry. This comparison is used to identify departures in level, coverage, and direction, rather than to replace the estimate. Where the two sources align, the registry strengthens confidence in the reported result; where they diverge, the difference is retained for qualification and further review.\n\nThe cross-check gives particular weight to the registry because its observations provide a structured basis for examining the relationship between outstanding obligations and recorded savings. The assessment therefore uses the source to test whether the independently derived estimate is plausible within the observed distribution. Results are interpreted conservatively, recognising that differences may reflect scope, timing, or reporting practices.\n\nFor implementation purposes, the registry serves as the central reference for this validation exercise. Its figures are incorporated into the evidence trail supporting the estimate, while unresolved discrepancies are noted in the assessment record and are not converted into additional assumptions.\n\nMention: \"Panel Debt and Savings Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the 2024 India database produced by the Central Bank to cross-check an independent estimate of household debt and savings.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:ed9f5c42531bc5c1ebc2297f", "occurrence_id": "c32783eee044cea72d6f46dd", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Clustered Student Transport Index (CSTI), an indicator used to support the review of student transport provision. The source is treated as a reference measure for comparing the distribution of services across the assessed groupings. Its use is limited to the figures reported in the source, without further transformation or extrapolation.\n\nThe assessment reproduces two recorded totals from the indicator. It reports 31 routes serving grouped schools and 18 routes serving dispersed schools. These figures are taken directly from the Clustered Student Transport Index (CSTI) and are presented as evidence that transport provision is uneven across the assessed groupings. The comparison supports the broader finding that service coverage varies materially according to the structure of demand.\n\nThe figures are interpreted descriptively and do not establish the causes of the observed difference. They provide a concise basis for assessing the distribution of transport services and for identifying areas requiring closer operational review. No additional calculation is applied to the cited values, and the indicator is not used to infer outcomes beyond the coverage pattern shown in the source.\n\nMention: \"Clustered Student Transport Index (CSTI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Clustered Student Transport Index (CSTI) indicator to report 31 routes and 18 routes and support a finding of uneven transport provision.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:f635196e7000881b21d1f0f5", "occurrence_id": "95b1b70542d39983c48c473d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Representative Transport Network Registry is a database produced by SECEX and referring to 2019. It provides the principal observed basis for the transport network measures used in this assessment. The registry is treated as a reference record rather than as a narrative account, allowing reported units and classifications to be read consistently across the analysis.\n\nAn estimate derived from a separate source was cross-checked against the registry. Agreement was assessed at the level of total network coverage and the distribution of recorded network elements. Where the estimate differed, the registry served as the controlling reference for checking transcription, classification, and aggregation. This procedure reduced the risk that an apparent difference reflected source treatment rather than a substantive difference in the underlying network.\n\nThe resulting figures are therefore grounded in the observed entries of the Representative Transport Network Registry, while the separate estimate is retained as a comparative measure. The comparison supports the reliability assessment without implying that the two sources have identical coverage or classification practices. Any residual discrepancy is interpreted cautiously and is not treated as evidence of change unless supported by the underlying records.\n\nMention: \"Representative Transport Network Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the 2019 Representative Transport Network Registry, a SECEX database, to cross-check an independent estimate of network coverage and composition.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:9b4fd996c392f4e4bdaae2bb", "occurrence_id": "c901b599b471cecd414988f5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Analytical Emission Inventory System (AEIS) is used as the administrative source for 2017. The record provides the principal quantitative basis for the assessment and is treated as the authoritative reference for the reported emissions figures. No supplementary source is introduced for the values presented in this section.\n\nTwo figures are reproduced directly from the AEIS record. Reported emissions amount to 12.4 thousand tonnes in the first measure and 8.7 thousand tonnes in the second measure. These values are retained without recalculation so that the assessment remains traceable to the cited source.\n\nThe figures support the interpretation of the emissions profile used in the implementation review. Their presentation is intended to distinguish the underlying reported quantities from subsequent analytical commentary, including comparisons, classifications, and implications for monitoring. Any interpretation should therefore be read alongside the two source values and should not be taken as a replacement for the administrative record.\n\nMention: \"Analytical Emission Inventory System (AEIS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the figures 12.4 thousand tonnes and 8.7 thousand tonnes taken directly from AEIS for 2017.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:8607c688984d0e8bf5fe8486", "occurrence_id": "b0b8c8d8a94b58184882743f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Coverage of occupational safety indicators was reviewed against the High-Resolution Occupational Safety System, an administrative source referring to 2017. The source is treated as the principal evidence for the narrow set of observations presented in this note. Its entries provide a consistent basis for reporting recorded events and completed supervisory actions without introducing estimates or adjustments.\n\nTwo figures are reproduced directly from the system. It reports 84 incidents and 67 inspections for the reference period. These values are presented as recorded totals, rather than as rates, projections, or results derived through comparison with another dataset. The distinction is material because the figures describe the administrative record available for review and do not establish the incidence of unreported events.\n\nThe 84 incidents indicate the number of cases entered in the system, while the 67 inspections indicate the number of completed inspection records. Read together, the figures support a limited description of documented activity. They should therefore be used for monitoring the completeness and operational use of the reporting mechanism, with interpretation confined to the observations contained in the source. No inference is made regarding conditions outside the recorded administrative coverage.\n\nMention: \"High-Resolution Occupational Safety System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the reported totals of 84 incidents and 67 inspections directly from the High-Resolution Occupational Safety System for 2017.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:5cbf3597a7a077edfc60fb98", "occurrence_id": "8db95e2546af32d10d85394e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Municipal Child Wellbeing Survey, a survey produced by the General Statistics Office, provides the central empirical basis for this assessment. Its observations are treated as evidence of changing child wellbeing conditions rather than as isolated reported figures. The analysis uses the source to derive comparable measures across the observed sequence and to establish the direction and magnitude of change.\n\nAverage levels are calculated from the survey observations after applying a consistent treatment to the available responses. These averages provide the principal measure used in the assessment, while the resulting trend indicates whether conditions have improved, deteriorated, or remained broadly stable. The calculations are intended to reduce the influence of individual observations and to support interpretation of the overall pattern.\n\nThe resulting estimates are used in the implementation assessment and in the prioritisation of follow-up actions. No inference is drawn from a single response or from a single reported value. Attention is instead given to the movement of the calculated average and to the persistence of that movement across the observations. This approach maintains a clear distinction between the underlying survey evidence and the analytical results derived from it.\n\nMention: \"Municipal Child Wellbeing Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The assessment derives average levels and trends from the Municipal Child Wellbeing Survey rather than quoting individual observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:0659c681980487f607be1d74", "occurrence_id": "e3d277237ad089c52d7b8b0d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Comparative Primary Care Survey is treated as the principal empirical input for the assessment. It is a 2023 survey produced by PISA, and its observations are used to construct comparable measures of primary care performance. The analysis does not reproduce individual results as isolated findings. Instead, it derives a central tendency and examines the direction of change across the observations reported in the source.\n\nFor each measure, the available responses are aligned and reviewed for consistency before calculation. The resulting averages provide the basis for interpreting service coverage, reported access, and continuity of care. Where observations differ, the analysis retains the underlying variation and uses the mean only as a summary measure. This approach limits the influence of any single response and supports a more stable assessment of overall performance.\n\nThe resulting trend is presented as an analytical estimate derived from the Comparative Primary Care Survey rather than as a direct quotation of its reported figures. Findings are therefore interpreted with attention to the breadth of the observations and to the limits of aggregation. The calculations support monitoring of relative movement and provide a consistent reference for subsequent implementation review. No separate institutional or geographic attribution is introduced, and the source remains the sole evidentiary basis for the estimates presented in this section.\n\nMention: \"Comparative Primary Care Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives a trend and average from the 2023 Comparative Primary Care Survey rather than reproducing individual observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:baef68ba9409d0f0b87e4814", "occurrence_id": "032b33f8d4ffed18cb7fafb9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Analytical Fuel and Energy Use Survey (AFEUS) is treated as the principal survey for assessing patterns in fuel and energy use. Its observations provide the central evidence for the assessment, while the accompanying interpretation remains limited to measures that can be derived consistently across the reported records.\n\nThe analysis calculates average use across the available observations and compares those averages in their recorded sequence. This approach indicates the direction and persistence of change without presenting any individual observation as representative of the full distribution. The resulting trend is therefore based on aggregation rather than on selected figures.\n\nThe calculated average shows a gradual movement in overall use, with differences between observations contributing to a moderate but discernible trend. Where variation is wider, the average is given greater weight than isolated changes, since the purpose is to identify the general pattern rather than short-term fluctuation. The evidence supports a cautious assessment of changing demand and does not establish a causal explanation.\n\nFor implementation purposes, the derived trend is used to inform monitoring and to provide a consistent basis for comparison in subsequent reporting. The AFEUS survey is consequently applied as an analytical input, with conclusions drawn from the computed averages and their direction over the observation sequence rather than from direct reproduction of source values.\n\nMention: \"Analytical Fuel and Energy Use Survey (AFEUS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses AFEUS to compute average fuel and energy use and derive a cautious trend for monitoring.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:d5817f3cee3d08c168ed4a69", "occurrence_id": "2d06b528c34f71a053770bb0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Data sources.** The Harmonised Hiring and Vacancies System is used as an administrative source produced by Small Arms Survey for 2019. Its recorded observations provide a structured reference for the employment and vacancy measures considered in the assessment, with definitions that support comparison across the relevant reporting categories.\n\nThe estimate presented in the analysis is cross-checked against the corresponding information in this system. The comparison is intended to test the plausibility of one claim concerning the observed level and distribution of vacancies, rather than to replace the estimate or extend the scope of the analysis. Differences are reviewed with attention to coverage, timing, and the treatment of incomplete records.\n\nThe result is used as a supporting verification within the broader paragraph. Where the two measures are consistent, the estimate is retained for interpretation. Where discrepancies arise, the administrative record is treated as an indication that the magnitude should be read with caution, without implying that either source provides a complete account of the underlying conditions. This use preserves the distinction between the principal estimate and the independent reference used for validation.\n\nMention: \"Harmonised Hiring and Vacancies System\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the 2019 administrative record produced by Small Arms Survey to cross-check a separate estimate.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:d57cb73bc0926d68cea60b27", "occurrence_id": "b872872047e608d3c95e46cf", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Local Early Grade Reading System, which OECD-PISA produces, is treated in this assessment as an administrative source for the measurement of early reading provision and performance. Its entries are used as recorded, without adjustment, inference, or supplementary classification. The source is therefore understood to provide the principal numerical basis for the observation reported below.\n\nTwo entries are reproduced directly from the system. The first records 31 grades within the assessed structure. The second records 68 percent of learners meeting the stated reading criterion. These figures are presented in their original form and constitute the central evidence for the assessment of coverage and attainment.\n\nThe reported distribution indicates that the system captures both the scale of the instructional structure and the proportion of learners reaching the specified benchmark. The two measures should not be interpreted as a calculated relationship, since no additional computation is applied in this note. Their value lies in providing a concise representation of the conditions documented in the source.\n\nFor implementation purposes, the findings support continued use of the system as a reference for monitoring early reading results. Subsequent review may examine changes in the recorded measures, but the present assessment remains limited to the two figures reproduced above and does not extend beyond the information contained in the Local Early Grade Reading System.\n\nMention: \"Local Early Grade Reading System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported figures of 31 grades and 68 percent from the Local Early Grade Reading System.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:ca66e20c972cf919755a7de9", "occurrence_id": "5ad22633cb55a69373868686", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The District Protracted Situation Survey (DPSS), a survey covering Sud and Nippes and referring to 2018, provides the principal empirical basis for the comparison presented below. Its district-level observations are treated as a consistent set of inputs for descriptive analysis rather than as isolated reported figures.\n\nThe assessment calculates average conditions across the covered area and derives a trend from the distribution of observations recorded in the DPSS. This procedure reduces the influence of individual entries and permits comparison of the general direction and magnitude of change within the survey evidence. The resulting measures are therefore analytical outputs generated from the source, not figures reproduced from a separate tabulation.\n\nThe computed average is used to establish the central level observed in 2018, while the calculated trend indicates whether conditions move in a common direction across Sud and Nippes. Differences between observations are retained in the calculation, but interpretation gives priority to the aggregate pattern. This approach supports a concise assessment of prevailing conditions while preserving the evidentiary role of the District Protracted Situation Survey (DPSS).\n\nThe findings should consequently be read as summary measures derived from the survey record. They describe the pattern obtained through calculation and provide a basis for interpreting the relative position of the covered districts within the available evidence.\n\nMention: \"District Protracted Situation Survey (DPSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the 2018 DPSS observations for Sud and Nippes to calculate an aggregate average and comparative trend.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:a4f537667ad769dc69a6353d", "occurrence_id": "68f0a91ff2178f50735ee006", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment relies on the Integrated Informal Sector System (IISS) as an administrative source covering 2022. The records provide the central evidence for examining observed levels of informal activity and for establishing a consistent basis for comparison across the reporting series. Their use is limited to the information contained in the source, with no adjustment for factors that cannot be identified in the underlying records.\n\nIndividual observations are not reproduced as reported values. Instead, the assessment aggregates the available entries and derives an average level for the period, together with a trend based on the ordering of observations within the records. This approach reduces the influence of isolated entries and supports a more stable interpretation of changes over time. The resulting measures are therefore analytical outputs calculated from the IISS material rather than direct quotations from it.\n\nThe calculated average is used to indicate the general scale of activity, while the derived trend provides an indication of direction and relative movement. These measures are presented as summary evidence and should not be interpreted as a complete account of all informal activity. Any differences in recording practice or coverage may affect the observed pattern, but the source remains sufficient for the comparative assessment undertaken here.\n\nMention: \"Integrated Informal Sector System (IISS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The assessment derives an average and a trend from the 2022 administrative records in the Integrated Informal Sector System (IISS).\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:cd3091d5bc147994c356c630", "occurrence_id": "0c96e63cda443823ac9f039b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The evidence base for the membership assessment is the Audited Cooperative Membership Census (ACMC), a census covering Ukraine and referring to 2017. The source is treated as the authoritative numerical reference for the descriptive statements in this section. Its scope is limited to recorded cooperative membership and the associated cooperative count, and no adjustment is applied to the reported totals.\n\nTwo figures are taken directly from the ACMC. It records 31,482 members and 1,126 cooperatives. These values are reproduced as reported, without interpolation, reclassification, or conversion into rates. They provide the central quantitative basis for the discussion of the scale of cooperative participation.\n\nThe figures should therefore be read as observations from the cited census rather than as estimates generated for this assessment. Their use supports a clear account of the recorded membership structure while preserving the source’s original units. Any interpretation of coverage or representativeness remains separate from the numerical statements themselves.\n\nMention: \"Audited Cooperative Membership Census (ACMC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the ACMC figures of 31,482 members and 1,126 cooperatives.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:d8ab89531b2055f10a93327e", "occurrence_id": "8ffbf7938f9cb282c8c57176", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Data sources.** The Diagnostic Immunisation Coverage Registry (DICR) database is the reference source for the coverage observations used in this assessment. Two values are reproduced directly from the database, with 86 percent recorded for completed schedules and 74 percent for timely initiation. The figures are retained without recalculation and provide the quantitative basis for the interpretation below.\n\nThe observed results indicate that continuity across the immunisation sequence remains incomplete. The difference between completion and timely initiation is consistent with a loss of coverage between the first contact and the final required contact. This pattern supports a broader finding that service utilisation is not translated uniformly into completion, even where initial access has been established.\n\nThe cited values are used as evidence for this finding rather than as a basis for constructing an additional series. They should therefore be read as fixed observations supporting the assessment of implementation performance. Other elements of the analysis are presented descriptively and do not alter the two figures drawn from the DICR database.\n\nMention: \"Diagnostic Immunisation Coverage Registry (DICR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage takes the 86 percent and 74 percent coverage figures directly from the Diagnostic Immunisation Coverage Registry (DICR) database.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:18f0f84befb361738744c2ab", "occurrence_id": "060568840446967b386436c6", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The evidence base is the Comparative Socioeconomic Survey (CSS), a survey produced by Frontex for 2023. The source is used as the principal statistical basis for the assessment, with its stated observations retained as the reference set for subsequent calculations. No additional source is introduced for the measures discussed below.\n\nThe analysis converts the observations into a comparable sequence and derives an average for the assessment period represented in the survey. It also calculates a trend across the ordered observations, thereby indicating the direction and relative stability of change rather than reproducing isolated values. This treatment provides a summary measure that is less sensitive to individual departures and permits the results to be reviewed against the common reference established by the CSS.\n\nThe resulting pattern is used to support the assessment of overall socioeconomic movement. The calculated average serves as the central level against which variation is considered, while the derived trend indicates whether the observed position is broadly improving, weakening, or remaining stable. The findings should therefore be interpreted as analytical results generated from the Comparative Socioeconomic Survey (CSS), rather than as figures quoted directly from the source. This distinction is material for interpretation, since the reported conclusions reflect aggregation and calculation applied to the survey observations.\n\nMention: \"Comparative Socioeconomic Survey (CSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage derives an average and trend from the 2023 Comparative Socioeconomic Survey (CSS) produced by Frontex.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:d30915f9baa315a637bdbbcb", "occurrence_id": "1ab7d8cd955134437740f230", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Quality-Assured Nutrition Surveillance System (QNSS), an administrative source for 2024. The source is used as the principal evidentiary basis for the nutrition status described below, with no adjustment to the reported values.\n\nTwo figures are taken directly from the QNSS record. It reports that 31.4 percent of children met the minimum dietary threshold, while 18.7 percent received a qualified nutrition contact. These values are reproduced as stated in the source and are not derived through an independent calculation.\n\nThe figures indicate that adequate dietary intake was more common than documented contact with nutrition services during the period covered by the source. This difference is material for implementation monitoring because it points to a gap between observed nutritional conditions and service interaction. The findings should therefore be read as a descriptive account of recorded observations rather than as an estimate of change over time.\n\nThe QNSS evidence provides a consistent reference for interpreting the reported coverage pattern. Its administrative character permits direct citation of the two figures while maintaining a clear distinction between source values and subsequent assessment commentary.\n\nMention: \"Quality-Assured Nutrition Surveillance System (QNSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported QNSS values of 31.4 percent and 18.7 percent and interprets their difference descriptively.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:594748b5a63250a781510a43", "occurrence_id": "bef47d671811c7d1bb9143c9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Data sources.** The Enumerative Population Grid Census is treated as the principal population reference for this assessment. It is a census produced by CGAP for 2021 and is used here without adjustment to the reported totals. The source provides a consistent basis for describing the enumerated population and household coverage within the analytical framework.\n\nTwo figures are reproduced directly from the Enumerative Population Grid Census. The reported population is 31,742 persons, while the corresponding household count is 8,416 households. These values are presented as recorded in the source and are not recalculated from subsidiary indicators or derived estimates.\n\nThe figures support the assessment of coverage and provide a fixed reference for subsequent interpretation. Their use is limited to establishing the magnitude recorded by the census. No inference is made regarding change over time, and no additional demographic measure is generated from the cited material.\n\nFor reporting purposes, the source values are retained in their original numerical form. This treatment preserves traceability between the evidence used in the assessment and the underlying statistical record.\n\nMention: \"Enumerative Population Grid Census\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the figures 31,742 persons and 8,416 households directly from the Enumerative Population Grid Census.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:e6269350e62bc9120f2120f6", "occurrence_id": "75851781881fd15a0aaf50c0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Independent Seasonal Forecast Collection is retained as an other source produced by SEEG for 2022. It provides the central evidence for the seasonal information reviewed under this assessment. The material is treated as a documentary input rather than as a basis for generating new estimates or assigning additional classifications.\n\nThe assessment team collected the available records, checked their internal consistency, and harmonised the presentation of the observations before inclusion in the working archive. Differences in wording and structure were regularised to support comparison across the retained entries, while the substantive content was preserved. This process provides a consistent evidentiary trail from the source material to the analytical tables.\n\nThe archived collection is used to document the information available at the time of review and to support verification of the reported seasonal signals. No extrapolation is applied beyond the material retained in the collection. The resulting record is therefore suitable for reference in implementation monitoring, subject to the limits of the source and the scope of the supporting documentation.\n\nMention: \"Independent Seasonal Forecast Collection\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The assessment collected, harmonised, and archived the Independent Seasonal Forecast Collection produced by SEEG for 2022 as an other source.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:37098f7ba1244a796478abdc", "occurrence_id": "7dada35da7f53c118a95e2b1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment considers short-term rainfall irregularity as a material factor in interpreting observed service conditions. The Rapid Rainfall Anomaly Report (RRAR), a report used for this assessment, provides the principal reference for the rainfall indicator applied in the analysis. Its terminology is retained to support consistency across the results discussion.\n\nTwo figures are reproduced directly from the report. The first records an anomaly of 18 percent, while the second records an anomaly of 27 percent. These values are presented without recalculation and are used only to support the broader finding that rainfall variability affected the reliability of reported outcomes during the assessment period.\n\nThe figures do not constitute a separate estimate or a substitute for the wider evidence base. They provide a limited quantitative basis for interpreting the pattern described in the surrounding analysis, particularly where changes in reported conditions coincide with marked departures from expected rainfall. The assessment therefore treats the RRAR evidence as corroborative rather than determinative.\n\nThis use of the source preserves the distinction between directly reported observations and conclusions drawn from their interpretation. No additional adjustment is applied to either figure, and the values remain attributable to the Rapid Rainfall Anomaly Report (RRAR) throughout the results section.\n\nMention: \"Rapid Rainfall Anomaly Report (RRAR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the RRAR figures of 18 percent and 27 percent to support the assessment of rainfall variability and reported outcomes.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:b18246a8448594f11fb45945", "occurrence_id": "6621fa9ead7a4c288bd3bb6d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis draws on the Endline Agri-Food Prices Survey, a survey designed to provide structured observations on prices across the monitored agri-food items. The source is treated as the central empirical basis for the assessment, while no additional institutional or geographic attribution is required for the interpretation presented here. Its records support a consistent comparison of observed prices and provide sufficient coverage for aggregation.\n\nFor analytical purposes, the observations are grouped by item and reporting interval, after which arithmetic averages are calculated to reduce the influence of isolated movements. The resulting series is used to derive a general trend in agri-food prices rather than to reproduce individual survey entries. This approach gives greater weight to the direction and persistence of change than to any single reported value.\n\nThe computed averages indicate the underlying movement in prices over the analytical sequence, including periods of relative stability and intervals of wider variation. Differences between items are retained in the calculations so that the aggregate result does not obscure divergent patterns. The estimates are therefore interpreted as summary measures derived from the Endline Agri-Food Prices Survey, not as direct quotations from it.\n\nThese results inform the assessment of price conditions and provide a reproducible basis for monitoring. They should be read as analytical estimates generated from the survey observations, with the calculated trend serving as the principal evidence for the conclusions in this section.\n\nMention: \"Endline Agri-Food Prices Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the Endline Agri-Food Prices Survey to calculate item-level averages and derive an aggregate trend in agri-food prices.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:539be7d6dc31d323b5520f2a", "occurrence_id": "65034808b217cab72e63305f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Follow-Up Household Income and Expenditure Survey (FHIES) is used as the survey for the household-level evidence presented in this assessment. Its reported results provide the principal quantitative basis for the discussion, while the surrounding text limits interpretation to the observations supported by the source.\n\nTwo values are reproduced directly from FHIES. The survey records 64.2 percent for households meeting the stated expenditure measure and 58.7 percent for households meeting the corresponding income measure. These figures are presented as reported, without recalculation, adjustment, or derivation of an additional indicator.\n\nThe evidence indicates a material difference between the two reported shares, but the difference should not be treated as a time-based movement or attributed to factors not measured in the survey. The figures are retained in their original percentage form to support verification and consistent use in the assessment.\n\nFurther analysis may examine the distribution of these results across relevant household categories, subject to the limits of the available observations. For the present purpose, the two quoted values constitute the central evidence drawn from the Follow-Up Household Income and Expenditure Survey (FHIES).\n\nMention: \"Follow-Up Household Income and Expenditure Survey (FHIES)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported values of 64.2 percent and 58.7 percent from the Follow-Up Household Income and Expenditure Survey (FHIES).\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:eae3f96b86d826997fcb5e62", "occurrence_id": "c7667e5e465603df15127f98", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Representative Transport Network Collection is classified as other and refers to 2024. It provides the principal empirical basis for this assessment, with structured observations on transport links and associated characteristics. The material is used to examine the distribution of network conditions and to support a consistent statistical treatment of the available records.\n\nThe assessment does not reproduce individual observations as reported values. Instead, the records are screened for internal consistency and aggregated to calculate average network measures. A directional trend is also derived by arranging the observations according to the analytical sequence applied in the assessment. This approach gives greater weight to the overall pattern than to any single entry and reduces the influence of isolated observations.\n\nThe resulting averages and trend estimates are used to describe the central tendency, dispersion, and relative concentration of the network characteristics. They inform the interpretation of implementation conditions and provide a reproducible basis for subsequent monitoring. The findings should therefore be understood as calculations generated from the collection rather than as figures quoted directly from it. Any variation within the underlying records remains relevant to interpretation, particularly where the calculated average may conceal substantial differences among observations.\n\nMention: \"Representative Transport Network Collection\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The assessment uses the 2024 Representative Transport Network Collection, classified as other, to calculate average network measures and derive a directional trend.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:1fc463b86b452c0bbe4f0d83", "occurrence_id": "ca592bd5a31213045fa255e1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Biennial Household Income and Expenditure Collection, identified as the Third Phase and classified as other supporting material. The underlying records were collected through the designated reporting process, reviewed for internal consistency, and harmonised to support comparison across the analytical tables. The resulting files were archived with the working documentation so that the derivation of the reported measures could be checked during implementation review.\n\nThis treatment supports the assessment finding that household expenditure measures are sufficiently coherent for use in the distributional analysis. Differences in recorded values are therefore interpreted with attention to coverage and reporting practices rather than treated as evidence of immediate structural change. The harmonised material also provides a consistent basis for checking aggregate totals against the tabulations presented in the main text.\n\nQuality assurance focused on preserving the substantive content of the collection while correcting inconsistent formats and removing duplicative entries. No adjustments were introduced solely to improve the apparent direction of the results. The archived version retains the information needed to reconstruct the calculations and to repeat the checks undertaken for this assessment. Taken together, these procedures strengthen the reliability of the cited evidence and clarify the limits of inference drawn from it.\n\nMention: \"Biennial Household Income and Expenditure Collection\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The passage identifies the Biennial Household Income and Expenditure Collection as Third Phase and other material and records its collection, harmonisation, and archiving.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:1871bbfc8886f3553d73cfe2", "occurrence_id": "2a611f6385b98079d3b11688", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Data sources.** The Clustered Road Segment Dataset (CRSD) is used as the principal geospatial source for the assessment of clustered road segments. Its structure permits a consistent reading of segment groupings and supports the descriptive account presented in this annex. No additional transformation is applied to the two figures reported below.\n\nTwo figures are reproduced directly from CRSD. The reviewed extract records 31 road clusters and 68 road segments. These values provide the central quantitative evidence for the assessment and establish the scale of the network represented in the dataset. They are presented as reported in the source, without extrapolation or recalculation.\n\nThe figures should be interpreted as descriptive measures of the observations contained in the dataset. They support comparison across the analytical tables and provide a fixed reference for subsequent discussion of coverage, grouping, and segment distribution. Their use is limited to the scope of the cited source and does not imply completeness beyond the records represented there.\n\nThe citation is retained with the evidence to ensure traceability in future review. In particular, the stated counts remain linked to the Clustered Road Segment Dataset (CRSD), rather than to a derived indicator or an independent estimate. This treatment is consistent with the reporting requirements for a statistical note and preserves the distinction between directly reported values and later analytical interpretation.\n\nMention: \"Clustered Road Segment Dataset (CRSD)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The author takes the directly reported figures of 31 road clusters and 68 road segments from the Clustered Road Segment Dataset (CRSD).\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:34c11c50adbb73dee722fb64", "occurrence_id": "9b31575a584cbfeb40b8cfc6", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Longitudinal Household Income and Expenditure Report (LHIER) report for 2016 as its principal source of household economic observations. The source provides a consistent basis for examining reported income and expenditure across the population covered by the assessment, while preserving the distinction between recorded values and the analytical measures derived from them.\n\nIncome and expenditure observations are arranged for comparison and used to calculate average levels and changes over the available sequence. The analysis therefore presents a trend derived from the LHIER material rather than repeating individual entries from the report. This approach reduces the influence of isolated household outcomes and supports a more stable interpretation of movement in household welfare indicators.\n\nThe resulting averages indicate the direction and scale of change relevant to the assessment, with expenditure considered alongside income to support interpretation of purchasing capacity and resource pressure. No single observation is treated as representative of the full distribution. The calculated measures are used as summary evidence, and their interpretation remains tied to the coverage and measurement conventions of the source. The LHIER report thus serves as the central empirical basis for the assessment, while the reported trends and averages constitute analytical results produced from the records it supplies.\n\nMention: \"Longitudinal Household Income and Expenditure Report (LHIER)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses 2016 LHIER income and expenditure observations to derive average levels and trends rather than reproduce individual reported figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:d22893ba79d0c4069c495718", "occurrence_id": "f2089210251dde2690aff403", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The principal evidence for the reproductive health results is the Audited Reproductive Health Survey, produced by TCC as a survey. The source is treated as the basis for the assessment of reported service conditions and outcome movement. Its use is subject to the ordinary requirements of traceability, internal consistency and review of the underlying observations.\n\nThe analysis does not reproduce isolated observations from the source. Instead, the observations are arranged in their recorded sequence, and an average is calculated to provide a stable measure for comparison. A trend is then derived from the movement of the resulting measures, with attention to persistence rather than to short-term variation. This approach permits the assessment to distinguish a sustained change from an isolated result.\n\nThe resulting estimates are used in the interpretation of implementation performance and in the identification of areas requiring follow-up. They should therefore be read as analytical measures derived from the Audited Reproductive Health Survey, rather than as figures transcribed directly from it. The calculation procedure is applied consistently across the available observations, and the reported conclusion reflects the direction and average level indicated by that procedure.\n\nMention: \"Audited Reproductive Health Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives an average and a trend from the Audited Reproductive Health Survey rather than reproducing isolated observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:55e1e657ffd89b65110d6bcd", "occurrence_id": "018d418aecdd5de3d692e93b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Continuous Transport Network Dataset (CTND) as a geospatial reference associated with 2018. Its role is limited to establishing background understanding of transport connectivity and continuity for interpreting the findings. The citation supports one contextual statement in the surrounding discussion rather than a separate analytical result.\n\nTransport access is considered alongside the institutional and operational factors examined in this section. The cited material provides a general basis for noting that the configuration of connected routes can influence how service coverage and movement constraints are understood. No observation, measurement, or estimate is taken from the dataset for the calculations presented here.\n\nResults reported in the following analysis derive from the assessment framework and its underlying evidence, not from a transcription of the cited dataset. The reference is retained to situate the discussion, clarify the scope of the transport-related statement, and avoid implying that a complete network inventory was undertaken.\n\nAccordingly, the source should be read as supporting context for a single claim within a broader paragraph. It does not supply a series, benchmark, or comparative value, and it is not used to generate a rate, ranking, or trend. This treatment preserves a clear distinction between contextual evidence and the materials used for substantive assessment.\n\nMention: \"Continuous Transport Network Dataset (CTND)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"CTND is used only to support a contextual claim about transport connectivity, with no dataset values incorporated into the analysis.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:8e22bcf4780200cb356b8103", "occurrence_id": "c564a1f7c5c2d09df201cc20", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Endline Socioeconomic System (ESS) is used as the central administrative source for the assessment. Its records provide the basis for summarising household conditions and for examining how measured outcomes vary across the reporting sequence. The source is treated as an evidentiary dataset rather than as a narrative account.\n\nThe analysis calculates average values from the ESS records and derives a trend from the resulting series. This approach reduces the influence of individual observations and supports comparison of the overall direction and level of the indicators. The reported findings therefore reflect transformations performed on the source material, including aggregation and estimation, rather than a reproduction of isolated entries.\n\nInterpretation remains subject to the coverage and consistency of the underlying records. Quality checks address missing observations, inconsistent classifications and unusually large deviations before the averages and trend estimates are produced. The resulting measures are used to support implementation assessment and to identify changes requiring further review. In this way, the ESS provides the principal empirical basis for the paragraph, while the reported statistics represent calculated summaries derived from it.\n\nMention: \"Endline Socioeconomic System (ESS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives averages and a trend from the ESS records rather than reproducing individual entries.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:f2791320368a261a37de242b", "occurrence_id": "45d75fcca28c5dbf824ac77f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment considers settlement distribution and elevation conditions as related dimensions of implementation risk. This framing is used to distinguish operational constraints from variations in service access and to support a cautious interpretation of observed differences across the reporting framework.\n\nThe National Settlement and Elevation Census (NSEC) is cited as a census relevant to this framing. Its inclusion indicates that settlement and elevation characteristics form part of the wider evidence base considered in the assessment, while keeping the discussion at a contextual level.\n\nNo values from the NSEC are incorporated into the analysis presented here. The census does not supply counts, rates, classifications, or measurements for the calculations, and no result in this passage depends on a value drawn from that source. Its role is limited to supporting the claim that physical and settlement conditions should be considered when interpreting implementation patterns.\n\nThe principal findings therefore remain based on the assessment evidence described elsewhere in the report. Reference to the NSEC provides supporting context for one aspect of the interpretation, without altering the reported estimates, thresholds, or conclusions.\n\nMention: \"National Settlement and Elevation Census (NSEC)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The passage uses the National Settlement and Elevation Census (NSEC) only to support contextual interpretation and takes no values from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:384a0721533c09b75fdf8fb9", "occurrence_id": "63b373e3c998629406e608d7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Joint Host Community Registry, a UNICEF database covering Haiti for 2019, provides the documentary basis for the household and population figures used in this assessment. It is treated as the primary statistical record for the profile presented below, with no adjustment applied to the reported totals.\n\nThe registry records 31 households within the assessed host-community caseload. This figure is reproduced directly from the database and is used to establish the scale of household coverage at the reference point. The count is presented as reported, rather than as an estimate derived from other information.\n\nA second figure reproduced directly from the same source is 47 individuals. Taken together, the two reported values indicate that the recorded caseload comprises 31 households and 47 individuals. The relationship between the counts is retained as stated in the source and is not interpreted as a measure of average household size.\n\nThese data provide a concise basis for planning and monitoring. Their use in this annex is limited to describing the recorded caseload, while preserving the source terminology and the numerical values contained in the Joint Host Community Registry.\n\nMention: \"Joint Host Community Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the figures of 31 households and 47 individuals directly from the Joint Host Community Registry.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:4a281e028e053ae9b20a34d2", "occurrence_id": "3513f5aa67cf33ca741b50c3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Metropolitan Household Consumption Survey is the principal survey used for the consumption evidence in this assessment. It was treated as a structured source for household expenditure observations, with attention to coverage, completeness, and internal consistency. The material supports a consolidated view of consumption patterns and provides the evidentiary basis for the statistical review presented in this annex.\n\nThe records were collected from the survey documentation and associated tabulations, then reviewed for consistency across reporting units and periods of observation. Differences in terminology, measurement units, and recording practices were harmonised before the information was incorporated into the analytical file. This process reduced avoidable discrepancies while preserving the distinctions required for interpretation of household consumption.\n\nFollowing review, the harmonised material was archived with its supporting documentation and processing notes. The archive provides a traceable basis for the tables and analytical statements derived from the survey. It also establishes a controlled reference for subsequent verification, allowing revisions to be assessed against the retained records without altering the underlying evidence. No additional source was used to replace observations unavailable in the survey.\n\nMention: \"Metropolitan Household Consumption Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The work taken from the source consisted of collecting, harmonising, and archiving household consumption records for analysis.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:0f416e57aae52575673e0e6b", "occurrence_id": "96579f544f7a8964cad6f873", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The evidence base for the distributional assessment includes the Cross-Sectoral Socioeconomic Census (CSC), identified as the Seventh Release of the census. The CSC was received through the established data intake process and retained as a structured source for subsequent review. Its inclusion permits the assessment to draw on a broad and internally consistent record of socioeconomic conditions.\n\nDuring preparation, the records were checked for completeness, harmonised across common categories, and archived with their original distinctions preserved. Duplicate entries and inconsistent representations were addressed through documented reconciliation procedures, while unresolved cases remained subject to review. These measures support reproducibility and reduce the risk that processing choices materially influence the reported results.\n\nThe resulting evidence supports the conclusion that observed differences are associated with variation in household circumstances rather than with inconsistencies in source treatment. The source is therefore used as supporting evidence within the wider assessment, while interpretation remains aligned with the limitations of a census-based record. Further analytical outputs rely on the harmonised archive and retain the provenance of the CSC material.\n\nMention: \"Cross-Sectoral Socioeconomic Census (CSC)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The passage identifies the Seventh Release of the Cross-Sectoral Socioeconomic Census (CSC) and describes its collection, harmonisation, archiving, and use as supporting evidence.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:80d12beabc23afe6430b00e9", "occurrence_id": "af1426eaa1a40dea01b3acc6", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Representative Inequality Survey (RIS) is the principal survey used for the distributional assessment. It supplies the observations required to examine changes in household outcomes across successive observations. The source is retained in its reported form, with documentation reviewed for coverage, response patterns, and comparability. No single reported observation is treated as sufficient evidence of movement in the indicators.\n\nFor this assessment, the RIS observations are aggregated to obtain comparable averages and to calculate the direction and magnitude of change over the available sequence. The resulting trend is based on repeated computation from the survey records rather than on a quotation of isolated values. Where the number of valid responses differs between indicators, the calculation uses the available responses for the relevant measure and records the resulting limitation in the interpretation.\n\nThe derived averages indicate a persistent but uneven distributional pattern. Changes are assessed against the calculated baseline, while fluctuations in individual observations are not interpreted as structural movement unless they are supported by the broader sequence. This treatment reduces the influence of exceptional responses and keeps the analysis aligned with the evidence contained in the RIS survey. The findings are therefore presented as analytical estimates generated from the source, with attention to sampling variation and incomplete reporting.\n\nMention: \"Representative Inequality Survey (RIS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives comparable averages and a trend in distributional outcomes from the Representative Inequality Survey (RIS).\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:232c25456e7fc023cf10c740", "occurrence_id": "bb4b52d423e7fffd0d9d52de", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Inter-Agency School Feeding Survey (ISFS) provides the principal evidence base for this assessment. The 2022 survey is used to examine the distribution of school-feeding provision across the observed reporting units and to establish a comparable basis for interpretation. Its records are treated as analytical inputs rather than as isolated figures reproduced in the text.\n\nThe assessment calculates an average level of provision from the observations contained in the ISFS and derives a trend by comparing the ordered results across the reporting sequence. This approach reduces the influence of individual high or low observations and supports a more stable reading of overall performance. The resulting measures indicate the general direction and central tendency of provision captured by the survey.\n\nThe analysis also considers the consistency of the reported pattern. Where observations differ, the calculation retains those differences while allowing the average to summarise the broader position. The trend is therefore presented as an analytical result derived from the source, not as a direct quotation of selected survey figures. This treatment supports cautious interpretation of change and avoids attributing undue significance to any single observation.\n\nThe ISFS evidence is consequently used to support the assessment of coverage, continuity, and relative performance. The calculated average and trend provide the basis for the findings reported in this section, while the underlying survey remains the reference point for verification and subsequent review.\n\nMention: \"Inter-Agency School Feeding Survey (ISFS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the 2022 ISFS survey to calculate an average and derive a trend in school-feeding provision.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:b3264f0ec7f6129098ba4345", "occurrence_id": "1502e639fbae2aa49fe841e3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Midline Digital Learning System (MDLS) as its principal administrative source for monitoring participation and delivery indicators. The system provides structured records at the unit level, allowing observations to be aligned across reporting periods and reviewed for completeness. Its use is appropriate for an implementation assessment because the records reflect routine programme transactions rather than a separate perception exercise.\n\nThe analysis derives comparable rates from the available records and calculates averages across units before examining movement between periods. Reported results therefore represent summaries generated from MDLS records, rather than individual figures reproduced from the system. The resulting trend is used to identify whether participation and delivery have strengthened, remained stable, or weakened over the observation sequence.\n\nData quality checks address duplicate entries, incomplete observations, and inconsistent coding within the extracted records. Where coverage differs across units, the calculations retain the common denominator and document the resulting limitation in interpretation. The evidence is consequently most suitable for describing aggregate direction and average performance, while individual records are not treated as independently representative. This approach supports the assessment of implementation progress using a consistent administrative basis and avoids attributing undue weight to isolated observations.\n\nMention: \"Midline Digital Learning System (MDLS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis calculates aggregate averages and trends from Midline Digital Learning System (MDLS) administrative records.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:3c39285137a6cdaa6303d30f", "occurrence_id": "4acdf9f88322180dc79e7132", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Follow-Up Output Area Index indicator covering Houston is used as the supporting data source for the assessment of local output conditions. Its entries provide a consistent basis for examining movement across successive observations while limiting reliance on isolated reported values. The source is retained in the working evidence because it permits comparison within the defined coverage area.\n\nFor the analysis, the observed entries are consolidated into an average and reviewed for the direction and persistence of change. This treatment produces a summary measure rather than a reproduction of individual source figures. The resulting trend is used to support the finding that output conditions changed over the period under review, while recognising that the indicator describes aggregate movement and does not establish the cause of that change.\n\nThe calculated average is therefore interpreted alongside the broader assessment of implementation performance. It provides a compact measure for checking whether the reported pattern is broadly stable, improving, or weakening, and it helps distinguish a sustained movement from a single irregular observation. The evidence is considered sufficient for this limited purpose, subject to the coverage and measurement boundaries inherent in the source.\n\nMention: \"Follow-Up Output Area Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the Follow-Up Output Area Index indicator for Houston to calculate an average and assess the direction of its trend.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:b0b47ff23441d809c1e5905f", "occurrence_id": "a5f8f8052e287c5e4322851e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Decennial Migration and Remittance Registry database produced by US Customs for 2016 as its central evidentiary record. The source is treated as an administrative basis for documenting migration and remittance observations, with the cited material retained in its original documentary relationship. No supplementary institutional designation, geographic attribution, or release classification is applied in this assessment.\n\nThe work comprised the collection of the available registry material, its harmonisation for consistent treatment, and its archiving for subsequent verification. During harmonisation, entries were reviewed for structural consistency, duplicated observations were addressed, and the resulting files were preserved with a clear link to the cited database. These procedures support traceability without introducing additional source identities or unsupported classifications.\n\nThe archived material provides the reference point for the statistical discussion and allows subsequent checks to be undertaken against the recorded observations. Its use is therefore limited to evidence supported by the Decennial Migration and Remittance Registry database produced by US Customs for 2016. The documentation approach prioritises reproducibility, controlled retention, and transparent treatment of the source throughout implementation reporting.\n\nMention: \"Decennial Migration and Remittance Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The passage records the collection, harmonisation, and archiving of the Decennial Migration and Remittance Registry database produced by US Customs for 2016.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:117fdb9d8811ff2099f35a62", "occurrence_id": "8f8bf00d64749bd1528023ec", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Quality-Assured Household Consumption Survey (QHCS) is the principal survey used for the consumption analysis covering Sud and Nippes and referring to 2024. Its household observations provide a consistent basis for examining expenditure patterns across the covered area. The survey is treated as an analytical input for producing summary measures rather than as a source of isolated reported figures.\n\nAverage consumption is calculated from the observations, with the calculation applied across the relevant household records. The resulting mean is used to represent the prevailing level of consumption in the covered area. Changes across the observed distribution are also consolidated into a trend measure, allowing the analysis to identify the direction and general magnitude of movement without relying on individual household values.\n\nThis treatment places the QHCS at the centre of the assessment evidence. The reported results therefore reflect derived statistics generated from the survey records, including the average and the associated trend. Interpretation remains limited to the population and coverage represented in the survey, and the derived measures are presented as analytical estimates rather than as direct quotations from the source.\n\nMention: \"Quality-Assured Household Consumption Survey (QHCS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses the QHCS records for Sud and Nippes in 2024 to calculate average consumption and derive a trend.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:62dfada9d71ec0160850ff8e", "occurrence_id": "f86169c87ec020a38b284663", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment considers the availability of labour-market information as an enabling condition for implementation. The data source is the Sentinel Job Search Survey (SJSS), a survey produced by Goskomstat, the organisation responsible for its production. The source is cited to support the broader observation that regular job-search monitoring can strengthen the interpretation of implementation conditions.\n\nThe cited material is used solely to frame that observation within the discussion of evidence quality and monitoring arrangements. No estimate, rate, count, or other numerical result in this assessment is calculated from the survey, and no finding is attributed to a value drawn from it. Its role is therefore limited to establishing the relevance of systematic information on job-search conditions to implementation review.\n\nThis distinction is material for interpretation. The assessment relies on the survey as a contextual reference while basing its operational conclusions on the evidence described in the surrounding analysis. The citation does not alter the reported results, create an additional indicator, or provide a separate basis for comparison. It supports one claim concerning the usefulness of sustained labour-market observation within a broader paragraph on monitoring and evidence management.\n\nMention: \"Sentinel Job Search Survey (SJSS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The source contributes no numerical values and is used to support a claim about the relevance of regular labour-market monitoring to implementation review.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:93465a0b93cbbc43374aeed6", "occurrence_id": "46eb72358c28a70e27436b2c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Joint Shock Coping Survey, a 2024 survey, is used as the principal empirical input for assessing changes in household coping conditions. The source provides repeated observations on reported responses to economic pressure and supports a consistent basis for comparison across the analytical sample. No individual response is treated as representative of the overall position.\n\nThe assessment derives average reported coping levels and measures the direction of change across the observed sequence. These calculations are used to identify whether conditions improved, deteriorated, or remained broadly stable. The resulting trend is the central evidence for the interpretation of changes in welfare-related behaviour, while variation around the average is retained to avoid overstating the precision of the findings.\n\nThe estimates are presented as analytical results generated from the survey rather than as reproduced survey figures. This approach permits the report to summarise the prevailing pattern while recognising differences among respondents. The calculated averages and associated trend provide the basis for monitoring movement over the reporting period and for determining whether subsequent implementation attention should be maintained or adjusted.\n\nMention: \"Joint Shock Coping Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives average coping levels and a trend in reported conditions from the Joint Shock Coping Survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:0aee6f981f51c30bbffaa68c", "occurrence_id": "a71128651073a1475a512ea4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Enumerative Early Childhood Registry (EECR) database produced by SHIP. The registry is used as a documentary basis for describing the observed scale of early childhood service provision. Its use is limited to figures reported in the source, without extending the evidence through additional estimation.\n\nThe EECR database records 31 enrolled children and 14 service facilities. These two figures are incorporated directly into the assessment and support the broader finding that provision remains limited relative to the population requiring access. The figures are presented as reported counts and are not recalculated or combined with information from another source.\n\nThe evidence indicates that service availability is uneven across the registry records, with the recorded number of facilities providing a useful indication of the operational base. The recorded enrolment figure provides a corresponding measure of the children reached. Taken together, the two observations support a cautious assessment of coverage and establish a factual reference point for implementation monitoring.\n\nMention: \"Enumerative Early Childhood Registry (EECR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the EECR database for the figures of 31 enrolled children and 14 service facilities.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:8215140b8e9f711d59c178d1", "occurrence_id": "8fbe882c5bd22ed98ab7afa9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses Pilot Public Works Microdata as its principal empirical basis. This microdata source covers Addis Ababa and refers to 2023. It provides structured observations on participant records, implementation outcomes, and household circumstances, allowing the analysis to examine differences within the covered population.\n\nThe evidence is processed rather than reproduced as isolated reported figures. Records are grouped into comparable observation periods, and average levels are calculated to reduce the influence of individual irregularities. The resulting trend provides the central basis for assessing whether participation and delivery outcomes remained stable or changed during the period under review. This approach also permits comparison across relevant population groups without attributing undue weight to single observations.\n\nThe calculated averages indicate the prevailing level of implementation, while the direction of change identifies areas requiring closer operational attention. Results are interpreted as evidence from the covered records and not as a general estimate beyond the stated geographic and temporal scope. Limitations arise from the completeness and consistency of recorded information, but the source remains sufficiently detailed for a descriptive assessment of programme performance. The analysis therefore relies on derived patterns and summary measures from Pilot Public Works Microdata, rather than on direct quotation of individual entries.\n\nMention: \"Pilot Public Works Microdata\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives average implementation levels and a trend for Addis Ababa in 2023 from Pilot Public Works Microdata.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:559d5baf67e97ff162ca9283", "occurrence_id": "3975a204709aac5fefb8cc84", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Follow-Up Informal Sector Survey, a survey covering Nigeria and referring to 2020. The source is treated as the principal empirical basis for examining reported informal activity and related operating conditions. Its observations are used within the analytical framework described below rather than reproduced as isolated quotations.\n\nThe assessment calculates mean values across the reported activity groups and compares their relative positions to identify a broad pattern of change. This treatment provides an indicative trend derived from the distribution of observations in the source. The resulting measures are intended to support interpretation of the sector profile and to reduce the emphasis placed on any single reported value.\n\nThe calculated averages indicate variation in the intensity and prevalence of the recorded characteristics. The direction of the differences is considered alongside the spread of observations, allowing the analysis to distinguish a general tendency from isolated cases. No inference is made beyond the coverage and reference period of the source.\n\nThese results are used to inform the assessment of implementation conditions and to establish a consistent statistical basis for subsequent discussion. The evidence is therefore presented as a derived summary of the survey material, with attention to comparability across the reported groups and to the limitations inherent in aggregation.\n\nMention: \"Follow-Up Informal Sector Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives grouped averages and an indicative trend from the Follow-Up Informal Sector Survey covering Nigeria in 2020.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:5a12514337a03efcec3082e9", "occurrence_id": "8a799eb36d65da2214a1c599", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Government Social Protection Survey is treated as a contextual data source in the assessment. It is a survey produced by Growth Lab at Harvard and is referenced to support the general observation that the interpretation of social protection performance benefits from attention to household access and institutional reach. The citation does not provide an input to the analysis or establish a quantitative result.\n\nThe assessment relies on the source for background framing within a broader discussion of implementation conditions. No values, estimates, distributions, or respondent-level information from the survey are reproduced, compared, or used in the findings. Its role is therefore limited to supporting one contextual claim concerning the relevance of social protection access to programme interpretation.\n\nThe remaining analysis is based on the evidence described in the relevant implementation sections and on the review of delivery arrangements. This separation preserves a clear distinction between contextual evidence and material used to assess performance, while allowing the cited survey to indicate the wider analytical basis for considering access, coverage, and institutional functioning.\n\nMention: \"Government Social Protection Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The source provides contextual support for one general claim about social protection access, without contributing survey values to the analysis.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:ef14187a995616968110f8d2", "occurrence_id": "be0e59550bfade8ee0f9e8a4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment examines changes in youth movement through education and work as part of the broader transition analysis. The evidence is used to support the finding that progress is uneven across successive stages, with movement into stable work occurring less consistently than movement through earlier stages.\n\nThe analysis draws on the Longitudinal Youth Transition Census, a census produced by WorldPop. Its records provide repeated observations on participation, completion and entry into work, allowing the assessment to calculate an average transition rate across the observed sequence. The resulting measure is treated as an analytical estimate rather than as a figure reproduced directly from the source.\n\nTo reduce the influence of isolated observations, the assessment derives a rolling average and compares the direction of change across successive observations. This procedure indicates a gradual improvement in participation alongside a weaker and more variable progression into stable employment. The pattern is consistent with the wider assessment, which identifies retention and delayed entry as continuing constraints.\n\nThe estimates are used only to substantiate this interpretation. They do not replace the underlying records or imply that every individual follows the average path. The derived trend is therefore presented as an indication of aggregate movement, subject to the coverage and measurement properties of the Longitudinal Youth Transition Census.\n\nMention: \"Longitudinal Youth Transition Census\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses the Longitudinal Youth Transition Census to calculate an average transition rate and derive a trend in youth participation and movement into stable work.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:badde1cde05b1cd632000954", "occurrence_id": "639e208e48952589bfc45a49", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Audited Nutrition Surveillance Index is used as an indicator for 2023 in assessing the reach of nutrition interventions. Its entries are considered sufficiently specific for descriptive reporting, while the assessment does not derive additional measures from the source. The index is therefore treated as a reference for the observed distribution of outcomes rather than as a basis for estimating changes over time.\n\nTwo figures are reproduced directly from the index. The reported coverage is 42 percent for children receiving the relevant service, while the corresponding figure for children meeting the specified nutrition condition is 18 percent. These values support the broader finding that programme reach remains uneven, with a substantial gap between service contact and the intended nutrition outcome.\n\nThe figures are presented without adjustment or reclassification. They should be read as surveillance observations for the stated period and not as estimates of the entire population. This limitation does not affect their use in the present paragraph, which relies on the reported levels to establish the scale of the implementation gap and to inform subsequent discussion of monitoring priorities.\n\nMention: \"Audited Nutrition Surveillance Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the directly reported figures of 42 percent and 18 percent from the Audited Nutrition Surveillance Index to support an assessment of uneven nutrition programme reach.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:f54679cfad296d45190ae07a", "occurrence_id": "0633aace7e1a7e67a81282ee", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Pilot Irrigated Area Index (PIAI) is used as an indicator for 2023 in the assessment of irrigated production. It provides a compact reference for interpreting reported changes in cultivated area and for checking the internal consistency of the associated narrative. No adjustment is applied to the values reproduced below.\n\nThe broader analysis supports the claim that irrigated cultivation remains material to production outcomes. In support of that claim, two figures are taken directly from the Pilot Irrigated Area Index (PIAI): 31 percent of the recorded area is identified as irrigated, while 18,400 hectares are reported under irrigation. These values are presented as source figures and are not derived estimates.\n\nThe figures are retained in their original units to preserve comparability within the assessment. Their use is limited to substantiating the stated claim and does not imply a recalculation of the indicator or an extension beyond the coverage reported in the source.\n\nMention: \"Pilot Irrigated Area Index (PIAI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the PIAI figures of 31 percent and 18,400 hectares from 2023 to support a claim about the significance of irrigated cultivation.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:05d293be80c2639c96eba37e", "occurrence_id": "0a9be565d0340313fc5e584c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Continuous Irrigated Area System (CIAS) is treated in this assessment as an administrative source for the measurement of irrigated land. Its records provide the principal quantitative basis for the observations presented in this annex and are retained as the reference for the figures reproduced below. The source is used without supplementary attribution or modification.\n\nTwo entries are reproduced directly from CIAS. The first records 31 hectares under continuous irrigation, while the second records 12 hectares under the same measure. These figures are presented in their reported form to preserve the evidentiary value of the underlying administrative record. The accompanying text does not expand the coverage of the source or introduce estimates derived from other materials.\n\nThe reported values provide a concise basis for reviewing the extent of continuously irrigated land within the assessed record. Their inclusion supports verification of the quantitative statements used in the implementation assessment and allows subsequent review to distinguish recorded observations from interpretation. Any comparison between the entries should therefore be understood as a reading of the source record rather than as a separately generated estimate. The CIAS entries remain the controlling evidence for the two quantities cited in this passage.\n\nMention: \"Continuous Irrigated Area System (CIAS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the figures 31 hectares and 12 hectares directly from the Continuous Irrigated Area System (CIAS).\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:2415fcef83376a5340fb280f", "occurrence_id": "362e5bf30eda7be09b823c78", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment relies on the Standardised Internal Movement Census, a census covering Central Asia and referring to 2024. The material is treated as the primary quantitative basis for the descriptive findings presented below. Its geographic scope and reference period are retained without adjustment so that the reported magnitudes remain traceable to the underlying record.\n\nTwo figures are taken directly from the census. It records 184,600 persons within the observed population, and it reports that 62.4 percent of households experienced an internal movement during the reference period. These values are reproduced as stated, without recalculation, interpolation, or comparison with another series.\n\nTogether, the reported values indicate a substantial volume of movement alongside a broad incidence across households. They provide the basis for the implementation assessment’s description of coverage and service pressure, while their differing units require separate reading. The first value expresses a count of persons, and the second expresses a share of households. No additional adjustment is introduced in this passage, and both values should be read as direct observations from the cited census.\n\nMention: \"Standardised Internal Movement Census\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Standardised Internal Movement Census to report 184,600 persons and 62.4 percent of households.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:b7c3274b77d82d69cb3f57d3", "occurrence_id": "622ca3636944859ca96072e7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Multi-Topic Road Segment Registry (MRSR) is used as the reference database for this assessment. It provides a consistent basis for reviewing the recorded characteristics of road segments and for checking the internal coherence of the supporting evidence. No adjustment to the source values is applied in the analysis presented here.\n\nThe assessment uses two figures reproduced directly from the database, namely 31 grades and 14 segments. These figures are cited to support the broader finding that the observed network has varied conditions and that the available records do not indicate a uniform level of service. The figures are retained in their reported form so that the interpretation remains traceable to the underlying registry.\n\nTaken together, the evidence supports continued use of a common classification approach in implementation monitoring. The results should nevertheless be read as a structured description of recorded conditions rather than as a complete measure of performance. Subsequent reporting may use the same basis to assess changes in coverage, condition, and consistency of reporting.\n\nMention: \"Multi-Topic Road Segment Registry (MRSR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the MRSR database for two directly reproduced figures, 31 grades and 14 segments.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:efe7ada14c3af2091205a185", "occurrence_id": "aab8a00a1eea3f80854ee3a0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Government Enterprise and Jobs Survey is used as the primary evidentiary source for the assessment. It is a 2024 survey produced by GBV IMS and is treated as the applicable reference for the reported enterprise and employment indicators. The source is cited in full to maintain traceability between the narrative findings and the underlying observations.\n\nTwo figures are drawn directly from the survey. The first indicates that 31 percent of surveyed enterprises reported an improvement in operating conditions. The second indicates that 47 percent reported that employment increased during the period covered by the survey. These values are reproduced without recalculation and are presented as reported measures rather than derived estimates.\n\nThe figures provide a concise basis for interpreting the direction of change captured in the source. They should not be read as a composite index or as evidence of a causal relationship between operating conditions and employment. Their principal use in this assessment is to establish the scale of the reported outcomes and to support comparison across the indicators covered by the survey. Any subsequent interpretation should retain the definitions and reporting conventions applied in the Government Enterprise and Jobs Survey.\n\nMention: \"Government Enterprise and Jobs Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage reports the directly sourced figures of 31 percent and 47 percent from the Government Enterprise and Jobs Survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:35111aa1c87a693266a4fbe1", "occurrence_id": "e93f27a888ef8471dac004fa", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Provincial Coastal Erosion Collection (PCEC) as other source material for the analysis of shoreline change. The 2023 collection is retained as a supporting input, with its observations aligned to the reporting framework and screened for internal consistency. No single entry is treated as determinative for the assessment finding.\n\nThe analysis derives an average level from the available observations and evaluates the direction of change across successive records. This calculation provides a measured indication of the underlying trend while limiting the influence of isolated fluctuations. The resulting measure is used to support the conclusion that erosion pressure is uneven over the observation period and should be interpreted through the combined pattern rather than through individual records.\n\nThe source therefore informs one component of the assessment without serving as the sole basis for the broader interpretation. Results are presented at an aggregated level, and the calculated average and trend are used to assess consistency with the wider evidence considered in this report. The treatment preserves the source’s role as supporting statistical material and avoids implying precision beyond that warranted by the observations.\n\nMention: \"Provincial Coastal Erosion Collection (PCEC)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the 2023 PCEC as other source material to calculate an average and assess a trend.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:e49ecc8073d4ffe37bb0a425", "occurrence_id": "8a05215acb8d22051464664a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment relies on the Continuous Camp Management Dataset (CCMD), a geospatial source produced by The World Bank for 2022. The dataset is used as the principal quantitative reference for the observations presented below, with its reported coverage and proportions retained without recalculation or adjustment. No supplementary source is introduced for the measures cited in this paragraph.\n\nThe analysis takes directly from the dataset a reported total of 31 monitored sites and a recorded share of 68 percent meeting the specified monitoring condition. These figures are reproduced as evidence of the scale of coverage and the observed level of compliance, respectively. Their inclusion provides the numerical basis for the assessment rather than an inferred estimate derived from other material.\n\nThe two measures should be read together. The first indicates the extent of the monitoring frame, while the second summarises the proportion within that frame satisfying the stated condition. This presentation preserves the distinction between an absolute count and a percentage, and avoids implying a change over time or a comparison beyond the information contained in the dataset.\n\nMention: \"Continuous Camp Management Dataset (CCMD)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the figures of 31 monitored sites and 68 percent meeting the specified monitoring condition directly from the cited source.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:72b37d59080be46ab4097f19", "occurrence_id": "e6c265817917504b890c5c3c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on administrative records, field verification, and routine monitoring materials to establish the basis for implementation review. These materials support a general description of coverage, operational conditions, and reporting arrangements. The Rapid Road Segment Registry is mentioned as a database relevant to the broader information environment, but it is not used to establish any finding in this assessment.\n\nThe registry is therefore treated as background documentation rather than as an evidentiary basis for measurement. No numerical value, classification, rate, or observation is taken from it, and no result in the assessment depends on its contents. Its mention indicates only that a structured source exists within the information system surrounding the work.\n\nThe analysis presented here relies on the records and verification procedures described in the relevant sections of this document. References to the Rapid Road Segment Registry do not alter the indicators, calculations, or interpretations reported elsewhere. The source is retained in the narrative to document the range of materials considered during preparation, while the assessment itself remains independent of any values that the database may contain.\n\nMention: \"Rapid Road Segment Registry\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The passage takes no values or derived results from the Rapid Road Segment Registry and uses it only for background context.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:7ce8a37a8c683674934381bc", "occurrence_id": "13e961d42664baac1b53ba41", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Recurrent Forest Products Index as a statistical reference for interpreting recurrent movements in forest-related outputs. The cited source is identified as the 2019 Third Version indicator, and its definitions are retained without modification in the discussion that follows.\n\nTwo figures are reproduced directly from this source. The reported level is 31 units, while the corresponding proportion is 62 percent. These values are presented as recorded in the indicator and are not recalculated, extrapolated, or combined with estimates from another source.\n\nThe figures provide a limited but sufficiently consistent basis for the assessment of observed conditions. The first value establishes the reported scale of activity, and the second expresses the associated share in percentage terms. Taken together, they support a cautious reading of the underlying pattern while avoiding claims beyond the coverage of the cited material.\n\nFor implementation purposes, the source is therefore treated as a fixed statistical point of reference. Any subsequent comparison should preserve the stated units, the reported percentage, and the definitions accompanying the Recurrent Forest Products Index. No adjustment is made to the two quoted values, since their evidentiary role depends on their direct reproduction from the cited indicator.\n\nMention: \"Recurrent Forest Products Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported figures of 31 units and 62 percent from the 2019 Third Version indicator.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:510ca67a217c454b3328420d", "occurrence_id": "a7e16004ccddd2dbc5273a58", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Cross-Sectoral Poverty and Vulnerability Index is cited as an indicator covering Russia in 2017. It is recorded in the assessment as a relevant source for understanding the broader framework in which poverty and vulnerability are considered. The citation supports orientation to the subject matter and does not form part of the evidentiary basis for the findings.\n\nThe reference is used in passing to establish a general statistical point of departure. No value from the indicator is reproduced, recalculated, compared, or assigned to a population group. The analysis therefore does not rely on the source for measurement, ranking, trend identification, or estimation of programme effects.\n\nThis treatment preserves a clear distinction between contextual documentation and analytical results. The cited material is acknowledged for its relevance to the topic, while the reported conclusions remain confined to the evidence presented elsewhere in the assessment. No interpretation is derived from the source beyond its identification as an indicator concerning Russia in 2017.\n\nMention: \"Cross-Sectoral Poverty and Vulnerability Index\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The passage identifies the Cross-Sectoral Poverty and Vulnerability Index as an indicator for Russia in 2017 and uses it only for contextual reference, without taking any values from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:9971a1782edb948e4fcd77f3", "occurrence_id": "7cf7ebcc6982477063c44d10", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Diagnostic Graduation Programmes Registry (DGPR) is used as the 2023 database for the descriptive evidence presented in this annex. The database reports 31 programme entries and 18 outcome entries; both figures are reproduced directly from the cited source. They provide the numerical basis for the limited comparison undertaken below.\n\nThe 31 programme entries indicate the recorded scale of activity within the registry, while the 18 outcome entries identify cases for which an outcome is separately reported. The figures are presented without adjustment, extrapolation, or reclassification. Their purpose is to establish the distribution captured in the source rather than to estimate coverage beyond the recorded observations.\n\nInterpretation remains subject to the scope and recording practices of the database. Differences between programme entries and outcome entries should not be read as evidence of implementation failure, since the two counts may reflect different reporting requirements. The comparison therefore supports a description of recorded activity only. It does not establish changes over time or permit conclusions about effectiveness.\n\nFor monitoring purposes, the two directly reproduced figures should be retained as reference points in subsequent reporting. Any future update should preserve the distinction between programme entries and outcome entries and should identify revisions to the underlying records. This treatment maintains traceability to the Diagnostic Graduation Programmes Registry (DGPR) while limiting the analysis to evidence explicitly contained in the 2023 database.\n\nMention: \"Diagnostic Graduation Programmes Registry (DGPR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes directly from the DGPR database the figures of 31 programme entries and 18 outcome entries.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:2a15ad2509d71a75ab0bf95f", "occurrence_id": "f23c6e0b796b2e304f34bce8", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment considers return and relocation within the broader framework of population mobility and administrative coordination. The Cross-Sectoral Return and Relocation Registry (CRRR) is cited as a database relevant to that framework. Its inclusion indicates that these movements are subject to structured recording and can be considered alongside the institutional arrangements discussed in the assessment.\n\nThe database is used to support the general observation that reliable registration contributes to the visibility of return and relocation processes. This observation forms one element of a wider discussion concerning planning, case management, and the consistency of administrative information. No tabulation, indicator, aggregate, or individual entry from the CRRR is reproduced or interpreted in the assessment.\n\nThe reference therefore has a limited evidentiary function. It establishes the relevance of an organised registration resource to the argument, while the substantive assessment rests on the report’s broader analytical treatment. The CRRR is not used to derive a distribution, compare categories, or establish a change over time. Its role is confined to supporting the stated claim regarding the importance of systematic recording.\n\nMention: \"Cross-Sectoral Return and Relocation Registry (CRRR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The passage uses the CRRR database only to support a general claim about the relevance of systematic recording and takes no values from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:fa519931b04cd811799637f2", "occurrence_id": "ea3ff086001fe6a8dfdda575", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment considers the institutional conditions affecting access to unemployment insurance at district level. The District Unemployment Insurance Index (DUII) is an indicator covering Viet Nam for 2018. Its scope is relevant to the interpretation of differences in protection and administrative reach across districts.\n\nWithin the broader discussion of labour-market resilience, the citation supports the claim that unemployment insurance conditions should be considered when reviewing the distribution of social protection. The reference helps establish the relevance of a district-level perspective, while the assessment does not rely on the index to quantify coverage, rank districts, or identify changes over time.\n\nThe DUII indicator is therefore used solely to frame the discussion surrounding that claim. No value, ranking, or calculation derived from the District Unemployment Insurance Index (DUII) is incorporated into the findings, and no estimate in the assessment is attributed to the source. This treatment preserves the source as supporting background for interpretation rather than as an input to measurement or comparison.\n\nMention: \"District Unemployment Insurance Index (DUII)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The passage uses the DUII indicator for Viet Nam in 2018 to frame a district-level claim without using its numerical values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:1ec5ec9ab74cd44ff7aa3b04", "occurrence_id": "287fa754b5c79d6c980a35ac", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis uses the Analytical Population Grid Census, a census produced by Latinobarómetro, as the underlying source for measuring the distribution of population across the analytical grid. The source is treated as a structured statistical record for comparison rather than as a narrative account. No additional institutional or geographic attribution is required for the assessment presented here.\n\nTo support the assessment of population concentration, the observations are grouped into comparable intervals and reviewed for consistency. The resulting measures indicate a gradual shift toward higher concentrations in the main settlement areas, while the dispersion of smaller observations remains material. This pattern supports the broader finding that service demand is unevenly distributed across the area under review.\n\nThe reported result is derived from the source through calculation of a trend and an average over the available observations. Individual values are therefore not reproduced as headline figures, and the analytical conclusion should not be read as a direct quotation from the census. The calculated average provides a stable reference point, while the trend indicates the direction and relative pace of change. Taken together, these measures provide a sufficiently robust basis for interpreting the distributional pattern used in the assessment.\n\nMention: \"Analytical Population Grid Census\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The analysis calculates an average and a population-concentration trend from the Analytical Population Grid Census rather than reproducing individual census values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:0044a5b58855a05da31f6051", "occurrence_id": "3f3681043419f7ebb2998eda", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment considers referral continuity as part of the wider service-delivery environment. Interpretation therefore takes account of institutional access, communication between facilities, and the practical conditions that shape whether a referral is completed. The discussion is intended to inform implementation decisions rather than to provide a separate estimate of referral performance.\n\nThe Provincial Referral Pathways Microdata is cited as a contextual source for this interpretation and is a microdata source referring to 2015. Its inclusion supports the claim that observed referral outcomes should be read in relation to the pathways through which clients move between services. The source is not used to introduce, reproduce, or calculate any values in this assessment.\n\nAccordingly, the findings presented here rely on the evidence discussed in the surrounding analysis, while the cited source provides background for the interpretation of referral processes. This distinction is maintained to avoid implying a level of measurement or comparability that is not established in the present assessment. The reference should therefore be understood as supporting one contextual claim within the broader implementation discussion, with no numerical extraction from the source.\n\nMention: \"Provincial Referral Pathways Microdata\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The source provides contextual support for interpreting referral pathways, while none of its values are used.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:2e741b5b0d5730823d72768d", "occurrence_id": "32bcc6ce2b529a4e97480076", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Quality-Assured Transport Network Registry database covers Cairo and refers to network conditions recorded in 2018. It is used as the principal evidentiary source for the transport coverage assessment, with the cited values retained in the form reported by the registry.\n\nThe assessment records 31 registered corridors and 184 route segments from the source. These figures are reproduced directly and are not recalculated, interpolated, or combined with observations from another source. Their use provides a consistent basis for describing the extent of the network within the stated coverage area.\n\nThe reported totals indicate a network with broad corridor coverage and a substantially larger number of component segments. Interpretation remains limited to the scope of the database and the reference period identified above. No adjustment is made for subsequent changes in routes, service patterns, or network classification.\n\nMention: \"Quality-Assured Transport Network Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported totals of 31 registered corridors and 184 route segments from the Quality-Assured Transport Network Registry database for Cairo in 2018.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:c12b58b8a955d7be62bd6f26", "occurrence_id": "27cccc7694a563592ecbbed8", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Urban Demographic and Health Report is cited as a 2020 report in the background documentation for the assessment. It is retained as a reference point for the demographic setting and for the general relevance of household and population conditions to the project’s analytical framework. No location, edition, producing organisation, or reporting round is specified in this citation.\n\nThe assessment relies on the project’s own monitoring arrangements and on records reviewed for implementation purposes. These materials support the description of service coverage, beneficiary identification, and reporting procedures. The cited report does not provide the basis for any calculation, comparison, target, or performance judgement presented in this annex.\n\nIts inclusion serves to acknowledge an established source of demographic information while keeping the evidentiary basis of the assessment separate. No figures, rates, percentages, tables, or indicators are reproduced from the Urban Demographic and Health Report. The reference therefore has a limited documentary function and should not be interpreted as a source for the results, estimates, or conclusions set out in the accompanying analysis.\n\nMention: \"Urban Demographic and Health Report\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The 2020 report is used only to frame the background, and no values or indicators are taken from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:4ef9f56a11ec46d7f2b1beba", "occurrence_id": "372a42eb05d6ca0d4afc632f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Inter-Agency Rural Livelihoods Census is treated as the principal statistical input for this assessment. It is a census covering the 2023 reference period and provides comparable observations on household livelihood conditions, productive activities, and access to selected services. The source is used as a common base for the calculations reported below.\n\nThe analysis does not reproduce individual records or present isolated entries as findings. Instead, observations are grouped by livelihood measure and converted into summary rates, proportions, and average values. These calculations indicate the relative concentration of activities, the distribution of access conditions, and the extent to which reported outcomes vary across the recorded groups.\n\nA composite trend is then derived from the resulting measures to identify areas of strengthening, stability, or decline within the livelihood profile. The trend is interpreted together with the calculated averages, allowing unusually high or low observations to be considered without allowing them to determine the overall result. This approach supports comparison across indicators while retaining the underlying variation captured by the census.\n\nThe resulting estimates are therefore analytical outputs based on the Inter-Agency Rural Livelihoods Census rather than figures transcribed directly from it. All reported changes and averages should be read as derived measures, with the census serving as the central evidence base for the assessment.\n\nMention: \"Inter-Agency Rural Livelihoods Census\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage derives comparative trends, rates, and averages from the 2023 census rather than reproducing individual records.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:dba446a2bf6b73f5729a0364", "occurrence_id": "4ba1093c7829f6c679cd2d73", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Follow-Up Digital Learning Registry database provides a supplementary basis for reviewing reported participation in digital learning activities. Its entries are treated as an administrative reference rather than as a complete measure of access, since registry coverage may not capture all relevant learners or delivery arrangements. The source is therefore used selectively in the assessment of the implementation results.\n\nThe registry evidence is used to cross-check the estimate of participation derived from a separate reporting stream. This comparison does not replace the principal estimate, but indicates whether the reported direction and approximate magnitude are consistent with an independent record. Where the two sources differ, the discrepancy is noted as a limitation of measurement rather than resolved through an unverified adjustment.\n\nWithin the broader assessment, this evidence supports the claim that recorded engagement increased during implementation. The registry is not used to establish a comprehensive trend or to infer outcomes beyond the reported activity. Its contribution is narrower, namely to provide a point of verification for one claim concerning participation and to strengthen the interpretation of the associated estimate. Any conclusions drawn from the comparison remain subject to the coverage and reporting constraints of both sources.\n\nMention: \"Follow-Up Digital Learning Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The author uses the Follow-Up Digital Learning Registry database to cross-check a participation estimate from a separate reporting stream and support one claim about increased engagement.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:5984d2eed0ec6ec465929d5c", "occurrence_id": "000a9a98775d8dd507ced79b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The High-Resolution Elevation Model Registry is cited as the database produced by GBV IMS for Mali and referring to 2023. It provides the principal documentary basis for the geographic coverage reported in this assessment. The citation is retained in the text to support verification of the reported quantities and to distinguish the underlying record from subsequent interpretation.\n\nTwo figures are transcribed directly from this database. The registry records 31 elevation records and 14 coverage records for the area under review. These values are presented without recalculation, adjustment, or aggregation, and they should therefore be read as reported entries rather than derived indicators. Their inclusion establishes the scale of the recorded material and provides a fixed reference for the accompanying assessment.\n\nThe evidence is limited to the information contained in the cited registry. No inference is made regarding changes beyond the stated period, and no additional classification is introduced. The figures are reproduced as documentary observations, with the database serving as the sole basis for the numerical statements in this passage.\n\nMention: \"High-Resolution Elevation Model Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the High-Resolution Elevation Model Registry to report 31 elevation records and 14 coverage records for Mali in 2023.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:02608e2d8632f99b5b8f8ca3", "occurrence_id": "f647bb253f3b5898eec5bb53", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Triennial Working Conditions Survey, a survey produced by CIESIN-CIAT, is used as the principal external reference for reviewing the reported employment estimate. Its tabulations provide a structured basis for examining the level and distribution of working conditions represented in the assessment. The source is treated as central evidence because it permits a direct comparison with an estimate derived from a separate source.\n\nThe comparison is undertaken at the aggregate level and, where the underlying classifications permit, across the principal groups used in the assessment. Differences between the two measures are reviewed for direction, magnitude, and possible implications for interpretation. The exercise is intended to identify material divergence rather than to replace the estimate used for implementation monitoring.\n\nResults from the Triennial Working Conditions Survey are therefore incorporated as a validation reference. Broad agreement is taken to support the reasonableness of the reported estimate, while material discrepancies are flagged for further examination of coverage, definitions, and measurement procedures. This approach limits reliance on a single statistical series and provides an additional basis for assessing the robustness of the findings.\n\nMention: \"Triennial Working Conditions Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The assessment uses the Triennial Working Conditions Survey from CIESIN-CIAT as an independent benchmark to cross-check a separate employment estimate.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:621d70df79c11215328fa982", "occurrence_id": "2d4f4191d50674aae109520b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Ministerial Time Use Survey is the principal source used for the time-allocation evidence in this note. It is a survey produced by Social Watch in 2023 and is treated as a direct record of reported activity rather than as a basis for further estimation. The source is cited here with its title, producing organisation, and year so that the origin of the observations remains identifiable in the assessment.\n\nTwo figures are carried forward directly from the source. It records that 47 percent of reported time was devoted to routine administrative work and that 18 percent was devoted to coordination activities. These values are presented in the same form as in the survey, with no recalculation, interpolation, or conversion. Their purpose is to anchor the discussion in observed distributions and to distinguish the reported shares from indicators derived elsewhere in the analysis.\n\nTaken together, the figures indicate that routine work accounted for the larger share, while coordination represented a materially smaller component. The difference is noted descriptively and is not interpreted as evidence of change over time. This limitation is appropriate because the available citation supports the two reported shares, but does not establish a sequence of observations. The findings should therefore be read as a concise account of the distribution documented in the source.\n\nMention: \"Ministerial Time Use Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the survey’s directly reported figures of 47 percent for routine administrative work and 18 percent for coordination activities.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:17eb11f4291c21045a0009fb", "occurrence_id": "95ccd5f631e27094a48210e7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Combined Flood Extent Index is used as an indicator of the spatial reach of reported flooding. It provides a supplementary reference for reviewing the estimate presented in this assessment and is not treated as a replacement for the principal measure. Its inclusion supports a focused check of whether the estimated extent is consistent with an independently derived signal.\n\nThe comparison is made at the level of the claim concerning affected extent. The estimate from the separate source is reviewed against the direction and relative magnitude indicated by the Combined Flood Extent Index. Agreement is interpreted as supporting evidence for the reported assessment, while divergence is retained as a qualification requiring further review. No adjustment to the principal estimate is made solely on the basis of this comparison.\n\nThe indicator is therefore used for verification within the broader analysis rather than as a standalone basis for attribution or measurement. The resulting evidence is considered adequate for the present assessment, subject to the limitations of the underlying observations and the differing methods applied by the sources. This treatment preserves the distinction between the reported estimate and the supporting cross-check.\n\nMention: \"Combined Flood Extent Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The author uses the Combined Flood Extent Index to cross-check the reported affected-extent estimate against an independent signal.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:86bd22d3bc4b329bc0ea2789", "occurrence_id": "f310755d4ee41702c3025d24", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Small-Area Public Works Survey is cited as a survey produced by World Bank-IFC. The source is used to situate the discussion of public works within the wider evidence base considered for the assessment. Its role is interpretive rather than quantitative, and it does not supply inputs to the tables, rates, totals, or classifications presented in this document.\n\nThe assessment focuses on implementation arrangements, the consistency of reporting, and the extent to which available records support a clear account of activities. Within that broader discussion, the survey supports one limited claim, namely that public works activity is a relevant contextual consideration when interpreting implementation evidence. The claim does not depend on any value reported in the source, and no extraction, recalculation, or comparison is undertaken from its contents.\n\nAccordingly, the citation should be read as supporting background for the stated interpretation, rather than as proof of a measured result. The analysis relies on the documentation reviewed for the assessment and retains the distinction between contextual material and evidence used to quantify performance. This treatment preserves a clear audit trail and avoids attributing precision to information that is not used in the reported results. The source is therefore acknowledged for its relevance to the framing of the issue, while the findings remain confined to the evidence directly assessed.\n\nMention: \"Small-Area Public Works Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The source contributed only contextual support for one claim, and no values were taken from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:c7092dc21b5c2d5e484ce5e6", "occurrence_id": "ee6a48bd1b242b64c31b21d4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Comparative Debt and Savings Index, an indicator produced by KTB in 2021, provides a consistent basis for examining changes in recorded debt and savings conditions. It is used as supporting evidence within the assessment of household financial resilience, rather than as a standalone measure of performance. The material is treated as an input to the analysis and is retained with its stated attribution.\n\nTo reduce the effect of short-term variation, the analysis derives an average and examines the direction of movement across the observations represented in the indicator. No single reported figure is reproduced as the basis for the conclusion. Instead, the calculated measure is used to identify whether the balance between debt exposure and savings capacity shows a broadly improving, stable, or weakening tendency.\n\nThis treatment supports the finding that financial resilience should be interpreted through the combined movement of debt and savings conditions. The resulting trend is considered alongside the wider assessment evidence and is used to qualify, rather than determine independently, the interpretation of changes in household capacity. The approach also limits the risk that an isolated observation will be given disproportionate weight in the implementation assessment.\n\nMention: \"Comparative Debt and Savings Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses the 2021 KTB indicator to calculate an average and trend supporting the assessment of financial resilience.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:859d88c5bfc62699ffce3f23", "occurrence_id": "200a4bad624701cf1025726b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The cited data source is the Urban Forced Population Index (UFPI), an indicator produced by PISA. It is used to support the assessment of population pressure and to provide a consistent reference for interpreting the distribution of observed needs. No supplementary designation is applied in this note, and the source is retained under its stated title.\n\nTwo figures are quoted directly from it, namely 62.4 percent and 14.8 percent. The first value is used to indicate the higher level observed in the principal assessment group, while the second records the lower level reported for the comparison group. The figures are reproduced without adjustment or recalculation.\n\nTaken together, these values support the broader finding that pressure is material but unevenly distributed across the assessment categories. This distinction is relevant to the sequencing of operational responses, since a single aggregate interpretation would obscure the difference between the groups. The indicator is therefore used as supporting evidence rather than as the sole basis for allocation decisions. The cited figures should be read alongside the wider findings and applied consistently in subsequent monitoring.\n\nMention: \"Urban Forced Population Index (UFPI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage reproduces 62.4 percent and 14.8 percent from the Urban Forced Population Index (UFPI) to support the assessment of uneven population pressure.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:96c0be735579d25c6a96ca9a", "occurrence_id": "a9210f3b4969afcdcff78251", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Small-Area Household Income and Expenditure Index (SHIEI) is an indicator produced by the World bank for 2022. It is used here as a consistent statistical reference for assessing changes in household economic conditions within the wider implementation review. No individual observation is treated as a standalone result.\n\nThe analysis derives an average from the available SHIEI observations and follows the direction of change across the series. This calculation provides a smoothed basis for comparison and reduces the influence of short-term variation in any single observation. The resulting pattern supports the assessment that income and expenditure conditions did not evolve uniformly over the period considered.\n\nThe source therefore informs one claim in the broader findings on household welfare, while the interpretation remains based on the calculated summary rather than on a direct reproduction of the source values. The derived movement is used to corroborate the reported distributional pattern and to guide the reading of related implementation results. Further conclusions are considered alongside the wider evidence presented in this assessment.\n\nMention: \"Small-Area Household Income and Expenditure Index (SHIEI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The passage identifies the 2022 SHIEI indicator produced by World bank and uses it to derive an average and trend for the assessment.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:be74ff7c2c75a4b378fcca70", "occurrence_id": "10c6624785c14df2a76195c7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Pilot Temperature Records Dataset is treated as the principal geospatial source for the assessment of temperature records referring to 2020. It is identified as a product of AICD and is cited in full to preserve the provenance of the observations used below. No adjustment, interpolation, or cross-source comparison is applied to the values reproduced from this record.\n\nTwo observations are carried into the analysis exactly as reported. The lower reported temperature is 18.4 degrees Celsius, while the higher reported temperature is 27.9 degrees Celsius. These figures are presented as direct readings from the dataset, rather than as estimates derived from a calculation. Their inclusion provides the numerical basis for the limited comparison made in this note.\n\nTaken together, the readings establish the range represented by the cited observations without supporting broader conclusions about conditions beyond the record. The evidence is therefore used for a bounded description of the reported temperatures only. Interpretation remains limited to the two values reproduced above, and no inference is made regarding an unobserved period or a wider distribution.\n\nMention: \"Pilot Temperature Records Dataset\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage reproduces 18.4 and 27.9 degrees Celsius as direct readings from the Pilot Temperature Records Dataset.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:f9a2bae6a9f4a05e2bd06dbf", "occurrence_id": "858b3d1612aeb1c154e27bad", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Provincial Water Body Census (PWBC), the census referring to 2015, is used as the documentary basis for the inventory assessment. The source is cited to establish the reported scale of the water body inventory and to support the interpretation of recorded assets in the wider assessment. No adjustment or reclassification is applied to the figures reproduced here.\n\nFor this purpose, two figures are reproduced directly from the PWBC. The census records 31 water bodies in the first reported category and 47 water bodies in the second reported category. These values are presented as quoted observations from the source and are not derived through calculation within the assessment. Their inclusion supports the broader claim that the documented inventory provides a material basis for planning, prioritisation, and subsequent review of coverage.\n\nThe figures should therefore be read as source observations rather than as indicators of change over time. They provide a concise numerical anchor for the discussion and allow the assessment to distinguish the reported scale of assets from later analytical interpretation. The PWBC remains the reference for both values, while the surrounding narrative uses them only to substantiate the stated inventory finding.\n\nMention: \"Provincial Water Body Census (PWBC)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the 2015 PWBC census and reproduces figures of 31 water bodies and 47 water bodies to support the inventory assessment.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:aab8c322752f5c565560976f", "occurrence_id": "efef14dd3896c60d0c5914c0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The principal empirical source for this assessment is the Enumerative Child Wellbeing Survey, a survey covering Croatia. Its records provide observations on selected dimensions of child wellbeing and establish the evidence base for the descriptive analysis presented below. No supplementary source is required for the calculations reported in this section.\n\nThe analysis consolidates the available observations into comparable measures and calculates mean levels for the selected indicators. It also derives a trend from successive observations, allowing the assessment to distinguish persistent movement from isolated variation. The resulting averages and trend estimates, rather than individual figures reproduced from the source, are used to support the interpretation of changes in child wellbeing.\n\nResults are reported at the level of the indicators for which the survey provides sufficiently consistent coverage. Where observations are incomplete, the calculation is limited to the information available and is interpreted with appropriate caution. This approach preserves the analytical value of the source while avoiding an unsupported degree of precision.\n\nThe derived measures are used in the assessment of distributional patterns, relative performance across indicators, and the direction of change over the observed period. They therefore function as summary evidence for implementation monitoring and provide a reproducible basis for subsequent discussion of outcomes.\n\nMention: \"Enumerative Child Wellbeing Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses observations from the Enumerative Child Wellbeing Survey to calculate indicator averages and derive a trend for Croatia.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:80cca948b93e80ca4d9e7cc0", "occurrence_id": "7946c025b0246079480fb308", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Urban Living Standards Survey (ULSS) is treated as the central empirical source for the assessment. It is a 2016 survey, and the analysis uses its observations to establish a consistent basis for comparison across the measured indicators. The source is cited for its analytical coverage rather than for isolated reported values.\n\nThe calculations derive average levels from the underlying records and use these averages to identify the direction and pace of change. A trend is estimated by comparing the calculated results across the relevant observation periods, with attention to the stability of the pattern and the extent of variation around the average. This approach reduces the risk that individual observations will be interpreted as representative outcomes.\n\nThe resulting estimates support the assessment of distributional change and provide the main quantitative basis for the accompanying interpretation. The findings should therefore be understood as derived measures based on the ULSS records, rather than as direct reproduction of figures presented in the survey. Where the calculated trend is weak or uneven, the assessment records this limitation and avoids drawing conclusions beyond the evidence.\n\nMention: \"Urban Living Standards Survey (ULSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses the 2016 ULSS survey to calculate average levels and derive a trend from its underlying observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:ffb433897937078a0eeaa7ec", "occurrence_id": "26abe37c8704d8a36f529d4f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Experimental Household Income and Expenditure Survey (EHIES) is the survey used as the documentary basis for the household measures presented in this assessment. Its reported observations provide the principal evidence for the limited comparison set out below, and no recalculation or adjustment has been applied to the quoted values. The source is therefore treated as the reference point for the descriptive findings.\n\nTwo figures are carried forward directly from the survey. It reports that 47.2 percent of households recorded expenditure within the specified consumption measure, while 18.6 percent reported income below the applicable threshold. These values are reproduced as stated in the source and are not presented as estimates derived from a separate calculation.\n\nThe figures support a cautious reading of household conditions. The difference between the reported proportions is noted for descriptive purposes only and should not be interpreted as evidence of a causal relationship. In view of the scope of the available observations, the results are most appropriately used to establish the distribution recorded in the survey and to inform subsequent review of the relevant indicators. Any further interpretation would require evidence beyond the two values cited here.\n\nMention: \"Experimental Household Income and Expenditure Survey (EHIES)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported figures of 47.2 percent and 18.6 percent from the EHIES survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:844a697705effb3ebde9ec12", "occurrence_id": "adc7ab4248175e7edeb32337", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Diagnostic Demographic and Health Index is cited as an indicator of the broader demographic and health information environment. It is included to situate the assessment of service coverage and population needs within the wider evidence base available for sector planning. The source is referenced in the same paragraph as the related interpretation and is not treated as a direct input to the reported estimates.\n\nThe citation supports the claim that conclusions about coverage should be read alongside the quality, scope, and accessibility of demographic and health information. This point is relevant to the assessment because gaps in information can affect the interpretation of observed differences across population groups and service areas. The index therefore provides background for the discussion rather than a basis for ranking performance or assigning outcomes.\n\nNo values from the Diagnostic Demographic and Health Index are reproduced, transformed, or used in the calculations presented here. The reference is limited to establishing the analytical setting for one claim within the broader paragraph. All quantitative findings in the assessment derive from the sources and procedures specified for those findings, while the cited indicator remains supplementary evidence for interpretation.\n\nMention: \"Diagnostic Demographic and Health Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The cited indicator provides contextual support for one claim about interpreting demographic and health information, while none of its values are used.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:12b6d1c1e33c21cf2c2d4a8a", "occurrence_id": "8f37ccb3c25339ffa69bd123", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Monthly Agricultural Household Report for Samoa in 2017, produced by SABESP, is treated as the report underlying the household estimates used in this assessment. It provides a contemporaneous reference point for interpreting the distribution of agricultural activity and for checking whether the narrative claim is supported by observed household counts. No adjustment, projection, or reconciliation is applied to the cited entries.\n\nThe analysis reproduces two figures directly from this source. It reports 31 households in the relevant category and 18 percent for the associated share. These values are presented as recorded, rather than recalculated, and they serve as the numerical basis for the statement that participation was concentrated within a limited segment of the agricultural household population.\n\nWithin the broader assessment, the figures support the conclusion that agricultural activity was unevenly distributed across households. The first figure establishes the observed count, while the second indicates the relative scale of that group. Together they provide a bounded empirical basis for the claim, while the surrounding discussion remains qualitative and does not infer changes beyond the period covered by the source. This use is limited to evidentiary support for the paragraph and does not treat the entries as a new survey result.\n\nMention: \"Monthly Agricultural Household Report\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the figures 31 households and 18 percent from the Monthly Agricultural Household Report.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:00c71b072c0aaac9071976cb", "occurrence_id": "fe7f9129fa8e0a4150ebf20a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Audited Poverty and Vulnerability Survey is a survey produced by the World Bank for 2022 and serves as the principal evidence for the welfare assessment. Its observations provide the basis for reviewing differences in reported conditions across the covered population, while preserving a consistent reference for subsequent calculations. All citation details remain associated with the source in the supporting documentation.\n\nThe assessment derives a trend from the ordered observations and computes an average for the relevant welfare measures. These calculations are used to summarise the direction and central tendency of the distribution, rather than to reproduce isolated survey figures. The resulting measures support comparison across population groups and provide a concise basis for interpreting changes in vulnerability.\n\nThe calculated trend is considered alongside the average to reduce the risk that a single observation or a small number of observations determines the assessment. The approach also permits the results to be incorporated into monitoring tables without implying a level of precision beyond that supported by the survey. Conclusions are therefore based on the pattern generated from the source and on the consistency of the derived measures.\n\nThe source remains the central statistical reference for this analysis. Any subsequent interpretation should retain the distinction between the underlying survey observations and the trend and average calculated from them, particularly when the indicators are used for implementation monitoring or resource prioritisation.\n\nMention: \"Audited Poverty and Vulnerability Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives a distributional trend and an average from the 2022 survey rather than reproducing isolated figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:664d37e714a5b4a777f0604f", "occurrence_id": "1e51ab44298a5df3171896e8", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Audited Rural Credit Registry, a database produced by Bureau Van Dijk, as the principal empirical basis for reviewing rural credit performance. The source provides a consistent set of records suitable for comparison across reporting periods and supports checks on the internal coherence of the observed values.\n\nThe analysis does not reproduce individual registry entries as reported results. Instead, it aggregates the relevant observations, calculates period averages, and derives a trend from changes in the resulting series. This treatment reduces the influence of isolated observations and provides a summary measure for assessing the direction and stability of rural credit activity.\n\nThe derived averages indicate the central level of recorded activity, while the calculated trend identifies whether that level has strengthened, weakened, or remained broadly stable. These measures are used in the assessment tables and inform the interpretation of changes in access, repayment performance, and portfolio quality. Any differences between the computed indicators and individual records are therefore attributable to aggregation and calculation procedures rather than to a separate data source.\n\nThe registry is retained as the primary evidentiary basis for this analysis. Recalculation using the same underlying records produced materially consistent results, supporting the use of the reported averages and trend measures in the implementation assessment.\n\nMention: \"Audited Rural Credit Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives period averages and a trend from the Audited Rural Credit Registry rather than quoting individual records.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:8e265f968dd74bb723babb47", "occurrence_id": "0f839a5dbfc6bb22f243579b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Monthly Settlement Location Index (MSLI) indicator for Kinshasa as its principal evidence on settlement distribution. The source is treated as a structured sequence of location observations, supporting comparison across successive monthly records without introducing additional geographic or institutional classifications.\n\nThe analytical series is derived by grouping the observations into consistent periods and calculating a rolling average for the observed settlement measure. Changes in this average are then used to identify the direction and persistence of the trend. The resulting pattern provides a basis for assessing whether observed movements are gradual, concentrated, or subject to short-term variation.\n\nThis approach avoids presenting isolated index readings as findings in themselves. Instead, the assessment uses the calculated averages and their change over time to support interpretation of settlement dynamics in Kinshasa. The MSLI therefore functions as the central empirical input to the trend assessment, while the reported conclusions reflect transformations performed on the source records rather than direct reproduction of individual entries.\n\nThe results are interpreted with attention to the coverage and regularity of the underlying observations. Where the calculated trend is stable, it is taken as evidence of a persistent pattern; where the average shifts materially between periods, the change is treated as a signal requiring further review. No separate estimate is introduced to replace the source-based calculation.\n\nMention: \"Monthly Settlement Location Index (MSLI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the Monthly Settlement Location Index (MSLI) indicator for Kinshasa to calculate rolling averages and assess settlement trends.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:ecf97875af2ea56ac2144a4a", "occurrence_id": "e8c422df5e20fb502ba64b4a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment reviews the information base used to frame the discussion of non-state education provision. The Independent Non-State Education Registry database for 2018 is cited as a contextual reference alongside the general description of the sector. Its mention indicates the availability of a structured administrative source relevant to the topic, but it does not establish a measure used in the assessment.\n\nThe findings and implementation considerations presented below are based on the evidence reviewed for the current assessment. No counts, rates, classifications, or other values are drawn from the registry. The source is therefore not used to calculate changes, compare categories, or support an inference about performance. Its role is limited to documenting the wider information environment in which the assessment was prepared.\n\nThis treatment avoids attributing conclusions to a source that is not analysed in the accompanying text. References to participation, service delivery, and institutional capacity should be read as findings of the assessment process rather than as statistics extracted from the registry. Any subsequent use of the database would require separate verification of coverage, definitions, and reporting consistency before quantitative findings could be derived.\n\nMention: \"Independent Non-State Education Registry\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The source provided contextual background only, and no values or derived results were taken from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:ee90a35ebe7e6beb3798a901", "occurrence_id": "174b43b8d74a4a243d8d2010", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Quality-Assured Seed Use System (QSUS), an administrative source maintained in the project archive. The records provide a supporting basis for the reported account of seed access and use, while their provenance and internal structure permit review of the underlying entries. No additional institutional attribution is required for the claim presented here.\n\nRecords from QSUS were collected through the established reporting process and screened for completeness before analytical use. Variations in spelling, category labels, and recording conventions were harmonised to support consistent aggregation. Entries that could not be reconciled were retained separately and excluded from summaries requiring uniform classification.\n\nThe harmonised material was archived with the associated documentation so that the reported result can be checked against the source records. This procedure supports the broader finding that administrative reporting provides a usable basis for describing patterns of seed use, while preserving a clear distinction between recorded information and subsequent interpretation. The source therefore functions as evidence for one claim within the assessment rather than as a substitute for the full body of project monitoring information.\n\nMention: \"Quality-Assured Seed Use System (QSUS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The passage identifies QSUS as an administrative source and records the collection, harmonisation, and archiving of its records to support one assessment claim.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:13af30bdeef06a704588ce79", "occurrence_id": "d07be961fa68541660539807", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Verified Emergency Care Survey (VECS) is the principal supporting source for the assessment of emergency care access and readiness. As a 2023 survey, it provides structured evidence on reported use of emergency services, observed availability of essential functions, and the consistency of service delivery across the assessed units.\n\nThe survey is used to cross-check an emergency-care utilization estimate derived from a different administrative source. The comparison is intended to identify material divergence between reported service use and the level implied by administrative records. Agreement between the two measures is treated as support for the reasonableness of the estimate, while substantial differences are retained as a qualification on interpretation.\n\nVECS findings also inform the assessment of service readiness by documenting recurring constraints in staffing, basic equipment, referral arrangements, and continuity of care. These observations are used to distinguish limitations in recorded activity from limitations in actual service capacity. The evidence is therefore applied as a validation reference rather than as a replacement for the underlying estimate.\n\nTaken together, the survey results strengthen the reliability of the assessment where the reported patterns are consistent with the independently derived measure. Where the evidence is not aligned, the assessment records the discrepancy and avoids attributing the difference to a single operational cause.\n\nMention: \"Verified Emergency Care Survey (VECS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the 2023 VECS survey to cross-check an independently derived emergency-care utilization estimate and assess service readiness.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:0247e786effd23e81d9fa202", "occurrence_id": "10b726c10b29ebfe56655efd", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Rapid Inequality Survey, a survey, is used as the principal source for the distributional observations presented in this assessment. No adjustment, extrapolation, or reclassification has been applied to the two figures reproduced below. The source is cited in the text to preserve a clear link between the reported values and the underlying evidence.\n\nThe survey records that 42.6 percent of respondents were in households reporting difficulty meeting essential expenditure. It also records that 17.3 percent of respondents reported a reduction in access to basic services. These figures are taken directly from the Rapid Inequality Survey and are presented without calculation of a change, ratio, or composite indicator. Their inclusion is intended to establish the scale of reported constraint and its incidence across the observed population.\n\nThe findings should be interpreted as descriptive evidence for implementation monitoring. They do not, on their own, establish causality or identify the effect of any individual intervention. Subsequent assessment may examine whether the reported conditions vary by household characteristics and whether the pattern is consistent with administrative records. For the purposes of this annex, however, the two reported percentages provide the central quantitative reference for discussion of inequality-related risks and service access.\n\nMention: \"Rapid Inequality Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The author takes the directly reported figures of 42.6 percent and 17.3 percent from the Rapid Inequality Survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:2d1b1c52ca2de26db49adf37", "occurrence_id": "3eef4b4cf1a40111eac4f131", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The cited source is the Sentinel Hiring and Vacancies Index, an indicator produced by Digital South Asia Library for 2019. It provides the central empirical basis for the assessment of hiring activity and reported vacancies. The source is used as a consistent reference for describing movements in recorded demand for labour, without introducing estimates beyond the published series.\n\nDuring preparation of the assessment, the available records were collected from the source and reviewed for completeness. Entries were harmonised to support comparison across reporting intervals, while differences in terminology and recording practice were resolved through a common treatment of hiring and vacancy observations. The resulting archive preserves the connection between the original records and the analytical tables used in this document.\n\nThe evidence is therefore presented as a documented statistical input rather than as an independently generated forecast. Its principal contribution is to provide a traceable basis for the reported assessment, with the 2019 observations retained in harmonised form for verification and subsequent use. No additional institutional, geographic, or temporal attribution is applied to the cited indicator.\n\nMention: \"Sentinel Hiring and Vacancies Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The assessment collected, harmonised, and archived hiring and vacancy records for 2019.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:9346e08e50503f094b5d274a", "occurrence_id": "ef649b05360577e24501c2ac", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Rural Supply Chain Survey is cited as a 2019 survey providing background for the assessment of supply conditions. Its inclusion establishes that the discussion is informed by an established body of survey evidence, while preserving a clear distinction between contextual material and evidence used for measurement. The source is identified in the same paragraph as its citation to maintain traceability.\n\nWithin the broader discussion, the survey supports the claim that supply-chain performance should be considered alongside the operational factors examined in the assessment. This claim is qualitative and does not depend on a reported estimate, comparison, or distribution drawn from the source. The reference therefore helps frame the subject without extending the evidentiary basis of the analysis beyond the materials directly examined.\n\nNo numerical or categorical values from the Rural Supply Chain Survey are incorporated into the findings, tables, or interpretation. The citation has a limited supporting function and should not be read as documentation of a calculated trend or as a basis for an additional result. This treatment is consistent with the use of background references in implementation reporting, where a source may substantiate the relevance of a topic without supplying inputs to the reported assessment.\n\nMention: \"Rural Supply Chain Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The source provides contextual support for one broader claim, while no values from it are used.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:7a3d0520186a57c404cfb209", "occurrence_id": "f90303cbe8a73bc49b5fe81d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment considers the available evidence as part of a broader review of household financial conditions. Interpretation is limited to patterns that can be supported by the documented evidence, while descriptive statements are separated from any conclusions requiring direct measurement. This approach is intended to preserve a clear distinction between contextual information and findings derived from the assessment.\n\nThe Harmonised Debt and Savings Registry (HDSR) database is cited for the 2020 reference period as a source of background on the subject under review. Its inclusion supports the general observation that debt and savings records form part of the relevant information environment. No figures, indicators, or other values from the registry are reproduced, recalculated, or incorporated into the assessment.\n\nThe citation therefore serves a narrowly defined supporting function within the broader discussion. It reinforces one contextual claim concerning the availability of organised financial records, but it does not determine the assessment results or provide a basis for comparison across periods. The conclusions reported here rely on the evidence presented directly in the surrounding analysis, with the registry reference retained solely to clarify the wider evidentiary setting.\n\nMention: \"Harmonised Debt and Savings Registry (HDSR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The HDSR database is used only to support a contextual claim, and no values are taken from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:f856374a06f2c48da4a1ba62", "occurrence_id": "30e61251d4d96c5994ab340c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Baseline Migration and Remittance Index is an indicator produced by the U.S. Census Bureau and is used here as the principal statistical reference for interpreting the reported movement in migration-related remittance conditions. Its designation is retained in full to support traceability and to distinguish the cited evidence from subsequent analytical judgments.\n\nTwo figures are taken directly from the indicator for the assessment. The first is 31.4 percent, which is used as the recorded level for the principal measure under review. The second is 18.7 percent, which is used as the corresponding level for the related measure. These values are reproduced without recalculation, adjustment, or conversion, and they constitute the central quantitative evidence in this note.\n\nInterpretation is limited to the relationship visible between the two reported figures. The difference between them is noted descriptively, without presenting an inferred time pattern or attributing changes to an unreported factor. This approach preserves the distinction between source evidence and analysis and allows the figures to be checked against the cited indicator. No additional classification is introduced, and no supplementary institutional or geographic attribution is required for the comparison.\n\nMention: \"Baseline Migration and Remittance Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Baseline Migration and Remittance Index for two directly reproduced figures, 31.4 percent and 18.7 percent.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:38c2a341cf2d0f827180e303", "occurrence_id": "daad8feafd11dbe07f96f74d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Nationwide Digital Learning Report (NDLR) is the report used for the limited quantitative evidence presented in this assessment. Two figures are reproduced directly from it. The report records 31 grades within the relevant learning structure and indicates that 68 percent of the assessed provision had access to digital learning materials. These figures are cited as reported and are not recalculated or combined with estimates from other sources.\n\nThe evidence supports the broader finding that digital learning provision remains uneven across the education system. The presence of materials across 31 grades indicates substantial structural reach, while the reported level of access, at 68 percent, points to a material gap between formal coverage and effective availability. The contrast is therefore relevant to implementation planning, particularly where delivery depends on regular access to appropriate content and supporting arrangements.\n\nInterpretation remains subject to the scope and definitions used in the report. The figures should be read as indicators of recorded provision rather than as a direct measure of learner use, learning quality or continuity of access. They nevertheless provide a concise basis for identifying coverage constraints and for prioritising further verification during implementation. Subsequent monitoring should preserve the same counting basis so that changes in reported access can be assessed consistently.\n\nMention: \"Nationwide Digital Learning Report (NDLR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the NDLR figures of 31 grades and 68 percent to support the assessment that digital learning provision remains uneven.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:ebb1ac6ceb3b4f1da526ddb8", "occurrence_id": "28ee1d4d4e1ec950a7e93d11", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Rolling Poverty and Vulnerability Survey (RPVS), a survey of household conditions and reported exposure to economic risks. The source provides repeated observations that support the analysis of changes in welfare, employment security, and access to basic services. Its coverage is treated as an input to the assessment rather than as a stand-alone estimate.\n\nFor the relevant indicators, the analysis derives a general trend from the successive observations and calculates an average movement across the available reporting periods. This approach reduces the influence of short-term fluctuations and avoids presenting any single observation as representative of the broader pattern. The resulting measures are used to support the finding that vulnerability has not evolved uniformly across the population.\n\nThe derived trend is consistent with the wider assessment, which identifies persistent differences in household resilience and uneven improvement in living conditions. Individual survey responses are therefore not reproduced in this section. Instead, the calculated averages provide a concise basis for comparison and inform the interpretation of related results reported elsewhere in the document. The estimates should be read as analytical summaries of the survey evidence, subject to the limitations associated with sampling, reporting, and periodic measurement.\n\nMention: \"Rolling Poverty and Vulnerability Survey (RPVS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives average trends in vulnerability and household resilience from the Rolling Poverty and Vulnerability Survey (RPVS).\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:fdc565c1daff2811fb584ebb", "occurrence_id": "6a2762ad05adf02e7d90ab58", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Record-Based Rainfall Anomaly System (RRAS), an administrative source maintained as a structured record of reported rainfall departures. Its inclusion provides the evidentiary basis for the observation that rainfall anomalies were documented with sufficient regularity to support comparison across reporting periods. The source is used as a record base rather than as an independent projection or modelled estimate.\n\nFor implementation purposes, the available entries were collected, reviewed for internal consistency, and harmonised before being incorporated into the analytical file. Differences in recording practice were reconciled through common treatment of dates, units, and anomaly descriptions, while entries lacking adequate supporting information were retained only where their interpretation remained clear. The resulting archive preserves the original reporting content while allowing the assessment to draw on a consistent body of observations.\n\nThis process strengthens the reliability of the reported pattern and supports the related finding that departures from expected rainfall conditions were not isolated reporting artefacts. The evidence is therefore used to substantiate the assessment narrative, while conclusions remain limited to what is documented in the archived records. No additional inference is attributed to the source beyond its contribution to the documented evidence base.\n\nMention: \"Record-Based Rainfall Anomaly System (RRAS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The passage records the collection, harmonisation, and archiving of rainfall anomaly records from the Record-Based Rainfall Anomaly System (RRAS).\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:401cca8875cd9d4b88a6474c", "occurrence_id": "fc37cbfff01de9029c0281a1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Continuous Exam Performance Registry (CEPR) as the principal database for the analysis, with 2015 identified as the year to which the registry evidence refers. The source is retained as the central basis for the performance calculations presented in this annex.\n\nRather than reproducing individual observations, the analysis arranges the recorded outcomes across successive examination stages and estimates the direction of change. It also derives an average performance measure from the same observations, allowing the assessment to distinguish broad movement from isolated results. These calculations provide a summary of the distribution and progression represented in the CEPR database.\n\nThe resulting trend and average are used to inform interpretation of implementation performance. They should be read as analytical measures derived from the registry evidence, rather than as figures copied directly from a reported table. The approach gives greater weight to the overall pattern contained in the records and reduces the influence of any single observation. Consistent treatment of the entries supports comparison within the assessed sequence and provides a reproducible basis for the findings reported here.\n\nMention: \"Continuous Exam Performance Registry (CEPR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The assessment derives a performance trend and average from the 2015 Continuous Exam Performance Registry (CEPR) database.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:0278bb47b7f43f44126f4a76", "occurrence_id": "a7c0f41a2fd9ae72d0f273b8", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Decennial Migration and Remittance Survey is recorded as a survey produced by GeoTerra Image in 2023. It is included in the source register to document a relevant body of information consulted during preparation of the assessment. The citation establishes provenance for the background material but does not define the measurement framework applied in this report.\n\nThe survey is referred to briefly alongside the broader evidence base. Its inclusion does not support an estimate, comparison, projection, or interpretation presented in the text. No tables, indicators, rates, totals, or respondent characteristics are reproduced from the source, and no calculation relies on its contents.\n\nFor purposes of this assessment, the reference serves only to acknowledge related documentation and to indicate that migration and remittance issues were considered during the review of available material. Findings reported below are based on the evidence specified in the relevant sections and should not be read as derived from the cited survey. The source is therefore retained as a contextual reference rather than an input to the reported results.\n\nMention: \"Decennial Migration and Remittance Survey\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"No numerical values or analytical results were taken from the Decennial Migration and Remittance Survey; it was retained solely as background reference.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:8a057507c3330b68faaff301", "occurrence_id": "c3b22ceea944fb335ac2f7e5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The population analysis supports the assessment of service demand, settlement pressure, and the distribution of households across the reporting framework. It is used as an input to derived indicators, with interpretation focused on comparable population levels rather than on isolated records.\n\nThe Audited Population Grid Dataset, a geospatial source produced by HCP for 2019, provides the underlying population surface for this exercise. Its observations are treated as a consistent analytical base, subject to the limitations associated with gridded estimation and the spatial allocation of population. No additional source characteristics are assumed beyond those required for the calculation.\n\nThe assessment computes an average population level and a trend from the dataset over the analytical intervals used in the results. Accordingly, the figures presented in the narrative are derived measures and should not be read as direct quotations of individual grid values. This treatment allows changes in the distribution and scale of population to be considered in relation to planning requirements, while reducing the influence of local irregularities in any single observation.\n\nThe resulting measures are used to support prioritisation and to establish a reproducible basis for monitoring. They should be interpreted as analytical estimates for comparison across the reporting framework, rather than as independently verified counts.\n\nMention: \"Audited Population Grid Dataset\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the 2019 HCP geospatial Audited Population Grid Dataset to compute an average and a population trend.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:138108af25489986152bedd3", "occurrence_id": "330c223e258850097e676b72", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment records selected quantitative evidence relevant to the review of forest-product activity. The Regional Forest Products Survey (RFPS) is identified as the survey source for 2015, and its results are used solely to preserve two reported magnitudes in the supporting narrative. No adjustment, projection, or supplementary classification is applied to these observations.\n\nThe first value reproduced from the source is 42 units, representing the reported quantity for the indicator under review. The second value is 67 units, likewise retained in the form presented by the survey. These figures are transcribed directly and are not recalculated, rounded, or interpreted as a time series. Their inclusion provides a concise numerical reference for the assessment without extending the evidentiary scope of the source.\n\nThe data are presented as documentary inputs rather than as independently verified estimates. Differences between the two reported quantities should therefore be read as a comparison of source figures only. The cited material supports the limited descriptive purpose of this annex and does not, by itself, establish causation, coverage beyond the reported observations, or a broader performance conclusion.\n\nMention: \"Regional Forest Products Survey (RFPS)\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"source\", \"summary\": \"The passage takes the figures 42 and 67 directly from the Regional Forest Products Survey (RFPS) for 2015.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:f6c28d4881a1ef85e5e1cf14", "occurrence_id": "1c5733c8dcf2cd61a54bdc28", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Rapid Enterprise and Jobs Microdata is the principal microdata source used for the assessment of enterprise activity and employment in the Republic of Zambia. It is produced by NSS and provides structured observations at the enterprise level, allowing the analysis to examine changes in business participation, staffing, and reported job creation. The source is treated as the evidentiary basis for the estimates presented below.\n\nThe analysis derives average employment and enterprise measures across the available observations and calculates the direction of change between successive reporting periods. This approach reduces the influence of individual records and supports a more stable assessment of aggregate movement. The resulting trend indicates whether enterprise activity and job creation are expanding, contracting, or remaining broadly unchanged over the period represented in the source.\n\nThe estimates are therefore analytical results derived from the Rapid Enterprise and Jobs Microdata rather than figures reproduced from a single entry. Particular attention is given to consistency across observations, with averages used to summarise the distribution and the calculated trend used to interpret overall performance. These results inform the assessment of implementation progress and provide the statistical basis for the related discussion of enterprise and employment outcomes.\n\nMention: \"Rapid Enterprise and Jobs Microdata\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses the Rapid Enterprise and Jobs Microdata to calculate average enterprise and employment measures and derive their overall trend.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:661142d2c98ab4414036b39a", "occurrence_id": "84c2bc93267a6589bbc1624e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Ministerial Immunisation Coverage Survey (MICS) as a contextual reference to the broader information environment relevant to immunisation planning. The source is identified as a survey and is treated as background material alongside the assessment’s own evidence. Its mention indicates the availability of a separate measurement exercise, but it does not establish any result reported in this document.\n\nNo figures, rates, totals, denominators, classifications, or other values are taken from the Ministerial Immunisation Coverage Survey (MICS). The analysis does not derive estimates, compare observations, or support conclusions through that source. Findings presented in the assessment rely on the evidence described in the relevant sections and on the procedures applied there.\n\nThe reference is retained to assist interpretation of the information base and to distinguish contextual documentation from evidence used in the calculations. Readers should therefore understand the citation as a signpost to related survey material rather than as a basis for any reported measurement. No further attribution or source detail is required for the results discussed in this annex.\n\nMention: \"Ministerial Immunisation Coverage Survey (MICS)\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"No values or analytical work are taken from the cited survey; it is included only as contextual background.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:d1df40f555ef6b23bf8634fb", "occurrence_id": "37d6d5b052d616de3cc78e41", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Annual Household Consumption Census is a census referring to 2015 and provides the principal empirical basis for the consumption analysis. Its records are used to establish comparable household groupings and to assess the distribution of reported consumption across the observation set. No supplementary source is required for the calculations presented in this section.\n\nThe analysis derives average consumption measures from the underlying records rather than reproducing individual reported figures. These averages are calculated for each household grouping and then compared to identify the direction and relative pace of change. The resulting trend is used as the central indicator for assessing changes in consumption conditions.\n\nResults are presented in aggregated form to reduce the influence of isolated observations and to support interpretation at the programme level. The calculations retain the internal structure of the census while applying consistent treatment to household records. Differences between group averages are therefore interpreted as analytical estimates derived from the source, rather than as figures directly quoted from its tables. The evidence supports a cautious assessment of variation in consumption and provides a basis for monitoring subsequent changes.\n\nMention: \"Annual Household Consumption Census\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses the 2015 Annual Household Consumption Census to calculate household-group averages and derive a consumption trend.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:1660a7efe9b4337020acbc7b", "occurrence_id": "7aff1ec6d59dcaaa290641b4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Independent Administrative Boundary Survey for Thailand as a survey supporting the review of territorial coverage. The source is treated as the controlling reference for the administrative units and boundary segments considered in this annex. No adjustment has been applied to the figures reproduced from the source.\n\nFor the present assessment, the reported totals of 31 administrative units and 74 boundary segments are reproduced directly. These two figures support the finding that the geographic basis for implementation is sufficiently differentiated to permit disaggregated planning and monitoring. The figures are used as reported and are not combined with estimates from other records.\n\nThe evidence also provides a consistent basis for checking the internal distribution of activities across the covered area. In operational terms, the delineation supports assignment of responsibilities, review of coverage, and identification of gaps during implementation. The cited source is therefore used to substantiate the boundary-related claim in this paragraph, while the remaining analysis concerns the implications of the reported structure rather than a recalculation of it.\n\nMention: \"Independent Administrative Boundary Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported figures of 31 administrative units and 74 boundary segments from the Independent Administrative Boundary Survey for Thailand.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:663057f51d528cbe8e4b24d8", "occurrence_id": "6854991082d940cc9a639d1d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Joint Settlement and Elevation Dataset (JSED) is used as a geospatial source for checking the internal plausibility of the settlement and elevation estimates reported in this assessment. Its use is limited to verification of a specific claim concerning the distribution of observed conditions, rather than to the construction of a separate monitoring series.\n\nThe estimate is first derived from the principal evidence reviewed for the assessment. It is then cross-checked against the JSED record to identify material discrepancies in the reported pattern and to assess whether the resulting interpretation remains credible. This comparison provides an additional control on the estimate while preserving the distinction between the underlying evidence and the supplementary check.\n\nThe cross-check indicates that the direction and broad magnitude of the claim are consistent with the independent geospatial evidence. Minor differences are not treated as evidence of error, since they may reflect differences in coverage, measurement procedures, or the timing of observation. The result is therefore used as a reasonableness test and not as a replacement for the primary estimate.\n\nFor implementation purposes, the finding supports retention of the reported estimate in the main analysis, subject to the limitations noted above. Future review should continue to apply the same verification principle where estimates rely on incomplete or unevenly distributed observations. The JSED citation is consequently retained in the supporting documentation as evidence of the cross-check undertaken.\n\nMention: \"Joint Settlement and Elevation Dataset (JSED)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the JSED geospatial source to cross-check an independent settlement and elevation estimate.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:91e88dcf4f951a266f4fcc77", "occurrence_id": "3573013f1421c740ad421c69", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: To support the assessment of distributional change, the Nationwide Output Area Index, an indicator produced by the Uganda Bureau of Statistics, is used as the underlying measure for one claim concerning relative output performance. The index is treated as a consistent analytical input rather than as a set of headline values reproduced in the text. Its role is to provide a common basis for examining movement across the reporting period and for checking whether the observed pattern is sustained.\n\nThe analysis calculates an average from the relevant observations and derives a trend from their sequence. These calculations are used to characterise the direction and approximate persistence of change, while limiting the risk that an isolated observation will determine the interpretation. The resulting summary supports the assessment that performance was uneven, with the central tendency and direction of movement considered together. No individual index reading is treated as conclusive evidence on its own.\n\nThis use of the indicator is aligned with the wider results framework, in which changes in output are interpreted alongside implementation progress and variation in recorded conditions. The computed measures therefore serve as supporting evidence for the claim rather than as a standalone outcome measure. Any difference between the derived average and the direction of the trend is retained in the assessment, since divergence may signal dispersion, timing effects, or changes in the composition of observations. The source is cited to ensure traceability of the calculation and to permit subsequent review of the underlying series.\n\nMention: \"Nationwide Output Area Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the Nationwide Output Area Index to derive an average and a trend supporting an assessment of uneven output performance.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:8093d2e071328889f10fe250", "occurrence_id": "0b7eafd613f2933460fd5c3d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Urban Remote Sensing Registry (URSR) database, covering Sierra Leone for 2019, is cited as a background reference for the statistical discussion. Its mention establishes the broader information environment in which the assessment is situated, without affecting the definitions, classifications, or estimates presented below.\n\nThe registry is not used as an evidentiary basis for the findings. No figures, tabulations, spatial measures, or derived indicators are taken from the source, and no comparison is made between its contents and the results reported in this document. The reference is retained solely to acknowledge a relevant source of remotely sensed information.\n\nThe analysis therefore relies on the measures and procedures described in the relevant sections of the report. Interpretation is limited to those materials, while the URSR citation provides passing background on available information concerning the area under review. This treatment avoids attributing any reported change, level, or distribution to the cited database.\n\nMention: \"Urban Remote Sensing Registry (URSR)\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"No values or analytical results were taken from the URSR source; it was cited only as background information.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:14c05ebb2c2c0396d1486ab0", "occurrence_id": "a6775f7031a6043edad0031e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Data sources.** The Enumerative Immunisation Coverage Registry (EICR) database, produced by WTO and referring to 2019, is used as the documentary basis for the coverage statement in this assessment. The source is treated as a direct administrative record for the period under review, with no adjustment applied to the reported observations.\n\nThe assessment records 87 percent for the eligible population receiving the relevant service and 64 percent for the corresponding population in the comparison category. These two percentages are reproduced directly from the EICR database and are used to substantiate the broader finding that reported coverage was uneven across the reviewed population. The figures are presented as published and should not be interpreted as a recalculated estimate.\n\nThe evidence supports continued attention to the completeness and consistency of immunisation reporting. In particular, the difference between the two reported levels indicates that aggregate performance can conceal material variation between population categories. The cited source therefore serves as supporting evidence for the assessment narrative rather than as the basis for a new trend calculation. Any interpretation remains limited to the information recorded in the database and to the period identified above.\n\nMention: \"Enumerative Immunisation Coverage Registry (EICR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage takes the 87 percent and 64 percent coverage figures directly from the 2019 EICR database.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:169cfd8753749280f7829f6b", "occurrence_id": "751ddf48843876ab71beced4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Statistical Unemployment Insurance Index (SUII) as a 2023 indicator for the review of income protection arrangements. The source is cited as the reference point for a limited comparison of reported performance, with its terminology retained in the accompanying tables and narrative. No adjustment is applied to the published scale, and the figures are treated as observations rather than estimates produced by the assessment.\n\nThe assessment quotes two figures directly from the Statistical Unemployment Insurance Index (SUII), namely 62.4 and 47.8. These values support the broader finding that access to unemployment insurance remains uneven across the measures considered. Their inclusion is intended to substantiate the statement on differing levels of protection, rather than to establish a separate analytical series or imply a change over time.\n\nThe figures are presented without additional aggregation. Interpretation is therefore limited to the relationship indicated in the source, while the surrounding discussion addresses administrative reach, eligibility conditions, and the practical availability of support. This treatment preserves the distinction between the cited observations and the assessment’s wider conclusions.\n\nThe Statistical Unemployment Insurance Index (SUII) is consequently used as documentary evidence for the stated comparison. The cited values provide a concise quantitative basis for the claim, while the remaining analysis relies on the broader evidence reviewed in the assessment.\n\nMention: \"Statistical Unemployment Insurance Index (SUII)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the SUII indicator’s 2023 figures of 62.4 and 47.8 to support a claim about uneven unemployment insurance protection.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:b344b80673ad2ad0d5e98bd7", "occurrence_id": "360dbf6f0349fac18a8c3b28", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the High-Resolution Water Body Dataset as its central geospatial source for establishing the underlying evidence on mapped water bodies. The material is treated as an administrative and analytical input rather than as a narrative account. Its coverage supports consistent identification of water-body presence, observed extent, continuity, and broad classification across the records reviewed.\n\nThe working files were collected from the source, checked for internal consistency, and harmonised to a common structure before analysis. Differences in naming, geometry, representation, and apparent duplication were reconciled through documented review. Where records could not be aligned without introducing an unsupported assumption, they were retained with an explicit quality notation and excluded from comparative aggregation.\n\nThe harmonised material was archived with its processing history, allowing the evidence used in the assessment to be traced from the original source through cleaning and consolidation. This procedure provides a stable basis for tabulation and map-based verification while limiting the risk that changes in formatting or interpretation are mistaken for changes in water-body conditions. Results presented in the accompanying analysis therefore reflect the archived, harmonised evidence drawn from the dataset, with no inference beyond the information contained in the source.\n\nMention: \"High-Resolution Water Body Dataset\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The work taken from the source is the collection, harmonisation, quality review, and archiving of evidence on water-body extent, continuity, presence, and classification.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:56bae4d670f2afd0c085a6e2", "occurrence_id": "c0e9d2775ea9766d5106acab", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis draws on the Longitudinal Unemployment Insurance Survey, identified as a survey for 2015. The material provides observations on unemployment duration, benefit receipt, and subsequent labour-market status. Its coverage and measurement conventions are retained for the analytical series, with no supplementary source introduced in this assessment.\n\nThe reported results are derived from the source rather than reproduced as individual survey figures. Observations are grouped into comparable intervals, and the resulting means are used to assess changes in unemployment duration and benefit receipt. The calculation also provides an estimated trend in the transition from unemployment to work, allowing movements across the observed population to be assessed on a consistent basis.\n\nThe derived averages indicate the central tendency of the recorded outcomes, while the trend measure summarises their direction over the analytical sequence. These calculations are treated as the principal evidence for interpreting changes in benefit reliance and labour-market re-entry. Minor differences between subgroup results are not presented as separate source estimates, but are incorporated into the aggregate calculations used in the assessment.\n\nMention: \"Longitudinal Unemployment Insurance Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses 2015 survey observations on unemployment duration, benefit receipt, and labour-market re-entry to calculate averages and an overall trend.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:8766ef3fb12e206acf99b0d4", "occurrence_id": "8b68cea18319133937291ac3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis uses the Representative Shock Coping Survey, a 2023 survey produced by the U.S. Census Bureau, as the underlying source for the coping indicator. The source provides observations on household responses to adverse conditions and is used to support the assessment of changes in reported coping behaviour.\n\nIndividual observations are not reproduced as standalone results. Instead, the analysis calculates an average across the available responses and compares the resulting values over the ordered reporting periods. This treatment reduces the influence of isolated responses and provides a consistent basis for identifying the direction and persistence of change.\n\nThe derived trend is used as supporting evidence for the broader finding that coping capacity varies over time and that reported responses do not move uniformly across the observation period. The calculation is therefore interpretive rather than a transcription of source figures. It does not imply that the source directly reports the trend or the average presented in this assessment; those measures are generated from the survey observations for analytical use. Results should be read alongside the associated indicators and methodological qualifications, with attention to the limits imposed by response coverage and the aggregation procedure.\n\nMention: \"Representative Shock Coping Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives an average and a trend from the 2023 Representative Shock Coping Survey rather than reproducing individual figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:6322148504b11b4886da4dba", "occurrence_id": "b9a2d500c55930ad07ffdf85", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws its central evidence from the Small-Area Inequality Registry (SIR) database. The source is used as a direct statistical reference for two reported measures, with no adjustment or recalculation applied in this note.\n\nThe first figure recorded in the registry is 18.4 percent for the indicator under review. The second figure is 27.1 percent for the corresponding comparison. These values are reproduced directly from the source and are presented to establish the observed level and distribution of the reported outcome. They should therefore be read as registry figures rather than as estimates generated through the analytical procedures described elsewhere in the report.\n\nThe two observations provide a compact basis for the assessment of variation within the reporting framework. Their use is limited to documenting the values recorded in the source. No additional interpretation is assigned beyond the distinction between the lower and higher reported figures, and no inference is made regarding changes over time. The registry remains the authoritative reference for both numbers cited in this paragraph, subject to the definitions and coverage applied in the underlying database.\n\nMention: \"Small-Area Inequality Registry (SIR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported figures of 18.4 percent and 27.1 percent from the SIR database.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:9987d463bd06d71517d834e4", "occurrence_id": "6886214a8d15994bcde7503a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The National Protracted Situation Registry database is used as the administrative reference for the assessment of recorded cases. Its entries provide a consistent basis for reviewing registration status, identifying unresolved records, and checking the internal coherence of reported totals. The source is treated as a documentary record rather than as an analytical estimate.\n\nThe registry records 31 registered cases and 18 pending cases. These figures are taken directly from the database and support the assessment that the recorded caseload includes a material unresolved component. The comparison is presented descriptively and does not imply a change over time or an estimate beyond the records themselves.\n\nThe information is sufficient for basic verification of coverage, status classification, and reconciliation of summary tables. Interpretation remains limited to the categories represented in the source, and no adjustment has been applied to the reported counts. On this basis, the database provides a suitable reference for monitoring implementation progress and documenting outstanding cases.\n\nMention: \"National Protracted Situation Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the National Protracted Situation Registry database for verification and reports 31 registered cases and 18 pending cases.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:41a033bbec6655da6cbee4fa", "occurrence_id": "ed61fc0a4aecb359a69aa591", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Household Building Footprint Census (HBFC) is used as the principal source for the built-environment indicators presented in this assessment. It is a census covering Cairo and referring to 2019. The source is cited for descriptive evidence only, with no adjustment to its reported units or geographic coverage.\n\nTwo figures are reproduced directly from the HBFC to support the assessment of concentrated urban development. The census reports 1.84 million building footprints and identifies 62 percent of mapped structures as located within the established built-up area. These values provide the quantitative basis for the claim that development is spatially concentrated rather than evenly distributed across the area covered.\n\nThe figures are presented as reported measures and should not be interpreted as a time series or as evidence of change between observation periods. Their use is limited to establishing the scale and distribution of the building stock recorded in the cited census.\n\nAccordingly, the HBFC serves as a supporting reference for the descriptive paragraph and does not provide a basis for estimating construction rates, population totals, or future demand. Any comparison with other evidence would require separate documentation of coverage, definitions, and measurement procedures.\n\nMention: \"Household Building Footprint Census (HBFC)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage takes 1.84 million building footprints and 62 percent of mapped structures directly from the HBFC to support the claim of concentrated development in Cairo.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:5e568528fc33714511693268", "occurrence_id": "3886a5c8406833a951df4fde", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment considers seed-use measurement as one element of the broader review of agricultural inputs. Interpretation is based on the reported structure of the monitoring framework and on the consistency of definitions applied across the available material. Particular attention is given to whether observed differences reflect substantive conditions or variation in coverage.\n\nThe Panel Seed Use Census is a census produced by DLA Piper and is used solely to provide contextual background for this assessment. Its values are not incorporated into the calculations, comparisons, or findings presented here. The source nevertheless supports the general observation that seed-use measurement requires clear coverage rules and consistent treatment of reporting units.\n\nThis qualification is material to the interpretation of the results. The evidence reviewed for implementation purposes remains limited to the material directly assessed under the project framework, while the cited census informs the framing of one claim concerning measurement comparability. No inference is drawn from the cited source beyond that contextual function, and no estimate or trend is attributed to it.\n\nMention: \"Panel Seed Use Census\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The source provides contextual support for one claim about measurement comparability, while none of its values are used.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:fc6beb39699567b833ca0753", "occurrence_id": "df4ede06e6066e3f500a3024", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment records that food security conditions vary across the project area and that service planning should therefore account for spatial differences. This observation is relevant to targeting, sequencing, and the interpretation of operational risks. It is presented as one element of a broader review covering implementation progress, service access, and monitoring arrangements.\n\nThe Statistical Food Security Dataset is cited as a geospatial source supporting this contextual observation. The reference indicates that spatially differentiated food security information is relevant to the assessment, but the dataset’s values are not incorporated into the tables, calculations, thresholds, or rankings presented here. No extraction, comparison, or recalculation from that source has been undertaken.\n\nThe finding is consequently used to frame the discussion rather than to establish a measured result. Programme responses should continue to be guided by verified monitoring evidence and documented implementation records. Any subsequent use of the cited source would require a separate review of coverage, comparability, and analytical suitability before its values could inform decisions or performance judgments.\n\nMention: \"Statistical Food Security Dataset\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The passage uses the Statistical Food Security Dataset only to support a contextual statement about spatial variation and extracts no values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:bcd73fa52176bc39d9d87f53", "occurrence_id": "2e1dc62748124b39a4fd9a00", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Metropolitan Forced Population Index indicator for 2023 as the central statistical reference for interpreting population pressure. The source provides a consistent basis for presenting the observed level without introducing supplementary classifications or external estimates.\n\nTwo values are reproduced directly from the indicator: 31.4 and 68.9. These figures constitute the principal evidence considered in this assessment. The first value is used to represent the lower recorded level, while the second identifies the higher recorded level within the material reviewed. No transformation, averaging, or recalculation has been applied to either figure.\n\nThe difference between the two values is therefore presented as a descriptive feature of the reported data rather than as a derived measure of change. This treatment preserves the source values and limits interpretation to the evidence available in the cited indicator. The figures support a cautious assessment of variation and provide a reference point for subsequent monitoring, subject to continued consistency in reporting and application.\n\nMention: \"Metropolitan Forced Population Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Metropolitan Forced Population Index indicator for 2023 and reproduces the figures 31.4 and 68.9.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:a3e3b0402911814e28a8b528", "occurrence_id": "0b4717bb31171748119a0ffa", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Continuous Apprenticeship Index (CAI) indicator for 2023 as a statistical input to the analysis of participation continuity. The series is considered relevant to the assessment because it provides a consistent basis for examining changes in apprenticeship engagement across the observations included in the source. No individual entry is treated as a standalone result, and the source is used only within the analytical framework described below.\n\nFor the purposes of the results section, the assessment computes an average from the CAI observations and derives a trend across the available sequence. These calculations are used to identify the general direction and level of continuity, rather than to reproduce or restate particular source values. The resulting measures support the finding that participation displays a discernible pattern over the period covered by the data.\n\nThe calculated average is used as a summary measure, while the trend provides an indication of movement over time. Together, they inform the interpretation of implementation performance and help distinguish broad changes from isolated fluctuations. The conclusion is therefore based on a derived assessment of the series, with the Continuous Apprenticeship Index (CAI) indicator serving as supporting evidence for the broader finding.\n\nMention: \"Continuous Apprenticeship Index (CAI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The assessment uses the 2023 CAI indicator to calculate an average and derive a trend in apprenticeship participation continuity.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:189a5638c7944eda23887185", "occurrence_id": "09360d95a0ff97bd6c868593", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Record-Based Apprenticeship Survey (RAS), a survey covering Sierra Leone in 2022, is used in this assessment as contextual evidence on the institutional relevance of apprenticeship arrangements. It supports the observation that apprenticeship activity should be considered when interpreting patterns in skills acquisition and transitions into work. The citation is limited to this contextual function, and no values from the source are used in the analysis.\n\nThe broader discussion draws on administrative and operational considerations to assess the consistency of reported participation, completion, and progression. Within that discussion, the RAS provides background for one claim concerning the presence of apprenticeship pathways in the skills system. It does not serve as a basis for calculating rates, comparing groups, or establishing changes over time.\n\nAccordingly, the source is retained in the evidentiary record but is not incorporated into any table, estimate, or derived indicator. Its role is to reinforce the interpretation that apprenticeship arrangements form part of the relevant delivery context, while the findings and numerical conclusions presented elsewhere rely on other material.\n\nMention: \"Record-Based Apprenticeship Survey (RAS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Record-Based Apprenticeship Survey (RAS) is used only to support a contextual claim about apprenticeship arrangements in Sierra Leone in 2022, without drawing values from the survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:f7b3aa56a074b6cf02aed9ec", "occurrence_id": "67f74d4f40dddc45f690be27", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The service-availability estimate is cross-checked against the Verified Mental Health System, an administrative source referring to 2024. The comparison is used to assess whether the estimate is consistent with recorded reporting patterns and to identify material divergence before the results are interpreted. The cited record supports the estimate as a reasonable representation of observed service coverage, while not being treated as a replacement for the underlying calculation.\n\nThe cross-check focuses on the presence of reported services and the distribution of recorded observations across reporting units. Differences between the estimate and the administrative record are examined for possible effects of incomplete reporting, timing differences, or variations in the definition of a functioning service. No adjustment is applied solely because the two measures are not identical.\n\nThe evidence is therefore used for validation within the broader assessment rather than as a separate outcome measure. Where the two sources indicate similar levels, confidence in the reported estimate is strengthened. Where they diverge, the result is retained with a qualification concerning measurement coverage and comparability. This treatment is consistent with the use of administrative evidence as a corroborating reference in implementation reporting.\n\nMention: \"Verified Mental Health System\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The author took a 2024 administrative cross-check of the service-availability estimate from the Verified Mental Health System.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:b6a48ef2d0b12be22bf98784", "occurrence_id": "547cd9ec506a262356ee99c2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis relies on the Follow-Up Social Protection Survey, a survey produced by oDesk, as its principal empirical source. The material is used to examine changes in reported access to support and the distribution of assistance across the observed population. No supplementary institutional source is introduced for this assessment.\n\nThe reported results are calculated from the observations contained in the survey rather than reproduced as isolated entries. Successive records are aligned on their common measures, and an average is derived for each comparison interval. These averages provide the basis for identifying the direction and relative strength of the underlying trend. The resulting series is therefore an analytical construction based on the source, not a transcription of selected responses.\n\nThe calculated pattern indicates gradual movement in the measured outcome, with variation between observations remaining material. The average is used to reduce the influence of individual responses and to support comparison across the available records. Interpretation remains limited to the information captured by the survey, including its stated coverage and response structure. The findings should consequently be read as estimates derived from reported responses rather than as a complete administrative account.\n\nOn this basis, the survey provides the central evidence for monitoring change, while the derived averages and trend supply the principal basis for assessment. This treatment preserves the source’s role as a consistent reference and makes explicit that the reported conclusion follows from aggregation and comparison of its contents.\n\nMention: \"Follow-Up Social Protection Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage derives comparative averages and a trend from the Follow-Up Social Protection Survey produced by oDesk.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:bd0ae084ae6e89642bd67108", "occurrence_id": "d2e2a59670e2c4dcde36e14b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The data source, the Quarterly Inequality Index (QII), is used as an indicator in the supporting statistical record. It provides a compact reference for reviewing distributional conditions and for maintaining consistency across the tables presented in this assessment. No further methodological interpretation is required for its limited use here.\n\nTwo figures are reproduced directly from the indicator, 0.42 and 0.47. They are included as reference points in the narrative and are not used to derive an additional measure, establish a trend, or support a separate estimation. Their role is limited to documenting the values reported by the cited source.\n\nThe figures are retained in their reported form to avoid introducing rounding or transformation that could obscure correspondence with the underlying record. The Quarterly Inequality Index (QII) therefore serves as a traceable statistical reference within the assessment, while the surrounding analysis remains based on the broader evidence presented in the relevant tables and discussion.\n\nMention: \"Quarterly Inequality Index (QII)\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"source\", \"summary\": \"The passage reproduces the two figures 0.42 and 0.47 from the QII indicator.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:f5f4649943740b8fe97aeef8", "occurrence_id": "898dc37ae27577f58d3175d9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Government Temperature Records Index is used as an indicator for 2016 in the assessment of reported temperature conditions. It provides a reference point for interpreting the numerical evidence presented in this section and supports a cautious reading of differences across reporting periods.\n\nTwo figures are reproduced directly from the index, namely 18.4 and 22.1. Their inclusion is limited to evidentiary support and does not imply a recalculation, projection, or adjustment of the underlying record. The figures are cited as reported so that the assessment remains traceable to the source identified above.\n\nTaken together, the figures support the broader claim that recorded temperature conditions are not uniform across the periods considered. This observation is used to qualify the interpretation of related results, rather than to establish a trend or attribute changes to a specific factor. The evidence therefore serves a bounded role within the paragraph and is not treated as a comprehensive measure of environmental conditions.\n\nThe source is retained in the project evidence base for reference during review of the statistical narrative. Any subsequent comparison should preserve the reported units and definitions associated with the Government Temperature Records Index and should avoid extending the two quoted figures beyond the claim they support.\n\nMention: \"Government Temperature Records Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Government Temperature Records Index as the 2016 indicator and reproduces the figures 18.4 and 22.1 to support a broader claim.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:22572c32b7204f0b88e94ebc", "occurrence_id": "67523d0108d492d9fe0bd1c1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Departmental Graduation Programmes Survey is treated as the principal empirical basis for this assessment. It is a 2019 survey produced by Central Electricity Boards and is used to characterise progression across the observed programme records. The source is retained as the reference point for the calculations reported below, with no additional institutional or geographic attribution required.\n\nThe analysis converts the reported observations into comparable annual measures and derives an overall average for the period represented in the survey. It also calculates the direction and magnitude of change between successive observations, allowing the assessment to distinguish a sustained trend from isolated variation. These results provide the basis for interpreting programme performance and for identifying whether observed movement is broadly stable, improving, or declining.\n\nThe resulting indicators are presented as analytical measures derived from the source rather than as reproduced entries. The average is used to establish the central level of performance, while the calculated trend provides an indication of consistency over time. This treatment supports comparison within the assessment framework and limits the influence of unusually high or low individual observations. Any conclusion concerning progress is therefore based on the aggregate pattern generated from the survey data.\n\nMention: \"Departmental Graduation Programmes Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the 2019 Departmental Graduation Programmes Survey produced by Central Electricity Boards to calculate an average and a trend.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:2981c6b7969edafde065a059", "occurrence_id": "6b5a58f6ac4750bc5cb8b229", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The High-Frequency Camp Management Registry is treated as a database produced by UNFCCC and serves as the principal reference for the comparison. Its recorded entries provide an administrative benchmark against which an estimate derived from a separate source is cross-checked. The exercise is intended to identify material divergence before the estimate is used in implementation reporting.\n\nThe registry evidence is reviewed for internal consistency across reported populations, operational status, and recorded service availability. Particular attention is given to duplicated entries, incomplete records, and changes in reporting coverage, since each may affect the relationship between the registry count and the independently derived estimate. Where the two measures differ, the discrepancy is retained for reconciliation rather than resolved through an unsupported adjustment.\n\nThe comparison indicates whether the independent estimate remains within a plausible range supported by administrative reporting. The registry is therefore used as corroborating evidence, not as a substitute for the separate estimation method. Results are incorporated into the assessment annex with a clear distinction between observed registry records and values inferred through estimation. This treatment preserves traceability and allows subsequent review of the basis for any adjustment to reported totals.\n\nMention: \"High-Frequency Camp Management Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the High-Frequency Camp Management Registry database produced by UNFCCC to cross-check an independent estimate against registry records.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:9699c562bf748dcf93565533", "occurrence_id": "e146aef56471a63055efd345", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Pilot Coastal Erosion Dataset (PCED), a geospatial source covering India for 2024. The dataset provides the principal empirical basis for the coastal exposure assessment and is used to support the aggregation of observations across the assessed shoreline. Its use is limited to the stated geographic and temporal coverage.\n\nFor analytical purposes, the observations are converted into comparable segment-level measures and reviewed for consistency before aggregation. The resulting series is used to calculate an average erosion measure and to identify the direction and approximate pace of change across the covered area. These calculations provide a summary signal for implementation planning rather than a reproduction of individual dataset entries.\n\nThe derived trend is assessed against the distribution of observations so that isolated high or low values do not determine the interpretation. Where coverage is uneven, the average is read as an indicator for the observed segments and not as a complete estimate for every coastal location. The analysis therefore treats the PCED as the central evidence for estimating spatial variation and for establishing the reference pattern used in subsequent monitoring. Findings should be understood as analytical results generated from the source, with conclusions conditioned by its coverage and measurement structure.\n\nMention: \"Pilot Coastal Erosion Dataset (PCED)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses the Pilot Coastal Erosion Dataset (PCED) for India in 2024 to derive an average erosion measure and trend.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:5ba8abd8f528dd47e443fb4c", "occurrence_id": "2d4e71918ec76a5c7ae5fe89", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Rapid Early Warning Index is cited as a 2020 indicator in the assessment framework. The reference is retained to acknowledge a relevant early warning measure and to situate the discussion within the broader evidence base. It is not used to establish a finding, define a benchmark, or support a comparison.\n\nThe analysis relies on the indicators and records specified for the present exercise. No numerical result, score, threshold, category, or classification from the Rapid Early Warning Index enters the tables, calculations, or narrative conclusions. Its mention therefore has no effect on the reported estimates or on the interpretation of implementation performance.\n\nFor purposes of traceability, the source is recorded in the data sources discussion alongside other background references. The citation is limited to recognition of its existence and stated analytical relevance. No extraction, transformation, validation, or reconciliation of its underlying observations was undertaken.\n\nThe findings and associated recommendations should consequently be read independently of the Rapid Early Warning Index. Any future use of that indicator would require a separate review of coverage, definitions, quality, and comparability before incorporation into monitoring or evaluation work.\n\nMention: \"Rapid Early Warning Index\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The document records the Rapid Early Warning Index as a 2020 indicator for contextual reference and uses none of its values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:dd086ad9f67a009f8641408a", "occurrence_id": "a6157a743158c0fa46aecf3d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The principal evidence for the assessment is the Rural Business Demography Survey (RBDS), a survey referring to 2017. The source is used to establish a consistent statistical basis for describing business entry, continuation and exit, while avoiding reliance on isolated administrative observations.\n\nThe assessment records the collection of the RBDS material, followed by harmonisation of classifications, coverage rules and reporting conventions. These procedures permit observations to be compared within a common analytical framework and reduce discrepancies arising from changes in recording practice. The documentation also indicates that the cleaned material was archived for controlled use during implementation and subsequent review.\n\nThe resulting evidence supports the interpretation of business demography patterns and provides a stable reference for monitoring outputs. No additional inference is assigned to the survey beyond the information retained through collection, harmonisation and archiving. The RBDS is therefore treated as the central empirical basis for the relevant findings, with its documented structure preserved throughout the assessment process.\n\nMention: \"Rural Business Demography Survey (RBDS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The passage identifies the 2017 RBDS survey and describes its collection, harmonisation and archiving as the central evidence for the assessment.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:8787724a64d6f0a18a6b2792", "occurrence_id": "47910af8126f0d2110b50c01", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Joint Service Catchments Dataset (JSCD), a geospatial source used to characterize the distribution of service catchments and the population reached by them. The dataset is treated as the principal evidentiary basis for the observations reported below, with no adjustment to the values reproduced from the source.\n\nThe JSCD records that 68 percent of the assessed catchments meet the minimum service-access criterion. This figure is reproduced directly from the dataset and is used to indicate the prevailing level of coverage in the observed service network. It should not be interpreted as an estimate derived through a separate calculation.\n\nThe same source records 41 percent of catchments with documented continuity of service throughout the reference period. This second figure is likewise taken directly from the JSCD and provides the corresponding measure of operational continuity. Together, the two observations support the assessment of service reach and reliability while preserving the distinctions made in the underlying record.\n\nThe results are reported for analytical consistency and should be read as source values rather than as independently modelled indicators. Any subsequent comparison should retain the definitions and coverage conditions applied in the Joint Service Catchments Dataset (JSCD).\n\nMention: \"Joint Service Catchments Dataset (JSCD)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported figures of 68 percent for minimum service access and 41 percent for continuity of service from the JSCD.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:12aa09e5e0f40f85e7cd52d4", "occurrence_id": "41a787ecfcf50c2e4df93ee1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Baseline Debt and Savings Collection as supporting evidence for the observed movement in household financial balances. This is an other source produced by CRU for 2019, and it is considered relevant to the interpretation of the broader results. Its role is limited to informing the assessment of change over time rather than defining the full evidence base.\n\nThe reported entries were reviewed for consistency and used to derive an indicative trend in debt and savings. The analysis also calculates an average across the available observations, providing a summary measure for comparison within the assessment. These derived measures are used to support the finding that financial conditions varied during the period under review, while avoiding reliance on any single recorded value.\n\nThe resulting estimates should therefore be read as analytical summaries of the source material. They support the broader paragraph on household financial resilience, but do not constitute an independent measurement framework. Interpretation remains subject to the coverage and internal consistency of the collection, and the calculated trend and average are presented only to clarify the direction and general level of the reported observations.\n\nMention: \"Baseline Debt and Savings Collection\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The passage derives a trend and an average from Baseline Debt and Savings Collection rather than reproducing individual entries.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:961a413f7a5e99a37c66ad5a", "occurrence_id": "6a13aae7404779c1b16dbbc6", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses Quarterly Forest Products Microdata as the principal empirical basis for the analysis. The source consists of 2017 microdata describing quarterly observations relevant to forest product activity. Its use is confined to the evidence required for constructing comparable measures across successive reporting periods.\n\nThe analysis derives a period trend from the observations and calculates an average for the reporting series. These calculations provide a consolidated indication of movement and central tendency, rather than a reproduction of isolated values from the source. Extreme observations are retained in the calculations unless they are inconsistent with the underlying measurement rules.\n\nThe resulting estimates are used to support the assessment of production stability and variation over time. Interpretation is limited to patterns that remain visible after aggregation, and no inference is made beyond the coverage represented in the source. The calculated trend and average are therefore presented as analytical results based on the microdata, not as figures directly reported by Quarterly Forest Products Microdata.\n\nMention: \"Quarterly Forest Products Microdata\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses the 2017 microdata to calculate a reporting-period trend and average.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:7ee18d01c83dc685e7e1619f", "occurrence_id": "4ba5b20bfae4b282a3485765", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Verified Farm Economy Registry (VFER), a database covering Eastern Europe and referring to 2018, as its central empirical source. The registry provides the evidence base for describing farm structure, production activity, labour allocation, and marketed output within the assessed economy. Its contents are treated as an administrative record suitable for structured review.\n\nThe underlying material was collected from reported farm records and consolidated into a common analytical arrangement. Differences in terminology, measurement units, geographic coverage, and reporting practice were harmonised before review. This process enabled comparable treatment of observations while preserving the distinctions required for interpretation. Missing observations were retained as gaps rather than replaced through unsupported assumptions.\n\nThe resulting archive supports the assessment of production patterns and the identification of concentrations across farm activities. Documentation accompanying the archived material records the procedures used to receive, check, reconcile, and store the records. The analysis therefore relies on the documented registry rather than on reconstructed estimates. Findings drawn from the source are presented as descriptive evidence and are not interpreted as forecasts or causal estimates. Subsequent use should maintain the same definitions and treatment rules to preserve consistency with the archived database.\n\nMention: \"Verified Farm Economy Registry (VFER)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The passage uses the VFER database for Eastern Europe in 2018 and describes its collection, harmonisation, documentation, and archiving.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:490f614d9da2800426bf0f61", "occurrence_id": "b04989ac520e862dc05e6f6e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Decennial Graduation Programmes Index is used as an indicator produced by SDC in 2021. It provides a concise reference point for the assessment of graduation programming and supports interpretation of the reported pattern across assessed categories. The source is cited in the evidence base because its entries are sufficiently specific for direct use in the narrative.\n\nTwo values are reproduced directly from the index. The first records 31 grades within the reported classification, while the second records 68 percent for the corresponding share. No recalculation or adjustment has been applied to either figure. Their inclusion is intended to anchor the broader finding that the programme framework is extensive, while the distribution of outcomes remains uneven across categories.\n\nThe figures are treated as descriptive evidence rather than as a basis for estimating change over time. They therefore support the assessment claim at the level of observed classification and reported proportion. Interpretation remains limited to what is contained in the cited indicator, with no inference made beyond the two values reproduced above.\n\nMention: \"Decennial Graduation Programmes Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage takes the figures 31 grades and 68 percent directly from the Decennial Graduation Programmes Index to support a broader assessment of programme coverage and uneven outcomes.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:2c68942570973b69fcaacfe5", "occurrence_id": "678626d9139d522cc421cbdf", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Civil Immunisation Coverage Survey (CICS), a survey of reported immunisation outcomes, provides the principal evidence for this assessment. Its observations are used to examine coverage performance across successive reporting periods and to support a consolidated interpretation of change. The source is treated as the underlying measurement record rather than as a set of isolated figures for reproduction.\n\nCoverage trajectories are derived by calculating period-to-period movement from the CICS observations. An average level is also calculated across the available reporting periods to reduce the influence of short-term variation and to provide a stable basis for comparison. These calculations indicate whether performance has improved, declined, or remained broadly unchanged over time, while avoiding undue emphasis on any single observation.\n\nThe resulting trend and average are used in the assessment of implementation progress. Differences between the calculated average and individual observations are interpreted as evidence of fluctuation rather than as inconsistencies in the source. Where the trajectory is weak or uneven, the finding is considered relevant to monitoring, follow-up, and the prioritisation of corrective action. The analysis therefore relies on the systematic treatment of CICS data, with conclusions based on derived patterns rather than direct repetition of reported values.\n\nMention: \"Civil Immunisation Coverage Survey (CICS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses CICS to calculate immunisation coverage trends and an average for assessing implementation progress.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:252983ef7a4d3c9e8f9ed410", "occurrence_id": "1c9f6b6cef1fbd99863895ba", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Continuous Deprivation and Hardship Index (CDHI) is an indicator produced by SIPRI for 2021. It is retained as a supplementary statistical reference for assessing the plausibility of the central estimate. The source is used in its reported form, without adjustment to the underlying definition or implied coverage.\n\nThe corresponding CDHI observation provides an external point of comparison for a separate estimate derived from another source. The comparison is intended to test the direction and approximate magnitude of the estimate rather than to replace the primary calculation. Agreement between the two measures is treated as supporting evidence for the assessment presented in the broader paragraph.\n\nThe result is therefore used to substantiate one claim concerning the persistence of hardship, while other elements of the assessment continue to rely on the principal evidence base. Differences between the measures are noted as a limitation where relevant, particularly because the two sources may apply different measurement procedures. On this basis, the CDHI is considered corroborative rather than determinative, and no broader inference is drawn from the comparison alone.\n\nMention: \"Continuous Deprivation and Hardship Index (CDHI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the 2021 CDHI indicator produced by SIPRI to cross-check a separate estimate and support one assessment claim.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:0a1394a77a0f2339a19336d2", "occurrence_id": "b0b0575180e6f2ad2155eb63", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: For the Myanmar coverage, the Cross-Sectoral Disability and Support Survey (CDSS) is identified as a survey for 2016, Edition 5, produced by NIRA. The source provides the relevant benchmark for interpreting reported conditions among persons requiring assistance and for distinguishing observed support gaps from administrative reporting differences.\n\nThe assessment reproduces two figures directly from this survey. It reports that 18.7 percent of respondents identified at least one functional difficulty, while 42.3 percent of those requiring assistance reported receiving support from an available household or community arrangement. These values are retained as reported and are not recalculated from underlying records.\n\nTaken together, the figures indicate a substantial difference between the reported prevalence of functional difficulty and the reach of available support. The comparison is used to inform implementation monitoring, particularly where service coverage may not correspond to identified needs. The evidence should be read as a descriptive benchmark, since the survey results capture reported conditions and support at the time covered by the source rather than subsequent changes in provision.\n\nMention: \"Cross-Sectoral Disability and Support Survey (CDSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the CDSS figures of 18.7 percent reporting functional difficulty and 42.3 percent reporting received support.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:4b9bd90a3123270c4bc46991", "occurrence_id": "2aae4b1f52580b09ca53ddcd", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Panel Water and Climate Report is a report used as the principal documentary basis for this verification. Its reported observations provide a stable reference for reviewing the estimate applied in the assessment, particularly where the underlying measurement is subject to incomplete coverage or differing collection practices.\n\nThe estimate was cross-checked against the evidence presented in the Panel Water and Climate Report. This comparison indicates that the direction and approximate scale of the estimate are consistent with the reported record. Differences are treated as a matter for reconciliation rather than as grounds for replacing the estimate, since the two sources may reflect distinct observation procedures and analytical treatments.\n\nFor implementation purposes, the report is therefore retained as a central validation source. It supports the plausibility of the estimate, helps identify material departures, and provides a documented basis for subsequent review. The cross-check does not imply that the reported observations and the estimate are identical; it establishes that the estimate remains within a defensible range when assessed against an independent source.\n\nMention: \"Panel Water and Climate Report\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the Panel Water and Climate Report to cross-check and validate the plausibility of an independent estimate.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:61290accc77c0c177dc1d52f", "occurrence_id": "19add3616cc677473b6dd303", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Data sources.** The assessment uses the Standard Inequality Survey (SIS), a survey referring to 2016, as the principal basis for the distributional baseline. The source provides observations on household welfare and the allocation of resources across the population, allowing the analysis to situate the reported pattern within a consistent statistical framework.\n\nThe figures are not reproduced as isolated quotations. Instead, the analysis calculates an average across the recorded distribution and derives a summary trend from the underlying observations. This treatment reduces the influence of individual responses and provides a more stable measure for comparison within the assessment. The resulting estimate supports the finding that welfare is distributed unevenly, with material differences across population groups.\n\nThe derived measure is used as one element of the broader assessment rather than as a standalone outcome indicator. It informs interpretation of access, vulnerability, and the concentration of benefits, while other evidence is considered for implementation performance. The calculation is therefore presented as an analytical input based on the Standard Inequality Survey (SIS), not as a direct restatement of any single value reported by the survey.\n\nMention: \"Standard Inequality Survey (SIS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses the 2016 Standard Inequality Survey (SIS) to calculate an average and derive a distributional trend.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:c6869e0b8362281a637597f7", "occurrence_id": "afa2d2bba59f768c1fa95655", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Progress in instructional delivery was assessed against the broader implementation results. The available evidence indicates that reported adherence to planned classroom routines is broadly consistent with the estimated improvement in observed practice, although the estimate remains subject to sampling and reporting limitations. This finding supports the conclusion that implementation quality improved during the period under review.\n\nThe Monthly Classroom Practice Registry database was used to cross-check that estimate against a separate source of classroom-level records. Its entries provide a direct operational reference for routine activities, including lesson preparation, use of learning materials, and completion of scheduled instructional tasks. The comparison did not replace the principal estimate, but it provided a corroborating basis for the claim concerning improved adherence to planned practice.\n\nThe cross-check was interpreted cautiously because administrative records may reflect uneven reporting and may not capture all classroom activity. Nevertheless, the direction of the evidence was consistent across the two sources. On this basis, the assessment retains the original estimate and treats the registry evidence as supporting evidence for the related implementation claim. No adjustment was made solely on account of the registry comparison.\n\nMention: \"Monthly Classroom Practice Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Monthly Classroom Practice Registry database provided corroborating evidence for an estimate of improved adherence to planned classroom practice.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:417a016700db06408b86c6c4", "occurrence_id": "7db266d04405a5ef87548994", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Periodic Labour Inspection Registry database for Mozambique and refers to the position recorded in 2023. The source is treated as the principal statistical basis for the observation of inspection activity presented below. Its coverage is understood to relate to registered labour inspection actions and associated findings recorded during the reference period.\n\nTwo reported values are reproduced without recalculation. The registry records 1,284 inspection visits and 37 cases in which a material non-compliance finding was entered. These figures are used as direct evidence of the scale of recorded enforcement activity and of the number of cases requiring follow-up within the reporting period.\n\nThe figures should be interpreted as administrative observations rather than as an estimate of all workplace conditions. They reflect entries available in the cited database and may therefore depend on the completeness and timing of registry updates. No adjustment has been applied for unregistered activity, duplicate records, or cases pending verification.\n\nFor purposes of this assessment, the two values provide a concise measure of recorded inspection volume and documented findings. Their use is limited to describing the administrative record for the stated coverage and period, without extending the interpretation to outcomes not represented in the source.\n\nMention: \"Periodic Labour Inspection Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Periodic Labour Inspection Registry figures of 1,284 inspection visits and 37 recorded non-compliance cases.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:6abd3228ea77502ad70d1d73", "occurrence_id": "d65276608c9876b0a4219f13", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Data sources.** The Record-Based Disease Surveillance Index is an indicator produced by WB/Makerere/FASID for 2020. It is included in the supporting material to provide a general point of reference on the availability and structure of disease surveillance information. The citation is not used to establish a finding in this assessment.\n\nThe index is relevant to the interpretation of the broader evidence base because record-based measures can assist in describing the reporting environment and the limits of administrative information. Its presence in the documentation therefore supports orientation to the subject matter, while leaving the analytical results dependent on the evidence reviewed for the assessment itself.\n\nNo index values, rankings, rates, or derived measures are incorporated into the tables, narrative findings, or conclusions. The source is mentioned in passing and does not provide the basis for a comparison, estimate, or trend. This treatment is intended to preserve a clear distinction between contextual reference material and evidence used directly in the assessment.\n\nAccordingly, the citation should be read as a supporting bibliographic reference rather than as an input to the reported results. The assessment neither recalculates the indicator nor interprets changes in its underlying observations. Its role is limited to acknowledging a relevant source while maintaining consistency in the presentation and interpretation of the evidence used.\n\nMention: \"Record-Based Disease Surveillance Index\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The source provided contextual orientation only, and no index values or derived measures were used in the assessment.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:5d84b952399fd97bc17a57be", "occurrence_id": "e4e6c1c1a1db714d726271f1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment relies on the Metropolitan Tertiary Education Index, a 2019 indicator produced by CEPAL. The source is used as a reference measure for interpreting differences in tertiary education participation and for situating the results reported in this annex. Its inclusion does not imply that the index replaces the project’s monitoring framework.\n\nTwo figures are reproduced directly from the indicator. The reported values are 64.2% for participation among the relevant population and 81.7% for completion among those entering tertiary education. Taken together, these figures support the broader finding that progression through tertiary education remains uneven, with participation not translating uniformly into completion.\n\nThe figures are presented without recalculation or extrapolation. They are retained in their published form to preserve comparability with the source and to avoid introducing assumptions about coverage, classification, or measurement. The evidence is therefore used narrowly, as supporting documentation for the assessment of access and progression rather than as a standalone estimate of project performance.\n\nOn this basis, the index provides a concise statistical benchmark for the discussion that follows. The two quoted values strengthen the interpretation of disparities in tertiary outcomes while leaving the principal conclusions dependent on the wider body of monitoring evidence.\n\nMention: \"Metropolitan Tertiary Education Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the 2019 Metropolitan Tertiary Education Index and quotes the source values 64.2% and 81.7% to support a claim about uneven tertiary progression.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:7d7627da601fe315ff1a88a1", "occurrence_id": "f5625c54d70ae4402afd5499", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Municipal Inequality Registry database for 2024 is used as the principal evidentiary basis for the distributional assessment. The database provides observations that permit comparison across municipal units and support a consistent treatment of the selected inequality measures. No individual record is reproduced in the assessment, and the source is not used merely to illustrate isolated cases.\n\nThe analysis calculates mean values from the recorded observations and derives a trend by comparing the resulting measures across the relevant categories. This procedure reduces the influence of isolated entries and provides a summary indication of whether inequality is increasing, declining, or remaining broadly stable. The calculated averages are therefore analytical outputs rather than figures transcribed from the source.\n\nResults are interpreted with attention to the coverage and comparability of the observations represented in the database. Where changes are modest, the assessment treats them as evidence of relative stability rather than as a material shift. The derived trend and average are used to inform the assessment of distributional conditions and to identify areas requiring continued monitoring. The Municipal Inequality Registry database remains the central source for this calculation, while the reported findings reflect the analytical treatment applied to its observations.\n\nMention: \"Municipal Inequality Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives average inequality measures and a comparative trend from the 2024 Municipal Inequality Registry database.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:60fb53ad2c3b328787a2a704", "occurrence_id": "52112607e1debb3d262d70d9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Urban Labour Force Census (ULFC) is treated as the principal census for 2016 and provides the central empirical basis for the labour-market assessment. Its observations are used to establish a consistent reference for the distribution of participation, employment, and non-participation across the measured population. The source is cited as the underlying statistical record, while the results presented here are derived measures rather than direct reproductions of individual entries.\n\nThe analysis calculates average outcomes from the ULFC records and examines the resulting trend across the reported labour-market groups. This approach reduces the influence of isolated observations and permits comparison of broad patterns within the census material. The calculated averages are used to describe the prevailing level of engagement with the labour market, while the trend indicates how that engagement varies between the categories recorded in the source.\n\nThe resulting estimates should therefore be interpreted as analytical summaries generated from the 2016 census. They are not presented as additional observations reported by the ULFC. This distinction is material for implementation monitoring, since the assessment relies on aggregation and comparison to identify persistent patterns, rather than on the reproduction of selected figures. The calculations provide the basis for the narrative findings and support a measured interpretation of labour-market conditions.\n\nMention: \"Urban Labour Force Census (ULFC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the 2016 ULFC census to calculate average labour-market outcomes and derive a trend across its reported groups.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:487beb917a86b442ac62cdd6", "occurrence_id": "8237302cfc8c7b8620f7e6b8", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Quality-Assured Rural Livelihoods Survey (QRLS) is used as the principal survey source for Switzerland in 2018. The source provides the evidentiary basis for the assessment of rural household conditions and is treated as the reference point for the observations presented below. No adjustment is applied to the reported values.\n\nThe survey reports that 62.4 percent of rural households maintained more than one livelihood activity during the reference period. This figure is quoted directly from the QRLS and is used to indicate the prevalence of livelihood diversification. A second figure taken directly from the same source shows that 47.8 percent of rural households reported income from agricultural production.\n\nTogether, these observations indicate that livelihood portfolios were differentiated across the covered population, while agriculture remained an important component of household income. The two reported proportions are presented as source figures rather than as estimates derived through recalculation. Their use is therefore limited to describing the distribution recorded by the survey and to supporting the assessment of rural economic conditions.\n\nMention: \"Quality-Assured Rural Livelihoods Survey (QRLS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the QRLS figures of 62.4 percent and 47.8 percent for Switzerland in 2018.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:b2bcfb709bbad7424c392cf9", "occurrence_id": "93cd4a51f6ab2922645b993e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Civil Reproductive Health Registry, a database produced by the Ministry of Investment, reports that 62 per cent of recorded consultations involved reproductive health services. The same registry records 14,280 consultations associated with those services. These figures are used directly in the assessment and provide the quantitative basis for the observation that service demand remains substantial while access is not uniform across reporting units.\n\nThe pattern is relevant to implementation monitoring because aggregate activity can conceal differences in registration, referral, and follow-up. The reported proportion indicates the relative weight of reproductive health within recorded consultations, while the recorded total provides an indication of workload. Taken together, the figures support a cautious interpretation of performance rather than a conclusion that all eligible users received timely care.\n\nThe registry is therefore treated as an administrative reference for this assessment. Its figures are presented without recalculation, and no adjustment is made for records that may have been omitted, duplicated, or entered after service delivery. This approach preserves the source values and limits the claim to what the database directly records. Further interpretation should distinguish reported activity from effective coverage and should consider the consistency of registration practices when reviewing implementation results.\n\nMention: \"Civil Reproductive Health Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Civil Reproductive Health Registry to cite 62 per cent and 14,280 consultations in support of a cautious assessment of service coverage.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:d058f2b1996a2e341adc9776", "occurrence_id": "d0cfd82644bab7dc143dcf73", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Recurrent Referral Pathways Report (RRPR) is a report produced by the Department of Defense. It provides a structured basis for examining recurring referral movements, with attention to the direction, frequency, and apparent continuity of recorded cases. The material is treated as supporting evidence rather than as a standalone measure of service performance.\n\nFor the claim concerning the persistence of referral patterns, the estimate used in this assessment is cross-checked against the RRPR. This comparison draws on a different source base and is intended to test whether the estimated pattern is consistent with an independently recorded sequence of referrals. The cross-check supports the claim within the broader discussion, while not replacing the primary estimate or extending its coverage.\n\nThe RRPR is read for recurring relationships among referral entries and for indications of concentration across the recorded pathway. Particular weight is given to repeated observations that correspond with the direction of the estimate, while isolated entries are not treated as sufficient evidence of a general tendency. Differences between the two sources are retained as a qualification on interpretation.\n\nAccordingly, the report is used as a corroborating reference for one analytical statement. It strengthens confidence in the reported pattern where the two records converge, but it does not establish causality, completeness, or comparability beyond the stated assessment.\n\nMention: \"Recurrent Referral Pathways Report (RRPR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The author uses the RRPR to cross-check an independent estimate and support one claim about recurring referral patterns.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:635bc4243e2e3c8dd7222e62", "occurrence_id": "be3644175ac86c7f2d012662", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Data sources.** The Local Rural Livelihoods Census (LRLC) is a census covering Sulaymaniya for 2021. It provides a household-level reference for the distribution of rural livelihoods and related population characteristics. The source is used here as supporting evidence for a single claim concerning the scale of the rural population represented in the surrounding analysis.\n\nThe estimate presented in the main text is drawn from a separate source and is not treated as a direct result of the LRLC. Instead, the LRLC is used to cross-check the magnitude of that estimate against an independently compiled census record. The comparison indicates that the estimate falls within a plausible range and does not materially alter the interpretation of the reported result.\n\nThe cross-check is therefore confirmatory rather than a basis for recalculation. Differences between the two sources may reflect coverage, timing, or measurement procedures, and the comparison is not intended to establish exact agreement. For reporting purposes, the LRLC is retained as a corroborating reference for the stated claim, while the estimate itself remains attributed to the separate source from which it was obtained.\n\nMention: \"Local Rural Livelihoods Census (LRLC)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the 2021 LRLC census for Sulaymaniya to cross-check a separate estimate.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:6daccdd3253891c3edb32fd3", "occurrence_id": "90586aa99b5e6180f76aa0c3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Midline Value Chain System as an administrative source produced by CLGF. The source provides a structured reference for reported participation, activity coverage, and recorded transactions within the programme monitoring framework. Its entries are treated as supporting evidence rather than as a replacement for the primary estimate.\n\nThe reported level of participation is estimated from a separate source and is cross-checked against the corresponding records in the Midline Value Chain System. This comparison supports the claim that observed coverage is consistent with the administrative record, while allowing for differences in scope, timing, and reporting practice. The result is used as a validation point within the broader assessment of implementation progress.\n\nThe comparison does not imply that the administrative record fully measures outcomes. It provides a traceable basis for reviewing the plausibility of the estimate and for identifying material divergence between reported implementation and recorded activity. Where the two sources differ, the estimate is retained with an appropriate qualification, and the administrative evidence is used to guide further review of the underlying observation.\n\nMention: \"Midline Value Chain System\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the Midline Value Chain System from CLGF to cross-check an independently derived participation estimate and assess the consistency of reported implementation with administrative records.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:3fb331a20debaaeeb8e2e553", "occurrence_id": "8c23347c7afbe62a45973d98", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: For the purposes of this assessment, the Ministerial Terrain and Elevation Census (MTEC) is treated as the reference census for 2021. It provides the underlying basis for the description of terrain and elevation conditions used in the analysis. The source is cited in the present section because its recorded categories offer a consistent point of reference for interpreting the distribution of observed conditions.\n\nTwo figures are reproduced directly from the MTEC without adjustment: 31 elevation grades and 18 terrain units. These values are reported as recorded in the source and are not presented as estimates, projections, or results of a separate calculation. Their inclusion is limited to documenting the scale and internal structure of the classification used for the assessment.\n\nTaken together, the cited figures support the broader finding that terrain conditions are differentiated across a substantial range of categories rather than represented by a single uniform profile. This interpretation remains descriptive and does not infer change over time. The MTEC therefore functions as supporting evidence for the claim concerning variation in terrain and elevation, while the surrounding analysis addresses the implications of that variation for implementation monitoring and reporting. No additional treatment of the source data is required for the purpose of this assessment.\n\nMention: \"Ministerial Terrain and Elevation Census (MTEC)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The assessment takes the directly recorded figures of 31 elevation grades and 18 terrain units from the 2021 Ministerial Terrain and Elevation Census (MTEC).\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:e2940a685c8f5f16c035f843", "occurrence_id": "0a896c1bdf460172181d9385", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Audited Mixed Movement Survey produced by PRIO/Uppsala. This 2018 survey is treated as the primary evidentiary source for the movement profile presented below. Its use is limited to the direct reproduction of two reported figures, with no additional estimation or adjustment introduced in the assessment.\n\nThe survey records 31 percent for the first measure reviewed. It also records 68 percent for the second measure. These values are reproduced as reported in the source and are used to anchor the accompanying discussion of observed conditions. The figures should therefore be read as source observations rather than as results generated through the present analysis.\n\nThe evidence provides a concise statistical basis for implementation monitoring. It supports comparison between the two reported measures while avoiding an inference about change over time. Any interpretation beyond the values stated above would require supporting information not contained in the cited survey. On this basis, the assessment retains the figures as descriptive evidence and does not derive a separate trend or composite indicator.\n\nMention: \"Audited Mixed Movement Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported values of 31 percent and 68 percent from the 2018 Audited Mixed Movement Survey produced by PRIO/Uppsala.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:43d5313d27c64de1a10bf294", "occurrence_id": "25ea0c4be95ff4d83a99cb8a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Clustered Early Childhood Index is an indicator produced by the IMPACT Initiative and referring to 2019. It is retained in the evidence base as a contextual reference for interpreting the assessment. No value, score, rate, rank, or other numerical observation from the indicator is used in the analysis.\n\nWithin the broader discussion of early childhood service conditions, the citation supports a single claim, namely that attention to early childhood conditions is relevant to implementation planning. The claim is presented as a framing consideration rather than as a finding derived from the Index. The source does not establish a target, a baseline, or a measure of change for this assessment.\n\nQuantitative findings in this annex rely on the assessment record and its internal checks. The cited indicator is not combined with those findings, used to calculate a rate, or used to rank units. This treatment preserves a distinction between contextual evidence and evidence directly supporting reported results. It also limits the citation to the point for which it is relevant, while avoiding any implication that the Index supplies the values presented elsewhere in the document.\n\nMention: \"Clustered Early Childhood Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The passage takes no values from the Clustered Early Childhood Index and uses it only to support a contextual claim about the relevance of early childhood conditions.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:bfc36aa83ccae80975619f82", "occurrence_id": "7ec3295d67f78baa845f4b64", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment records continued variation in household access to productive and durable assets. This pattern is relevant to the interpretation of welfare outcomes, but it should not be treated as a direct measure of programme performance. The evidence is considered alongside administrative and operational findings presented elsewhere in this report.\n\nThe Endline Asset Ownership Survey (EAOS), a survey, is cited to situate the discussion of asset access within the broader evidence base. Its role is limited to supporting the descriptive claim that ownership conditions warrant consideration when implementation results are reviewed. No figures, distributions, or calculated indicators from the survey are incorporated into the assessment.\n\nThe analysis therefore relies on the findings reported in the main results section and uses the cited source only to frame one element of the interpretation. This treatment avoids implying a level of precision that is not required for the present assessment and maintains a clear distinction between contextual evidence and the results used for reporting.\n\nMention: \"Endline Asset Ownership Survey (EAOS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The passage uses the EAOS survey to frame one assessment claim and incorporates none of its values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:c1c502c98e82a2f5b9d25853", "occurrence_id": "1dab6a9b6e3a7c24fe8c5766", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis uses the Provincial Vocational Training Survey, a survey conducted in 2015, as the principal empirical basis for assessing participation in vocational training. The source is treated as a consistent observation of enrolment, completion, and programme coverage, with attention to the comparability of measures across the reporting units represented in the survey. No additional source is introduced for the calculations presented here.\n\nRather than reproducing isolated figures, the assessment derives an average level of participation from the reported observations and uses the resulting measure to establish a broad trend. This approach reduces the influence of individual reporting variations and provides a more stable indication of the distribution of training activity. The calculated average is used as an analytical benchmark, while the direction of change is interpreted cautiously where the underlying observations show uneven movement.\n\nThe findings therefore represent a transformation of the information contained in the Provincial Vocational Training Survey rather than a direct quotation of its reported values. The resulting trend is used to support discussion of access and completion, subject to the limitations inherent in a single survey observation and in the aggregation of heterogeneous training activities. These estimates are intended for comparative assessment and should not be read as independently verified administrative totals.\n\nMention: \"Provincial Vocational Training Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The assessment calculates an average and derives a broad participation trend from the Provincial Vocational Training Survey rather than reproducing its reported figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:517f11ed209a0a57784bbbe4", "occurrence_id": "2ccfdd76a125178731aa9f3e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment considers the quality and continuity of information used in operational planning. Within this broader discussion, the Panel Seasonal Forecast Registry is cited as a database produced by the National Statistics Office. The reference provides institutional context for the availability of organised forecasting information and supports the narrower observation that formal records can inform the description of planning arrangements.\n\nNo numerical, categorical, or temporal entries from the registry are incorporated into the assessment. The analysis relies on the surrounding evidence and does not derive indicators, comparisons, or conclusions from the database. Its mention therefore clarifies the information environment rather than establishing a result through direct measurement.\n\nThe cited material is relevant to the assessment of documentation practices, particularly where recurring forecasts are maintained for administrative use. This role is limited to supporting one contextual claim within the broader paragraph. The registry is not treated as an analytical dataset for the purposes of the findings, and no inference is drawn from any individual record or aggregate contained in it.\n\nMention: \"Panel Seasonal Forecast Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The passage takes no values from the Panel Seasonal Forecast Registry and uses it only to support a contextual claim about organised forecasting information.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:c4d8d07b3c58d3c89e0e49ed", "occurrence_id": "4cdff0fd91b105eb64a75650", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Inter-Agency Time Use System as its administrative source for Cambodia. The system is treated as the principal evidentiary basis for the time-use observations presented in this annex. Its coverage is used without adding a second source or substituting estimates generated through the project.\n\nTwo figures are reproduced directly from the system. The record indicates 4.2 hours per day devoted to unpaid domestic work by women in the assessed population, while the corresponding figure for men is 3.1 hours per day. These values are transcribed as reported, with no averaging, adjustment, or derivation undertaken for this note. They provide the central quantitative evidence for the discussion of the distribution of unpaid work.\n\nThe figures should be read as descriptive measures associated with the coverage of the administrative record. They support comparison between the two reported groups but do not, on their own, establish causality or explain differences in participation. The analysis therefore retains the source wording and units, and limits interpretation to the contrast visible in the two reported values.\n\nFor monitoring purposes, the pair is carried forward unchanged in the results framework and may be used to check consistency in subsequent tables. Any future update should preserve the same unit and population definitions so that the comparison remains transparent.\n\nMention: \"Inter-Agency Time Use System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage reproduces the Inter-Agency Time Use System values of 4.2 hours and 3.1 hours for unpaid domestic work in Cambodia.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:3ba93286c1cc560b15ffe07f", "occurrence_id": "4126fa2ef00a6ebad0b2a2a5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Quarterly Graduation Programmes Microdata (QGPM) microdata produced by ProGres are used as a supporting source for the assessment of programme participation and completion. The source provides records at an operational level, permitting the review of distributions across relevant participant categories and the identification of discrepancies in reported outcomes. Its use is limited to the verification of one estimate within the broader analysis.\n\nThe QGPM observations are used to cross-check the participation estimate obtained from a separate source. The comparison is based on compatible measures and does not imply that the two sources have identical coverage or reporting procedures. Differences are therefore treated as an indication for review rather than as evidence of a change in programme performance.\n\nFor the purposes of this assessment, the source is retained as a consistency check on the reported estimate and as a supplementary basis for examining completion patterns. The resulting interpretation gives priority to measures that remain broadly consistent across the available evidence, while recording material divergence for follow-up during implementation review.\n\nMention: \"Quarterly Graduation Programmes Microdata (QGPM)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The passage uses ProGres microdata from Quarterly Graduation Programmes Microdata (QGPM) to examine completion patterns and cross-check a participation estimate from a separate source.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:848b81857ec477e82f21a90e", "occurrence_id": "485139328e5addc255a448df", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Monthly Apprenticeship Registry database, produced by УВКБ ООН, REACH, provides the principal evidence for the apprenticeship analysis. Its records are treated as an administrative series rather than as isolated observations. The assessment uses the entries to examine changes in registration activity and to establish a consistent basis for comparison across reporting periods.\n\nFor analytical purposes, the observations are reviewed for completeness and arranged in their recorded sequence. An average is calculated from the available registration entries, and the resulting measure is used to moderate short-term fluctuations. A trend is then derived from the same series by comparing the direction and persistence of changes over time. No single entry is treated as representative of the overall pattern.\n\nThe calculated results indicate variation in apprenticeship activity, with the underlying movement assessed through the combined evidence of the registry. This approach supports interpretation of the scale and direction of change while limiting reliance on exceptional observations. The findings are therefore presented as derived indicators based on the database, rather than as figures reproduced directly from individual records. The measure is suitable for monitoring purposes and can be updated when additional registry entries become available.\n\nMention: \"Monthly Apprenticeship Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives an average and a directional trend in apprenticeship registrations from the Monthly Apprenticeship Registry database.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:887fc28fc4798fc60f6fec01", "occurrence_id": "78bfba8b422b301c8b72ffdf", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Biennial Household Consumption Survey is used as the principal external benchmark for assessing the reported consumption estimate. The survey covers Africa and South Asia and refers to 2017. Its coverage and timing provide a suitable basis for reviewing whether the estimate is consistent with observed household expenditure patterns.\n\nThe comparison is conducted at the level of aggregate consumption rather than individual records. Results from the Biennial Household Consumption Survey are aligned with the estimate derived from a separate source, using common concepts of household consumption and comparable population coverage. Differences are treated as an indication of possible measurement variation and are not interpreted as evidence of a change in welfare.\n\nThe survey therefore serves a cross-checking function within the assessment. Where the independent estimate falls within the range supported by the Biennial Household Consumption Survey, the reported result is considered plausible for analytical purposes. Where material divergence is observed, the discrepancy is recorded for further review of coverage, recall effects, and the treatment of non-cash consumption. This procedure improves the transparency of the estimate without implying that either source provides a complete measure on its own.\n\nMention: \"Biennial Household Consumption Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The author uses the 2017 Biennial Household Consumption Survey covering Africa and South Asia to cross-check a separate household consumption estimate.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:e9d6fc71042a042ab96af1d4", "occurrence_id": "5568fbc956c1988fba2b774a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment considers household conditions as a relevant dimension of municipal implementation capacity. The Municipal Household Income and Expenditure Survey is a survey covering Slovakia for 2022. Its inclusion provides background for interpreting the wider setting in which local activities are planned and monitored.\n\nThe source supports the contextual observation that household income and expenditure conditions are pertinent to the assessment of implementation risks. This observation is one element of a broader discussion covering institutional arrangements, service access, and the practical constraints faced by municipalities. No numerical result, distribution, or household-level measure from the survey is incorporated into the analysis.\n\nThe cited material therefore serves a limited evidentiary function. It indicates the relevance of household conditions without establishing a separate estimate or changing the analytical basis of the findings. Conclusions in this section rely on the documented implementation evidence and on the assessment framework, while the survey remains a contextual reference for the single claim concerning the importance of household conditions.\n\nMention: \"Municipal Household Income and Expenditure Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The source provides contextual support for one claim about the relevance of household conditions, while none of its numerical values are used.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:d523e6be67961161400073e5", "occurrence_id": "cc46a09f004c01204b1b27ca", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Combined Collective Bargaining Index is used as an indicator of the extent and consistency of collective bargaining arrangements. It provides a structured basis for comparing observations across the analytical series and supports the interpretation of variation in bargaining conditions. Its use is limited to the claim examined in this section and does not substitute for the wider institutional assessment.\n\nFor the purposes of the assessment, the underlying observations are reviewed for consistency and then combined into an average. The analysis also traces the direction of change across the available sequence, rather than reproducing any single reported figure. This procedure reduces the influence of isolated movements and provides a more stable measure for the comparison presented in the results section.\n\nThe calculated average and derived trend support the finding that collective bargaining conditions differ materially across the observations under review. The indicator is therefore used as an analytical input to the broader conclusion, while the reported interpretation remains based on the computed series rather than on a direct quotation from the source. Any changes in the number or composition of observations are considered when interpreting the resulting measure.\n\nMention: \"Combined Collective Bargaining Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the Combined Collective Bargaining Index as an indicator and derives an average and trend from its observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:3061b8ece229b0b50aa55607", "occurrence_id": "762e205a9744a694cf6b8a95", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Comparative Settlement Location Registry (CSLR) is used as the principal evidence for assessing changes in settlement location patterns. The CSLR database records observations for 2017 and provides a consistent basis for comparing the distribution of settlements across the reporting framework. Its entries are treated as source material for calculation rather than as standalone findings.\n\nThe analysis aggregates the recorded locations and derives average values for the relevant settlement measures. These calculations reduce the influence of individual records and permit a consolidated assessment of concentration, dispersion, and change. The resulting estimates are presented as analytical outputs generated from the registry, not as figures reproduced directly from it.\n\nThe computed series indicates the direction and relative pace of change within the observed settlement pattern. Interpretation is limited to the measures supported by the database, with attention to the distinction between recorded observations and values calculated for reporting purposes. This approach supports comparison across the assessed categories while preserving a clear link between the reported trend and its underlying evidence.\n\nMention: \"Comparative Settlement Location Registry (CSLR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses the 2017 CSLR database to calculate average settlement measures and identify a trend rather than quote individual records.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:aa35fd6da07365e59f77378a", "occurrence_id": "b371dc53f8acdc6c550d4f3a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Data sources.** The Triennial Built-Up Surface Census is a 2018 census covering Azerbaijan and is used as a reference point for the assessment of built-up surface. Its recorded extent provides an independently compiled observation against which the estimate used in this section can be examined.\n\nThe estimate is not treated as a direct replacement for the cited material. Rather, the census evidence is used to cross-check the magnitude and general direction of the estimate obtained from a separate statistical source. Agreement between the two measures is taken as support for the reported claim, while any material divergence is retained as an indication of measurement uncertainty.\n\nThe comparison is limited to the aspect of the analysis concerned with built-up surface. It does not imply that the two sources apply identical definitions, coverage rules, or measurement procedures. Differences may reflect classification thresholds, spatial resolution, or the treatment of marginal areas. The cited census therefore backs one claim within the broader assessment, while the principal estimate continues to be reported from its original basis. This approach permits a cautious validation of the result without presenting the cross-check as a reconciliation of the underlying datasets.\n\nMention: \"Triennial Built-Up Surface Census\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the 2018 Triennial Built-Up Surface Census for Azerbaijan to cross-check a separate built-up-surface estimate.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:24e46903311df66b56e623fc", "occurrence_id": "97da66a69292f2a7220b8057", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The principal empirical basis for this assessment is the Baseline School Feeding Survey (BSFS), a survey covering Rio de Janeiro and referring to 2016. The source is used to establish the distribution of reported feeding conditions across the observed school population and to support comparisons within the available records.\n\nRather than reproducing individual observations, the analysis calculates an average across the reported measures and derives a trend across the relevant school groups. These computations provide a compact indication of the prevailing pattern and reduce the influence of isolated responses. The resulting estimates are used as the central evidence for interpreting coverage, regularity, and reported access within the survey population.\n\nThe calculated average is treated as an analytical measure rather than as a value stated directly by the source. Similarly, the identified trend reflects the aggregation and ordering of observations in the BSFS records, not a quoted conclusion. Findings are therefore presented as derived results, with attention to the limits imposed by the coverage and reference period of the survey.\n\nMention: \"Baseline School Feeding Survey (BSFS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives an average and a trend from the Baseline School Feeding Survey (BSFS) covering Rio de Janeiro in 2016.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:0ad8252cf4ca86df803f27d0", "occurrence_id": "ba491ce1b962b2fc1c29fbe4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Combined Job Search Survey, a 2022 survey produced by CONARE, provides the principal empirical basis for the employment analysis. Its observations are treated as a consistent series for assessing changes in reported job-search conditions and related outcomes. The source is used to support estimation rather than to reproduce isolated reported figures.\n\nThe analysis calculates a trend across the available observations and derives an average to represent the underlying level over the period covered by the survey. These calculations reduce the influence of short-term variation and provide a more stable basis for comparison. The resulting measures are used in the assessment of progress, with attention to the direction and persistence of change.\n\nInterpretation remains conditional on the coverage and response characteristics of the source. Nevertheless, the calculated trend and average provide the clearest summary of the evidence available for this assessment. Individual observations are retained for verification, while the reported findings rely on the derived measures rather than on selected values from the survey.\n\nMention: \"Combined Job Search Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives a trend and average from the Combined Job Search Survey rather than reproducing individual reported figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:4fc33a03f688e555570858ec", "occurrence_id": "18cd0c0c814f8247844fbd4d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Triennial Apprenticeship Survey (TAS), a 2024 survey of apprenticeship participation and completion. The source provides the central evidence for estimating movement over time, while the surrounding discussion records the implications for implementation monitoring. Its figures are treated as observations for analysis rather than as isolated values for direct reproduction.\n\nRates were arranged in their reported sequence and used to calculate an average annual change. The resulting trend indicates the direction and approximate pace of change across the observation period, with the average reducing the influence of short-term variation. This approach supports comparison of overall performance without attributing undue weight to any single reported figure.\n\nThe estimate should therefore be read as an analytical summary derived from the TAS survey. It is used to assess whether participation and completion are moving consistently, to identify the broad magnitude of change, and to inform subsequent monitoring. The calculation does not imply that every interval followed the same path, and the reported average should not be interpreted as a direct quotation from the source.\n\nMention: \"Triennial Apprenticeship Survey (TAS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives a multi-period trend and an average annual change from the 2024 TAS survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:7e2a1d4f044ed4ec813d33b2", "occurrence_id": "73a5b84a0e9f03874aad9277", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Endline Teacher Absence Survey is a survey produced by INDEC and is used as a supporting source for the assessment. Its inclusion provides a direct reference for the reported measures and permits the figures to be read without additional adjustment or interpretation.\n\nThe text records an absence rate of 31 percent and a response rate of 68 percent from the Endline Teacher Absence Survey. These figures are reproduced as reported in the source and are presented solely to document the underlying statistical reference.\n\nThe source is treated as descriptive evidence within the implementation record. No further aggregation, comparison, or recalculation is undertaken in this section. The cited measures therefore retain the definitions and scope established by the survey, while their use remains limited to the numerical statements set out above.\n\nThis treatment is consistent with the reporting approach applied to supporting statistical materials. The figures provide a concise indication of the results documented by INDEC and do not constitute a separate estimate generated for the assessment.\n\nMention: \"Endline Teacher Absence Survey\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"source\", \"summary\": \"The passage takes the 31 percent absence rate and 68 percent response rate directly from the Endline Teacher Absence Survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:a1839d392a20fad32b5bfbe3", "occurrence_id": "07dfe5221966131848fd66b9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Linked Settlement and Elevation Index (LSEI) is treated in this assessment as an indicator covering Serbia and produced by Germanwatch e.V. It provides a structured basis for reviewing the distribution of settlements across elevation categories and for identifying areas in which exposure may be concentrated. The index is used as supporting evidence for the assessment claim concerning the association between settlement location and elevation.\n\nThe reported estimate in the main analysis is derived from a separate source. The LSEI is applied as an independent cross-check rather than as a replacement for that estimate. Agreement in the direction and broad scale of the observed pattern is taken to strengthen confidence in the claim, while any divergence is retained as a limitation requiring further review.\n\nFor implementation purposes, the indicator is therefore used selectively within the evidence base. It informs the verification of the relevant estimate and assists in interpreting the spatial pattern, but it is not used to establish additional trends or to assign causal effects. The resulting comparison is recorded with the assessment findings so that the basis for the claim remains transparent and proportionate to the available evidence.\n\nMention: \"Linked Settlement and Elevation Index (LSEI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the LSEI indicator for Serbia, produced by Germanwatch e.V., to cross-check a separate estimate concerning settlement distribution and elevation.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:a660415ca07813799599eda4", "occurrence_id": "ae0d1656db83c2ac2a6825e7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Monthly Household Income and Expenditure Survey is treated as the principal empirical basis for the income and expenditure assessment. It is a 2019 survey, and its observations provide the underlying material for the statistical summaries reported in this annex. The source is used in its stated form, with attention to the comparability of the recorded income and expenditure measures.\n\nThe analysis calculates average monthly household income and average monthly household expenditure across the observed reporting periods. It then derives a simple trend from the sequence of calculated averages, allowing the direction and relative pace of change to be assessed without relying on any single observation. This treatment places emphasis on the distribution of reported results over time rather than on isolated amounts.\n\nAccordingly, the figures presented in the assessment are analytical outputs derived from the source. They should not be read as direct reproductions of individual entries in the survey. The calculated averages support comparison between income and expenditure measures, while the derived trend provides a concise indication of movement across the reporting period. Any interpretation is therefore based on the consistency of the aggregated pattern and not on selected observations.\n\nThis approach is retained for the main results because it provides a transparent basis for describing household conditions while limiting the influence of unusually high or low records. The resulting estimates are used as summary evidence for the assessment and as a reference point for subsequent discussion of changes in household economic conditions.\n\nMention: \"Monthly Household Income and Expenditure Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The work derives comparative averages and a trend measure from the 2019 survey rather than reproducing individual observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:bf718417a8f5b262cffdcaed", "occurrence_id": "b6217439533b74275f16465e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Linked Job Search System, an administrative source produced by Relief International for 2024, is used as the principal evidence for activity recorded through the employment support mechanism. Its coverage is understood as a record of reported transactions rather than as a population survey, and the figures are therefore interpreted as administrative counts subject to the completeness of reporting.\n\nTwo figures are taken directly from the system. The record shows 1,284 individuals registered for job search assistance and 412 placements recorded during the reporting period. These values are presented without recalculation and are used to indicate the observed scale of participation and the resulting employment outcome. The difference between the two counts is not treated as a measure of non-placement, since the source does not establish that every registered individual remained active throughout the period.\n\nThe evidence supports a limited assessment of implementation progress. It confirms that the mechanism generated a substantial volume of registrations and that a smaller number of placements was recorded in the same administrative record. Interpretation should remain cautious because the source does not, by itself, identify duration of participation, characteristics of individuals, or outcomes after placement. The two reported figures nevertheless provide a consistent basis for describing recorded service use and immediate results.\n\nMention: \"Linked Job Search System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Linked Job Search System to report 1,284 registered individuals and 412 recorded placements.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:06a0180cdcee0568f90c4bb3", "occurrence_id": "34e65f1e6c1d12289553ca66", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Pilot Storm Surge Index is used in this assessment as an indicator for 2018. It provides the central quantitative evidence for the limited comparison presented below. The source is cited by name at the point of use, and its reported values are retained in the form in which they appear. No adjustment, recalculation, or conversion is applied to the figures reproduced in this note.\n\nTwo entries are taken directly from the source. The first recorded value is 12.4, while the second recorded value is 18.7. These figures are presented as reported observations rather than as estimates generated through the assessment process. Their inclusion supports a concise description of the measured conditions and permits the results to be reviewed against the cited indicator without introducing an additional calculation.\n\nThe difference between the two entries is not interpreted as a trend, since the available material does not establish a broader sequence of observations. The figures are therefore used for documentation and comparison within the defined assessment context only. This treatment preserves the evidentiary role of the Pilot Storm Surge Index and avoids attributing to it conclusions that are not supported by the two directly reproduced values. The citation and the numerical entries should be read together when reviewing the relevant findings.\n\nMention: \"Pilot Storm Surge Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Pilot Storm Surge Index indicator for 2018 and reproduces its two direct values, 12.4 and 18.7.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:dbbb2055d9dcfc6e2e15000d", "occurrence_id": "f0ead0418695c1b614e608eb", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The principal empirical basis for the assessment is the Analytical Road Segment Index, an indicator produced by MoF. It provides a structured basis for examining the condition and distribution of road segments through a consistent set of observations. The index is treated as the sole cited data source for the quantitative discussion in this section, and its entries are interpreted as measurements rather than as narrative descriptions.\n\nThe analysis derives a weighted average from the observations in the Analytical Road Segment Index and uses the resulting measure to assess the prevailing pattern across the recorded segments. This calculation permits comparison of relative performance without assigning significance to isolated entries. The resulting average is therefore presented as an analytical estimate based on the source, rather than as a value reproduced directly from it.\n\nA directional trend is also calculated from the ordered observations. The trend is used to identify whether the measured condition is improving, stable, or deteriorating over the sequence represented in the data. Attention is given to the consistency of the movement and to the effect of unusually high or low observations on the aggregate result. On this basis, the source supports an evidence-based assessment while leaving interpretation tied to the indicators recorded by MoF.\n\nMention: \"Analytical Road Segment Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives a weighted average and a directional trend from the Analytical Road Segment Index.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:275ca14be42285e59f5756e0", "occurrence_id": "feb91ff75dbc797fdffe43c5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Municipal Informal Trade Survey (MITS) is the cited survey for the statistical observation used in this annex. Produced by the Bank of Italy and referring to 2017, it provides the evidentiary basis for the two quantitative statements reported below.\n\nThe assessment takes two figures directly from the survey. The first is 38 percent, representing the reported share of informal activity. The second is 1.24 million transactions, representing the corresponding recorded volume. Both figures are reproduced as stated, without recalculation, interpolation, or adjustment. Their role is to anchor the discussion in observed magnitudes rather than inferred estimates.\n\nTaken together, these figures establish a baseline for implementation monitoring and support a limited assessment of the scale of activity covered by the statistical observation. The percentage indicates the relative prevalence of informal activity, while the transaction count indicates its measured volume. The figures are therefore used as direct evidence for the assessment, with no additional trend or comparative claim derived from them.\n\nMention: \"Municipal Informal Trade Survey (MITS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the figures 38 percent and 1.24 million transactions directly from the Municipal Informal Trade Survey (MITS).\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:8f0fd767495b3d5c2d49313d", "occurrence_id": "0140373a61703509ca2c5089", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The wage analysis uses the Inter-Agency Wage Structure Index as an indicator for describing the distribution of graded remuneration. The source is retained in the form cited for the assessment, and its reported values are treated as the primary evidence for the comparison presented below.\n\nTwo figures are quoted directly from the indicator. The structure covers 31 grades, while the recorded difference between the relevant compensation points is 14.6 percent. These values are reproduced without recalculation or interpolation and are presented as reported in the source.\n\nThe figures provide a concise basis for assessing the internal consistency of the wage structure. The number of grades indicates the extent of the classification framework, while the percentage identifies the measured separation within that framework. Taken together, they support a descriptive reading of the distribution rather than an estimate of change over time.\n\nThe assessment therefore uses the cited indicator for reference and limits the associated conclusion to the two reported values. No broader inference is drawn from the figures beyond the observed structure and the stated difference.\n\nMention: \"Inter-Agency Wage Structure Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Inter-Agency Wage Structure Index to report 31 grades and a 14.6 percent difference.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:68a6c24c8878663f8b8e4901", "occurrence_id": "21d4c6a267c26405c283c5af", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the administrative source Sentinel Migration and Remittance System (SMRS), which covers Rio de Janeiro and refers to 2021. The source is treated as the primary record for the descriptive statements in this section, with no adjustment to its reported units or coverage.\n\nTwo figures are reproduced directly from the SMRS. The system records 184,200 remittance-linked registrations for the period under review. It also reports that 37.4 percent of those registrations involved households receiving transfers at least once during the reference period. These values provide the principal quantitative basis for the accompanying interpretation.\n\nThe figures are presented as reported in the administrative record and should not be read as estimates generated through an independent calculation. Differences between registration activity and household conditions may therefore reflect the coverage and reporting practices embedded in the source. Within those limitations, the two reported measures establish the scale of recorded activity and the share associated with recurrent receipt.\n\nMention: \"Sentinel Migration and Remittance System (SMRS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses SMRS figures of 184,200 registrations and 37.4 percent of registrations involving recurrent receipt.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:7043d8164f9279dc15849db6", "occurrence_id": "2e494f43ccb3d432a03677ad", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Rolling Hiring and Vacancies System (RHVS) is an administrative source covering Addis Ababa for 2017. It provides a direct record of staffing demand and recruitment outcomes within the area covered, with the reported values retained without recalculation.\n\nTwo figures reported directly in the source are used as the central evidence for this assessment. The record shows 4,862 vacancies and 1,174 filled posts. These values are presented as stated in the system and are not derived from a comparison with another dataset.\n\nThe figures indicate a substantial difference between recorded staffing requirements and positions filled during the period. This gap is relevant to implementation monitoring because it identifies the scale of unmet staffing demand while preserving the distinction between vacancies and completed recruitment. The evidence should therefore be read as an administrative account of reported positions rather than as an estimate of total labour demand.\n\nFor consistency, the two values are retained in their original numerical form throughout the assessment. No additional adjustment is applied, and the source is cited as the basis for the quantitative statement concerning staffing requirements and filled posts.\n\nMention: \"Rolling Hiring and Vacancies System (RHVS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes 4,862 vacancies and 1,174 filled posts directly from the RHVS.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:7f43e64db3d7f3b45d108a8b", "occurrence_id": "9e0e9de53498b771f8a7ed8a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis uses the Civil Storm Surge Dataset, a geospatial source produced by WALS, as the central evidence for assessing the distribution and persistence of observed surge conditions. The source is incorporated into the analytical file without reproducing individual records as reported values. Its role is to support a consistent comparison across the observation set and to provide a common basis for summary measures.\n\nA mean level is calculated from the source observations after applying the same inclusion rules across the analytical sample. The resulting average is used to describe the general scale of exposure, while reducing the influence of isolated observations. A directional trend is also estimated from the ordered records. This calculation indicates whether the measured conditions show a sustained increase, decline, or limited movement over the assessment sequence.\n\nThe findings should therefore be interpreted as derived statistical results rather than as figures copied directly from the dataset. The analysis retains the source definition and applies transparent aggregation procedures, with checks for missing or inconsistent entries before computation. The resulting measures are used in the assessment as evidence for prioritising monitoring and for informing subsequent implementation decisions. They do not substitute for site-level verification, but provide a reproducible basis for comparing conditions and identifying patterns requiring further review.\n\nMention: \"Civil Storm Surge Dataset\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives an average level and a directional trend from the Civil Storm Surge Dataset.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:80b4240709ca71f13b6437d2", "occurrence_id": "7949c681ac80094d2f2be479", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Independent Population Grid Dataset (IPGD), a geospatial source covering Colombia for 2021. The dataset provides a consistent spatial basis for examining the distribution of residents across populated areas and for identifying differences between densely settled and sparsely settled zones. Its use supports a comparable reading of population concentration without relying on isolated administrative observations.\n\nThe analysis derives the central evidence from the grid rather than reproducing individual source figures. Population values are aggregated across successive spatial units, and the resulting series is used to calculate an average level and a directional trend. This procedure reduces the influence of local irregularities and places emphasis on the pattern observed across the full coverage. The calculated measures indicate a gradual shift toward greater concentration in already populated areas, while maintaining substantial variation between locations.\n\nThese results are used to inform the assessment of service requirements and the interpretation of geographic disparities. The average provides a reference point for comparing areas, whereas the trend indicates whether observed differences are becoming more or less pronounced over time. Because both measures are computed from the same geospatial base, the findings remain internally consistent and can be reviewed against the accompanying maps and tables. The source is therefore treated as the principal analytical input for the population distribution assessment.\n\nMention: \"Independent Population Grid Dataset (IPGD)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis computes a spatial average and trend from the IPGD population data for Colombia in 2021.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:323aef49dbf7cc616e155641", "occurrence_id": "9773b7bfe874b3b4b674245b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Baseline Migration and Remittance Survey, a 2024 survey, is cited to provide background for the discussion of migration-related household conditions. Its inclusion supports one contextual statement in the broader assessment and does not constitute a basis for the estimates, classifications, or comparisons presented elsewhere in this section. The source is therefore treated as supporting documentation rather than as an input to the reported results.\n\nThe principal findings rely on the evidence and procedures described in the assessment, including the stated coverage, review criteria, and interpretation of observed conditions. No numerical result, distribution, rate, or other measured value from the Baseline Migration and Remittance Survey is incorporated into those findings. The citation serves only to situate the subject matter and to indicate that relevant survey evidence exists alongside the material examined directly.\n\nThis treatment preserves a clear separation between contextual reference and analytical evidence. Readers should not infer that the survey supplied a denominator, a benchmark, or a time series for the conclusions reported here. Any interpretation requiring quantitative support is addressed through the information documented in the assessment itself, while the cited survey remains a background reference for the single claim concerning the relevance of migration and remittance conditions.\n\nMention: \"Baseline Migration and Remittance Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Baseline Migration and Remittance Survey provides contextual support for one claim, while none of its values are used.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:f474609525d1148435864505", "occurrence_id": "b7c5801bd2e0d1d09a140d44", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The monitoring results indicate that reported changes were uneven across the observations rather than uniform. This assessment is supported by the High-Frequency Agri-Food Prices Survey (HAPS), a survey covering 2024, which provides a consistent reference for interpreting the reported movements. The source is used as supporting evidence for the broader finding that conditions differed materially across the period under review.\n\nTwo figures are reproduced directly from the HAPS survey. The first records 31 percent, while the second records 18 percent. These values are presented without recalculation and are retained in their reported form to preserve comparability with the underlying observations. Taken together, they support the conclusion that the magnitude of change was not distributed evenly.\n\nThe figures should therefore be read as evidence within the wider assessment rather than as a standalone estimate. Their use strengthens the statistical note by linking the narrative finding to directly reported survey results and by limiting interpretation to what is observable in the cited source. No additional adjustment is applied to either value.\n\nMention: \"High-Frequency Agri-Food Prices Survey (HAPS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage takes the 31 percent and 18 percent observations directly from the 2024 HAPS survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:0696189e44a9a4dc04a4d108", "occurrence_id": "27956435677ab19a9e7a9649", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Provincial Nighttime Lights Dataset is cited as a geospatial source covering Thailand to provide background for the interpretation of settlement concentration. Its inclusion establishes a relevant spatial reference for the discussion of uneven access to services and economic activity.\n\nThe reference supports the limited observation that areas associated with greater nighttime illumination may also exhibit denser settlement and more active service locations. This observation is used to frame the broader assessment of connectivity, local market function, and the distribution of public facilities.\n\nThe dataset does not provide the basis for a calculation in this assessment. No figures, classifications, or derived indicators are transferred from the source into the analysis, and no comparison across reporting periods is made on its basis.\n\nThe substantive findings rely on the project evidence described elsewhere in the report. The cited material is retained as corroborating background for one spatial interpretation and does not alter the measurement framework, the coverage of the assessment, or the conclusions concerning implementation progress.\n\nMention: \"Provincial Nighttime Lights Dataset\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The passage takes no numerical values from the Provincial Nighttime Lights Dataset and uses it to support a contextual claim about settlement patterns.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:1fcd47862f2673791c4635f4", "occurrence_id": "0fe1ee539653483405ab3a0b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Coordinated Occupational Safety Survey (COSS) provides the principal survey basis for the assessment of occupational safety conditions. Its observations are used to support a consistent description of reported incidents, preventive practices, and workplace coverage across successive reporting periods. The source is treated as the central evidence for the quantitative assessment rather than as a repository of isolated quotations.\n\nFor analytical purposes, the reported observations are consolidated and reviewed for comparability. Average levels are calculated across the available observations, and changes between successive periods are used to derive a trend. This approach reduces the influence of individual fluctuations and gives greater weight to the overall direction of occupational safety performance. The resulting measures are therefore analytical estimates derived from the COSS material, not figures reproduced without interpretation.\n\nThe assessment uses these calculations to identify whether conditions are improving, remaining stable, or deteriorating. Particular attention is given to the persistence of patterns across periods, since a single reported value may reflect temporary circumstances. The calculated average and trend are presented as the basis for monitoring progress and informing subsequent implementation decisions. No separate data source is introduced, and the interpretation remains limited to the evidence generated through the Coordinated Occupational Safety Survey (COSS).\n\nMention: \"Coordinated Occupational Safety Survey (COSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives average levels and a reporting-period trend from the Coordinated Occupational Safety Survey (COSS).\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:177304b41cd970e0a405f86b", "occurrence_id": "faecb0187c8c69a27cfeb8bc", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis draws on the Quality-Assured Social Protection Survey, a survey produced by the National Statistics Office for 2022. The source provides the principal observations used to assess coverage, receipt, and reported access across the social protection measures considered in this assessment. Its records are treated as the empirical basis for the summary estimates, rather than as isolated quotations.\n\nTo reduce the influence of individual responses, the assessment calculates averages over the relevant observations and then compares those averages across the measures. The resulting pattern is used to identify the direction of variation in reported outcomes and to support a concise trend assessment. No single response is presented as representative; the reported values are generated through aggregation of the source observations.\n\nInterpretation therefore focuses on the level and distribution implied by the calculated summaries. Differences are considered meaningful only when they persist after aggregation, and apparent deviations are reviewed against the spread of responses. This treatment preserves the analytical value of the source while avoiding an unsupported reading of any individual record. The estimates should be read as derived results from the survey, suitable for monitoring and implementation decisions, rather than as direct statements reproduced from the source.\n\nMention: \"Quality-Assured Social Protection Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses the 2022 survey produced by the National Statistics Office to calculate aggregated averages and a comparative trend.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:3c50741bd7c7a3760daacdec", "occurrence_id": "310a8b643a651f1dea20bba5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Household welfare conditions are assessed using the Panel Poverty and Vulnerability Index, produced by CBS as an indicator. The source is used to anchor the assessment of exposure to poverty and to distinguish the scale of current deprivation from the risk of deterioration. Its role is evidentiary rather than methodological, and the figures are retained in the form reported.\n\nTwo values are taken directly from the source. It records 31 percent of households as currently exposed to poverty and 18 percent as vulnerable to a further decline in welfare. These values support the broader finding that economic insecurity extends beyond households already classified as poor. They are presented without recalculation, interpolation, or comparison with an unreported benchmark.\n\nThe results indicate that programme attention should combine immediate support with measures that reduce the likelihood of additional welfare loss. The distinction between present poverty and prospective vulnerability is material for targeting, since a narrow focus on the first group would omit a substantial share of households facing adverse outcomes. The cited values are therefore used to substantiate the assessment of broad and differentiated need, while the source remains the sole basis for the numerical statements in this passage. This treatment is consistent with the limited purpose of the citation.\n\nMention: \"Panel Poverty and Vulnerability Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the source’s direct figures of 31 percent currently exposed to poverty and 18 percent vulnerable to further welfare decline.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:3866f5c504d3215ab373384e", "occurrence_id": "93967aaf998b1be7f7d61556", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Assessment of working conditions relies on a combination of administrative records, field observations, and structured review of institutional procedures. The evidence indicates that reporting arrangements provide a relevant basis for interpreting differences in recorded working conditions, while not constituting a complete measure of employment quality.\n\nThe Regional Working Conditions Registry is a database produced by ActivityInfo and is used here as a contextual reference for that observation. Its inclusion supports the general finding that registry-based reporting can inform the interpretation of working conditions within a broader assessment. The source does not supply figures, classifications, or other values used in the analysis.\n\nThe assessment therefore treats the registry as corroborating material rather than as a dataset from which results are derived. Conclusions are based on the documented review and on the consistency of findings across the available evidence. This approach limits the role of the cited source to one supporting claim and avoids attributing analytical precision to information that is not directly incorporated into the calculations.\n\nMention: \"Regional Working Conditions Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The passage uses the Regional Working Conditions Registry, a database produced by ActivityInfo, only to support a contextual claim and takes no values from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:b05fd83290fecbe083262aa8", "occurrence_id": "7202289ad7ce0bdbbb01b53c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis uses the Coordinated Settlement and Elevation System as an administrative source for 2016. The material provides the underlying observations for the assessment of settlement and elevation patterns, with the source retained as the principal evidentiary basis for the calculations presented below.\n\nThe observations are consolidated into a comparable series before calculation. The resulting measures indicate the direction of change across the recorded units and provide an average level for the period under review. This treatment reduces the influence of isolated entries and supports comparison across the observations without reproducing individual source values as quoted findings.\n\nThe calculated trend is used to assess whether the recorded pattern is broadly stable, increasing, or declining, while the average supplies a reference point for interpreting deviations within the series. Results are reported as derived measures, with attention given to the consistency of coverage and to the effect of unusually high or low observations on the aggregate. The Coordinated Settlement and Elevation System therefore functions as the central empirical basis for the assessment, while the interpretation rests on the trend and average calculated from its contents.\n\nMention: \"Coordinated Settlement and Elevation System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The work derives a trend and average from the 2016 administrative Coordinated Settlement and Elevation System.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:409d53ee52c654f45746ae58", "occurrence_id": "25926285a7b3a9d1505a01d3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Data sources.** The Midline Rural Livelihoods Survey is identified in the assessment as a 2017 survey produced by Fed. It is retained as a supporting reference for the interpretation of implementation conditions and is not treated as a source of estimates for the tables or narrative calculations. The citation is therefore descriptive rather than evidentiary for measured levels, changes, or distributions.\n\nThe broader discussion considers whether livelihood conditions affect the feasibility and continuity of project activities. Within that discussion, the survey supports the limited claim that household economic circumstances are relevant to implementation performance. This function is confined to framing the issue; no observation, total, percentage, rate, index, or other value from the source is reproduced, transformed, compared, or used to derive an indicator.\n\nConsequently, the findings reported in this annex do not depend on extraction from the cited material. The source remains visible so that readers can distinguish the contextual basis for the claim from the evidence used for assessment judgments. Its inclusion does not imply validation of any reported estimate, nor does it establish a separate measurement series. The treatment is consistent with a concise reference in a broader paragraph, where the survey helps define relevance but does not alter the reported results.\n\nMention: \"Midline Rural Livelihoods Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The passage uses the Midline Rural Livelihoods Survey as contextual support for one general implementation claim and takes no values from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:953587e6fad37be888797be1", "occurrence_id": "a206965c8a9f45af4657f793", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Comparative Clinic Quality Survey as its central empirical source. This survey covers Formosa and provides observations on clinic readiness, staffing, essential equipment, and continuity of basic services. The source is cited for the structure and coverage of the underlying observations rather than for isolated reported figures.\n\nFor the purposes of this assessment, the observations are organised into comparable service indicators and reviewed across successive reporting periods. Indicator scores are combined into period-specific averages, with equal treatment applied to clinics for which the relevant observations are available. The resulting averages are then compared sequentially to identify the direction and approximate strength of change. This procedure permits a summary trend to be presented without implying greater precision than the survey design supports.\n\nThe calculated series indicates gradual improvement in readiness and continuity, although progress is uneven across the measured dimensions. Staffing-related indicators show less consistent movement than equipment and basic service indicators, and the average masks differences among individual clinics. These results are therefore interpreted as evidence of broad change in observed clinic quality, not as a direct estimate for every facility. The analysis relies on the internal comparability of the Comparative Clinic Quality Survey and on the consistent treatment of its observations throughout the calculation.\n\nMention: \"Comparative Clinic Quality Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage derives comparable averages and a gradual quality trend from the Comparative Clinic Quality Survey rather than reproducing its figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:1583ea5e8db98472dcdf7868", "occurrence_id": "05437cfb5a428e5af5747e84", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment refers to the Nationwide Hiring and Vacancies Survey, a survey produced by SDC in 2022. The source is noted in the supporting evidence record as a background reference relevant to labour market conditions and vacancy reporting. Its inclusion provides a general indication of the information environment in which the assessment was prepared.\n\nThe survey is not used to establish benchmarks, calculate changes, or substantiate findings in the assessment. No estimates, counts, rates, or other numerical results from the source enter the analysis. The reference therefore serves a limited documentary function and does not affect the interpretation of the reported indicators.\n\nThe discussion relies on the assessment evidence and its stated analytical procedures rather than on figures attributed to the Nationwide Hiring and Vacancies Survey. This treatment is consistent with a cautious approach to supporting material, under which a source may be acknowledged for relevance while remaining outside the computation of results. The citation is retained to identify related evidence, not to extend the scope of the analysis or introduce an additional statistical basis.\n\nMention: \"Nationwide Hiring and Vacancies Survey\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The 2022 Nationwide Hiring and Vacancies Survey produced by SDC is used only as general background, with none of its numerical values entering the analysis.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:acfc846ed163c434c66778db", "occurrence_id": "6e6ebf1a71cd0b59d2990eff", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Rolling Non-State Education Index (RNEI) is treated as the principal indicator for the assessment. The 2020 observation provides the reference point for the computation, with the source retained in its stated form and not supplemented by an external classification. This treatment preserves the distinction between the cited indicator and the measures derived for reporting.\n\nFor analytical purposes, the available RNEI entries are consolidated into an average rather than reproduced as individual reported figures. The calculation also derives a trend from the sequence of observations, allowing the assessment to identify the direction and relative pace of change. The resulting measures are therefore analytical outputs based on the Rolling Non-State Education Index (RNEI), not values presented as direct quotations from the source.\n\nThe derived average is used to support comparison across the assessment categories and to reduce the influence of isolated observations. The trend measure provides an additional basis for interpreting movement over the reporting sequence, while retaining the underlying indicator as the central evidence. No adjustment is applied beyond the stated aggregation procedure, and the results are presented as calculated summaries of the source data.\n\nMention: \"Rolling Non-State Education Index (RNEI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the 2020 RNEI indicator to calculate an average and derive a trend rather than reproduce quoted figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:a4ac3f637410ff435a8015b3", "occurrence_id": "599b8fe37e4cdceae7812abc", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: For the Iraq service-delivery assessment, the Combined Patient Flow Registry database is used as the principal statistical source for 2021. It provides the documentary basis for the patient-flow figures presented in this note and is treated as the authoritative record for the reporting period. No adjustment or external validation is applied to the two counts reproduced below.\n\nThe registry records 31 admissions during the period under review. It also records 12 referrals, with both figures reproduced directly from the Combined Patient Flow Registry database. The counts are presented as reported and are not converted into rates, shares, or estimates. Their inclusion is intended to preserve the scale indicated by the underlying record and to support a transparent reading of service activity.\n\nTaken together, the figures indicate a documented volume of activity across the recorded patient pathway. They should be interpreted as registry counts rather than as a complete measure of demand, coverage, or outcome. The evidence is therefore sufficient for descriptive reporting, while conclusions requiring broader population denominators remain outside the scope of this assessment.\n\nMention: \"Combined Patient Flow Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage reports 31 admissions and 12 referrals taken directly from the Combined Patient Flow Registry database for Iraq in 2021.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:c2b006a14600a990c6a782ec", "occurrence_id": "9c96d3b9888ad03a1be155f8", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the National Road Segment Registry (NRSR) database as the underlying source for selected indicators of network condition and coverage. The registry provides a consistent set of segment-level observations that supports comparison across the reporting period. Its use is limited to the analytical component described below, while interpretation remains subject to the coverage and classification conventions applied in the registry.\n\nFor the purposes of this assessment, the observations are grouped into comparable categories and reviewed for internal consistency. The analysis computes an average across recorded segments and derives a trend from the grouped results, rather than reproducing individual entries or treating any single observation as representative. This processing provides the basis for the assessment’s conclusion that network condition varies materially across the observed categories.\n\nThe resulting indicator is used as one supporting element in the broader review of implementation performance. It is not presented as a direct quotation from the source, and no inference is made beyond the distribution captured in the database. Differences in the number of recorded segments may affect the calculated average and should therefore be considered when interpreting changes over time. The NRSR results are retained in the working tables to permit verification of the aggregation and to support subsequent updates using the same approach.\n\nMention: \"National Road Segment Registry (NRSR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the NRSR database to calculate an average and derive a trend from grouped road-segment observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:5499a47e72400f2ed84fbe87", "occurrence_id": "d6e2ca2d918587eef8395584", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Combined Clinic Quality Survey (CCQS) is a survey produced by oDesk covering Nigeria. It is cited here as contextual evidence for the limited observation that conditions relevant to service delivery are not uniform across facilities. The citation is not used to supply measurements, rankings, rates, or other numerical content to the assessment, and no result from the source is incorporated into the calculations presented below.\n\nWithin the broader discussion, this observation helps frame the interpretation of implementation constraints. The assessment relies on its own documented evidence for any statements about coverage, performance, or change. The CCQS reference therefore supports one general claim about variation in operating conditions, while the analytical conclusions remain separate from the survey record. This treatment avoids implying that the cited material establishes the magnitude, distribution, or direction of any reported result.\n\nReaders should consequently treat the citation as background documentation rather than as a dataset used for estimation. It identifies a relevant source for the Nigeria setting and indicates why differences in facility conditions warrant attention in the narrative. No extracted value, facility count, percentage, or aggregate from the survey is reported here. The source is retained solely to anchor the stated contextual point.\n\nMention: \"Combined Clinic Quality Survey (CCQS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The report takes contextual support for one general observation from the survey and takes no numerical values from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:a236ed9299f100aa7f145c66", "occurrence_id": "b9e629b049129f95a0ed531a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Regional Asylum and Protection Report is a report covering Lao PDR and is used as the principal evidence for the assessment of recorded asylum and protection activity. Its entries provide a consistent basis for examining changes in applications, decisions, and persons receiving protection across successive reporting periods.\n\nThe analysis does not reproduce individual figures as isolated observations. Instead, the recorded values are arranged in sequence and used to calculate an indicative trend for each measure. A mean is also derived for the observed periods to reduce the influence of short-term variation and to support comparison across the reporting interval. These calculations provide the basis for the interpretation presented below.\n\nThe resulting pattern indicates the direction and approximate scale of change in recorded activity, while the calculated averages establish a reference level for monitoring. Differences between measures are retained in the analysis, since applications, decisions, and protection outcomes do not necessarily move in parallel. The results should therefore be read as summary indicators derived from the report rather than as direct quotations of its individual entries. This treatment supports a cautious assessment of operational pressure and changes in recorded protection needs.\n\nMention: \"Regional Asylum and Protection Report\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives period trends and average levels for asylum and protection activity in Lao PDR from the cited report.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:e31dad423476a9eef86d7abf", "occurrence_id": "5f0bbbaa8361688975d48c32", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Implementation progress was assessed against the broader operational framework, which considers whether access to appropriate equipment is reflected in routine service arrangements. This framing supports the claim that mechanisation is a relevant contextual factor in interpreting reported implementation constraints, without treating it as an outcome measure.\n\nThe District Mechanisation Dataset is a geospatial source referring to 2023. It is cited only to situate that claim within the available evidence; its values are not reproduced, compared, aggregated, or used in any calculation. No inference about coverage, utilisation, or performance is drawn from it.\n\nThe assessment therefore relies on the narrative and administrative evidence described elsewhere in this section. The dataset provides background orientation rather than a measurement base, and its citation should not be read as evidence of a quantified change or attribution of results. This distinction preserves the separation between contextual reference and reported implementation findings.\n\nMention: \"District Mechanisation Dataset\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The source provided contextual support for the relevance of mechanisation, while no dataset values were used.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:32484bde11cd2e4806d4e818", "occurrence_id": "624c7f7a5b5df2db67f0daa1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment relies on the Local Remote Sensing System (LRSS) as an administrative source for Lebanon covering 2022. Its records provide a consistent basis for describing the distribution of reported conditions across the areas examined. The source is used to support the finding that observed coverage can be compared across reporting areas without introducing a separate measurement framework.\n\nThe underlying material was collected from available remote observations and accompanying administrative records. It was then harmonised to align geographic references, reporting periods, and category definitions used in the assessment. Incomplete entries were retained where their existence was relevant to coverage, while inconsistent entries were reviewed against the surrounding record before inclusion.\n\nFollowing review, the harmonised material was archived with the supporting documentation and retained for verification during implementation supervision. This process provides traceability for the claim presented in the broader assessment and permits subsequent reconciliation if reporting boundaries or classifications are revised. No inference is made beyond the recorded coverage, and the source is not used to establish changes outside 2022.\n\nMention: \"Local Remote Sensing System (LRSS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The passage records the use of the 2022 administrative source for Lebanon and the collection, harmonisation, and archiving of its records.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:08b9fa6a7468bd7ed804068a", "occurrence_id": "27a4b56a345a3bf6c5d4d2b3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis draws on the Quarterly Service Catchments Census (QSCC), a census of reported service access conditions. The source provides a consistent basis for examining changes in recorded coverage, staffing availability, and operating continuity across successive reporting periods. Its use is limited to the measurement of broad service patterns rather than the verification of individual records.\n\nFor this assessment, the reported observations were consolidated and averaged across the available periods. The resulting series indicates a gradual improvement in overall coverage, although the pace of change remains uneven among service categories. This finding is derived from the distribution of observations in the census and should not be interpreted as a direct quotation of any single reported value.\n\nThe calculated trend supports the assessment that access expanded during the observation period, while persistent gaps remained in continuity of provision. Differences in reporting density and local operating conditions may affect the magnitude of the estimated change. Accordingly, the results are used as an indicative measure for implementation monitoring and are considered alongside operational evidence in the broader assessment.\n\nMention: \"Quarterly Service Catchments Census (QSCC)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives an average and trend in service coverage from the Quarterly Service Catchments Census (QSCC).\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:cd78ff1dda0db37569cadf2f", "occurrence_id": "c21a2e47df91a63125e79cb1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Linked Public Works Survey is cited as a background source for the institutional and statistical setting of the assessment. It is a 2021 survey produced by INFRAS/IWW and is referenced solely to indicate that a relevant body of information exists alongside the materials reviewed for this report.\n\nThe survey is not used to establish a baseline, define an indicator, or support an estimate. No observations, totals, rates, classifications, or other values reported through the source are incorporated into the analysis. Its mention therefore has no effect on the results, the interpretation of findings, or the conclusions set out below.\n\nThe assessment relies on the evidence presented in the report and on the procedures described for reviewing that evidence. The cited source serves only as a brief point of orientation for readers seeking to understand the wider information environment. It is not compared with the assessment results, and no inference is drawn from its contents. This treatment preserves a clear distinction between referenced background material and the evidence used for findings.\n\nMention: \"Linked Public Works Survey\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The passage takes no values from the Linked Public Works Survey and uses it only as background reference.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:31697338ff6809066f91173d", "occurrence_id": "47c94b1a6891a3f526c4b95f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Small-Area Apprenticeship System, an administrative source covering Malaysia and referring to 2024. The source provides observations for small geographic units and is treated as the principal evidence for assessing the distribution of apprenticeship activity. Its records are used in their reported structure, with no supplementary source introduced for the estimates presented here.\n\nThe analysis does not reproduce individual entries as quoted findings. Instead, observations are grouped by small area and converted into comparable measures. An average is calculated across the relevant areas to indicate the typical level of activity, while differences between areas are examined to identify the direction and strength of the observed trend. This procedure gives greater weight to the collective pattern in the records than to any single reported value.\n\nThe resulting measures support a cautious interpretation of geographic variation. The average provides a summary benchmark, and the calculated trend indicates whether activity is concentrated, dispersed, or broadly stable across the covered units. Because the evidence is administrative, the results are interpreted as measures of recorded activity rather than as a complete representation of all apprenticeship participation. The source nevertheless provides the consistent base required for comparison and for monitoring changes in the distribution of activity.\n\nMention: \"Small-Area Apprenticeship System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the Small-Area Apprenticeship System for Malaysia in 2024 to calculate an area-level average and trend from administrative records.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:09a4afb174ceefb6c9117206", "occurrence_id": "9c6ac75ad5db11f891f72f3b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Household Heat and Cold Extremes Collection, an other source referring to 2015. The material is used to support the finding that exposure to thermal extremes varies over the observation period. Its inclusion provides a consistent basis for comparing the reported household conditions without assigning the source a broader institutional role.\n\nRather than reproducing individual entries, the analysis consolidates the available observations and calculates an average level of exposure. It also derives a trend from the sequence of observations, allowing changes in the measured condition to be considered alongside the wider results. These calculations are treated as analytical outputs and are not presented as figures reported directly by the collection.\n\nThe resulting pattern is consistent with a gradual change in the incidence of heat and cold extremes and supports the interpretation presented in the main assessment. Differences in individual observations remain relevant for quality review, but they do not alter the aggregate direction used for this claim. The source is therefore retained as supporting evidence within the broader analysis, with the derived average and trend carrying the substantive weight.\n\nMention: \"Household Heat and Cold Extremes Collection\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The assessment derives an average and a trend from the Household Heat and Cold Extremes Collection for the 2015 reference period.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:06b37f75705d7ef99b05e9c5", "occurrence_id": "c19d653f772910d7724ec320", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Rural Protection Incident Registry database covering Morocco in 2023 is used as the principal source for the protection assessment. It provides a structured record of reported events and affected households, with sufficient detail to support a limited description of the observed caseload. The source is treated as an administrative record rather than as an estimate of unreported conditions.\n\nTwo figures are taken directly from the registry. The recorded caseload comprises 31 incidents, while 14 affected households are identified in the same source. These values are presented without adjustment, extrapolation, or calculation of a rate. Their purpose is to establish the scale of documented protection concerns within the coverage of the registry.\n\nInterpretation remains subject to the scope and completeness of the underlying records. The figures should therefore be read as documented occurrences and documented household impacts, rather than as a comprehensive measure of all protection risks. They nevertheless provide a consistent reference for monitoring implementation, reviewing the distribution of reported cases, and assessing whether follow-up actions correspond to the level of need indicated by the available evidence.\n\nMention: \"Rural Protection Incident Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Rural Protection Incident Registry database for Morocco in 2023 and reports 31 incidents and 14 affected households.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:fadc23df9e5c3416a690a411", "occurrence_id": "4306c64d68e4394c16d20796", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The implementation assessment considers minimum-needs conditions as one element of the broader discussion on service access, household welfare, and operational priorities. The Independent Minimum Needs Survey is cited as a survey referring to 2023 and is used to support the general observation that minimum-needs considerations remain relevant to the interpretation of implementation constraints. Its reported values are not reproduced in this assessment.\n\nThe reference therefore provides background for a single statement within the wider analytical narrative. No estimate, rate, distribution, or other numerical result from the source is incorporated into the findings, and no comparison is made between its observations and the results presented elsewhere in this document. The discussion remains focused on the implications for implementation rather than on measurement of the underlying conditions.\n\nThis treatment preserves the distinction between contextual evidence and evidence used for calculation. The source is acknowledged to indicate the basis for the general interpretation, while the assessment’s conclusions rely on the analysis described in the surrounding sections. The citation should consequently not be read as a presentation of survey results or as an independent estimate of project performance.\n\nMention: \"Independent Minimum Needs Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Independent Minimum Needs Survey was used only to support a contextual claim, and none of its values were incorporated.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:ae72dde9ad7536949dcafead", "occurrence_id": "e53e061795fe4f724be5f07c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Joint Movement Intentions System is used as an administrative source produced by CONAPO for 2023. Its records provide a structured benchmark for assessing the plausibility of the estimate reported in this assessment. The source is treated as the central supporting evidence for the comparison, rather than as a descriptive reference.\n\nThe estimate is cross-checked against the recorded movement intentions in the system. Differences between the two measures are reviewed for consistency in magnitude and direction, while allowing for the distinct coverage and timing of the underlying sources. This comparison reduces the risk that the assessment relies on a single measurement process.\n\nThe results indicate broad correspondence between the estimate and the administrative record, although the comparison does not establish exact equivalence. Divergences are retained in the assessment because they may reflect reporting practices, incomplete registration, or differences in the populations represented. The Joint Movement Intentions System therefore functions as an independent validation point for the estimate and supports a cautious interpretation of the reported results.\n\nMention: \"Joint Movement Intentions System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the 2023 administrative material produced by CONAPO in the Joint Movement Intentions System to cross-check an independent estimate.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:67fa88ecb1c013958f73596e", "occurrence_id": "156ea6e247104f129fdd23df", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Inter-Agency Storage and Loss Microdata, a microdata source covering recorded storage conditions and reported losses. The material is treated as the central analytical input for examining variation across observations, with attention to the consistency of reporting and the distribution of loss outcomes. No individual record is presented as a stand-alone result.\n\nFor each storage measure, the analysis derives an average across the available observations and compares the resulting values in sequence. This calculation indicates a gradual reduction in recorded losses alongside a more stable pattern in storage duration. The direction is not uniform across all measures, but the combined results support a modest improvement in operational performance. The computed averages are used to distinguish persistent movement from isolated reports.\n\nThe findings should be interpreted as statistical estimates derived from the cited microdata rather than as figures reproduced from a single record. Differences in coverage and reporting completeness may affect the magnitude of the calculated changes, particularly where observations are sparse. Even with these limitations, the source provides sufficient evidence for monitoring the overall trajectory and for identifying measures requiring closer review. The results are therefore retained as a baseline for implementation follow-up and subsequent analytical updates.\n\nMention: \"Inter-Agency Storage and Loss Microdata\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives averages and a sequential trend in storage duration and recorded losses from the Inter-Agency Storage and Loss Microdata.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:2fdccadcaad014968438d011", "occurrence_id": "9b74e06f270c90250c3b7bce", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Municipal Mental Health Index (MMHI) is used as the principal indicator for assessing the availability and distribution of mental health support. The index provides a consolidated reference for interpreting reported service conditions and for identifying differences in coverage across the units reviewed. Its use in this assessment is limited to the evidence recorded in the source.\n\nTwo figures are reproduced directly from the Municipal Mental Health Index (MMHI). The source records 31 service units within the assessed system and reports 14 units with an established referral arrangement. These figures are presented without recalculation, adjustment, or extrapolation. They therefore provide the central quantitative basis for the observations set out below.\n\nThe reported distribution indicates that service presence is broader than formal referral capacity. This distinction is relevant for implementation monitoring because the existence of a service unit does not, by itself, demonstrate continuity of care or access to onward support. The figures should consequently be read as descriptive measures of recorded provision rather than as estimates of population need.\n\nFurther interpretation would require additional evidence on utilisation, staffing, quality, and outcomes. In the absence of such evidence, the MMHI remains an appropriate reference for documenting the observed position and for establishing a baseline against which subsequent administrative reporting may be reviewed.\n\nMention: \"Municipal Mental Health Index (MMHI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Municipal Mental Health Index (MMHI) indicator and takes the figures 31 and 14 directly from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:4081cc4ef5470802c87e03e7", "occurrence_id": "27bc2b93feea06da1b35cc9f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis draws on the Independent Time Use Survey, a survey designed to measure the distribution of daily activities across population groups. The source provides the principal evidence for assessing changes in unpaid work, paid work, care responsibilities, and residual time. Its observations are treated as analytical inputs rather than as isolated reported figures.\n\nThe estimates are harmonised across activity categories and weighted to support comparison between population groups. From these records, the analysis derives average time allocations and identifies the direction and magnitude of the trend in unpaid and paid work. The resulting measures are therefore calculated indicators based on the source, not direct reproductions of individual survey results.\n\nThe computed averages indicate a persistent concentration of unpaid care among those with primary household responsibilities, alongside a gradual reallocation of time toward paid work. The trend is interpreted cautiously because aggregate movements may conceal variation across demographic groups and activity types. Nevertheless, the consistency of the derived measures supports use of the survey as the central empirical basis for monitoring changes in time use.\n\nThe findings are presented as summary statistics for implementation monitoring. They provide a reproducible basis for assessing whether observed shifts in daily activity are broad-based or limited to selected categories, while avoiding claims that cannot be supported by the underlying observations.\n\nMention: \"Independent Time Use Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives weighted averages and a trend in time allocation from the Independent Time Use Survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:f8089639cb8ebe867157cd2e", "occurrence_id": "e2a6d0b9fdfa46d4a9918f56", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Decennial Aquaculture Output Survey is the principal statistical input for the production analysis. It is a survey produced by Ministère des Finances and contains observations suitable for examining changes in reported aquaculture output. The material is treated as a source for aggregation and comparison within the assessment, with attention to consistency across the recorded observations.\n\nThe analysis derives an average output measure from the survey and calculates the direction and rate of change across the available observations. These calculations provide the central basis for assessing whether production has expanded, contracted, or remained broadly stable. Individual reported quantities are therefore used in the construction of summary measures, rather than presented as stand-alone evidence.\n\nThe calculated average reduces the effect of irregular observations, while the resulting trend indicates the underlying movement in output. This approach supports a cautious interpretation where variation between observations is material. The results should be read as an analytical reconstruction from the Decennial Aquaculture Output Survey, with conclusions limited to the information contained in that source.\n\nMention: \"Decennial Aquaculture Output Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses the Decennial Aquaculture Output Survey to calculate an average and derive a trend in aquaculture output.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:2c0cb9276afacf7f7c5890d7", "occurrence_id": "338334bc56e7ec26e1fd859a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Decennial Inequality Survey is a 2019 survey cited in this assessment as background for interpreting distributional conditions. Its inclusion indicates that inequality is considered alongside the institutional and demographic factors relevant to the analysis. The source is not treated as a basis for calculating indicators or establishing numerical benchmarks.\n\nThe broader discussion supports the claim that observed differences in welfare should be read with attention to the conditions under which household information is generated and compared. This claim is situated within the assessment of measurement quality and does not depend on figures drawn from the cited survey. The reference therefore provides documentary orientation while leaving the reported estimates unchanged.\n\nFor implementation purposes, the assessment retains the source as a supporting reference for the interpretation of inequality-related findings. No values, classifications, or derived measures are taken from the Decennial Inequality Survey. The cited material is used solely to situate one statement within the wider analytical narrative, while the remaining discussion relies on the evidence and methods described elsewhere in the report.\n\nMention: \"Decennial Inequality Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The cited survey provides contextual support for interpreting inequality, while no numerical values or derived measures are taken from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:310b82c31e46e234fd735435", "occurrence_id": "6b7ea1463eca1fa82f9cef0a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment considers the institutional conditions affecting access to social protection and the consistency of related implementation arrangements. The discussion is intended to situate the findings within the available evidence base while maintaining a clear distinction between contextual material and information used for measurement.\n\nThe District Social Protection Survey (DSPS) is a 2016 survey produced by IMSS. It is cited to support the observation that access to social protection is influenced by institutional and household conditions. The reference therefore provides background for one claim within the broader assessment rather than serving as a source for reported levels, group comparisons, or calculated changes.\n\nNo values from the DSPS are incorporated into the analysis, and no estimate, rate, or distribution is derived from the survey. Its role is limited to reinforcing the stated interpretation of implementation conditions. This treatment preserves the separation between descriptive context and the evidence used to assess results, while retaining a traceable reference for the relevant observation.\n\nMention: \"District Social Protection Survey (DSPS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The passage uses the 2016 IMSS District Social Protection Survey (DSPS) to support one contextual claim without using its values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:3c4834f48b4b5c06f81a3b8a", "occurrence_id": "5f051ee90174528aa8a957f2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Coordinated Temperature Records Collection, classified as other, to support the analysis of temperature variation. The records provide a consistent basis for examining changes across the observation sequence and for checking whether the broader findings are compatible with the underlying measurements. Their role is evidentiary within the wider assessment and does not determine the interpretation on its own.\n\nIndividual readings are not reproduced as reported figures. Instead, the observations are screened for comparability, grouped according to the analytical intervals, and used to derive an average path of change. This calculation reduces the influence of short-term fluctuations and permits a more stable comparison across the sequence. The resulting trend is treated as an analytical estimate and not as a direct quotation from the collection.\n\nThe derived pattern supports the assessment claim that temperature conditions changed progressively rather than remaining constant. The finding is considered alongside the other indicators used in the review, with attention to the limits imposed by record coverage and measurement consistency. On this basis, the collection provides supporting evidence for the stated conclusion while leaving the principal interpretation dependent on the combined evidence.\n\nMention: \"Coordinated Temperature Records Collection\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the Coordinated Temperature Records Collection, classified as other, to derive an average temperature trend supporting a broader assessment claim.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:3b4ddbbad6e95d2baac545fa", "occurrence_id": "16a44bc1a493b5ac66e322e0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Local Labour Inspection Index (LLII) is treated as a 2020 indicator of institutional inspection capacity. It provides a structured reference for assessing whether reported changes in labour oversight are consistent with observed administrative conditions. Its use is limited to supporting the interpretation of one claim concerning the reliability of the underlying estimate.\n\nThe estimate was obtained from a separate statistical source and is subject to measurement constraints arising from incomplete reporting and variation in recording practices. The LLII is therefore used as a cross-check rather than as a substitute for the primary calculation. Where the direction and relative magnitude of the estimate are compatible with the indicator, the finding is considered more credible; where they diverge, the result is retained but interpreted with caution.\n\nThe comparison does not establish causality and should not be read as a direct validation of all reported values. It provides a supplementary consistency test within the broader assessment, particularly for the claim that inspection capacity changed in line with the estimated pattern. The evidence is consequently weighted alongside other available observations, with the LLII serving as a corroborating measure for this specific inference.\n\nMention: \"Local Labour Inspection Index (LLII)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the 2020 LLII indicator to cross-check a separate estimate and support one claim about inspection capacity.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:671bf955291fd88bcbeb612e", "occurrence_id": "c90d744dbc5dfc6e24421cab", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Stratified Vaccination Campaign Survey provides the central survey evidence for this assessment. It is treated as an operational record of campaign observations rather than as a narrative account. The material supports review of reported coverage, delivery reach, and the completeness of reporting across the units represented in the source. No inference is made beyond the recorded observations.\n\nData were collected through the survey instruments and transferred into a consolidated working file. Entries were checked for internal consistency, aligned to common labels, and harmonised where equivalent responses had been recorded in different forms. Duplicate records and incomplete entries were retained in the archive with their status documented, allowing subsequent review of decisions made during preparation. This procedure preserves traceability between the source material and the analytical file.\n\nThe resulting archive serves as the evidentiary base for the assessment annex. Its contents are used to describe reported implementation patterns and to identify limitations in the completeness or comparability of observations. The documentation accompanying the archive records the processing steps, while preserving the distinction between information directly reported through the survey and conclusions drawn during analysis.\n\nMention: \"Stratified Vaccination Campaign Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The passage records the collection, harmonisation, quality checking, and archiving of evidence from the Stratified Vaccination Campaign Survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:3cefd0a5c4b6aa700c5206c1", "occurrence_id": "b6d7be81f2c31b6c89423d95", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Household Time Use Registry (HTUR) database produced by TRMM is used as the sole statistical source for the descriptive observations in this section. It supplies structured records on the daily allocation of time and permits the reported measures to be presented without additional adjustment. The source provides the central evidence for the assessment of participation and duration.\n\nTwo figures are taken directly from the database. The recorded average for unpaid domestic activity is 42 minutes per day, while 68 percent of observed records indicate participation in such activity. These figures are reproduced as reported, with no recalculation, projection, or inferred change. Their purpose is to establish the scale of the measured activity and to distinguish the incidence of participation from the time allocated by participants.\n\nThe figures should be read as descriptive evidence rather than as an estimate of causality or a basis for extrapolation. The first measure concerns duration, whereas the second concerns the share of records reporting participation; the two statistics therefore address related but distinct dimensions of activity. Taken together, they provide a concise basis for monitoring the distribution of time use and for identifying the magnitude of the activity captured in the registry. No additional interpretation is required for the values reproduced here.\n\nMention: \"Household Time Use Registry (HTUR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the reported figures of 42 minutes and 68 percent directly from the source.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:2af2ea677fa4d48d36ac1942", "occurrence_id": "9e32f1a3761fd8023f508e44", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment considers disparities in access and participation across population groups. It finds that unequal outcomes are not confined to a single measure, but appear across several dimensions of civil inclusion. This finding is treated as a supporting observation within the broader review of service access and social participation.\n\nThe Civil Inequality Index is an indicator produced by Afristat and used to substantiate this assessment. The underlying records were collected, harmonised, and archived before tabulation. Processing reconciled differences in terminology, removed duplicate entries, and applied a common structure to the retained observations. The resulting series was reviewed for internal consistency and used as a supporting data source rather than as the sole basis for conclusions.\n\nThe source contributes comparable evidence on the distribution of civil inequality among population groups. Its use strengthens the assessment by providing a consistent reference for interpreting differences in participation and access. Where observations were incomplete, the analysis retained the documented limitations and avoided extending the indicator beyond the coverage supported by the archived records. The results are therefore presented as descriptive evidence, with interpretation confined to the patterns visible in the harmonised material.\n\nMention: \"Civil Inequality Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The passage records the collection, harmonisation, archiving, and use of the Civil Inequality Index as an indicator produced by Afristat.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:7c83c70acee82df64c0bee15", "occurrence_id": "989b295c92dd974281da007b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Follow-Up Elevation Model Registry, a 2020 database, as the principal empirical basis for estimating changes in measured elevation. The registry provides observations that are sufficiently consistent for aggregation across the analytical units used in the assessment. No individual observation is treated as determinative, and isolated deviations are retained only insofar as they contribute to the aggregate calculation.\n\nThe reported results are derived by calculating average elevation values and comparing the resulting aggregates across the observation sequence. This procedure produces a trend measure from the database rather than reproducing selected entries. The trend is used to identify the direction and relative magnitude of change, while the average provides a stable reference for interpreting variation within the recorded observations. Calculations apply the same treatment to all eligible records and exclude entries that cannot support a comparable measurement.\n\nThe resulting indicators should therefore be read as summary estimates based on the registry’s recorded coverage and internal consistency. They are suitable for monitoring broad movement in elevation measures and for supporting subsequent assessment of implementation performance. The analysis does not assign causal significance to the observed pattern, and the derived averages should not be interpreted as direct measurements of any single observation. Replication can be undertaken using the same database and aggregation procedure, subject to the availability and integrity of the underlying records.\n\nMention: \"Follow-Up Elevation Model Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives average elevation values and a trend measure from the 2020 database.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:3067f3eadce0d0be151fad73", "occurrence_id": "1151f485fa22a0b7ecfa7767", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment examines the condition and distribution of civil transport assets as part of the implementation review. The evidence is used to support the finding that incomplete network records can constrain the prioritisation of routine maintenance and the interpretation of reported service coverage.\n\nThe Civil Transport Network System is treated as an administrative source for the 2023 registry. Records were collected through the established reporting process, harmonised to align terminology and coverage, and archived for controlled use in the assessment. This treatment supports comparability across reporting units without introducing additional estimates or inferred observations.\n\nThe archived material provides a structured basis for checking whether recorded links, facilities, and associated attributes are represented consistently. Where entries required reconciliation, the review retained the reported observation and documented the adjustment in the working record. The resulting file was used to verify the cited claim and to identify gaps requiring attention during implementation.\n\nInterpretation remains limited to the information preserved in the source. The assessment therefore distinguishes between recorded assets and conditions that could not be confirmed from the registry, while maintaining a traceable link between the evidence reviewed and the conclusions presented.\n\nMention: \"Civil Transport Network System\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The passage records use of the 2023 administrative registry and the collection, harmonisation, archiving, and verification work undertaken with it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:f3c37dc78f497bec567bc4c9", "occurrence_id": "ce1ddeff88d058d4687512ba", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment reviews the information base used to frame the cadastral discussion. It distinguishes descriptive material from evidence applied in the analytical tables and maintains a clear separation between contextual references and reported measurements. This approach supports consistent interpretation of the findings and limits the risk that background material is treated as an observation.\n\nThe Pilot Cadastral Frame Dataset is cited as a geospatial source referring to 2015. Its role in this assessment is limited to situating the discussion of cadastral coverage and the general structure of the information environment. No value, estimate, classification, or record from the dataset is incorporated into the calculations, comparisons, or tabulations presented here. The citation supports the claim that the assessment is informed by a defined cadastral reference while leaving the results dependent on the evidence specified for each analysis.\n\nThe broader review therefore relies on documented procedures for validation, reconciliation, and interpretation. Contextual sources are retained to clarify the scope of the discussion, but they do not substitute for the records used to produce reported findings. This distinction is material for reproducibility, since the inclusion of a source in the narrative does not imply that its contents were processed or that its measurements contributed to any conclusion. The cited dataset is accordingly treated as background documentation within the evidentiary framework.\n\nMention: \"Pilot Cadastral Frame Dataset\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The source provides only contextual identification of a geospatial reference for 2015, with no values taken into the analysis.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:709ca68b5f6e1a68eadf126c", "occurrence_id": "ac051aa77dddcf8eff4cd0d0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Follow-Up Host Community Registry (FHCR) database for 2023 is used as the reference source for the household reach estimate presented in this assessment. Its use provides a consistent basis for reporting the observed scale of assistance and for distinguishing recorded coverage from broader population assumptions.\n\nThe assessment draws directly on two figures in this database, 412 households recorded as reached and a 68 percent coverage rate. Taken together, these values support the broader conclusion that delivery extended beyond a limited beneficiary group, while retaining a measurable basis for the coverage claim. The figures are reproduced as reported and are not recalculated for this discussion.\n\nThis evidence is used within the wider assessment of implementation performance. It indicates that recorded delivery was substantial relative to the reference population, although the figures alone do not establish the quality, continuity, or longer-term effects of assistance. The registry therefore serves as a supporting source for the reach finding rather than as the sole basis for overall performance judgement.\n\nMention: \"Follow-Up Host Community Registry (FHCR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage takes 412 households and a 68 percent coverage rate directly from the 2023 FHCR database to support the finding that assistance reached a substantial beneficiary group.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:4eff4c98c2d2413ff158201b", "occurrence_id": "f99a129b9f411b25878e1470", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Joint Landslide Hazard Index (JLHI) is used as an indicator for 2018 in the assessment of exposure conditions. It provides a common basis for comparing relative hazard levels across the observations used in the analysis. The index is incorporated into the results as supporting evidence for the assessment of changes in hazard conditions over time.\n\nThe analysis derives an average from the JLHI observations and uses the resulting series to estimate a broad trend. Individual index values are therefore not reproduced as standalone findings. Instead, the calculation aggregates the available observations, reduces the influence of isolated values, and provides a summary measure for comparison across reporting periods. This treatment supports the interpretation that hazard conditions were not uniform throughout the period considered.\n\nThe resulting trend is used to substantiate the broader finding that areas with higher indexed hazard require closer attention in planning and monitoring. The calculation is presented as an analytical use of the cited source rather than as a direct transcription of its underlying values. Conclusions remain limited to the coverage and comparability of the observations represented in the JLHI, and the derived average should be read as a summary statistic rather than a direct measure of realised losses.\n\nMention: \"Joint Landslide Hazard Index (JLHI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The passage identifies the Joint Landslide Hazard Index (JLHI) as an indicator for 2018 and explains that an average and trend are derived from its observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:6cc910c076611808f9f61788", "occurrence_id": "e94c77163071820616543b41", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Follow-Up Non-State Education Survey is cited as the survey produced by Enerdata for Brazil and referring to 2021. It is treated as the principal statistical basis for the paragraph because its reported coverage corresponds to the population examined in the assessment. No adjustment, projection, or reconciliation is applied to the values reproduced below.\n\nThe survey gives a total of 4,218 institutions and 612,407 enrolled learners. These two figures are quoted directly from the source and constitute the central evidence used to describe the scale of non-state education. The first figure indicates the number of institutions within the reported coverage, while the second records the learners associated with them.\n\nTaken together, the reported totals provide a bounded reference for interpreting participation and provision. They are presented as source values rather than estimates generated for this assessment, and no inference is made about change over time. The figures should therefore be read as a descriptive account of the coverage reported in the survey, with the source designation and date retained for traceability.\n\nMention: \"Follow-Up Non-State Education Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the source’s directly reported totals of 4,218 institutions and 612,407 enrolled learners.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:94302057777763198a52c8c9", "occurrence_id": "46d31d04e8deba1472e46ccc", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on Audited Supply Chain Microdata (ASCM), a microdata source used to examine the recorded structure and operational status of supply chain entries. The material is treated as a direct evidentiary input for the assessment and is not supplemented by a second administrative source. Its definitions and reported values are retained without adjustment in the discussion below.\n\nThe ASCM records 31 grades within the reviewed supply chain classification. It also reports that 68 percent of entries satisfy the stated documentation conditions. These figures support the broader finding that compliance is substantial but not universal, with residual gaps concentrated in the documentation of lower-rated entries. The figures are reproduced directly from ASCM and are presented as reported rather than as estimates derived through further calculation.\n\nInterpretation remains limited to the coverage and classifications represented in the source. The assessment therefore uses the two reported values to substantiate the compliance finding, while avoiding claims about change over time or conditions outside the recorded entries. Any comparison across categories is descriptive and follows the classification applied in the microdata. This treatment is consistent with the requirements for traceability, reproducibility, and conservative use of administrative evidence in implementation reporting.\n\nMention: \"Audited Supply Chain Microdata (ASCM)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the directly reported values of 31 grades and 68 percent from ASCM to support the finding that compliance is substantial but not universal.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:152345b6ea78dcf2826acd0c", "occurrence_id": "c1c477aeef68d7e8274ff6c3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Follow-Up Settlement and Elevation Registry, a database produced by NRS, as the underlying reference for settlement and elevation observations. The material is used to support interpretation of the indicator reported below and to maintain consistency across the analytical series.\n\nThe reported result is calculated from the observations in the registry rather than reproduced as a quoted figure. An average is derived across the available records, and the resulting sequence is examined for its direction and relative stability. This procedure reduces the influence of isolated observations and provides a more suitable basis for assessing whether the indicator is broadly increasing, declining, or unchanged.\n\nThe derived average and trend are used as supporting evidence for the assessment of settlement and elevation conditions. They do not constitute a separate measurement exercise, and no inference is made beyond the coverage and internal consistency of the recorded observations. The calculation is retained in the working analysis so that the reported claim can be reviewed against the same database and applied consistently in subsequent monitoring.\n\nMention: \"Follow-Up Settlement and Elevation Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the NRS database to calculate an average and assess a trend in settlement and elevation observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:3abee2dfa675d9aa42e460c8", "occurrence_id": "c06a3b05c94e73e6a11f990a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis uses the Comparative Reproductive Health Survey (CRHS), a survey conducted in 2015, as its principal empirical source. The material is treated as a consistent basis for descriptive measurement, with attention to the comparability of responses and the internal distribution of reported outcomes. No supplementary source is introduced in the assessment.\n\nReported observations are grouped into analytically relevant categories and converted into summary measures. The resulting averages indicate the central level of each outcome, while the ordering of those averages provides an empirical trend across the categories. This procedure reduces the influence of isolated responses and supports a more stable interpretation of the underlying pattern.\n\nThe estimates presented in the annex therefore represent calculations derived from the survey material rather than direct reproduction of individual reported figures. Differences between categories are assessed through the calculated averages, and the direction of change is used to identify the principal trend. The approach is intended to support monitoring and comparison within a single evidentiary framework, while avoiding precision that cannot be justified by the available observations.\n\nMention: \"Comparative Reproductive Health Survey (CRHS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses the 2015 CRHS survey to calculate grouped averages and identify an ordered trend.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:96ef6324c76bcbc970b93cec", "occurrence_id": "c51e01586f4cb819ff3e9558", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Panel Social Protection Survey, a 2018 survey used to support the estimate of household access to basic support. The source provides a consistent reference for interpreting reported receipt, non-receipt, and stated barriers across the analytical tables. Its evidence is treated as complementary to the broader results rather than as a standalone measure of programme performance.\n\nThe underlying material was collected through structured interviews and subsequently harmonised to improve comparability across response categories. Attention was given to common definitions, consistent treatment of missing responses, and alignment between reported household conditions and the analytical framework. The resulting archive was retained with the supporting documentation so that the tabulations can be reviewed and reproduced.\n\nThe survey evidence indicates that access remains uneven among households with similar stated needs, a finding that informs the discussion of implementation constraints. This interpretation is subject to the limitations of self-reported information and the coverage of the source material. Nevertheless, the harmonised records provide a useful basis for checking whether the patterns observed in the assessment are consistent with respondent accounts.\n\nFor reporting purposes, the source has therefore been incorporated into the evidence base accompanying the relevant tables. No attempt is made to infer changes over time from this material alone. Its principal contribution is to preserve comparable observations and document the procedures used for collection, harmonisation, and archiving.\n\nMention: \"Panel Social Protection Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The analysis used harmonised household responses from the 2018 survey and documented their collection and archiving.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:837460b4ceb7f145dee8ded3", "occurrence_id": "a9f0a26edcb94c04126c064d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Regional Literacy and Numeracy Survey (RLNS), a survey produced by Small Arms Survey, is cited in this assessment as a contextual reference for interpreting patterns discussed in the surrounding analysis. Its inclusion provides background on the broader evidence environment relevant to learning outcomes and supports a measured reading of the findings.\n\nThe source supports one claim concerning the need to interpret reported literacy and numeracy conditions alongside wider social and institutional factors. It is not used to derive estimates, establish rankings, or supply numerical inputs for the assessment. No values from the survey are incorporated into the calculations or attributed to the results presented here.\n\nThe principal evidence for the assessment is evaluated through the methods and records described in the relevant sections. The RLNS reference therefore serves a limited corroborative function within a broader paragraph, allowing the discussion to acknowledge a related evidence base without extending the scope of the analysis. This treatment preserves the distinction between contextual documentation and evidence directly used for measurement.\n\nMention: \"Regional Literacy and Numeracy Survey (RLNS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The author uses the Regional Literacy and Numeracy Survey (RLNS) only to support contextual interpretation of one claim and takes no values from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:1d2288f5dc70271c9f6f1679", "occurrence_id": "ee13bce44f45fbdb78455487", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Rapid Disease Surveillance Registry (RDSR) database serves as the principal evidentiary base for the assessment. It provides a consolidated record of reported disease events and related surveillance observations. The material is used to establish a consistent reference for reviewing coverage, completeness, and internal coherence across the records retained for analysis.\n\nRecords were collected through a structured transfer and review process, with attention to duplicated entries, missing values, inconsistent labels, and changes in reporting practice. The resulting files were harmonised before analysis so that equivalent observations could be compared under common definitions. Where discrepancies could not be resolved, the original entries were preserved and the limitation was recorded in the accompanying documentation.\n\nThe harmonised material was then archived in a controlled repository, together with a description of the cleaning decisions and the checks applied during preparation. This procedure maintains a traceable link between the reported observations and the analytical tables derived from them. The RDSR database therefore functions not only as a record of surveillance activity, but also as the central evidence supporting the assessment’s statistical observations and reproducibility. Interpretation remains subject to the completeness of reporting and to the quality of the underlying records.\n\nMention: \"Rapid Disease Surveillance Registry (RDSR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The passage records the collection, harmonisation, and archiving of the Rapid Disease Surveillance Registry (RDSR) database.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:15383517a16fdeabaac3c3d3", "occurrence_id": "72e99ae8a63072c84fd97561", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Representative Water and Climate Registry (RWCR), a database produced by NSB for 2018. The source provides the principal evidence for the analysis of water and climate indicators, with records reviewed for consistency and completeness before calculation. No additional source is used to establish the baseline presented in this annex.\n\nThe reported measures are derived from the database rather than reproduced as individual registry entries. Available observations are aggregated to obtain average values, and the resulting series is examined for directional movement across the reporting categories. This treatment supports comparison of relative conditions while limiting the influence of isolated observations.\n\nThe calculated averages and trend measures are used to inform the assessment of implementation performance. Results are presented as analytical outputs based on the RWCR records, not as a transcription of figures appearing in the source. The method therefore preserves the evidentiary role of the registry while making the source suitable for summary interpretation.\n\nMention: \"Representative Water and Climate Registry (RWCR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses the 2018 NSB RWCR database to calculate averages and trend measures for the assessment.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:4d223b5074ee4416de7ab342", "occurrence_id": "d0e4e27df8330aec59bd7d36", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment relies on the Verified Administrative Boundary Dataset (VABD), a geospatial source covering Syria and referring to 2020. It provides the central spatial evidence for the delineation of administrative areas used in the analysis. The dataset is treated as a consistent reference for interpreting territorial coverage and for maintaining comparability across the reported observations.\n\nThe work included collecting the available boundary material, reviewing its internal consistency, and harmonising the representation of administrative units and boundary geometries. Differences in naming, geometry, and spatial detail were reconciled through a documented preparation process. The resulting material was archived in a form suitable for controlled consultation, reproduction of the analytical maps, and subsequent administrative review.\n\nUse of the Verified Administrative Boundary Dataset (VABD) is therefore limited to the evidentiary role assigned to the geospatial record. No additional territorial interpretation is introduced beyond the boundaries retained in the archive. The reference supports a transparent account of the geographic basis for the assessment while preserving the distinction between collected source material and analytical findings.\n\nMention: \"Verified Administrative Boundary Dataset (VABD)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The passage records the collection, harmonisation, and archiving of the geospatial VABD covering Syria in 2020.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:2b096bd14022e9f2774901f1", "occurrence_id": "bb025e5a497a8e656300c66c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The District Service Catchments Dataset (DSCD) is treated as the principal geospatial source for the assessment. Its records provide the base for examining service coverage across the observed catchments and for establishing a consistent denominator for comparison. The analysis uses the source as an analytical input rather than reproducing isolated entries or presenting individual observations as findings.\n\nMeasures are calculated from the available records after applying a common treatment to incomplete and non-comparable observations. The resulting average coverage is used to describe the central level of service access, while changes across successive observations are used to derive a trend. This approach reduces the influence of exceptional catchments and provides a more stable indication of overall movement.\n\nThe computed results indicate the direction and magnitude of change in service coverage within the assessment framework. They should therefore be interpreted as estimates derived from the DSCD, not as figures directly reported by the source. Any differences between the calculated measures and individual records reflect aggregation, weighting, and the treatment of missing observations. The dataset remains the central evidentiary basis for the analysis, with the derived average and trend supporting the assessment of implementation progress.\n\nMention: \"District Service Catchments Dataset (DSCD)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses the District Service Catchments Dataset (DSCD) as a geospatial source to calculate an average and derive a trend in service coverage.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:71fee4967c73d772824ccd80", "occurrence_id": "e2c388de31828759be280077", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Clustered Land Use and Tenure Registry database as the principal source for the land-use and tenure indicators presented in this annex. The record is used to establish the distribution of registered holdings and the reported status of tenure documentation. No adjustment is applied to the figures reproduced below.\n\nThe database directly reports 31 registered categories and a coverage rate of 68 percent for the observations included in the assessment. These figures are reproduced as recorded and are not estimates generated through the analytical procedures described elsewhere in the report. They provide the quantitative basis for the finding that registration coverage remains uneven across the recorded categories.\n\nThe evidence should therefore be interpreted as a description of recorded administrative status rather than as a complete measure of all land-use arrangements. The registry supports comparison of the documented categories, while the assessment’s broader conclusions remain subject to the scope and completeness of the underlying records. The two reported figures are retained to permit verification against the cited source and to maintain consistency between the statistical note and the accompanying discussion.\n\nMention: \"Clustered Land Use and Tenure Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Clustered Land Use and Tenure Registry database and reproduces its figures of 31 registered categories and 68 percent coverage.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:808707dd4f06e65069601e65", "occurrence_id": "4d5257c0c860e35020798cb1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment finds that constraints affecting informal activity remain material to the implementation environment. They influence the reliability of administrative reporting, the reach of formal support measures, and the interpretation of observed changes in participation. The evidence is therefore used to qualify the broader finding that improvements in recorded performance do not necessarily indicate a corresponding reduction in informality.\n\nFor the 2016 reference year, the Biennial Informal Sector Index is used as an indicator produced by Mercer. Two figures are drawn directly from this source. The reported values of 17.4 and 42.8 are cited in support of the assessment’s finding on the uneven incidence of informal activity. They are presented as source values and are not recalculated or combined with other observations.\n\nThese figures provide a limited but relevant basis for comparison within the paragraph. Their principal use is evidentiary, allowing the assessment to substantiate the stated claim while retaining a cautious interpretation of the underlying conditions. The figures should not be read as establishing causality, and they do not replace the administrative evidence used elsewhere in the assessment. Any apparent difference between the two values is consequently treated as descriptive rather than as a measured trend. The source is retained in the reference material to permit verification of the quoted values and to maintain consistency in subsequent discussion.\n\nMention: \"Biennial Informal Sector Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the figures 17.4 and 42.8 from the 2016 Biennial Informal Sector Index produced by Mercer to support a claim about uneven informal activity.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:8147c760d18daeedaecc088c", "occurrence_id": "786c357608c1d53c58454dcb", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment relies on the Cross-Sectoral Terrain and Elevation Dataset as a geospatial source for describing physical conditions relevant to implementation. The dataset provides the central evidence for interpreting terrain variation and elevation differences across the analytical framework. Its use supports a consistent basis for reviewing spatial constraints and for relating observed conditions to planned activities.\n\nThe underlying records were collected from available technical materials, harmonised through common definitions and processing rules, and archived for controlled use during implementation review. This treatment reduced inconsistencies in coverage, terminology, and measurement while preserving the distinctions required for subsequent analysis. The resulting archive was retained as a reference set rather than treated as a standalone estimate of project performance.\n\nAnalysis drew on the harmonised records to identify broad patterns in terrain and elevation and to inform the screening of operational risks. The evidence was used descriptively, with interpretation limited to the information contained in the source and to the transformations applied during preparation. No additional institutional attribution, geographic designation, or temporal classification is required for the findings presented here.\n\nThe dataset therefore functions both as a documented evidence base and as a reproducible input to implementation assessment. Its collection, harmonisation, and archiving provide continuity between technical review, quality assurance, and later updates to the analytical record.\n\nMention: \"Cross-Sectoral Terrain and Elevation Dataset\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The work taken from the source is the collection, harmonisation, archiving, and descriptive use of terrain and elevation records.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:8bb4447741e6efaddd6c9a69", "occurrence_id": "e7577fd6122288e1a23b84f3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis uses the Quarterly Collective Bargaining Registry (QCBR) database as a supporting source for the assessment of bargaining activity. The database covers CHT and Sylhet divisions and refers to 2015 records. It is produced by SECEX and is cited here to establish the underlying coverage of the observations used in the analysis.\n\nRather than reproducing individual registry entries, the assessment calculates an average level of recorded activity and derives a trend from the available observations. This treatment provides a consistent basis for comparing the incidence of collective bargaining across the covered reporting units and supports the broader finding that activity was unevenly distributed during the period under review.\n\nThe resulting measure is used as one input to the interpretation of institutional performance. It should therefore be read as an analytical summary derived from the Quarterly Collective Bargaining Registry (QCBR), not as a direct quotation of a reported statistic. The approach also limits the influence of isolated entries and places greater weight on the overall pattern contained in the database.\n\nMention: \"Quarterly Collective Bargaining Registry (QCBR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the SECEX-produced Quarterly Collective Bargaining Registry (QCBR) database for CHT and Sylhet divisions in 2015 and derives an average and trend from its observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:5903363a39e72b7a96e6453a", "occurrence_id": "763090517b6077ff7005225a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The principal empirical source for this assessment is the Audited Apprenticeship Survey, a survey produced by INSTAT. The source provides structured observations on apprenticeship participation, completion, duration, and the characteristics of training arrangements. Its use supports a consistent basis for examining variation across the observed records.\n\nThe analysis does not reproduce individual reported values as headline results. Instead, the observations are consolidated into comparable measures, and arithmetic averages are calculated for the principal indicators. Changes across the ordered observations are then used to derive a trend in participation and completion, with attention to movements that persist across the dataset rather than isolated fluctuations.\n\nThese calculations provide the central evidence for the assessment of implementation performance. The resulting measures are interpreted as descriptive estimates of the observed apprenticeship system and are not treated as forecasts. Where the number of observations is limited, the findings are presented with appropriate caution, while the common definitions applied in the survey permit comparisons across the main dimensions of provision. The derived trend and averages are used to inform the discussion of outputs, retention, and completion, and to identify areas requiring closer operational review.\n\nMention: \"Audited Apprenticeship Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage derives comparative averages and an observed trend in apprenticeship participation, retention, and completion from the Audited Apprenticeship Survey produced by INSTAT.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:2dcf77e83be42b81d2f07409", "occurrence_id": "0cc915f7ec2cf1a4b069f0c9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Small-Area Remote Sensing Registry (SRSR) database as the reference for classifying observed surface conditions. The cited material refers to 2020 and is identified as Round 4, providing a consistent basis for the descriptive statement on coverage and classification used in this annex. It is used to support that statement within the wider review of measurement quality and operational readiness.\n\nFor operational use, the underlying records were collected from submitted observation materials, harmonised to a common classification structure, and archived with the associated documentation. This process reduced variation in labels and supported the comparison of records across reporting units without introducing additional estimates. The resulting archive was retained as the working reference for the assessment and for subsequent verification of selected entries.\n\nThe source therefore serves a documentary and quality-assurance function rather than a basis for projecting future conditions. Its inclusion permits the reported classification to be traced to a defined round and reference year, while the harmonised archive provides an orderly record for review. Any limitations arising from incomplete observations or inconsistent source materials remain relevant to interpretation and are addressed through the validation procedures described elsewhere in the report.\n\nMention: \"Small-Area Remote Sensing Registry (SRSR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The passage uses the 2020 Round 4 SRSR database and describes collecting, harmonising, and archiving its records.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:728fd7986f1b8db5dd2f1c7d", "occurrence_id": "160b21604e25397931f3de3c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Baseline Shock Coping Index (BSCI) indicator as a reference measure for household resilience. The indicator refers to 2019, and the Penn World table produces it for comparative statistical use. Its inclusion provides a consistent basis for interpreting reported differences in the capacity to absorb economic pressure.\n\nThe broader assessment finds that coping capacity is uneven across the observed population. Two figures are reproduced directly from the indicator for this purpose. The first is 42 percent, and the second is 57 percent. These values support the conclusion that a substantial gap remains between lower and higher levels of reported resilience, without requiring an additional calculation from the underlying series.\n\nThe figures are used as evidence within the broader argument rather than as a standalone estimate of programme performance. Their role is limited to anchoring the discussion in a published statistical measure and indicating the scale of variation relevant to implementation planning. Interpretation should therefore remain aligned with the scope of the indicator and should not be extended to conditions not represented in the cited measure.\n\nMention: \"Baseline Shock Coping Index (BSCI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage reproduces the 42 percent and 57 percent figures directly from the 2019 BSCI indicator.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:e6381e97d477842f5abfe1b7", "occurrence_id": "4c76352e51019c5f23ffe562", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Decennial Emergency Care System (DECS) is treated as an administrative source for the assessment of emergency care capacity. Its records provide the central quantitative evidence for the paragraph and are used without adjustment or extrapolation. The source is cited in the text so that the basis of the reported observations remains identifiable.\n\nThe assessment reproduces a reported total of 84 emergency care units. This figure is taken directly from the DECS record and is presented as an observed count rather than as an estimate generated for the assessment. The same source reports 31 units with continuous emergency coverage, and that figure is likewise retained in its reported form.\n\nTaken together, these two reported counts indicate the scale of the service network and the more limited extent of continuous coverage within it. No additional calculation is required to establish the distinction between the overall number of units and the subset meeting the specified coverage condition. The figures are therefore used as direct evidence in the implementation assessment, while interpretation is limited to the relationship visible in the source record.\n\nMention: \"Decennial Emergency Care System (DECS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage draws directly from DECS, reporting 84 emergency care units and 31 units with continuous emergency coverage.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:6f13065c14e9dcbbc0cb182b", "occurrence_id": "700a146a138003159a955f90", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Biennial Unemployment Insurance Survey is the principal statistical source used for the assessment of unemployment insurance conditions in Nepal. The survey provides a consistent basis for describing reported access and recorded participation, while limiting interpretation to the measures directly presented in the source. Its coverage is treated as representative of the population addressed by the instrument.\n\nTwo figures are taken directly from the survey and are used as the central evidence for this assessment. The reported participation rate is 42.6 percent, while the reported access rate is 18.4 percent. These values are reproduced without recalculation, adjustment, or derivation from other observations.\n\nThe difference between the two measures indicates that participation is materially broader than effective access under the arrangements observed. This finding should be read as a descriptive result rather than as evidence of a causal relationship. It nevertheless provides a concise benchmark for assessing implementation performance and identifying areas requiring further review.\n\nThe figures should be retained as reference points in subsequent monitoring. Any comparison with later results should preserve the definitions and population coverage applied by the survey so that apparent changes are not introduced by measurement differences.\n\nMention: \"Biennial Unemployment Insurance Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Biennial Unemployment Insurance Survey figures of 42.6 percent for participation and 18.4 percent for access in Nepal.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:ff51d5baa5b3167b94d9ac1d", "occurrence_id": "e2d542a332df49f0e4199d6b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Urban Land Cover Change System is used as an administrative source produced by Teikoku Databank for 2020. Its records provide the principal documentary basis for the land-cover evidence reviewed in this assessment, with the source designation retained in the working archive.\n\nThe records were collected, harmonised, and archived through a common processing procedure. Entries were checked for internal consistency, aligned to the reporting structure, and preserved without introducing supplementary institutional or geographic identifiers. This treatment supports comparison across the compiled material while maintaining a clear link to the original administrative record.\n\nThe resulting archive is used to substantiate the assessment narrative and to document the provenance of the reported observations. No additional estimation is attributed to the source in this passage. The cited material therefore serves as the central evidence base, while the processing record establishes how the information was prepared for review and subsequent use in the implementation file.\n\nMention: \"Urban Land Cover Change System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The work taken from the source is the collection, harmonisation, and archiving of its 2020 administrative records.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:44fad0b9c72ded41cef53c36", "occurrence_id": "d21d8ceb40df4311c8cefb8a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Administrative monitoring indicates that informal economic activity continues to be recorded unevenly across reporting units. Differences in registration practice, record retention, and the treatment of small operators limit direct comparison between administrative returns. The assessment therefore gives priority to consistent definitions and traceable supporting documentation when interpreting changes in participation and compliance.\n\nThe Municipal Informal Sector Survey is a 2024 survey used to support the finding that registration gaps remain material among smaller operators. The assessment documents how the underlying records were collected, harmonised across reporting forms, and archived for subsequent verification. These procedures improve the comparability of observations while preserving the distinction between reported activity and activity inferred from administrative records. No adjustment is made where the available documentation does not permit a reliable reconciliation.\n\nThe evidence is used as a supporting measure rather than as a complete account of informal activity. It strengthens the assessment of reporting coverage and provides a basis for reviewing whether observed differences reflect actual changes or variation in recording practice. Continued maintenance of the archive and periodic review of harmonised classifications are recommended for future monitoring.\n\nMention: \"Municipal Informal Sector Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The passage uses the 2024 survey to support a finding on registration gaps and records its collection, harmonisation, and archiving.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:847cc1f084cbfc5dfb42b1d5", "occurrence_id": "c4e2210d8145dfd280bcd349", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Panel Unemployment Insurance Survey as its central empirical basis. This 2018 survey, produced by ICAO, provides the observations used to examine changes in reported unemployment insurance conditions. The source is treated as a consistent statistical record for analytical purposes, while individual responses are not reproduced as isolated findings.\n\nThe assessment derives a trend from the available observations by arranging them in their reported sequence and calculating comparable changes across successive periods. It also computes an average to summarise the general level represented in the survey. These calculations provide the principal evidence for the findings and are preferred to reliance on selected figures that could give disproportionate weight to an exceptional observation.\n\nThe resulting measures are used to describe the direction and overall scale of change. Interpretation remains limited to patterns supported by the survey, with no inference made beyond the information captured in the source. Where the calculated trend and average point in the same direction, the conclusion is treated as more stable for implementation reporting. Where they diverge, the assessment retains the distinction and avoids presenting the average as a substitute for the underlying pattern. This approach supports a concise statistical account while preserving the analytical role of the source.\n\nMention: \"Panel Unemployment Insurance Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The assessment derives a trend and average from the 2018 ICAO survey rather than quoting individual observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:aafc08f14f0555c211abbbba", "occurrence_id": "e71cb1bb7010e6017eb5eb8a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Metropolitan Settlement Location Report (MSLR) is the principal report used for the settlement-location evidence reviewed in this assessment. Its tabulations provide a consistent basis for describing the observed distribution and for distinguishing reported counts from analytical interpretation.\n\nThe report records that 31 percent of the settlements fall within the specified classification. This figure is reproduced directly from the MSLR and is treated as an observed proportion rather than as a result generated through subsequent analysis. The same source records 418 settlements in the corresponding tabulation, a count that is likewise reproduced without adjustment.\n\nTaken together, these two figures establish the quantitative reference point for the discussion that follows. They indicate the scale of the recorded settlement set and the share assigned to the relevant classification, while leaving broader explanations to the assessment. No additional inference is attached to the figures in this passage. The MSLR report is therefore used as a documentary source for the two reported values, with the surrounding text limited to their administrative and statistical significance.\n\nFor purposes of review, the figures should be read as source observations. Any later comparison or assessment may draw on them, but such work would constitute a separate analytical step and should not be understood as a revision of the values reported in the MSLR.\n\nMention: \"Metropolitan Settlement Location Report (MSLR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported figures of 31 percent and 418 settlements from the MSLR report.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:360f2cf10d73daf040ffb976", "occurrence_id": "1b299197aacc5da3fc4aff1f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Within the assessment, the Small-Area Soil Fertility Registry (SSFR) is cited as a database relevant to the discussion of soil fertility information. Its role is limited to establishing that a structured source exists for the subject under review. The citation therefore supports a general statement about the availability and organisation of evidence, rather than supplying an observation for the calculations.\n\nThe registry is used only to frame one claim in a broader paragraph concerning the interpretation of local agricultural conditions. No value from the database is transcribed, compared, aggregated, or used to derive a rate, estimate, classification, or conclusion. The assessment does not rely on the source for a measurement, and the citation should not be read as evidence that any particular level or change was observed.\n\nAccordingly, the SSFR reference has a documentary and contextual function. It signals a relevant basis for the narrative while leaving the quantitative content of the assessment unchanged. The findings reported in this section are presented independently of the registry’s entries, and no inference is attributed to records that are not reproduced here. This treatment preserves the distinction between acknowledging a supporting source and incorporating its data into the analysis.\n\nMention: \"Small-Area Soil Fertility Registry (SSFR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The passage takes from the SSFR only contextual support for one broader claim and incorporates no values from the database into the assessment.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:3adae737e354f86d4387d0ad", "occurrence_id": "b5a0b16e774a4eef8c6772ba", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Subnational Farm Economy Index is treated as an indicator for the assessment of subnational agricultural conditions. It provides a structured basis for comparing reported dimensions of farm activity across administrative observations and for identifying persistent differences in performance. The source is retained as the central quantitative input to the analysis.\n\nThe assessment does not reproduce individual index entries as standalone findings. Instead, the observations are consolidated into arithmetic averages and examined over successive reporting periods to derive a trend. This calculation reduces the influence of isolated movements and gives greater weight to the general direction of change. The resulting series is used to distinguish broad improvement, stagnation, and deterioration in farm-economy conditions.\n\nInterpretation remains cautious where the underlying observations are sparse or unevenly distributed. The computed average is therefore presented as an analytical measure rather than a direct statement about every reporting unit. Comparisons are based on consistent treatment of the available records, with exclusions and missing entries handled before aggregation. The resulting trend supports the assessment of changes in agricultural performance and informs the prioritisation of follow-up analysis.\n\nMention: \"Subnational Farm Economy Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the Subnational Farm Economy Index to calculate averages and derive a trend in farm-economy conditions rather than reproduce individual entries.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:9008e58bfa21068e7a8ad98b", "occurrence_id": "dff3e0399510f88a95696374", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Local Immunisation Coverage Registry is a database produced by INE and serves as the principal source for the coverage observations presented in this assessment. Its use permits the results to be linked to recorded service delivery rather than to estimates derived from secondary reporting. No adjustment has been applied to the values reproduced below.\n\nTwo figures reported directly in the database are used as the central evidence. Coverage among eligible children is recorded at 78 percent, while coverage among adolescents is recorded at 64 percent. These values are presented as reported and should not be interpreted as a calculated trend or as a comparison generated through a separate analytical procedure.\n\nThe difference between the two observations indicates a lower recorded level for adolescents within the reported coverage categories. The result supports continued attention to registration completeness, follow-up of missed vaccinations, and routine review of delivery performance. Interpretation remains limited to the measures available in the registry and does not imply that additional population characteristics were assessed.\n\nMention: \"Local Immunisation Coverage Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Local Immunisation Coverage Registry for two directly reported figures, 78 percent for eligible children and 64 percent for adolescents.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:5ce090d2101689402e366405", "occurrence_id": "01369fc1d9f98330b50ae139", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the National Wildfire Burn Index as an indicator in reviewing the reported exposure estimate. The source provides a standardized reference for interpreting the relative incidence of wildfire conditions and supports a consistency check on the magnitude presented in the results. No adjustment is made to the underlying estimate on this basis.\n\nThe estimate derived from the household questionnaire is cross-checked against the National Wildfire Burn Index to assess whether the direction and broad scale of the reported finding are plausible. This comparison supports the claim that areas with higher recorded exposure also exhibit greater wildfire-related pressure. The comparison is used for validation of that claim within the wider assessment and is not treated as a replacement for the primary estimate.\n\nThe results should therefore be read as an integrated assessment based on the questionnaire evidence, with the indicator serving a corroborative function. Differences between the two measures may reflect differences in coverage, timing, reporting practices, or the unit of observation. These limitations do not invalidate the comparison, but they constrain the interpretation of small differences and preclude direct aggregation of the measures. The cited source is retained in the supporting evidence to improve transparency and facilitate subsequent review.\n\nMention: \"National Wildfire Burn Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the National Wildfire Burn Index indicator to cross-check a questionnaire-based estimate and support one broader assessment claim.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:c45b91404ded0e7eb55ec0bd", "occurrence_id": "8ec39eb37e22ddc1e008d9a3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Quality-Assured Living Standards Survey is the survey used for the supporting observations in this section. It provides a direct basis for assessing household welfare conditions and for checking whether the reported pattern is consistent across the indicators reviewed.\n\nThe survey reports that “62 percent” of households experienced the condition examined in the assessment. This figure is reproduced directly from the source and is treated as a descriptive observation rather than as an independently estimated result.\n\nThe same source records “14.7 percent” for the related measure. This second figure is also reproduced directly and, together with the first, supports the broader finding that access and welfare outcomes remain uneven among households covered by the assessment.\n\nThe two observations are presented without further adjustment, aggregation, or extrapolation. Their role is limited to substantiating the assessment's central claim, while the interpretation of underlying causes remains subject to the wider evidence considered in the report.\n\nMention: \"Quality-Assured Living Standards Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the survey's directly reported figures of 62 percent and 14.7 percent to support the finding of uneven household welfare outcomes.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:3d7dd277cd7a9a0352d7bda1", "occurrence_id": "bc008123afbb8a95132ad5d3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Endline Literacy and Numeracy Registry database for 2018, which provides the recorded observations used in the learning analysis. The source is treated as a structured administrative record and not as a standalone statement of the findings. Its entries support the calculation of summary measures for literacy and numeracy performance.\n\nThe analysis aggregates the available observations and derives an average across the relevant records. It also examines the ordering of recorded results to identify the direction and consistency of the observed trend. These calculations provide an analytical basis for the finding that performance was uneven across the assessed learning domains, while avoiding reliance on any single reported value.\n\nThe derived measures are used as one element of the assessment of implementation results. They inform interpretation of the overall pattern, but do not by themselves establish the causes of variation or the durability of observed gains. Results are therefore presented with appropriate caution, particularly where the number of usable records differs between literacy and numeracy measures. The registry remains the underlying reference for the calculation, while the reported averages and trend are analytical outputs generated from its contents.\n\nMention: \"Endline Literacy and Numeracy Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the Endline Literacy and Numeracy Registry database for 2018 to calculate an average and derive a trend supporting the assessment of learning performance.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:86daf6754f1f5f65be8946ea", "occurrence_id": "1ad8abe27c38840296c75e5c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis draws on the Pilot Health and Household Registry, a database produced by the Fiscal Affairs Department. The source is used as a descriptive reference for assessing coverage and the distribution of reported household characteristics. Its contents are treated as administrative evidence rather than as an independent estimate of outcomes.\n\nFor the implementation assessment, the reported figures are reproduced directly from the database. The registry indicates that 68 percent of recorded households received the specified service, while 4,216 households were included in the corresponding count. These figures support the broader finding that operational reach was substantial but not uniform across the registry records.\n\nThe figures are presented without recalculation or adjustment. They should therefore be read as reported measures within the source and not as a basis for inferring changes over time. Further interpretation remains subject to the coverage and recording practices applied to the registry.\n\nMention: \"Pilot Health and Household Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Pilot Health and Household Registry to reproduce the directly reported figures of 68 percent and 4,216 households in support of an implementation assessment.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:075e0bf7a86da80cc9ef538b", "occurrence_id": "94a4f0bd5735614164f3d9af", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the National Storm Surge Registry, a database, as the underlying source for the analysis of recorded surge events. The registry provides a consistent basis for comparing observations across successive reporting periods and for identifying changes in the frequency of recorded events. Its entries are treated as administrative observations rather than as independent estimates of exposure or loss.\n\nFor the purpose of this assessment, the source is used to derive an average event rate and a directional trend over the observation period. These measures are calculated from the available records after routine checks for duplicate entries and incomplete observations. Individual registry entries are therefore not reproduced as stand-alone evidence; instead, their aggregate pattern supports the assessment of changing event frequency.\n\nThe resulting trend is used as one component of the wider risk analysis. It indicates whether recorded activity has increased, declined, or remained broadly stable, while the calculated average provides a reference level for interpreting deviations in individual periods. The results should be read as evidence of reporting patterns captured in the database and not as a complete measure of all underlying occurrences.\n\nMention: \"National Storm Surge Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the National Storm Surge Registry database to calculate an average event rate and a trend in recorded storm surge activity.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:14fa853ba9d6d1e95fe4f086", "occurrence_id": "afef8e313c4d61c583e0f00d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Inter-Agency Cash Transfer Registry Survey is the principal empirical basis for this assessment. The survey covers Sierra Leone and refers to 2016, providing a consolidated record of registered cash transfer activity for the period under review. Its entries are used to establish the analytical baseline and to support comparison across the observed cases.\n\nThe analysis does not reproduce isolated figures from the source. Instead, the recorded observations are combined to calculate an average level of coverage and to derive a trend across the available distribution. This treatment reduces the influence of individual entries and provides a more stable indication of the pattern represented in the registry. The resulting measures are used as summary evidence rather than as reported quotations.\n\nInterpretation remains subject to the scope and consistency of the underlying records. Differences in registration practices, timing, and completeness may affect the calculated results, and the derived trend should therefore be read as an analytical estimate. Within these limitations, the Inter-Agency Cash Transfer Registry Survey supplies the central evidence for assessing the direction and typical level of recorded activity in Sierra Leone in 2016.\n\nMention: \"Inter-Agency Cash Transfer Registry Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses the Inter-Agency Cash Transfer Registry Survey for Sierra Leone in 2016 to calculate an average and derive a trend rather than quote individual figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:70da987bf6ab44cf9fee7e6a", "occurrence_id": "003abbffa0b62fc0aa9168a5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Record-Based Debt and Savings Survey is treated as a survey produced by the Penn World table and is used as a supporting reference for the assessment. Its record-based observations provide an independent basis for checking the estimate of household saving presented in the surrounding analysis.\n\nThe comparison is limited to the claim concerning the direction and approximate magnitude of the saving estimate. Results in the survey are reviewed against an estimate derived from an administrative compilation. Agreement between the two sources is taken as evidence that the reported pattern is plausible, while differences are retained as a reason for cautious interpretation rather than reconciled through an unsupported adjustment.\n\nThis use of the source supports one claim within the broader paragraph and does not replace the principal evidence on which the assessment rests. The cross-check also provides a transparent indication of uncertainty for readers reviewing the calculation. No inference is drawn beyond the variables needed for that comparison, and the source is not treated as establishing causality. The cited material therefore functions as corroborating evidence for the stated estimate.\n\nMention: \"Record-Based Debt and Savings Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the Record-Based Debt and Savings Survey as a survey produced by the Penn World table to cross-check a household saving estimate against an administrative compilation.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:6b92294b2e7ea85368e11f0f", "occurrence_id": "88f4ddf52d127f572f5836ca", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The National Inequality Census (NIC) is a census covering Turkey and produced by the European Space Agency. It is referenced here as a documentary source for the institutional setting in which inequality is considered. The citation is retained with the source description to support traceability and to distinguish background evidence from material used in the assessment.\n\nWithin the broader discussion of distributional conditions, this source supports the claim that inequality measurement is relevant to the assessment framework. The reference provides contextual grounding for that claim, while the analysis itself relies on the evidence described in the surrounding sections. No values reported by the National Inequality Census (NIC) are incorporated into the calculations, comparisons, or findings presented here.\n\nThis treatment limits the role of the census to background documentation. It does not imply that the source supplies the indicators used for estimation, nor that its classifications determine the analytical categories applied in this report. The citation is consequently interpretive rather than quantitative, and its inclusion should not be read as evidence of a separate estimate or an additional measurement exercise. The distinction preserves a clear boundary between contextual material and the evidence directly used to support the assessment.\n\nMention: \"National Inequality Census (NIC)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The source is used to support a contextual claim about inequality measurement, with none of its values used in the assessment.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:1216fb1f95a07e75c1a5bd43", "occurrence_id": "9aa1f7659c6c2f1d7e35aa4c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The cited data source is the Provincial Ward Outline Dataset (PWOD), a geospatial source produced by SECEX for 2020. It provides the underlying spatial observations used to support the assessment of distributional patterns. The dataset is treated as the principal evidence for the calculations reported below, rather than as a source from which isolated values are reproduced.\n\nThe analysis consolidates the available observations into comparable measures and derives an average across the relevant units. It also estimates a directional trend by examining how the measured characteristics vary across the dataset. These results are therefore analytical outputs generated from PWOD records, not figures copied directly from the source. The procedure preserves the relative structure of the observations while reducing the effect of individual irregularities.\n\nThe resulting measures are used to identify broad differences in concentration, continuity, and spatial extent. Interpretation remains limited to relationships supported by the dataset, and no inference is made beyond the information encoded in its geospatial records. The average is reported as a summary statistic, while the trend is used to indicate the prevailing direction of variation. Together, these derived measures provide the basis for the implementation assessment and support consistent comparison within the 2020 reference frame.\n\nMention: \"Provincial Ward Outline Dataset (PWOD)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the 2020 PWOD geospatial records produced by SECEX to derive an average and a directional trend.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:b18715bd69670d7146f5452b", "occurrence_id": "fbdf7457bc02a2f8bb75d6a3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis draws on the Annual Nutrition and Mortality Registry (ANMR), a database covering Nigeria. The source provides a structured basis for examining nutrition outcomes and mortality patterns across the reporting record. Its contents are treated as administrative evidence for the assessment rather than as a set of isolated figures reproduced in the text.\n\nFor the present analysis, ANMR observations are reviewed for consistency and combined into comparable measures. The calculation produces an indicative trend in the relevant rates and an average level across the available observations. This approach reduces the influence of individual fluctuations and supports interpretation of the underlying direction of change. The resulting measures are used to substantiate the assessment of persistent pressure on nutrition and survival outcomes.\n\nThe findings should be read as analytical summaries derived from the database, not as direct quotations from individual registry entries. They provide a concise basis for comparing the general level and movement of the indicators within Nigeria, while recognising that coverage and reporting completeness may vary across observations. Detailed source entries are retained for verification, but the report presents only the calculated pattern and average needed for this assessment.\n\nMention: \"Annual Nutrition and Mortality Registry (ANMR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The passage states that a trend and average for nutrition and mortality indicators in Nigeria were calculated from the ANMR database.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:641ac6a65c82f11cb0c43d83", "occurrence_id": "33fe17c507d17efa8406ebe9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Continuous Built-Up Surface Dataset is used as a geospatial source for 2024. The assessment team collected the available record, harmonised its representation with the analytical framework, and archived the resulting material with the supporting documentation. These procedures were applied to preserve the source structure while enabling consistent interpretation across the assessment files.\n\nThe source provides a spatial basis for distinguishing established built-up surfaces from areas that remain outside the mapped settlement footprint. Its use therefore supports the broader finding that observed differences in developed coverage reflect both the distribution of built surfaces and the limits of the underlying delineation. The dataset is treated as an input to interpretation rather than as an independent estimate of population, service access, or economic activity.\n\nQuality checks focused on internal consistency, continuity of the mapped surface, and compatibility between the archived record and the formats used for review. Any apparent discontinuity was retained for inspection rather than adjusted without documentation. This approach reduces the risk that processing choices are mistaken for changes in the built environment.\n\nThe resulting archive is referenced in the assessment record as supporting evidence for the spatial analysis. It remains subject to the coverage, classification, and resolution characteristics inherent in the source, and conclusions drawn from it are accordingly presented at the level supported by the mapped evidence.\n\nMention: \"Continuous Built-Up Surface Dataset\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The passage records the collection, harmonisation, and archiving of the Continuous Built-Up Surface Dataset as a geospatial source for 2024.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:95918776f412bc1442a958c1", "occurrence_id": "5a830a1e5a71b40ec2f89049", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Experimental Vocational Training Survey, a survey produced by the U.S. Census Bureau and referring to 2018. The source is used to support the finding that reported participation varies across the observed training categories and that the available records require consistent treatment before comparison.\n\nFor operational use, the underlying material was collected from the source files, harmonised against the analytical definitions applied in this assessment, and archived with the supporting documentation. Particular attention was given to consistent coding of participation status, training completion, and reported respondent characteristics. These procedures reduced discrepancies arising from differing labels and facilitated review of the derived tabulations.\n\nThe resulting file was retained as a reference dataset for verification of the estimates presented in the assessment. It does not replace the source material; rather, it provides a controlled basis for reproducing the relevant calculations and checking the interpretation of the reported patterns. The citation is therefore maintained with the processing record and the associated tables, while the source description remains limited to the information required for identification and traceability.\n\nMention: \"Experimental Vocational Training Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The source was collected, harmonised, and archived to support verification of reported training participation patterns.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:deaab77fe117aaf020db95ff", "occurrence_id": "5b1087c08b8f39ac79da49b4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Nationwide Clinic Quality Registry (NCQR) is a database produced by LRMIS in 2015. It is used here as the principal empirical reference for assessing the reported level of clinic quality. The registry supplies a structured set of observations that can be aligned with the measure under review without introducing an additional classification or adjustment.\n\nThe estimate presented in this assessment is cross-checked against the NCQR record. The comparison is intended to test whether the magnitude of the estimate is consistent with the observed distribution in the registry, rather than to replace the estimate or infer a broader time pattern. Where the two sources differ, the registry evidence provides the basis for reviewing the calculation and identifying possible measurement sensitivity.\n\nThe resulting assessment gives greater weight to the estimate where its direction and scale are supported by the NCQR observations. This use of the database also improves traceability, since the cited source provides an independent reference for interpreting the reported result. No additional inference is drawn beyond the evidence required for this cross-check.\n\nMention: \"Nationwide Clinic Quality Registry (NCQR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the 2015 NCQR database produced by LRMIS to cross-check the estimate against registry observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:7e27e9d3a964b0dccf67bdad", "occurrence_id": "7bcd02fcc85007df7ecedace", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Standardised Arrival and Registration Report is the 2020 report used for the assessment of registration completeness. It provides a bounded reference for interpreting the administrative record and supports the finding that recorded arrivals were not uniformly documented across the reporting period. The source is treated as a reference record rather than as an independently adjusted estimate.\n\nTwo figures are reproduced directly from the report. It records 31 registrations in the specified reporting category and indicates that 68 percent of recorded arrivals had a corresponding registration entry. These values are presented as reported and have not been recalculated, extrapolated, or combined with information from another source.\n\nTaken together, the figures support the assessment that registration coverage was substantial but incomplete. The difference between the total recorded registrations and the reported percentage is not interpreted as evidence of a change over time, since the cited material does not provide a basis for estimating a trend. The figures instead serve to anchor the broader observation that administrative completeness remained an operational constraint.\n\nThe source is retained in the supporting documentation to permit verification of the two quoted values and to preserve the evidentiary basis for the assessment. No additional adjustment is applied in the analysis, and the reported figures are used only within the stated interpretive scope.\n\nMention: \"Standardised Arrival and Registration Report\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported figures of 31 registrations and 68 percent from the source to support a claim that registration coverage was substantial but incomplete.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:e64196e8d17bdf20f6ed5ad7", "occurrence_id": "fbb46c62de812ea75287c522", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Biennial Access and Documentation Registry database for 2015 as the principal evidentiary basis for reviewing changes in recorded access and documentation conditions. The source is treated as a structured statistical input, with observations retained for comparability across successive reporting periods.\n\nThe analysis does not reproduce individual registry entries. Instead, the recorded measures are consolidated into comparable groups, and arithmetic averages are calculated to reduce the influence of isolated observations. The resulting series is then examined for its direction and consistency over time, allowing the assessment to distinguish a sustained movement from short-term variation.\n\nThis treatment provides a more stable basis for implementation reporting than reliance on selected records. The derived averages indicate the prevailing level of access and documentation, while the calculated trend signals whether performance is improving, deteriorating, or remaining broadly unchanged. Interpretation is limited to patterns supported by repeated observations, and no inference is drawn from an isolated value. The Biennial Access and Documentation Registry database for 2015 therefore serves as the central source for a derived trend and average, rather than as a set of figures reproduced directly in the text.\n\nMention: \"Biennial Access and Documentation Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage derives comparable averages and a directional trend from the 2015 database rather than reproducing individual registry entries.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:6b8f2337154802085ef0879a", "occurrence_id": "54ce43b45c93b8ee344bd773", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: For the boundary review, the Independent Administrative Boundary Dataset is used as the central geospatial evidence for 2019. The dataset supplies a consistent representation of administrative units and their recorded boundaries, allowing the assessment to establish a common spatial basis for tabulation. Its use is limited to the reference information contained in the source, with no inference about changes beyond the stated reference.\n\nSource preparation involved collecting the available records, checking their internal completeness, and harmonising names, identifiers, and boundary geometries across the retained material. The harmonisation procedure resolved duplicate unit descriptions, aligned coding conventions, and preserved the distinction between an administrative unit and its boundary representation. Where records required interpretation, the adopted treatment was documented in the working archive so that subsequent tables could be reproduced from the same evidence.\n\nThe resulting archive provides the base for summary counts, spatial overlays, and quality checks reported in this assessment. It also permits cross-checking between the unit inventory and the mapped boundary record without introducing a second reference source. Retention of the original material alongside the harmonised files supports auditability and reduces the risk that later processing will obscure omissions, duplicated records, or inconsistencies in the source. The dataset is therefore considered the principal evidentiary input for the boundary-related results.\n\nMention: \"Independent Administrative Boundary Dataset\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The passage uses the Independent Administrative Boundary Dataset as geospatial evidence for 2019 and records its collection, harmonisation, and archiving.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:fb7e6f11f0f70058cf7ccbbb", "occurrence_id": "a056284f47fbdb2557236812", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Regional Business Demography Index (RBDI) is used here as an indicator of business demography in Africa, with the reference period identified as 2020. It provides a concise statistical basis for describing business entry and exit conditions within the assessment framework. The source is cited in its stated form, and no adjustment is applied to the reported values.\n\nThe RBDI indicator is treated as a descriptive measure rather than as a complete account of enterprise performance. Its use supports a focused reading of business turnover, while recognising that the indicator does not by itself explain the causes of observed movements. The resulting interpretation is therefore limited to the pattern represented by the published measure and avoids extending the evidence beyond that scope.\n\nTwo figures are reproduced directly from the source, namely a business entry rate of 12.6 per cent and a business exit rate of 8.4 per cent. These values support the broader assessment that business turnover remains material within the observed business population. They are presented as reported figures and are not recalculated, combined, or converted into a separate trend estimate. Their inclusion provides a factual anchor for the surrounding discussion while preserving the distinction between the cited indicator and the wider assessment.\n\nMention: \"Regional Business Demography Index (RBDI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage takes the 12.6 per cent entry rate and 8.4 per cent exit rate directly from the RBDI indicator for Africa in 2020.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:2b2b7fc70c9dbf1606cb6a75", "occurrence_id": "f0e6fad468e8a50f911aa028", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment draws on the Ministerial Disease Surveillance Survey (MDSS), a survey produced by IOM for 2021. The source is identified as a supporting reference for the interpretation of surveillance conditions relevant to the assessment. No geographic scope, edition, or collection round is assigned to the citation.\n\nThe reference is used within a broader discussion of information quality and institutional reporting. It supports the claim that disease surveillance material provides relevant background when interpreting implementation conditions and assessing the consistency of reported patterns. The citation does not extend the evidentiary basis of the analysis beyond that contextual function.\n\nNo figures, rates, counts, distributions, or other values from the MDSS are incorporated into the assessment. The source is not used to calculate a change, establish a benchmark, or compare results across reporting periods. Its role is limited to confirming the relevance of surveillance evidence to the surrounding narrative, while the substantive findings remain based on the evidence described elsewhere in the report.\n\nMention: \"Ministerial Disease Surveillance Survey (MDSS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The 2021 IOM-produced MDSS survey is used only as contextual support for one claim, with no values taken from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:03922eff9c58779d9ae7dd0d", "occurrence_id": "b214cc490506faf847e0738a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Household Place Name Index is treated as an indicator for monitoring the stability and distribution of recorded household place names. It provides the central evidence for the assessment because its entries can be compared across successive observations using a consistent counting procedure. No individual entry is treated as a substantive outcome in isolation.\n\nFor analytical purposes, the recorded counts are aggregated and examined as a sequence. The resulting calculations produce an average level and a trend measure, allowing the assessment to distinguish persistent movement from isolated variation. This treatment supports interpretation of changes in coverage, concentration, and continuity without relying on selected examples from the underlying register.\n\nThe derived measures are reported in the main results table and used to support implementation judgments. Differences between successive observations are reviewed alongside the calculated average, while unusually large changes are subject to verification before being incorporated into the assessment. The approach therefore uses the Household Place Name Index as an analytical base rather than reproducing its entries as quoted figures. Any limitations in completeness or consistency affect the derived measures and are considered when interpreting the findings.\n\nMention: \"Household Place Name Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives an average and a trend measure from the Household Place Name Index rather than reproducing individual entries.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:e5c29e7d068272b502cb57cd", "occurrence_id": "dd6c0906fa89bc95feaec5be", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Civil Public Works Registry (CPWR) database for 2021 provides the underlying observations used in this assessment. It is treated as the reference record for the distribution of reported public works activity, with attention to consistency across the entries retained for analysis. No additional attribution or classification is introduced in this paragraph.\n\nThe analysis does not reproduce isolated figures from the registry. Instead, the recorded observations are aggregated to derive an average level of activity and a directional trend across the relevant categories. This treatment reduces the influence of individual entries and supports comparison within the broader assessment of implementation performance.\n\nThe resulting measure is used as one supporting claim in the assessment rather than as a standalone estimate of outcomes. Interpretation remains limited to the information available in the Civil Public Works Registry (CPWR) database, and the calculated average and trend are presented as analytical summaries of that record. These summaries inform the discussion of implementation consistency and provide a basis for identifying whether activity was broadly stable, increasing, or declining during the period under review.\n\nMention: \"Civil Public Works Registry (CPWR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the 2021 CPWR database to calculate an average and a directional trend rather than reproduce individual reported figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:4b1278c3dc37896558743876", "occurrence_id": "d00d0d182c67323bb3b6e820", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Quality-Assured Service Catchments Index, an indicator covering Xinjiang province, is cited to situate the discussion of service access within the broader assessment framework. Its inclusion provides a recognised point of reference for the interpretation of catchment conditions and supports the contextual statement presented in this section.\n\nThe cited material is not used to report, reproduce, or calculate any of its values. No numerical result from the indicator is incorporated into the assessment, and no inference is drawn from its underlying measurements. The reference therefore serves a limited evidentiary function rather than forming a basis for comparison or trend analysis.\n\nThe principal findings continue to rely on the assessment evidence described elsewhere in the report. Within that broader paragraph, the source supports one claim concerning the relevance of service catchment conditions to implementation performance. This use preserves the distinction between contextual documentation and analytical inputs, while allowing the discussion to acknowledge the wider information base associated with Xinjiang province.\n\nMention: \"Quality-Assured Service Catchments Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The passage uses the Quality-Assured Service Catchments Index to support one contextual claim about service catchments in Xinjiang province and takes no numerical values from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:ce7c14ece02847ecef2ae1b9", "occurrence_id": "de1deaf72007993fda2def83", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis draws on the Departmental Housing Conditions System, an administrative source referring to 2016. The material is treated as a structured record of housing conditions and is used to support the assessment of changes in household circumstances. No separate institutional attribution, geographic qualifier, or release designation is attached to the source in this assessment.\n\nThe source contributes to the estimation of a general movement in observed conditions rather than to the reproduction of individual records. Measures are grouped across the available observations, and an average level is calculated to reduce the influence of isolated entries. The resulting estimate is interpreted as an indicative pattern for the assessment and not as a direct quotation from the underlying system.\n\nThis treatment supports the broader finding that reported housing conditions vary across the observations considered. The calculated average and the resulting direction of change are used as one element of the evidence base, alongside the internal consistency checks applied to the compiled material. Differences in reporting intensity may affect the magnitude of the estimate, but they do not alter its use as a summary measure for implementation review.\n\nMention: \"Departmental Housing Conditions System\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the administrative Departmental Housing Conditions System for 2016 to calculate an average and assess the direction of change in housing conditions.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:2b5e9c582d9095e9d70441a4", "occurrence_id": "3a4b98abd14a7089f2b69fea", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Data sources.** The Linked Sea Level Report (LSLR) is used as the report for 2018. It provides the evidentiary basis for the quantitative statement presented in this section, while the surrounding discussion interprets the figures within the broader assessment of observed conditions and associated planning needs.\n\nTwo figures are quoted directly from the report, 12.4 and 18.7. They are reproduced without recalculation and are retained at the precision given in the source. The first figure establishes the lower reference point used in the discussion, while the second indicates the higher recorded value relevant to the comparison.\n\nTaken together, these observations support the claim that the reported range is material for prioritising follow-up analysis and for determining whether additional measures are warranted. The figures are not treated as a complete assessment on their own; rather, they substantiate the specific comparison made in the paragraph and provide a transparent basis for the stated conclusion.\n\nThe citation is therefore limited to the direct use of the two reported values. No further characteristics of the source are inferred, and no adjustment is applied to the underlying numbers. This treatment is consistent with the requirements for traceable use of externally reported evidence in implementation monitoring and preserves a clear distinction between source data and the interpretation applied in the assessment.\n\nMention: \"Linked Sea Level Report (LSLR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported figures 12.4 and 18.7 from the Linked Sea Level Report (LSLR) for 2018 to support a comparison in the assessment.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:86040427038f0d46c0ad9e0d", "occurrence_id": "e1b673ca0da44fb9d02d5421", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment reviews the institutional basis for social protection reporting and considers the extent to which available records can inform the description of system arrangements. The discussion remains descriptive and does not derive indicators or numerical findings from the cited material.\n\nWithin this broader assessment, the Biennial Social Protection System is cited as an administrative source produced by WTO. The reference supports the limited claim that organized reporting arrangements provide relevant background for understanding the structure of social protection information. Its role is confined to establishing that background and does not extend to the extraction, comparison, or interpretation of any values.\n\nThe source is therefore treated as documentary support for one statement in the surrounding analysis rather than as evidence for a calculated result. No measure, distribution, or change is attributed to it, and no conclusion depends on figures drawn from the system. This treatment preserves the distinction between contextual documentation and the evidence used for substantive assessment.\n\nThe citation is retained in the assessment record to identify the basis for the institutional observation. Other findings are presented separately and are not represented as having been generated from the Biennial Social Protection System.\n\nMention: \"Biennial Social Protection System\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The source provides contextual support for an institutional observation, with no values taken from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:5c7654d5491766caa5bd2bc4", "occurrence_id": "e324f8a20dfa8afab3b8e076", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Audited Internal Movement Registry (AIMR) database covering Lagos State is used as the primary administrative reference for the movement profile presented in this annex. The source is cited in its stated form because the reported values are reproduced without recalculation or adjustment. Its use provides a consistent basis for describing the recorded scale of internal movement within the covered area.\n\nTwo figures are taken directly from the database. It records 18,730 internal movements during the reporting period, while 62.4 percent of recorded movements are associated with arrivals in receiving areas. These values are reproduced as reported and provide the central quantitative evidence for the assessment. They are not presented as an estimated population total or as a measure of movements outside the registry’s coverage.\n\nThe figures should therefore be interpreted as administrative counts subject to the scope and recording practices of the registry. Differences in registration access, reporting completeness, or timing may affect the observed distribution. Within these limitations, the two reported values support a concise assessment of movement volume and concentration in Lagos State and establish the statistical reference used in the surrounding analysis.\n\nMention: \"Audited Internal Movement Registry (AIMR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Audited Internal Movement Registry (AIMR) database to report 18,730 internal movements and a 62.4 percent share for Lagos State.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:fda9496aa9da7bbeda0740be", "occurrence_id": "75451d2b8cde6a722cab7389", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment of labour-market implementation conditions draws on the Small-Area Hiring and Vacancies Registry (SHVR) database as a cited administrative source. The database is used to anchor a limited observation on the balance between reported hiring activity and unfilled demand. Its inclusion provides a traceable basis for the numerical statements that follow, while the surrounding interpretation remains qualitative.\n\nRecruitment pressure remains material where the number of open positions exceeds completed hiring activity. In the SHVR database, 31 vacancies and 18 hires are reproduced directly as recorded figures. These observations support the assessment that available staffing has not fully matched stated requirements. They are presented as reported levels rather than as an estimate of coverage or a measure of change over time.\n\nThe figures should therefore be read as support for the specific implementation finding that unmet staffing needs persist, rather than as a comprehensive measure of labour-market performance. The source is retained in the reference record under its full title and acronym, enabling verification of the two cited values. No additional aggregation is applied in this passage, and the interpretation is limited to the relationship between vacancies and hires shown in the database.\n\nMention: \"Small-Area Hiring and Vacancies Registry (SHVR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the SHVR database for the directly reported counts of 31 vacancies and 18 hires.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:5e3fbdf649f5f88772358107", "occurrence_id": "e7f94555952e773c29584a43", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment considers the available evidence on service delivery conditions and operational readiness. The Enumerative Service Delivery and Readiness Survey (ESDRS) is cited as a survey relevant to this evidence base and is noted among the materials consulted for general orientation.\n\nThe reference is used to indicate the broader information setting in which the assessment was undertaken. It does not provide the basis for reported levels, comparisons, rankings, or derived estimates in this passage. No result from the survey is interpreted as an observation for the assessment population.\n\nThe analysis presented here relies on the documented assessment procedures and the records identified for that purpose. The ESDRS is therefore retained as a supporting reference rather than as an input to calculation. Its mention helps situate the discussion of readiness and service delivery without extending the scope of the findings or affecting the interpretation of the reported evidence.\n\nMention: \"Enumerative Service Delivery and Readiness Survey (ESDRS)\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"No numerical values or analytical results were taken from the ESDRS; it served only as contextual background.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:83bd4ffff98f25f000e20fe6", "occurrence_id": "ce6f29612ccec360d4a29a3e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Evidence from the Monthly Adaptation Finance Dataset (MAFD) is used as a geospatial source for 2019. The assessment does not reproduce individual monthly observations from the dataset. Instead, the reported pattern is derived through a consistent calculation applied across the available monthly entries, thereby retaining the source’s temporal structure while avoiding reliance on a single reported value.\n\nFor analytical purposes, the observations are ordered chronologically and reviewed for continuity. A mean is calculated to represent the typical monthly level, and a trend measure is derived to indicate the direction and pace of change over the period. These calculations provide the basis for the narrative assessment and are treated as estimates generated from the MAFD material rather than as figures quoted directly from it.\n\nThe resulting interpretation indicates whether activity was broadly stable, increasing, or declining during the period under review. Short-term fluctuations are therefore considered in relation to the calculated average, while the trend measure is used to distinguish persistent movement from isolated variation. This approach supports comparison within the dataset and provides a transparent basis for the assessment’s findings. Any conclusion is consequently dependent on the completeness and internal consistency of the monthly observations available in the cited source.\n\nMention: \"Monthly Adaptation Finance Dataset (MAFD)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses MAFD to calculate a monthly average and trend for 2019 rather than quote individual observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:bbbf5969a5b85ffdd6d33354", "occurrence_id": "0ca0130bb4da09d7b7a44968", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Household Rainfall Anomaly Registry database provides the reference record for the Republic of Zambia covering 2015. It is used here to anchor the assessment of reported rainfall-related household effects and to maintain consistency in the interpretation of the underlying observations. No adjustment or supplementary estimation is applied to the values reproduced from the source.\n\nThe database records 31 households reporting a rainfall anomaly during the reference period. This figure is reproduced directly from the registry and is treated as an observed count rather than as an inferred estimate. The same source records 18 percent of assessed households as having experienced a related disruption. That percentage is also taken directly from the registry and is retained without recalculation.\n\nTogether, these two figures provide the central quantitative basis for the discussion. They indicate the scale of reported exposure and the proportion associated with a recorded disruption, while preserving the distinctions used in the source database. The figures should therefore be read as registry observations for the specified coverage and period, rather than as projections or measures of change over time. Their use in this assessment is limited to documenting the reported distribution of effects within the source record.\n\nMention: \"Household Rainfall Anomaly Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The author uses the source’s direct figures of 31 households and 18 percent to document reported rainfall-related exposure and disruption.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:f55b67ae8a394c3142f800a7", "occurrence_id": "8c6562156019331ef9c51a76", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Joint Seasonal Forecast Index is used here as an indicator of the information available for anticipating seasonal conditions. It provides a consistent reference for the assessment of planning assumptions and supports the finding that preparedness measures were aligned more closely with anticipated conditions than with retrospective observations. The index is considered alongside the broader evidence, rather than treated as a complete measure of implementation performance.\n\nFor operational use, the underlying records were collected from the relevant reporting materials, checked for internal consistency, harmonised across reporting units, and archived with the supporting documentation. This process reduced differences in terminology and presentation that could otherwise affect comparison. The resulting series was retained as a traceable input to the assessment and was not extended beyond the observations available in the source.\n\nThe evidence is therefore used to substantiate one claim concerning the relationship between forecast information and planning decisions. It does not, on its own, establish the effectiveness of those decisions, the accuracy of subsequent conditions, or the distribution of outcomes. Interpretation remains subject to the limitations of coverage, reporting continuity, and the consistency of the records supplied for consolidation. These qualifications are reflected in the assessment narrative and in the treatment of the indicator as supporting evidence rather than as a standalone performance measure.\n\nMention: \"Joint Seasonal Forecast Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The author collected, harmonised, and archived the Joint Seasonal Forecast Index as an indicator supporting a claim about forecast information and planning decisions.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:f22d853a8dbbf59025a17bdb", "occurrence_id": "bdf1d6abae0a4a07be14283a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment of classroom practice draws on the Biennial Classroom Practice Registry (BCPR) database covering Bhopal in 2022. The source provides a structured record of observed instructional routines, classroom organisation, use of learning materials, and teacher–pupil interaction. It is used here to support the finding that implementation conditions varied across classrooms, particularly in the regularity with which planned activities were documented.\n\nRecords were collected through the designated reporting process and reviewed for completeness before analysis. Entries with differing terminology were harmonised to permit comparison across classroom observations, while retained identifiers were archived with the supporting documentation. This treatment reduced duplication and limited the risk that variations in recording practice would be interpreted as substantive differences in classroom conditions.\n\nThe resulting evidence is used as a corroborating source rather than as a standalone measure of performance. It indicates that the presence of planned instruction did not invariably correspond to consistent classroom delivery, and it provides a basis for interpreting the implementation results reported in the assessment. The database also permits subsequent verification of the underlying observation records, subject to the preservation of the archived materials and associated documentation.\n\nMention: \"Biennial Classroom Practice Registry (BCPR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The passage uses the BCPR database for Bhopal in 2022 and records the collection, harmonisation, and archiving of its classroom-practice data.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:96c8b2247fcb1e6d05a5f4b4", "occurrence_id": "21a8da0066f8ff88268a7d33", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Record-Based Coastal Erosion Census as the principal empirical reference for the reported coastal exposure. It is a census produced by the National Treasury, covering Bhopal and referring to 2022. The source is treated as the controlling record for the two quantitative observations presented below.\n\nThe census records 31 affected shoreline segments within the covered area. It also reports 14.6 kilometres of exposed coast associated with those segments. These figures are reproduced directly from the record and are not recalculated for this assessment.\n\nThe reported values provide the basis for the scale statement used in the results section. They indicate a measurable concentration of shoreline exposure and support the classification of the affected area as requiring continued observation. No adjustment has been applied to either figure, and no supplementary estimate is introduced. Interpretation is therefore limited to the scope and coverage stated by the Record-Based Coastal Erosion Census.\n\nMention: \"Record-Based Coastal Erosion Census\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the census figures of 31 affected shoreline segments and 14.6 kilometres of exposed coast for Bhopal in 2022.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:077951ba93acece113fe9658", "occurrence_id": "b91b46b604f2e44d92324cf5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Municipal Fuel and Energy Use Registry (MFEUR) database provides the underlying record for the assessment of municipal fuel and energy consumption. Its coverage of reported use supports the finding that observed differences across municipal services reflect both variation in demand and differences in the completeness of reporting.\n\nThe records were collected from administrative returns and harmonised to improve comparability between reporting units and periods. Units of measurement, fuel categories, and entries with inconsistent descriptions were reviewed through a common processing procedure. Where duplicate or incomplete entries were identified, they were retained in the archive with the relevant qualification rather than removed without documentation.\n\nThe resulting archive was used to prepare the analytical tables and to verify the consistency of aggregate totals presented in the assessment. The source is therefore treated as a structured evidentiary database rather than as an independent estimate of consumption. Its use permits the assessment to distinguish reported changes in fuel and energy use from changes attributable to data quality or classification practices.\n\nThe MFEUR database also provides a traceable basis for subsequent updating, subject to the continuation of comparable reporting and the preservation of the harmonised records.\n\nMention: \"Municipal Fuel and Energy Use Registry (MFEUR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The passage draws on the collection, harmonisation, archiving, and verification of municipal fuel and energy use records in the MFEUR database.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:42bd7c1e57d3c69d0eda0063", "occurrence_id": "6ca804f8b2b0a535623b7bed", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment considers whether basic service conditions are described consistently across the implementation narrative. The evidence is interpreted conservatively, since the discussion combines administrative observations with operational explanations rather than presenting a standalone statistical series.\n\nReference is made to the Local Output Area System (LOAS), an administrative source produced by UDAPV, solely to frame the discussion of how local output information can support interpretation of the observed pattern. The citation does not introduce any values from the system into the analysis, and no calculation, comparison, or measurement is derived from it.\n\nWithin the broader discussion, the source supports one limited statement concerning the availability of structured output information. This statement provides background for the interpretation of implementation conditions but does not alter the assessment or supply evidence for additional findings. The analysis therefore remains based on the material presented elsewhere in the report, with the cited source serving only to establish the relevance of the information category under review.\n\nMention: \"Local Output Area System (LOAS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The passage uses the Local Output Area System (LOAS), produced by UDAPV, only to provide contextual support for one claim.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:b2cea5ea0ce2416c5c66967c", "occurrence_id": "c9dd54b2d973f54a554503db", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Linked Shock Coping Survey (LSCS) is the survey used for the assessment of reported coping responses. Its use is limited to the direct presentation of selected findings, with no adjustment, extrapolation, or construction of a derived indicator undertaken in this note. The source provides a consistent basis for describing the incidence of reported responses within the assessed population.\n\nTwo figures are reproduced directly from the LSCS. The survey records that 54 percent of respondents reported reducing non-essential expenditure following a shock. It also records that 27 percent reported drawing on savings or other readily available reserves. These figures are presented as reported observations and are not interpreted as estimates of change over time.\n\nThe two measures provide the principal quantitative evidence for the discussion of immediate coping behaviour. The first indicates the prevalence of expenditure reduction, while the second captures reliance on available financial reserves. Their differing levels suggest that coping responses were not uniform across respondents, although the survey alone does not establish the reasons for that difference.\n\nFor implementation purposes, the LSCS evidence is therefore used descriptively. The figures should be read as source values reproduced without recalculation, and any comparison with subsequent monitoring results would require a separate assessment of coverage, comparability, and measurement conditions.\n\nMention: \"Linked Shock Coping Survey (LSCS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes two reported figures from the Linked Shock Coping Survey (LSCS), namely 54 percent reducing non-essential expenditure and 27 percent drawing on savings or reserves.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:bdb807d5bec2fb0276c636a6", "occurrence_id": "d9475919e16fbe871de84db6", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Civil Business Demography Registry is a database produced by NCAER covering Eritrea. It provides an administrative reference for assessing the distribution and continuity of registered business activity. The source is used here to support the interpretation of observed establishment patterns, while maintaining a clear distinction between recorded counts and broader analytical conclusions.\n\nThe reported totals reproduced directly from the database are 31 active establishments and 14 recorded closures. These figures indicate that business entry and exit occur within the same observed population, although the counts alone do not establish the reasons for changes in status. They therefore support a limited assessment of business continuity rather than a causal explanation of performance.\n\nThe evidence is treated as descriptive and is presented without extrapolation beyond the recorded units. In particular, the figures provide a compact basis for discussing the scale of formal activity and the incidence of closure in the covered population. Interpretation should remain proportionate to the available observations, with attention to possible differences in registration status, reporting completeness and the timing of administrative updates. The registry nevertheless offers a consistent source for the two figures used in this assessment and strengthens the transparency of the supporting evidence.\n\nMention: \"Civil Business Demography Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Civil Business Demography Registry for Eritrea and reports 31 active establishments and 14 recorded closures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:97f071d181d7a75eee62cf67", "occurrence_id": "14d8c1a5dc4cf8d7eb7ede63", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Metropolitan Housing Conditions Survey, a 2015 survey, provides supporting evidence for the assessment of housing conditions. Its reported observations are used to substantiate the finding that housing circumstances vary across household groups and that recorded deficiencies require differentiated attention. The source is treated as empirical background to the analysis rather than as a basis for generating a new time series.\n\nFor the purposes of this assessment, the underlying records were collected, reviewed, and harmonised with the classification applied to the other materials examined. Differences in terminology, coverage of indicators, and recording practices were documented before the information was entered into the working archive. Where a direct correspondence could not be established, the relevant observation was retained as a source note and excluded from aggregate interpretation.\n\nThe archived material supports quality assurance by allowing reported patterns to be checked against the original survey evidence. It also provides a traceable basis for interpreting the housing-related statements presented in the results section. No adjustment has been made to the reported observations beyond the harmonisation required to align definitions and remove duplicate entries. The resulting use of the source is therefore limited, transparent, and consistent with the evidentiary role assigned to supporting survey material.\n\nMention: \"Metropolitan Housing Conditions Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The passage records the collection, harmonisation, and archiving of evidence from the 2015 survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:05bcf6b1adca47d41b5fa5ed", "occurrence_id": "3b86d337a767e0f510425a05", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis draws on the Periodic Time Use System (PTUS), treated as an administrative source for measuring how available time is distributed across routine activities. The records provide the principal empirical basis for the estimates presented in this annex. Their use supports consistent treatment of activity categories and permits comparison across the reporting series without introducing information from other datasets.\n\nRather than reproducing isolated observations, the analysis aggregates recorded durations and calculates average time allocations for the population groups under review. It then derives a trend from the resulting averages, with attention to the direction and persistence of change across successive observations. The reported pattern therefore represents a calculation based on PTUS records and not a quotation of a single reported figure. This approach reduces the influence of short-term irregularities and provides a more stable basis for interpretation.\n\nResults are presented as summary measures suitable for monitoring implementation. The average indicates the typical allocation of time, while the trend identifies whether that allocation is increasing, declining, or remaining broadly stable. Apparent movement is interpreted with caution where the underlying number of observations is limited or where activity classification may affect comparability. The source remains central to the assessment, and the calculations are retained so that the reported measures can be reviewed and reproduced from the same administrative records.\n\nMention: \"Periodic Time Use System (PTUS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage describes calculated average time allocations and trends derived from Periodic Time Use System (PTUS) administrative records.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:bf11652ad74bd1187d78e951", "occurrence_id": "f3455fe368c65a0bccfbbd5d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Household consumption measures are used to establish the distributional profile underlying the assessment. The analysis relies on the Standardised Household Consumption Census (SHCC), a census covering Riga and referring to 2017. The Ministry of Investment produces the source, which provides the principal observations for the consumption indicators used in this section.\n\nThe source supports the finding that reported consumption varies materially across household groups and that the resulting profile is suitable for interpreting differences in access to basic goods and services. Its use is limited to this evidentiary purpose; the figures are not presented as a separate projection or as a measure of subsequent change.\n\nThe documentation records the collection of the SHCC material, followed by harmonisation of household identifiers, expenditure categories, and measurement units. Archiving procedures preserve the original observations and the transformed analytical files, permitting the tabulations to be reproduced from the same underlying records. These procedures also provide a basis for checking whether missing responses and inconsistent classifications were treated uniformly.\n\nIn the assessment, the harmonised census material is therefore used to substantiate the stated consumption pattern while maintaining a clear distinction between observed household conditions and the interpretation applied to them.\n\nMention: \"Standardised Household Consumption Census (SHCC)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The passage identifies the Standardised Household Consumption Census (SHCC) as a 2017 census for Riga produced by the Ministry of Investment and notes its collection, harmonisation, and archiving.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:8efa3ca58b62f4de8f718f25", "occurrence_id": "8a3aac501a0653b1747897f1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Follow-Up Value Chain Registry is a database produced by UNHCR & WFP covering Accra for 2018. It is used here as the documentary basis for a limited assessment of participation patterns, with the source attribution retained in the accompanying text rather than separated from the evidence.\n\nThe assessment reproduces two values directly from the database, namely 31 registered operators and 47 participating workers. No recalculation, projection, or adjustment is applied to these figures. Their inclusion provides a narrow quantitative basis for the observation that participation was concentrated among a relatively limited group of operators and workers.\n\nThe figures are presented as descriptive evidence rather than as an estimate of wider activity. They support the paragraph’s principal claim that recorded participation was uneven within the covered area, while leaving broader conclusions outside the scope of the available record. Interpretation is therefore limited to the reported counts and to the period identified by the source.\n\nMention: \"Follow-Up Value Chain Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the directly reported figures of 31 registered operators and 47 participating workers from the Follow-Up Value Chain Registry to support a claim about concentrated participation.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:1b6b7b55c3b8050e9602a884", "occurrence_id": "7c907841eef53007096e9989", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Clustered Business Demography System (CBDS) is treated as an administrative source produced by the Ministry of Interior for 2017. Its records provide the principal quantitative basis for the limited business-demography assessment presented below. The source is cited in its institutional form to preserve the attribution of the underlying observations.\n\nThe reported records show 31 newly registered enterprises and 14 enterprise closures. These two figures are taken directly from CBDS and are reproduced without recalculation, adjustment, or interpretation beyond their presentation as recorded totals. They therefore provide the central evidence for the comparison in this section.\n\nThe difference between the two reported counts is used only to describe the balance reflected in the source. No additional estimate is derived, and no inference is made regarding causes, coverage, or subsequent developments. The figures should consequently be read as administrative counts attributable to the Ministry of Interior record for 2017.\n\nMention: \"Clustered Business Demography System (CBDS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the CBDS figures of 31 newly registered enterprises and 14 enterprise closures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:4df865d777d02ffe32b38875", "occurrence_id": "c0dce1cf7a7259e3999383c8", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Integrated Asset Ownership Survey (IAOS) is the survey used to establish the 2016 baseline for the asset ownership analysis. Its observations provide the principal empirical basis for assessing differences in reported ownership across the analytical categories used in this assessment. No additional source description is required for the calculations presented below.\n\nThe analysis derives average ownership rates from the IAOS observations and compares those averages across categories. This treatment supports an assessment of the direction and relative strength of the observed pattern, while avoiding reliance on isolated figures reproduced from the source. The resulting measures are therefore analytical estimates based on the survey material rather than direct quotations of its reported values.\n\nThe calculated averages are used to identify categories with comparatively higher or lower ownership and to describe the trend across the distribution. Interpretive weight is assigned to the pattern that remains after aggregation, rather than to individual responses. The results should consequently be read as summary measures generated from the Integrated Asset Ownership Survey (IAOS), with the 2016 survey serving as the evidentiary foundation for the assessment.\n\nMention: \"Integrated Asset Ownership Survey (IAOS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses 2016 IAOS observations to calculate average ownership rates and assess category trends.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:56f11739c9e204c5abf47984", "occurrence_id": "2f760b8a2596a192145c19b5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Pilot Student Transport System (PSTS) is cited as an administrative source recorded for 2019 in relation to the assessment of student transport arrangements. Its inclusion indicates that transport provision is a relevant operational consideration for the analysis, while leaving the source in a supporting rather than evidentiary role.\n\nThe citation supports one claim within the broader discussion of implementation conditions. It helps establish the institutional relevance of student transport to the subject under review, but it does not supply a basis for comparison, estimation, or measurement in this passage. No value from the source is reproduced, recalculated, or used to derive a quantitative result.\n\nAccordingly, the reference is retained to document the background for the stated claim and to make the evidentiary boundary clear. The analysis relies on the surrounding discussion for its interpretation, whereas the Pilot Student Transport System (PSTS) serves only to confirm that the topic has an administrative record associated with 2019. No additional attributes of the source are required for the conclusion presented here.\n\nMention: \"Pilot Student Transport System (PSTS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The passage uses the 2019 administrative PSTS source only for contextual support of one claim and takes none of its values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:ba17c2685294b8ef4dd1ae1b", "occurrence_id": "bc24a5d525b01ed944ee3e6e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Biennial Durable Solutions Report (BDSR) is used as the report supporting the assessment of recorded outcomes. Its tabulations provide a documentary basis for the interpretation of coverage, classification, and reported status. No adjustment, extrapolation, or reconstruction is applied to the figures reproduced here.\n\nFor the indicator under review, the report gives 31 grades and 14 households. These two values are reproduced directly from the BDSR and are retained in their reported form. They support the broader finding that the recorded caseload is distributed across distinct administrative categories rather than concentrated in a single status.\n\nThe figures are used descriptively and do not constitute an independent estimate of population size or programme performance. Their role is to substantiate the assessment statement concerning the composition of the recorded caseload. Other interpretations remain subject to the definitions and coverage applied in the source material.\n\nAccordingly, the BDSR is cited as the underlying documentary reference for this part of the assessment, while the surrounding analysis confines itself to the two reported values. This approach preserves traceability and avoids implying a calculation that is not presented in the source.\n\nMention: \"Biennial Durable Solutions Report (BDSR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported figures of 31 grades and 14 households from the BDSR to support a broader assessment of caseload composition.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:e782b19f7a5af2e2e2cbb363", "occurrence_id": "c2822f84bbe35223fa63e2f2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: To support the assessment of deprivation patterns, the Audited Multidimensional Poverty Census (AMPC) is used as the principal census for the household indicators examined in this annex. The records provide a consistent basis for reviewing the incidence and composition of multidimensional poverty across the analytical series. No individual observation is treated as a standalone estimate of current conditions; the evidence is considered in aggregate.\n\nThe reported direction of change is derived by calculating a period-to-period trend from the AMPC observations and by estimating an average across the available observations. This treatment reduces the influence of isolated fluctuations and supports the interpretation of gradual movement in deprivation. The result is therefore an analytical measure generated from the census, rather than a figure reproduced from a single entry or presented as a direct quotation.\n\nOn this basis, the assessment finds a broad reduction in the combined deprivation measure, while noting that progress is uneven among the component dimensions. The finding is used to substantiate the wider conclusion that improvements in aggregate welfare may coexist with persistent gaps in selected indicators. Further interpretation should remain aligned with the coverage and measurement rules applied to the AMPC census.\n\nMention: \"Audited Multidimensional Poverty Census (AMPC)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The analysis derives an aggregate deprivation trend and average from the Audited Multidimensional Poverty Census (AMPC).\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:f68c59ee89a32b97f16d7943", "occurrence_id": "24549a4abea6708c5c2bd31a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Subnational Settlement and Elevation Index (SSEI) is used as an indicator for 2016 in support of the settlement and terrain measures presented in this assessment. The source is cited as a reference for the underlying observations and is not treated as a substitute for the project’s monitoring framework.\n\nTwo figures are reproduced directly from the SSEI record. The reported settlement measure is 62.4 percent, while the corresponding elevation measure is 1,240 metres. These values are retained in their stated form and are not recalculated, reclassified, or combined with other observations.\n\nThe figures provide a limited statistical reference for interpreting variation across the records reviewed. Their inclusion supports consistency between the narrative discussion and the cited source, while avoiding additional assumptions about measurement coverage or comparability. The SSEI values should therefore be read as quoted observations used to document the assessment, rather than as estimates generated through the present analysis.\n\nMention: \"Subnational Settlement and Elevation Index (SSEI)\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"source\", \"summary\": \"The passage cites the SSEI indicator for 2016 and reproduces the figures 62.4 percent and 1,240 metres directly from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:bd9618c620bd39631655e0e2", "occurrence_id": "06e0941d6d884a71b01d3a69", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Local Poverty and Vulnerability Registry database for Viet Nam in 2020 as its principal empirical source. The database provides records on household conditions, exposure to shocks, and access to basic services. These records are treated as the basis for aggregate measurement rather than as isolated case observations.\n\nIndicators were calculated from the available records after applying common definitions to the relevant household characteristics. The analysis derives average values for the principal vulnerability measures and computes changes across the recorded categories. This approach supports comparison of relative conditions while limiting the influence of unusually large or small observations.\n\nThe resulting estimates are used to identify the direction and magnitude of variation in vulnerability. No single registry entry is presented as representative of the population. Instead, the calculated averages and derived trend provide the evidence for assessing distributional differences and for establishing a consistent reference point for subsequent monitoring. The results should therefore be interpreted as analytical measures constructed from the database, rather than as figures reproduced directly from individual records.\n\nMention: \"Local Poverty and Vulnerability Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the Local Poverty and Vulnerability Registry database for Viet Nam in 2020 to calculate averages and derive vulnerability trends.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:79277c3d731aa587cd02af33", "occurrence_id": "30867980a3e115e13c4e756d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Linked Time Use Survey is a survey produced by the World Bank Group and is used here as the sole statistical source for the assessment. Its tabulations provide a consistent basis for describing the distribution of daily activities and for anchoring the discussion in observed rather than assumed conditions. No adjustments or supplementary estimates are introduced in this passage.\n\nTwo figures are reproduced directly from the Linked Time Use Survey. The first indicates that 46 percent of respondents reported undertaking the activity under review. The second records an average of 2.3 hours per day devoted to that activity. Together, these figures provide the central evidence for the assessment and establish both the extent of participation and the associated time burden.\n\nThe figures are presented without extrapolation. They should therefore be read as reported measures within the survey framework, rather than as projections or administrative totals. Their use in the assessment is limited to supporting comparison across the analytical categories already defined in the project work, while preserving the source’s original measurement basis.\n\nMention: \"Linked Time Use Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the 46 percent participation figure and the 2.3 hours per day estimate directly from the Linked Time Use Survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:267812d6916a8af18a56b5f8", "occurrence_id": "b19b57db2586326b37e64ede", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis uses the Quarterly Service Catchments Registry database, Version 9, covering Afghanistan, as the underlying observation frame for service-access assessment. The source provides catchment-level records that support comparison across reporting intervals and administrative units.\n\nThe assessment does not reproduce individual registry entries as reported values. Instead, observations are aggregated to derive average coverage levels and to estimate the direction and consistency of change across successive intervals. This treatment reduces the influence of isolated catchment movements and supports a more stable interpretation of service availability.\n\nThe resulting trend indicates whether access expanded, remained broadly unchanged, or weakened over the observation sequence. Differences between catchments are retained in the calculations, while the reported summary reflects the average pattern rather than any single record. The estimates are therefore used as an analytical input to the broader implementation assessment, not as a standalone measure of performance.\n\nInterpretation remains subject to the scope and internal consistency of the registry records. The analysis gives greater weight to persistent movements observed across multiple intervals and treats short-term deviations cautiously. On this basis, the derived averages provide a concise measure for comparison in the accompanying assessment.\n\nMention: \"Quarterly Service Catchments Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The analysis uses Version 9 of the Quarterly Service Catchments Registry database for Afghanistan to derive average coverage and a service-access trend.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:1688c91299c4c94961dedf43", "occurrence_id": "0f78e0d21d171fdd49031945", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Quarterly Migration and Remittance System is an administrative source covering Tigray Region. It provides the principal empirical basis for assessing movements in recorded migration and remittance activity within the project’s monitoring framework. The source is treated as a continuous body of observations rather than as a collection of isolated reported figures.\n\nFor this assessment, the observations are arranged in their recorded sequence and used to derive an overall trend. An average is also calculated to indicate the prevailing level across the available reporting period, reducing the influence of short-term variation. These calculations support comparison between the general direction of change and the underlying central tendency without attributing significance to any single observation.\n\nThe resulting measures are used to inform interpretation of demand, transfer activity, and related implementation pressures. They are presented as analytical outputs derived from the source, rather than as figures reproduced directly from it. This approach is intended to maintain consistency in the assessment and to provide a concise basis for monitoring changes in the covered area. Limitations remain where reporting frequency or coverage varies, and the calculated measures should therefore be read as summary indicators for implementation review.\n\nMention: \"Quarterly Migration and Remittance System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage derives an overall trend and an average from migration and remittance observations in the Quarterly Migration and Remittance System.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:3f82b1e221c6ae912b496d36", "occurrence_id": "2b04cdddda52daf4007c2c54", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The National Wage Structure Index (NWSI) is used as the indicator for 2020 in assessing the distribution of remuneration across established wage grades. It provides a standardized reference for reviewing the internal structure of pay and for interpreting differences between lower and higher positions. The source is cited in support of the broader finding that the wage system remains differentiated rather than uniform.\n\nTwo values are reproduced directly from the indicator, namely 31 grades and a central value of 8,450 units. These figures are presented as reported, without recalculation or adjustment, and serve to anchor the discussion of the scale and segmentation of the wage structure. Their use is limited to documenting the observed parameters underlying the assessment.\n\nThe cited values do not, on their own, establish changes over time or explain the causes of variation between grades. They nevertheless provide a consistent numerical basis for the implementation review and help distinguish the reported structure from assumptions introduced during analysis. The National Wage Structure Index (NWSI) therefore functions as a supporting statistical source for the paragraph’s assessment of wage differentiation.\n\nMention: \"National Wage Structure Index (NWSI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the 2020 NWSI indicator and takes the figures 31 grades and 8,450 units directly from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:1c9f6974c8f697b19089763e", "occurrence_id": "1631479e1190aae85bf4fe98", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Implementation monitoring indicates that service delivery should be assessed alongside the practical demands placed on households. The review therefore considers whether operational arrangements are compatible with the time required for access, participation, and follow-up. This consideration is treated as a supporting dimension of performance rather than as a separate outcome measure.\n\nThe Nationwide Time Use Survey produced by SEP is cited as a survey supporting the contextual observation that the allocation of time can affect the interpretation of service use. The survey is used to situate this observation within the broader assessment, and no values from it are incorporated into the analysis or used to establish a trend. Its role is limited to supporting this single contextual claim within the wider discussion of implementation conditions.\n\nThe assessment otherwise relies on programme records, monitoring observations, and reported implementation experience. These materials are reviewed for consistency in coverage, timeliness, and responsiveness. Findings are presented in qualitative terms where the available evidence does not support a reliable quantitative comparison. This approach is intended to avoid attributing measured changes to factors that were not directly examined, while retaining a clear account of the operational issues relevant to implementation. The resulting interpretation should be read as an assessment of delivery conditions and analytical limitations, rather than as an estimate of changes in time allocation.\n\nMention: \"Nationwide Time Use Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Nationwide Time Use Survey produced by SEP is used only to support a contextual statement about the relevance of time allocation, with none of its values used in the analysis.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:bdc36637c2d4a9c98dd2fc10", "occurrence_id": "dc8601319b9394cec516ab7b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Collective bargaining coverage is treated as a relevant dimension of labour-market performance in the assessment. The Nationwide Collective Bargaining Index (NCBI) indicator for 2021 provides the cited basis for the comparison. It is used as a descriptive benchmark rather than as a substitute for the broader set of implementation measures.\n\nThe reported distribution shows substantial variation in the reach of negotiated arrangements. The assessment reproduces two figures reported directly by this indicator, namely a coverage rate of 62.4 percent in one observation and 47.8 percent in another. The figures are presented without adjustment and are retained in their reported form for purposes of comparability.\n\nTaken together, the observations support the finding that collective bargaining coverage is uneven, with a material gap between the higher and lower recorded rates. This distinction is relevant to implementation because aggregate progress may conceal differences in access to negotiated protections. The evidence is therefore used to qualify the assessment of institutional reach and to indicate that improvements should be monitored across the distribution rather than through a single average. No additional inference is drawn from the indicator beyond this bounded comparison.\n\nMention: \"Nationwide Collective Bargaining Index (NCBI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the 2021 NCBI indicator and reports 62.4 percent and 47.8 percent to support a claim about uneven collective bargaining coverage.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:cfa1794feeec3699f7137d83", "occurrence_id": "5b653f341b7093a5e6820e1c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Provincial Poverty and Vulnerability Collection is identified as a CIA-produced other source relevant to the general framing of household exposure and welfare conditions. It is recorded in the evidence base to indicate the availability of related material and to support a broad understanding of the subject addressed by this report.\n\nThe reference is used for orientation only. No counts, rates, shares, classifications, or individual records from the collection are incorporated into the assessment, and no conclusion is derived from its contents. The collection therefore does not affect the estimates, comparisons, or prioritisation presented in the main text.\n\nThis limited treatment is consistent with the evidentiary requirements of the exercise. The analysis relies on the information specified elsewhere in the report, while the cited material is retained as a background reference for readers reviewing the basis of the discussion. Its inclusion should not be interpreted as a source of measured results or as support for any particular finding.\n\nMention: \"Provincial Poverty and Vulnerability Collection\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The source contributes background orientation only, and no values or analytical results are taken from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:a3472350f44c946c1bc602ee", "occurrence_id": "d466ab9a77c66c8e7673fb0e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The discussion of farm conditions draws on the Triennial Farm Economy System, identified here as an administrative source. The reference is retained to indicate the type of evidence available for the sector and to support a single contextual statement concerning the existence of an established reporting basis. No value, estimate, category, or observation from that system is reproduced or entered into the analysis. Its role is therefore limited to framing the discussion rather than informing a calculation.\n\nThe assessment proceeds primarily through the review of documented findings, operational records, and measures presented elsewhere in the report. These materials are considered in relation to implementation progress, observed constraints, and the consistency of reported results. The cited system does not determine the measurement framework, supply a benchmark, or establish a denominator for any indicator discussed below. Its mention serves only to situate the narrative within a wider information environment.\n\nThis treatment is consistent with the limited evidentiary function assigned to the source. The report does not infer change, compare periods, or derive classifications from its contents. Any conclusions regarding performance are based on the evidence explicitly examined in the relevant sections and should not be interpreted as estimates generated from the cited system. The distinction is maintained to avoid attributing analytical weight to material that is included solely to substantiate the surrounding narrative.\n\nMention: \"Triennial Farm Economy System\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The cited administrative source provides background for one contextual claim, while none of its values enter the analysis.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:5cd4a3a6485536aa4420aff0", "occurrence_id": "6d1a348e360459eae58cb77e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses a consolidated description of operating conditions and reporting arrangements. Attention is given to the consistency of terminology, the scope of recorded activities, and the limitations affecting interpretation of administrative information. These considerations support a cautious reading of the findings and do not constitute an independent estimate of supply performance.\n\nThe Rolling Supply Chain Registry is a database covering Switzerland for 2023. It is mentioned as a background reference to indicate the existence of a relevant administrative record. No entries, counts, rates, classifications, or other values from the registry are incorporated into the assessment, and no inference is drawn from its contents.\n\nThe analysis relies on the evidence presented in the project records and on the procedures described for review, validation, and consolidation. Where information is incomplete or not directly comparable, the results are treated as indicative rather than definitive. This approach preserves a clear distinction between documented observations and contextual material that is not used in the analytical calculations.\n\nAccordingly, the cited registry has no role in determining reported totals, differences, or conclusions. Its inclusion serves only to identify a potentially relevant source for readers reviewing the information environment surrounding the assessment.\n\nMention: \"Rolling Supply Chain Registry\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The source supplies background reference only, and no values are drawn from it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:deca3143304776840b24e3bf", "occurrence_id": "2d99bf006eadab2936b8b4d6", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: **Data sources.** The Representative Cash Transfer Registry Survey (RCTRS) is cited as a survey relating to 2015. It is retained in the supporting evidence to indicate the background against which registry-based measures may be read, while no observations, totals, rates, or other values from the survey enter the assessment.\n\nThe reference provides a limited point of orientation for the discussion of administrative information and beneficiary records. It does not provide the basis for an estimate, a comparison, a forecast, or a conclusion presented in this report. Its inclusion therefore has no effect on the reported calculations or on the interpretation of results.\n\nThe assessment relies on the evidence described in the surrounding sections and treats the RCTRS reference as supplementary documentation only. No numerical result is attributed to the Representative Cash Transfer Registry Survey, and no finding is derived from its contents. The citation is maintained to acknowledge relevant background material without extending the analytical scope of the report.\n\nMention: \"Representative Cash Transfer Registry Survey (RCTRS)\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The RCTRS is used only for background, with no values or analytical results taken from the 2015 survey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:6bd206b5fecdcd7f68ee87b9", "occurrence_id": "d7e75296b173804bc7a7e013", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Follow-Up Flood Extent Index (FFEI) indicator as a supporting data source for the interpretation of reported flood exposure. The source is treated as a direct reference for the quantitative observations presented below, rather than as a basis for additional estimation or recalculation.\n\nTwo figures are taken directly from the indicator, namely 62 percent and 14.7 square kilometres. These values are reproduced in the assessment to support the finding that the recorded extent was material and required targeted attention in implementation planning. No adjustment is applied to either figure, and the units are retained as reported by the source.\n\nThe figures are used together because they describe complementary dimensions of the same operational concern. The percentage provides a measure of relative incidence, while the area provides a corresponding measure of physical extent. Read jointly, they support a cautious interpretation of the scale of exposure and provide a consistent basis for the narrative assessment.\n\nThe cited values should therefore be understood as reported observations from the Follow-Up Flood Extent Index (FFEI) indicator. Their use is limited to supporting the stated claim, with no inference beyond the information conveyed by the two reproduced figures.\n\nMention: \"Follow-Up Flood Extent Index (FFEI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage takes the figures 62 percent and 14.7 square kilometres directly from the Follow-Up Flood Extent Index (FFEI) indicator to support the assessment of flood exposure.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:183f595d7b8bdbec91546004", "occurrence_id": "3eb89ce8f9c643904ee7d1b6", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The population evidence used in this assessment is the Baseline Population Grid Census, a census produced by Geostat for Sindh and referring to 2023. It provides a gridded basis for reviewing the distribution of residents and households at a level suitable for operational planning. The source is treated as the reference point for the numerical statements presented here, without adjustment or extrapolation. It reports 2,450,000 residents and 1,180,000 households. Both figures are reproduced directly from the source.\n\nTaken together, these counts support the broader assessment that population coverage is sufficiently extensive to justify structured service planning and the prioritisation of locations with higher concentrations of residents. The figures are used to substantiate this conclusion rather than to establish a change over time. They provide a common numerical basis for interpreting the scale of the planning requirement and for comparing the relative weight of household and resident coverage.\n\nThe evidence should nevertheless be read as a population reference for implementation planning. It does not, on its own, establish service quality, access conditions, or the distribution of needs among households. Further operational decisions therefore require review against the relevant planning criteria. Within that limitation, the cited counts provide a clear and reproducible basis for the assessment's central population-related claim.\n\nMention: \"Baseline Population Grid Census\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Baseline Population Grid Census and directly reproduces figures of 2,450,000 residents and 1,180,000 households to support a service-planning claim.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:26c7da19abd83d4058ce9dda", "occurrence_id": "e4ca6673644c13f4b0cdc207", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Harmonised Ward Outline Dataset is used as the geospatial reference for Lagos State for 2018. It provides a consistent spatial basis for interpreting the distribution of administrative areas and for checking whether reported programme coverage corresponds to the delineated units. The source is treated as a reference layer rather than as an analytical estimate.\n\nFor this assessment, two counts are reproduced directly from the dataset without recalculation. The record identifies 245 ward outlines and 20 local government areas. These figures support the broader finding that the reporting frame provides comprehensive territorial coverage for the assessment area, while avoiding any inference about changes over time.\n\nThe spatial reference was applied to verify the assignment of reported activities to the relevant administrative units. Where a reported activity could not be matched to a delineated unit, the case was retained for review rather than reassigned. This procedure limits the risk that apparent gaps reflect inconsistent boundary interpretation rather than an actual absence of implementation.\n\nThe resulting checks are intended to strengthen the internal consistency of the monitoring record. They do not constitute a population estimate, a service-access measure, or an assessment of boundary quality beyond the documented matching exercise. The two reported counts are therefore used as descriptive benchmarks for coverage and administrative reconciliation.\n\nMention: \"Harmonised Ward Outline Dataset\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Harmonised Ward Outline Dataset to reproduce the directly reported counts of 245 ward outlines and 20 local government areas for Lagos State in 2018.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:929a039602d60d8f652298de", "occurrence_id": "b7b77f09152f77c4b87c7dd9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis uses the District Multidimensional Poverty Survey (DMPS), a survey that provides the principal empirical basis for the assessment. The records are treated as comparable observations for the purpose of examining variation in multidimensional poverty and related welfare conditions. No additional institutional attribution, geographic coverage, or temporal designation is assigned in this note.\n\nThe assessment derives summary measures from the DMPS rather than reproducing isolated reported figures. It calculates average levels across the available observations and uses those calculated values to identify the direction and relative stability of the observed trend. This approach reduces the influence of short-term irregularities and provides a more consistent basis for comparison.\n\nThe resulting averages are used in the interpretation of distributional change and in the identification of areas requiring further analytical attention. The findings should therefore be read as estimates generated from the survey records, not as a transcription of individual entries. Any difference between a calculated average and a reported observation reflects the aggregation procedure applied for this assessment.\n\nMention: \"District Multidimensional Poverty Survey (DMPS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the District Multidimensional Poverty Survey (DMPS) to calculate averages and identify a trend from its observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:8186cfc0a2a03a2d3b38b5f4", "occurrence_id": "160511dc49c9753dc09ca92b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Transport connectivity is assessed using the High-Frequency Transport Network Dataset (HTND), a geospatial source covering observations for 2022. The dataset supplies repeated measures of network conditions at regular observation points and permits comparison across the reporting period. Its use is limited to the construction of summary indicators for this assessment; no individual record is treated as a standalone estimate of the overall situation.\n\nThe analysis aggregates the available observations into period averages and calculates the change between successive periods. The resulting trend is based on the arithmetic movement of the aggregated measure, rather than on a quotation of isolated values from HTND. This procedure reduces the influence of short-term fluctuations and provides a consistent basis for judging whether network conditions improved, remained stable, or deteriorated during the period under review. All calculations use the same inclusion rules and retain observations with sufficient continuity for comparison.\n\nResults derived from HTND are presented as analytical estimates and should therefore be interpreted as measures constructed from the source, not as figures reproduced verbatim. The average level is used to establish the central tendency, while the calculated trend indicates the direction and pace of change. These outputs support monitoring of implementation progress and provide a reproducible basis for subsequent review. Any difference between a source observation and a reported summary reflects aggregation or calculation, rather than a correction to the underlying dataset.\n\nMention: \"High-Frequency Transport Network Dataset (HTND)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage derives period averages and a trend for 2022 from the HTND geospatial dataset.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:5b6a0e86f052746f88b3a43f", "occurrence_id": "f2daa367cf8b18ceb3ab9254", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Midline Social Protection Survey is cited as the principal household survey source covering Spain in 2024. Its reported distribution of access to social protection is used to assess the plausibility of the estimate presented in the accompanying analysis. The survey provides an independent reference point for interpreting observed differences in coverage across households and for identifying whether the estimated level is within a reasonable range.\n\nThe comparison is limited to the indicators required for the assessment and does not imply that the two measures are identical. Differences in definitions, reporting conditions, and population coverage may account for part of the observed variation. Nevertheless, the direction and approximate magnitude of the survey results support the central estimate used in this section.\n\nOn this basis, the Midline Social Protection Survey is treated as a cross-check rather than as the underlying source of the estimate. The evidence is therefore used to substantiate one claim concerning the scale of coverage, while the broader interpretation continues to rely on the separate source from which the estimate was drawn. Any discrepancy is recorded as a qualification to the finding rather than as evidence of a change in the underlying pattern.\n\nMention: \"Midline Social Protection Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Spain 2024 survey is used to cross-check an independently derived estimate of social protection coverage.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:6792873d816beb0e72ab13de", "occurrence_id": "4afcce9da6f281484e2ab470", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment reviews the status of alert registration and verification within the project’s monitoring framework. The analysis focuses on the consistency of recorded events, the completeness of supporting documentation, and the extent to which registered alerts can be used for timely operational decisions. These considerations are relevant to the interpretation of reported outputs and to the prioritisation of follow-up actions.\n\nTwo figures are reproduced directly from the Record-Based Early Warning Registry database, namely 31 registered alerts and 18 verified alerts. The figures are retained as reported and are not supplemented by estimates or recalculated indicators. Their inclusion supports the assessment that registration is progressing more rapidly than verification, leaving a material requirement for document review and closure of outstanding records.\n\nThe result should be read as an implementation finding rather than as a measure of underlying incidence. Differences between registered and verified records may reflect delays in validation, incomplete evidence, or other procedural constraints. Continued reconciliation of the registry with case documentation is therefore warranted before the figures are used for resource allocation or performance comparison. The reported counts nevertheless provide a concise basis for monitoring the movement of records through the alert management process.\n\nMention: \"Record-Based Early Warning Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses 31 registered alerts and 18 verified alerts from the Record-Based Early Warning Registry database to support a finding that verification lags registration.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:d12bfbf34a14b247efcad37a", "occurrence_id": "17f09e5b0d8560fa76a2e35d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The National Socioeconomic Survey (NSS) is treated as the principal survey for the analysis of household welfare conditions. The NSS provides the observations used to construct comparable measures of participation, access, and material security. No individual observation is presented as a reported estimate. Instead, the available records are grouped into consistent analytical intervals and assigned common weights before calculation.\n\nThe resulting series indicates a gradual improvement in average access to basic services, although the pace is uneven across population groups. A separate calculation of mean household consumption shows a modest upward movement, with larger gains concentrated among households initially below the central distribution. The analysis also identifies a narrowing difference in participation between the lower and upper groups, while the remaining gap in service access remains material.\n\nThese results are derived from successive averages and changes between those averages, rather than from isolated figures reproduced from the source. Sensitivity checks using alternative weighting assumptions do not alter the direction of the trend, but they reduce its measured magnitude. The estimates should therefore be interpreted as analytical summaries of the NSS evidence and not as direct quotations from the survey. This treatment supports monitoring of broad welfare movement while retaining caution regarding subgroup comparisons and the precision of individual estimates.\n\nMention: \"National Socioeconomic Survey (NSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage derives weighted averages, welfare trends, and subgroup changes from the NSS rather than quoting individual survey figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:863a0e49d3464475f0322b8d", "occurrence_id": "69f2f4969c586693e66f2d00", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The principal reference for the education coverage assessment is OCHA’s administrative Cross-Sectoral Non-State Education System, which covers Zatari camp and refers to 2021. The source is used as the governing record for interpreting the reported structure and scale of provision within the defined coverage area. Its administrative character is material because the figures represent recorded service information rather than an independently generated survey estimate.\n\nThe assessment reproduces two figures directly from this source. It records “31 grades” and “8,412 learners” as stated in the Cross-Sectoral Non-State Education System. These figures are presented without recalculation or extrapolation and are retained in their reported units to preserve comparability with the underlying administrative record. The first indicates the breadth of grade-level provision, while the second indicates the corresponding number of learners captured by the source.\n\nTaken together, the figures provide a concise baseline for reviewing education provision in Zatari camp during 2021. They support an assessment of coverage and scale, but they should not be interpreted as evidence of attendance, completion, or change over time. The analysis therefore limits its use of the source to the two reported quantities and avoids deriving additional indicators from them. This treatment is consistent with the evidentiary role assigned to the administrative record and maintains a clear distinction between directly reported information and subsequent assessment judgment.\n\nMention: \"Cross-Sectoral Non-State Education System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported figures 31 grades and 8,412 learners from OCHA’s administrative Cross-Sectoral Non-State Education System for Zatari camp in 2021.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:a4b46f3883835c62e9ba0819", "occurrence_id": "b6b13d5b9f29a94ce2311b18", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The implementation discussion treats apprenticeship as a relevant component of workforce preparation and considers the institutional conditions that may affect participation, completion, and transition into employment. The Combined Apprenticeship Survey is a survey produced by IMSS and referring to 2021. It is cited to support the general observation that apprenticeship arrangements form part of the evidence base for assessing implementation conditions.\n\nThe source is used for orientation within the broader discussion rather than for measurement. No counts, rates, distributions, or other values from the survey are incorporated into the assessment, and no indicator, comparison, or estimate is derived from it. The citation therefore provides background support for one claim concerning the relevance of apprenticeship arrangements while leaving the quantitative findings of the report independent of that source.\n\nThis treatment is consistent with the limited evidentiary role assigned to the reference. The surrounding assessment relies on its own documented observations and analytical procedures, while the cited survey serves only to establish that apprenticeship is an appropriate subject for consideration. The distinction is maintained to avoid implying that the source supplies numerical confirmation for the findings presented elsewhere in the report.\n\nMention: \"Combined Apprenticeship Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Combined Apprenticeship Survey is used as contextual support for one claim, with no values taken from the 2021 survey produced by IMSS.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:56110f6840618b793bf96d8f", "occurrence_id": "8e01105582a4bb98119b17d0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Monitoring of household hardship relies on the Recurrent Deprivation and Hardship Index indicator for Mexico in 2020, produced by UNU-WIDER. The measure is used as a descriptive input for interpreting the distribution of material constraints and for identifying the scale of deprivation relevant to implementation planning.\n\nWithin this assessment, the indicator reports 12.4 percent for the first recorded hardship measure and 27.8 percent for the second. These figures are reproduced directly from the cited indicator and support the assessment that deprivation remains material across the observed population. The values are presented without adjustment, reclassification, or extrapolation, so that the reported position can be reviewed against the source.\n\nTaken together, the figures provide a bounded reference for assessing the persistence of hardship in the implementation context. They support a broader finding that deprivation is not confined to a small residual group and that responses should account for differing levels of material constraint. The indicator is therefore used to anchor the discussion, while conclusions remain limited to the evidence represented by the cited values.\n\nMention: \"Recurrent Deprivation and Hardship Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Recurrent Deprivation and Hardship Index to report 12.4 percent and 27.8 percent for Mexico in 2020.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:7b7d6df1cd533dfaabf7f1d4", "occurrence_id": "4dee18f71a12ac72617f187b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Cross-Sectoral Administrative Boundary Index indicator for Azerbaijan in 2019 is used as the primary reference for assessing the consistency of administrative classifications across sectors. The source provides a structured basis for reviewing the degree to which territorial groupings can support comparable reporting and interpretation. Its use in this assessment is limited to the values recorded for the specified coverage and period.\n\nFor the purposes of the assessment, the reported entries are reproduced without adjustment. The index records 31 grades and 14 boundaries. These two figures constitute the central quantitative evidence and are retained in their source form to avoid introducing a separate calculation or an additional interpretation of the underlying classification.\n\nThe first figure indicates the number of grades recorded in the index, while the second identifies the number of boundaries associated with the reviewed classification. Read together, they provide a concise measure of the structure represented in the source and support the comparison of administrative arrangements within the assessment framework. No extrapolation is applied beyond these reported values.\n\nMention: \"Cross-Sectoral Administrative Boundary Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The assessment takes the source-reported values of 31 grades and 14 boundaries from the Cross-Sectoral Administrative Boundary Index.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:19748e5a4115a4e66c7061c0", "occurrence_id": "7ef0726fbc5b113b601f2f49", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Integrated Shock Coping Survey is used as an input to the assessment of short-term household adjustment. It is a 2021 survey providing observations on reported responses to economic pressure and changes in access to basic needs. The material is treated as evidence for a broader assessment of coping capacity rather than as a standalone estimate.\n\nTo support the analysis, the assessment derives average levels across the reported observations and compares their direction over the available period. This calculation reduces the influence of isolated responses and provides a more stable indication of prevailing behaviour. No single observation is presented as representative of the full population, and the source is not reproduced through selected quotations.\n\nThe resulting average supports the finding that coping responses were uneven and that more constrained households relied on a narrower range of adjustments. The evidence is therefore used to substantiate the broader discussion of vulnerability, while the interpretation remains based on the calculated pattern rather than on any one reported value. Differences in response intensity are considered alongside the wider results, with the source serving as a supporting input to the assessment rather than its sole basis.\n\nMention: \"Integrated Shock Coping Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The assessment derives average levels and a directional pattern from the Integrated Shock Coping Survey to support its finding on uneven coping responses.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:f492a6c9acd32118c79f62a4", "occurrence_id": "24dcdac6f4bb4ba9860f9918", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Harmonised Heat and Cold Extremes Dataset (HHCED) is used as the geospatial source for the 2015 reference assessment. It provides a consistent basis for interpreting reported exposure to temperature extremes and for checking whether the observed burden is material for planning purposes. The source is cited in the evidence paragraph rather than treated as a separate methodological result.\n\nTwo values are reproduced directly from the dataset. The record indicates 14.2 days with extreme heat and 3.7 days with extreme cold during the reference period. These figures are presented without recalculation, adjustment, or aggregation, and are retained as reported to preserve traceability between the assessment narrative and the underlying evidence.\n\nTaken together, the observations support the broader finding that exposure to both heat and cold extremes should be incorporated into screening, prioritisation, and the design of resilience measures. The evidence does not establish a causal estimate, but it provides a sufficiently clear indication of material exposure for implementation planning and routine monitoring. Subsequent reviews may use the same source to maintain comparability across reporting cycles.\n\nMention: \"Harmonised Heat and Cold Extremes Dataset (HHCED)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The assessment takes the directly reported values of 14.2 days with extreme heat and 3.7 days with extreme cold from HHCED for its 2015 exposure finding.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:2685a617075d8f066238f885", "occurrence_id": "d4ac4ef6a663c7352c9cb0be", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment considers settlement conditions through a broader review of access, occupancy, and service constraints. Within that discussion, the Metropolitan Settlement Location Survey (MSLS) is cited as a 2016 survey to support the general observation that the location of settlements is relevant to the interpretation of observed service and infrastructure gaps. The reference is therefore used to establish background for one statement in the assessment rather than to introduce a separate analytical series.\n\nNo numerical result, classification, or estimate from the MSLS is incorporated into the calculations reported here. The citation does not alter the measurement framework, the allocation criteria, or the interpretation of the indicators. Its function is limited to confirming that settlement location constitutes a material consideration when comparing conditions across the assessment framework.\n\nThis treatment preserves a clear distinction between evidence used for quantitative findings and material used to frame their interpretation. The cited survey provides supporting context for the stated relationship between location and service conditions, while the findings and conclusions remain based on the measures presented in the assessment. Accordingly, the MSLS reference should not be read as a source for any value reproduced in the accompanying analysis.\n\nMention: \"Metropolitan Settlement Location Survey (MSLS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The passage uses the 2016 MSLS survey only to support the contextual claim that settlement location matters, without incorporating its values into the analysis.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:f0afe7d14c58d8d932508a62", "occurrence_id": "b0239807c0b27c27c408bffb", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Baseline Labour Force Survey (BLFS) is treated as the principal survey for the analysis. Its observations provide the central evidence for assessing changes in labour force conditions, with attention to consistency across the available records. The source is used as a basis for measurement rather than as a descriptive reference alone.\n\nThe analysis derives an average from the observations and uses that result to establish the prevailing level of participation. It also compares the successive values to identify the direction and strength of the underlying trend. These calculations reduce the influence of isolated movements and support a more stable interpretation of the evidence.\n\nThe resulting measures are presented as analytical outputs derived from the BLFS, not as figures reproduced directly from the source. This approach permits the assessment to distinguish broad movement from short-term variation while retaining a clear link to the underlying survey evidence. The calculated average and trend are therefore used throughout the assessment as the principal basis for interpreting labour market developments.\n\nMention: \"Baseline Labour Force Survey (BLFS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The passage uses the BLFS survey to calculate an average and identify a trend from its observations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:d63ba76f7d7ff2cc32cd0570", "occurrence_id": "bbea6b1d0467716d44b50b7b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Quality-Assured Minimum Needs Survey (QMNS), a 2023 survey, is used as the sole reference for the two quantitative observations reported in this assessment. The source is cited in the analytical narrative to anchor the discussion of minimum access conditions and reported service gaps.\n\nThe analysis quotes two figures directly from the survey. It reports that 47 percent of assessed households meet the stated minimum access threshold, while 18 percent remain below that threshold. Both proportions are reproduced as reported, without recalculation, interpolation, or adjustment.\n\nTaken together, these figures support the assessment that access remains uneven and that a substantial minority continues to require targeted support. The evidence also indicates that aggregate coverage should not be interpreted as uniform achievement across the population. This finding is relevant to the prioritisation of follow-up measures and to the monitoring of progress against minimum needs.\n\nThe QMNS evidence is therefore used narrowly, to substantiate the access claim within the broader assessment rather than to establish a time series or infer changes beyond the observations recorded in the survey. The two reported proportions provide a consistent statistical basis for the conclusion presented above.\n\nMention: \"Quality-Assured Minimum Needs Survey (QMNS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage takes the figures 47 percent and 18 percent directly from the Quality-Assured Minimum Needs Survey (QMNS).\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:98d085eecf1407469b540ab6", "occurrence_id": "c63bf1b7f236b035e83539b5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Comparative Transport Network Census (CTNC) is a census produced by the Ministry of Education for 2016. It provides the underlying observations used to assess the recorded structure of the transport network and to support the related finding on the distribution of network links across reporting categories.\n\nFor the assessment, the available CTNC records were collected and transferred into a common working archive. Entries were harmonised to align naming conventions, remove duplicate observations, and preserve the original classification structure. The resulting archive was reviewed for internal consistency before being used in the tabulations presented in this report.\n\nThe source supports the interpretation that observed differences reflect variation in recorded network coverage rather than changes introduced during processing. No additional assumptions were applied to extend the census beyond its stated coverage. The harmonised file was retained as a reference dataset for verification of the reported results and for reconciliation of subsequent analytical outputs.\n\nMention: \"Comparative Transport Network Census (CTNC)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The passage identifies the 2016 census produced by the Ministry of Education and describes the collection, harmonisation, archiving, and use of CTNC records.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:59544d802a6e9460c6f3bece", "occurrence_id": "9023165b167f55d176ec832e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Quarterly Temperature Records Dataset is used as the primary evidentiary source for the temperature assessment. It is a geospatial source produced by CONARE and is treated as an underlying record rather than as a secondary interpretation. The available material is sufficiently structured to support direct reference to individual observations.\n\nTwo temperature figures are reproduced directly from the dataset. One observation records 31 degrees Celsius, while another records 14 degrees Celsius. These values are retained in the form presented by the source, without adjustment, interpolation, or conversion. Their inclusion provides the central quantitative basis for the accompanying assessment and allows the reported conditions to be reviewed against the source record.\n\nThe figures should therefore be read as documented observations rather than as a calculated average or an inferred tendency. The dataset is cited for the values themselves, and the surrounding text limits interpretation to what can be supported by those entries. This approach preserves traceability and avoids introducing estimates that are not contained in the cited material.\n\nMention: \"Quarterly Temperature Records Dataset\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage uses the Quarterly Temperature Records Dataset for its direct observations of 31 degrees Celsius and 14 degrees Celsius.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:ad756f9a60a2e9adca5066e0", "occurrence_id": "8fdbc3a1da127bd4b35807fd", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis uses the Midline Learning and Equity Registry database for Kosovo covering 2020 as a supporting statistical source. The source is consulted for descriptive evidence on learning outcomes and equity patterns, with its contents treated as reported observations rather than as a basis for recalculation or projection.\n\nThe broader assessment finds that learning performance is uneven across learner groups and that the distribution of results warrants attention in implementation planning. Two figures are reproduced directly from the database to support this finding. It reports that 64 percent of assessed learners reached the stated benchmark, while 37 percent reached the benchmark among learners in the lower-performing group. These figures are presented as complementary measures of the same distribution and are not interpreted as evidence of change over time.\n\nThe difference between the two reported shares indicates a material equity gap within the observed population. The result should be read alongside the assessment’s qualitative findings and operational constraints, while retaining the original scope of the registry. No adjustment is applied to either figure, and no additional estimate is derived from the cited source. The evidence therefore supports the conclusion that implementation measures should address both overall attainment and unequal results among learners.\n\nMention: \"Midline Learning and Equity Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage takes the directly reported figures of 64 percent and 37 percent from the Midline Learning and Equity Registry to support a claim about unequal learning outcomes.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:56c7e56d79038eba3cd5ee30", "occurrence_id": "8b8d932a9c76e47878f156e4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The assessment uses the Household Land Cover Change System as an administrative source referring to 2023. The reference establishes that land-cover information is maintained through a formal reporting arrangement relevant to the subject under review. No values from the system are reproduced, recalculated, or used to derive rates, totals, classifications, or comparisons.\n\nThe broader analysis addresses the adequacy of evidence for monitoring changes in household land cover and the consistency of reporting procedures. Within that discussion, the cited system supports the limited claim that an established information basis exists for the topic. This function is contextual rather than evidentiary for the numerical findings presented elsewhere in the assessment.\n\nInterpretation therefore remains dependent on the observations and checks described in the surrounding sections. The Household Land Cover Change System is not treated as a dataset from which results are extracted for this passage, and no additional attributes are assigned to it. Its inclusion records the relevant administrative reference while maintaining a clear separation between background documentation and the evidence used for assessment conclusions.\n\nMention: \"Household Land Cover Change System\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The source provides contextual support for the existence of an administrative information basis and contributes no numerical or categorical values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:942e5d42c00420f4a870e6a1", "occurrence_id": "db0c78f235c7430538ae57a3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis uses the Combined Fuel and Energy Use Index indicator produced by Amadeus as a supporting measure of changes in aggregate energy use. The indicator is treated as a consistent statistical input for comparison across the observations reviewed in this assessment. Its inclusion supports the interpretation of movements in the broader consumption series and provides a basis for assessing the direction and persistence of change.\n\nRather than reproducing individual entries, the analysis derives a period average and calculates a simple trend from the reported index values. This treatment reduces the influence of isolated movements and permits the results to be considered alongside the other measures used in the assessment. The resulting summary is used to support the finding that energy use changed progressively over the observation sequence, while avoiding an inference that each recorded movement represents a structural shift.\n\nThe measure is therefore used as corroborating evidence rather than as a standalone outcome indicator. Differences between the calculated average and individual observations are retained in the review, since dispersion around the central value remains relevant to interpretation. The source supports one claim concerning the direction of change, while the conclusion rests on the combined reading of the index and the accompanying statistical evidence.\n\nMention: \"Combined Fuel and Energy Use Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The analysis calculates an average and trend from the Combined Fuel and Energy Use Index indicator produced by Amadeus to support a claim about the direction of energy-use change.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "val"}
{"origin": "authored", "source_key": "authored:2ae9169bc5e6cf9d8cb06044", "occurrence_id": "265d27b028cd4125603db68b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Decennial Tertiary Education System (DTES) provides an administrative basis for reviewing the reported pattern in tertiary provision. Its records are used to verify the internal consistency of the estimate and to identify possible differences arising from coverage, classification, or reporting practices. The source is treated as a reference point rather than as a substitute for the principal estimate.\n\nThe comparison indicates that the estimate is broadly compatible with the administrative record, although minor discrepancies remain in the distribution of institutions and enrolled learners. These differences do not alter the direction of the finding, but they warrant caution when interpreting small changes across categories or periods. The DTES evidence therefore strengthens the specific claim concerning the scale of tertiary participation within the broader assessment.\n\nThis cross-check is limited to the claim under review and does not imply that all measures in the surrounding analysis are derived from the same source. The administrative information is used to assess plausibility, clarify the scope of the estimate, and support a measured interpretation of the reported results. Any residual divergence is noted as a limitation of comparison rather than treated as evidence of a substantive reversal.\n\nMention: \"Decennial Tertiary Education System (DTES)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The passage uses the Decennial Tertiary Education System (DTES) administrative source to cross-check one independent estimate supporting a broader finding.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:3524dcead0451a916d3c5f40", "occurrence_id": "29730acdd7e5c218ad6b0011", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The analysis draws on the Household Poverty and Vulnerability Survey (HPVS), a survey designed to support monitoring of household welfare and exposure to economic stress. The source provides the central evidence for the assessment of changes in poverty and vulnerability across the observation period.\n\nThe reported results are derived from the survey records through calculation of average rates and comparison of successive observations. This treatment identifies the direction and approximate pace of change while reducing the influence of short-term variation in individual responses. The resulting measures are therefore analytical estimates based on the HPVS rather than figures reproduced directly from a single table.\n\nThe analysis indicates a gradual movement in the combined welfare indicators, with vulnerability remaining material among households near the poverty threshold. Differences between observations are interpreted cautiously, since sampling variation and changes in household circumstances may affect the measured levels. The computed averages nevertheless provide a consistent basis for assessing the underlying trend and for framing implementation priorities.\n\nFor reporting purposes, the derived series are retained as the principal quantitative reference. They support comparison across reporting intervals and permit the assessment to distinguish persistent patterns from isolated movements in the survey results. No additional administrative source is required for the calculation presented here.\n\nMention: \"Household Poverty and Vulnerability Survey (HPVS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The analysis computes average poverty and vulnerability trends from the Household Poverty and Vulnerability Survey (HPVS).\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:3cab25c51e2254fa8adee704", "occurrence_id": "49c42e0b2756877f3dd87eaa", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The evidence base for this assessment is the Metropolitan Fuel and Energy Use Survey (MFEUS), a survey covering Saint Lucia and referring to 2019. The source is retained as the principal reference for the household energy indicators presented in this section, with its reported values reproduced without adjustment.\n\nThe survey reports that 62 percent of households used electricity as their main source of cooking energy. This figure is taken directly from the MFEUS tabulation and is presented as a descriptive measure rather than as an independently derived estimate. The same source records that 18 percent of households relied mainly on liquefied petroleum gas, providing the second figure used in the analysis.\n\nTaken together, these observations indicate a marked difference in the reported prevalence of the two principal household cooking fuels. No interpolation, trend calculation, or harmonisation is applied to the cited values. They are included to establish the distribution recorded by the source and to support comparison with the accompanying discussion of household energy use. The figures should therefore be read as survey results for the stated coverage and reference period, rather than as projections or administrative counts.\n\nMention: \"Metropolitan Fuel and Energy Use Survey (MFEUS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The passage takes the 62 percent electricity figure and the 18 percent liquefied petroleum gas figure directly from the MFEUS survey for Saint Lucia in 2019.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:0f29032659bc5f2e561ee13b", "occurrence_id": "3d553509373bedffd14d7d66", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The cited Metropolitan Heat and Cold Extremes Report is a report produced by UNICEF for the 2015 observation year. It is used as a supporting statistical source for the assessment of exposure to unusually severe temperature conditions. No adjustment, interpolation, or reclassification is applied to the values reproduced from this source.\n\nTwo figures are taken directly from the report and retained in their reported form. The first indicates a maximum heat departure of 4.6 degrees Celsius, while the second records a minimum cold departure of 7.8 degrees Celsius. These values are presented as observations rather than as estimates generated by the assessment.\n\nTaken together, the figures support the broader finding that exposure is uneven across the observations considered and that both heat and cold conditions require attention in implementation planning. Their use is limited to substantiating this claim; they are not combined into a new index, compared across additional periods, or used to infer a trend. The cited report therefore provides a narrow evidentiary basis for the paragraph while preserving the original measurements and their stated units.\n\nMention: \"Metropolitan Heat and Cold Extremes Report\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses UNICEF's 2015 report and reproduces its directly reported figures of 4.6 degrees Celsius and 7.8 degrees Celsius to support a finding on uneven heat and cold exposure.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "holdout"}
{"origin": "authored", "source_key": "authored:f59f4f658032023da917ca93", "occurrence_id": "be9c78247448d2945deb3972", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The cited data source is the Provincial Poverty and Vulnerability Survey (PPVS), a survey produced by NBRM for 2017. It provides the underlying observations used to assess household exposure to economic stress and to establish a consistent basis for interpreting vulnerability across the analytical population.\n\nFor the purposes of the assessment, the available records were collected and reviewed for completeness before being harmonised with the structure used in the analysis. Definitions were reconciled across the retained materials, and the resulting files were archived with sufficient documentation to support subsequent verification and controlled reuse. These procedures reduced the risk that differences in recording practice would be interpreted as substantive differences in household conditions.\n\nThe PPVS evidence supports the finding that vulnerability is associated with the combined effect of constrained resources and limited capacity to absorb shocks. It is therefore used to inform the broader assessment rather than treated as a stand-alone measure of welfare. Results derived from the source are interpreted alongside the stated analytical framework, with attention to coverage, comparability, and the limits imposed by the available observations. The documentation retained for the survey also provides a traceable basis for reviewing the assumptions applied in the assessment and for maintaining continuity between the evidence base and the reported conclusions.\n\nMention: \"Provincial Poverty and Vulnerability Survey (PPVS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The assessment uses the 2017 PPVS survey produced by NBRM and documents its collection, harmonisation, and archiving.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:4e2de79d6abe92b7789bf9a8", "occurrence_id": "628110dba3d1004f7bbccc39", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Multi-Topic Job Search Survey (MJSS) is a survey produced by VIIRS and is used as a supporting source for the assessment. Its responses provide comparable information on job-search activity and related participation patterns. The citation is retained in the evidence register so that the basis for the reported interpretation remains identifiable.\n\nDuring preparation, the available survey material was collected and checked for internal consistency. Responses were harmonised to support comparison across the analytical categories used in the assessment, while the accompanying documentation was archived with the processed material. These procedures provide a traceable basis for the use of the source and reduce the risk that differences in recording practice are interpreted as substantive differences in reported activity.\n\nWithin the broader analysis, the source supports the observation that reported job-search activity is uneven across respondent groups. The MJSS is not used as a stand-alone basis for the overall assessment, but as corroborating evidence alongside the other material reviewed. Its contribution is therefore limited to the interpretation supported by the recorded responses, with the preservation of the source documentation allowing subsequent verification of the analytical treatment.\n\nMention: \"Multi-Topic Job Search Survey (MJSS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The passage records the collection, harmonisation, and archiving of MJSS material and uses it to support a claim about job-search activity.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:c264e4357ad97ca38463ab19", "occurrence_id": "e8e6d26bc66e29d8d2d5e94d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The Civil Primary Care Registry (CPCR) database for 2023 is used as the administrative reference for the service-availability assessment. The record is treated as a source of reported counts and proportions rather than as an independently verified estimate. Interpretation in this section remains limited to the indicators recorded in that database.\n\nDirectly reported entries give 31 facilities with a recorded primary-care function and indicate that 68 percent of facilities met the stated minimum availability criterion. These two figures support the broader finding that coverage is substantial but not uniform across the observed service network. They are reproduced without recalculation, adjustment, or extrapolation.\n\nThe figures anchor the assessment of implementation status, while the surrounding discussion addresses operational implications at aggregate level. No inference is made about facilities not represented in the CPCR database, and no causal conclusion is drawn from the difference between the count and the percentage. The evidence therefore supports a bounded conclusion concerning recorded availability in 2023, with the registry serving as the cited basis for the two numerical statements.\n\nMention: \"Civil Primary Care Registry (CPCR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The passage uses the 2023 CPCR database to report 31 facilities and 68 percent meeting the stated availability criterion.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "openrouter/openai/gpt-5.6-luna", "split": "train"}
{"origin": "authored", "source_key": "authored:b56855ebd515d359560cafe6", "occurrence_id": "8d1520a32b4ff18f2ab90648", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Operations timeliness was recomputed for this review from the Continuous Operations Monitoring Index (COMI), Round 9, which covers Accra in 2020, and the figures below are the review's own rather than the index's.\n\nEach return in the index was dated against its deadline and the share arriving within 10 days worked out: 63 per cent in the first half of the window and 78 per cent in the second. The index prints a composite score for each period, so a timeliness measure of this kind is not something it carries.\n\nThe exercise supports one claim in the review, that reporting improved after the submission portal opened. The remaining chapters rest on the mission's own returns, and nothing else was taken from the index.\n\nReturns without a legible receipt date were left out, 3 per cent of the window, and every share is printed with the number of returns behind it. The dating rule and the rebuilt series are in the annex.\n\nMention: \"Continuous Operations Monitoring Index (COMI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Continuous Operations Monitoring Index provides the submission records from which reporting timeliness was rebuilt.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:06e795cc15055baceec18ac5", "occurrence_id": "f0eaae5903271f629ee3d667", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Service availability was recomputed for this assessment from the Harmonised Service Delivery Index (HSDI), which reports for Burkina Faso in 2018, and the series below is the assessment's construction rather than the index's.\n\nCommunes were grouped by travel time to the nearest facility and the share with any facility within 5 kilometres computed: 31 per cent in the nearest group and 7 per cent in the farthest. The index publishes a single score per commune, so no distance-based split of this kind appears in it.\n\nThat comparison supports one claim, that the programme's first sites should be the farthest communes. Everything else in the assessment rests on the household survey described in the annex, and the index supplied nothing further.\n\nCommunes whose coordinates could not be resolved were excluded, 4 per cent of the total, and each share carries the commune count behind it. The grouping rule is filed in the annex.\n\nMention: \"Harmonised Service Delivery Index (HSDI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Harmonised Service Delivery Index supplies the commune records used to compare facility availability by travel time.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:1ae7f861aa54f93085b4389e", "occurrence_id": "4a4cd0d210efa514252322f8", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Antenatal coverage was worked out for this chapter from the Regional Maternal and Newborn Survey (RMNS), Sixth Round, fielded in Slovakia in 2019, and the estimates below come from the exercise rather than from the survey's tables.\n\nThe team counted the antenatal visits each mother reported and derived the share with four or more: 72 per cent in urban areas against 54 per cent in rural ones. The survey reports the mean number of visits by area, so no threshold share of this kind exists in its outputs.\n\nThe comparison supports one claim, that the outreach visits recommended later in the chapter should begin in the rural districts. The rest of the chapter's health figures come from the ministry's facility returns, and the survey supplied nothing else.\n\nMothers reporting a first visit after the first trimester were kept in the count, and the 6 per cent with an incomplete visit history were left out. Each share carries the number of mothers behind it, and the derivation is in the annex.\n\nMention: \"Regional Maternal and Newborn Survey (RMNS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Regional Maternal and Newborn Survey supports the rural and urban comparison of antenatal visit coverage.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:056310dc0862c0eab6869c36", "occurrence_id": "3822558f506f573a57fe09cd", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Field-level water access for the irrigation component is reported from the Follow-Up Market Access Dataset (FMAD), Third Edition, the geospatial dataset DWASA maintains, and the figures below are given as the dataset's grid gives them.\n\nThe dataset places 38 per cent of the mapped cropland more than 2 kilometres from a water point and shows 12 per cent of the area inside a flood-bounded zone. Both are carried over as published, and the cell size and the distance rule the dataset applies are printed with them.\n\nNo zone was merged to simplify the map and no cell was re-scored to agree with the national drought series. Where the dataset marks a cell as unclassified it stays in the denominator and is shown as unclassified.\n\nBecause the component's siting of storage works rests on the dataset, the annex lists the layers used, the thresholds behind each share, and the area covered by the extract.\n\nMention: \"Follow-Up Market Access Dataset (FMAD)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Follow-Up Market Access Dataset supplies the field-level water access and flood exposure shares behind the siting decisions.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:ab5cd946a182efd152042af1", "occurrence_id": "25e320d2d7c1ef1f43703bb4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Client throughput was rebuilt for this review from the Regional Service Register System (RSRS), the administrative system NSS maintains for 2020, and what follows is the review's arithmetic on those records.\n\nEach registration was followed from first contact to closure and the elapsed time converted to days, giving a median of 14 days in the first quarter and 22 days in the last. The system reports intake and closure counts, so no elapsed-time figure exists in what it publishes.\n\nSplitting the same records by case type placed the median at 8 days for the simplest category and 41 days for the most complex, the spread the review treats as the reason the queue grows despite stable intake.\n\nRecords with a closure date before their intake date were dropped as transcription errors, 2 per cent of the year, and every median is printed with the case count behind it. The tracking rule is in the annex.\n\nMention: \"Regional Service Register System (RSRS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Regional Service Register System supplies the registration records from which case processing times were rebuilt.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:31e65acecb3114ec377d4484", "occurrence_id": "1fafa577caa3748f60947229", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The post-harvest loss figure in this chapter is the programme's own calculation, and the Municipal Storage and Loss Registry (MSLR), compiled by JP Morgan for 2023, is cited to test it.\n\nThe programme puts storage loss at 13 per cent of marketed grain. The registry's tables give 16 per cent for the same commodity group and the same period. Programme staff built the first figure from warehouse returns while the registry built its own from licensed store submissions, and no records are shared between the two routes.\n\nThe distance between 13 and 16 sits inside the margin the registry publishes, so the programme's figure stands and the registry is treated as corroborating it rather than replacing it. No registry value enters any other table here.\n\nThe annex records how each source defines a marketed tonne and which stores each counts, since the comparison holds only where those definitions align, and periods that could not be aligned are omitted.\n\nMention: \"Municipal Storage and Loss Registry (MSLR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Municipal Storage and Loss Registry corroborates the programme's storage-loss figure, which is therefore reported unchanged.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:8f8f4b9f31bfa3b4255aceff", "occurrence_id": "b9e065183221830d7b1cd34a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Deprivation was measured for this assessment from the Representative Multidimensional Poverty Survey (RMPS), which INFRAS/IWW fielded in Fiji, and the figures below come from that work rather than from any table the survey prints.\n\nThe team counted the deprivation items each household lacked and derived the share lacking three or more: 26 per cent of households on the outer islands against 11 per cent on the main island. The survey's tabulations give the items one at a time, so a combined count of this kind is not available from it.\n\nHouseholds were then separated by whether their main income came from fishing, and the share lacking three or more items is 31 per cent in that group against 17 per cent elsewhere, which is the gap this assessment treats as the reason the programme is proposed for the outer islands.\n\nHouseholds with an incomplete item list were excluded, 5 per cent of the sample, and each share carries the margin around it. The item count and the resulting series are recorded in the annex.\n\nMention: \"Representative Multidimensional Poverty Survey (RMPS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Representative Multidimensional Poverty Survey provides the records used to derive combined deprivation by island and livelihood.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:fbec70b864a529549bc43365", "occurrence_id": "5fbb60386015a03364dc80ae", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The records task in this component concerns the District Drought and Resilience Collection (DDRC), Edition 7, the collection Geostat maintains, and what is described is the handling of its entries rather than anything drawn from them.\n\nEntries arrive from district offices on different schedules and use different codes for the same crop, so the task is to bring them under one identifier: codes are mapped to the standard classification, entries recorded under two varieties of the same crop are joined, and each decision is written to a log kept with the file.\n\nNo coverage figure and no seasonality figure was produced during this work. The dry-season shares quoted elsewhere in the component come from the mission's own returns rather than from this exercise.\n\nThe merged entries, the identifier crosswalk and the review log are deposited together, entered in the national catalogue and placed under annual refresh, so that a later update begins from one coded collection instead of parallel district submissions.\n\nMention: \"District Drought and Resilience Collection (DDRC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The District Drought and Resilience Collection is the collection whose crop entries are being coded, joined and catalogued.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:b51fd47c8749126153a98535", "occurrence_id": "71532e1fcbf8753ab3079c24", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Working arrangements in informal trade were analysed for this assessment from the Audited Informal Trade Survey, which FISE fielded across Eastern Europe in 2015, and the figures below are this assessment's calculations rather than published values.\n\nTraders were grouped by whether they held any written agreement and the share working more than six days a week computed: 48 per cent among those without an agreement against 27 per cent among those with one. The survey reports hours by trade type, so the split by agreement does not exist in its tables.\n\nWeekly earnings were then compared between the two groups, and traders without an agreement take 22 per cent less for the same number of days worked, the difference this assessment treats as the cost of the missing paper.\n\nTraders reporting no sales in the reference week were excluded, 6 per cent of the sample, and each share is printed with the number of traders behind it. The grouping rules are in the annex.\n\nMention: \"Audited Informal Trade Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Audited Informal Trade Survey provides the trader records used to compare working days and earnings by written agreement.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:4d24b4361dfad04e7b4edd57", "occurrence_id": "c4d4c9f7c2053f4e24559f39", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Ward coverage was measured for this assessment from the Coordinated Ward Outline Dataset (CWOD), Third Collection, the geospatial dataset compiled for Panama in 2016, and the figures below were computed from its geometry rather than read from it.\n\nEach service point was matched to the ward polygon containing it and the share of points falling outside every ward computed: 9 per cent overall, rising to 17 per cent in the coastal wards. The dataset holds the outlines and nothing about service points, so no such share exists in it.\n\nWards were then ordered by area and the share of their households more than 5 kilometres from a matched point compared between the largest and smallest, 24 per cent against 11 per cent, which is the distance problem this assessment treats as the reason the coastal wards are proposed first.\n\nWards whose polygons failed the topology check were excluded, 3 per cent of the total, and each share carries the household count behind it. The matching rule and the resulting table are in the annex.\n\nMention: \"Coordinated Ward Outline Dataset (CWOD)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Coordinated Ward Outline Dataset provides the ward geometry used to measure service-point coverage and distance.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:0feb35384b62574026ef6769", "occurrence_id": "8aa2e3643ceccf7a00a49f58", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Household energy costs were analysed for this assessment from the Representative Fuel and Energy Use Survey (RFEUS), Phase 4, fielded in Malaysia in 2016, and the figures below are this assessment's calculations rather than published values.\n\nHouseholds were ranked by total spending and the share of spending absorbed by fuel and electricity computed: 17 per cent in the bottom quintile against 4 per cent in the top. The survey's tables report monthly expenditure by category, so no share of this kind appears in them.\n\nSpending was then held constant and the burden still differs by 9 percentage points, which is the part this assessment attributes to the smaller volume purchases and higher unit prices that low-income households face.\n\nHouseholds reporting no energy spending in the reference month were excluded, 4 per cent of the sample, and each share is printed with the number of households behind it. The ranking rule and the resulting series are in the annex.\n\nMention: \"Representative Fuel and Energy Use Survey (RFEUS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Representative Fuel and Energy Use Survey provides the records used to derive the energy burden by spending quintile.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:93f7cea8e4466d984984488a", "occurrence_id": "56ad6df3f15190e6435a93bf", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The registration figure quoted in this chapter is the office's own estimate, and the National Administrative Survey Index (NASI), Fourth Collection, produced by MEF, is cited to test it.\n\nThe office puts adult registration at 79 per cent for its reporting period. The index gives 77 per cent for the same period on its own construction. The office reached its figure from registration counters while the index reached its own from household submissions, and no records pass between the two.\n\nThe distance between 79 and 77 falls inside the tolerance agreed for the indicator, so the office's estimate is printed as calculated and the index supplies no adjustment and fills no gap in the office's series.\n\nThe annex sets out how each source defines a registered adult and which populations each counts, together with the periods that could not be aligned on that basis and were therefore left out of the comparison.\n\nMention: \"National Administrative Survey Index (NASI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The National Administrative Survey Index tests the unit's registration figure, which the agreement leaves unaltered.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:42137948adf64e9ae99f55a9", "occurrence_id": "8dafc39cca5b6354f1173751", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Reporting completeness was rebuilt for this review from the Provincial Programme Monitoring Index (PPMI), which covers Jordan in 2017, and the series below is the review's own rather than anything the index publishes.\n\nEach facility return in the index was checked for the fields its form requires and the share with every required field filled computed: 68 per cent in the first two quarters of the year and 81 per cent in the last two. The index prints a composite completeness score per province, so a field-level measure of this kind is not available from it.\n\nThe same returns were then split by whether the submitting facility had more than one site, and completeness runs 12 percentage points lower among those, which is the weakness the review puts its follow-up training against.\n\nReturns that arrived blank were excluded rather than counted as incomplete, 4 per cent of the year, and each share carries the number of returns behind it. The checking rule is in the annex.\n\nMention: \"Provincial Programme Monitoring Index (PPMI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Provincial Programme Monitoring Index provides the return records from which reporting completeness was rebuilt.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:1d99271c72f6e25d2d49497d", "occurrence_id": "0be4352a7af2173d3a8015d4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Bargaining coverage was examined for this assessment from the Panel Collective Bargaining Survey (PCBS), which covers Albania in 2021, and the figures below were derived from its records rather than read from its tables.\n\nWorkplaces were grouped by whether they provided any funded training and the share covered by a collective agreement computed: 62 per cent among those that did against 23 per cent among those that did not. The survey reports coverage and training provision in separate tabulations, so the crossing of the two does not exist in its outputs.\n\nEstablishment size was then held constant and the difference narrows to 21 percentage points, which is the part this assessment attributes to bargaining itself rather than to the size of the workplace.\n\nWorkplaces with no staff count recorded were excluded, 5 per cent of the panel, and each share is printed with the number of workplaces behind it. The grouping rule and the resulting table are in the annex.\n\nMention: \"Panel Collective Bargaining Survey (PCBS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Panel Collective Bargaining Survey supplies the workplace records used to compare coverage by training provision.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:c4e192e8df7c1e791c3b0dd7", "occurrence_id": "13ddf0f4ebfbffeadd016bf3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Transfer receipt was measured for this assessment from the Government Social Protection Survey, which PRIO/Uppsala fielded across Africa and South Asia in 2016, and the estimates below come from that work rather than from the survey's publications.\n\nHouseholds were grouped by the area they cultivate and the share receiving any social transfer computed: 38 per cent among those farming less than half a hectare against 15 per cent among those farming more than two. The survey tabulates transfer receipt by region, so no split by farm size appears in it.\n\nHouseholds with a member absent for work were then compared separately, and receipt among that group is 9 percentage points lower than among households with no absent member, a gap the assessment treats as the reason the smaller farms are proposed for first registration.\n\nHouseholds reporting no land at all were kept out of the grouping rather than placed in the smallest class, which removes 11 per cent of the sample. The grouping rule is stored in the annex.\n\nMention: \"Government Social Protection Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Government Social Protection Survey provides the household records used to compare transfer receipt by farm size.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:d023db54a8089548622966c5", "occurrence_id": "e296cc26bc52dc4c0b5765cd", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Transport connectivity in this section is reported from the Combined Transport Network Registry, the register Fitch keeps for Sri Lanka as of 2019, and the figures are given as the register's tables give them.\n\nThe register counts 14 600 kilometres of classified road and 2 310 structures it classes as bridges, with 7 per cent of those structures carrying a load restriction. Nothing was recalculated here, and the register's definitions of a classified road and of a restricted structure are printed beside the figures.\n\nNo route was moved between classes to improve the reported totals, and no restricted structure was dropped from the denominator to lower the restricted share. Where the register marks a segment as under survey, it is shown as under survey rather than assumed passable.\n\nBecause the section's conclusions about which corridors need reopening first rest on these counts, the annex records the extract date, the tables used, with the segments excluded from them alongside.\n\nMention: \"Combined Transport Network Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Combined Transport Network Registry supplies the route and bridge counts behind the connectivity figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:b79b1b10434a4d817eb2f905", "occurrence_id": "d744313c15f07963d865233d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Enrolment and dropout were rebuilt for this review from the Multi-Topic Programme Monitoring System (MPMS), the administrative system maintained for India in 2022, and the figures below are the review's arithmetic on those records.\n\nEach beneficiary was followed from enrolment to either renewal or exit, and the share exiting within six months came out at 19 per cent, with 27 per cent among those enrolled at a walk-in point and 12 per cent among those enrolled through a field visit. The system reports monthly enrolment and exit counts, so a within-cohort measure of this kind does not exist in its outputs.\n\nExits concentrated in the second month after enrolment, which is where the review places the follow-up contact it recommends.\n\nBeneficiaries whose exit date could not be read were censored at their last recorded month rather than counted as exits, and every share carries the beneficiary count behind it. The cohort rule is in the annex.\n\nMention: \"Multi-Topic Programme Monitoring System (MPMS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Multi-Topic Programme Monitoring System provides the enrolment records from which dropout timing was rebuilt.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:18f0b07b64420efe3a371b51", "occurrence_id": "e15bfd46e4827ff4a08c45a1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Flood exposure and access to care were measured for this assessment from the Urban Flood Extent Dataset (UFED), Second Collection, the geospatial dataset Bloomberg maintains, and the figures below come from that measurement rather than from anything the dataset reports.\n\nEach residence was checked against the modelled extent and the share standing in the mapped area computed: 22 per cent citywide, reaching 41 per cent along the eastern channel. The dataset holds the extent alone, so no share of residences appears in it.\n\nThe same residences were then matched to the nearest clinic and the share more than 1 kilometre from one compared between flooded and unflooded areas, 18 per cent against 7 per cent, which is the access problem this assessment treats as the health consequence of the extent.\n\nResidences whose address could not be geocoded were excluded, 6 per cent of the total, and each share carries the residence count behind it. The matching rule and the resulting table are in the annex.\n\nMention: \"Urban Flood Extent Dataset (UFED)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Urban Flood Extent Dataset provides the flooded-area geometry used to relate exposure to clinic distance.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:10d79dec5d968b03efdf3eeb", "occurrence_id": "db3f8e874fac423015d0be5b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The share of household spending taken by food in this chapter is the unit's own calculation, and the Follow-Up Household Consumption Survey (FHCS), Seventh Collection, compiled by Angus Maddison, is cited to check it.\n\nThe unit puts the food share at 46 per cent for its reporting period. The survey's tables give 49 per cent for the same period on their own construction. The unit built its figure from price series and administrative returns while the survey built its own from household diaries, and the two share no inputs.\n\nThe distance between 46 and 49 sits inside the tolerance agreed for the indicator, so the unit's figure is reported as calculated, and the survey neither adjusts it nor supplies values for the periods the unit's series lacks.\n\nThe annex sets out the spending items each source counts as food and the household types each includes, since the comparison holds only where those definitions align, and periods that could not be aligned are omitted.\n\nMention: \"Follow-Up Household Consumption Survey (FHCS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Follow-Up Household Consumption Survey checks the unit's expenditure-share figure, which the agreement keeps.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:6cc4156bea64831863f3fc33", "occurrence_id": "1545712c3873863ab2890e1a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Employment among people with a disability was measured for this assessment from the Municipal Disability and Support Survey (MDSS), which USAID fielded in Kyrgyzstan, and the figures below are the product of that measurement rather than of the survey's tables.\n\nThe team separated respondents reporting a long-standing difficulty from those who did not and computed the share currently working: 34 per cent against 61 per cent. The survey reports employment for the sample as a whole, so nothing like that contrast exists in its own outputs.\n\nAmong respondents reporting a difficulty, the share working rises to 47 per cent where a support payment is received and stands at 26 per cent where none is, the difference this assessment treats as the reason the payment is recommended.\n\nRespondents who declined the difficulty questions were excluded, 4 per cent of the sample, and each share carries the number of respondents behind it. The definitions used are listed in the annex.\n\nMention: \"Municipal Disability and Support Survey (MDSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Municipal Disability and Support Survey provides the records used to compare employment by support receipt.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:303efd8570e1bd002aa2ca3c", "occurrence_id": "56d6257a3a0ab67efeffb829", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The enterprise count in this section is the council's own, taken from its licensing ledger, and the Local Business Demography System, maintained by Michelin for Riga in 2020, is cited to test it.\n\nThe council counts 12 400 active enterprises for the year. The system's tables give 11 700 for the same year on its own counting. The council's ledger records every licence issued, while the system counts only enterprises that filed a return, and no list is shared between them.\n\nThe distance between 12 400 and 11 700 is 6 per cent of the council's figure, inside the tolerance set for the indicator, so the ledger count is printed as it stands and the system is not used to restate it or to extend it to other years.\n\nThe annex records what each source treats as active and which sectors each covers, since the two populations are not identical, and the comparison is confined to the enterprises they share.\n\nMention: \"Local Business Demography System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Local Business Demography System checks the council's enterprise-count figure, which the agreement leaves as calculated.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:3d01614aed24ed6c8c298f2a", "occurrence_id": "b5eab86c3ad62994d6653f29", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The ownership analysis in this chapter is built from the land office's register, and the Quality-Assured Asset Ownership Survey, which the Latinobarómetro Corporation fielded in Sud and Nippes in 2024, is quoted only to place those figures in context.\n\nThe survey reports that 41 per cent of households hold a document naming the owner and that 12 per cent of those documents were registered in the last five years. Both are given here as the survey prints them, with its own wording for what counts as a document.\n\nNothing was derived from the survey, and Its responses do not appear in any chapter table. Where the register and the survey appear to describe the same quantity differently, the difference is noted rather than reconciled, since the register covers all parcels and the survey only its sample.\n\nThe annex gives the survey questions behind the two quoted figures, the date of extraction and the sample it covers.\n\nMention: \"Quality-Assured Asset Ownership Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Quality-Assured Asset Ownership Survey is quoted for context while the chapter's own analysis rests on other records.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:9a6abeb0df56494e913b7071", "occurrence_id": "2815d259dacf7207f93903fa", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Service coverage in this section is reported from the Experimental Service Register Index, the indicator series Dealogic compiled for Lebanon in 2016, and the figures appear as the index publishes them.\n\nThe index puts registration in the service population at 74 per cent and the response rate among those contacted at 61 per cent, with the lower response concentrated among the youngest age group. Both are carried over unchanged, together with the definitions the index uses for registration and for a completed response.\n\nNo age group was removed from the denominator to raise the reported response, and no registration was credited on the basis of a contact that produced no answer. Where the index marks a district as partially enumerated, the mark travels with the figure.\n\nBecause the section's conclusions about which districts need outreach first rest on this index, the annex lists the tables used, the fields behind each figure and, alongside it, the date of extraction.\n\nMention: \"Experimental Service Register Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Experimental Service Register Index supplies the coverage and response shares reported for the service population.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:c5678b8f3d59f4a66a7f5599", "occurrence_id": "cb8cfef7c65c3de7384cc500", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Illness reporting was examined for this assessment from the Harmonised Disease Surveillance Survey (HDSS), Edition 3, which UNFCCC maintains, and the measures below were produced by that examination rather than published by the survey.\n\nFacility records were grouped by the week's maximum temperature and the share of attendances coded as heat-related computed: 6 per cent in weeks below 30 degrees and 19 per cent in weeks above 35. The survey tabulates attendances by diagnosis alone, so no temperature-classed share of this kind appears in it.\n\nRecords were then restricted to the two diagnoses most often coded in the hot weeks, and the rise holds at 11 percentage points, which is the part this assessment treats as a genuine heat effect rather than as a change in coding practice.\n\nWeeks with attendance reported for fewer than half the reporting facilities were excluded, 7 per cent of the window, and each share carries the attendance count behind it. The grouping rule is in the annex.\n\nMention: \"Harmonised Disease Surveillance Survey (HDSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Harmonised Disease Surveillance Survey provides the records used to relate reported illness to heat exposure.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:f5fb4a945f92c87175271c2a", "occurrence_id": "619afd215d19d600b2168ef9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two coverage figures in the annex table are drawn from the Linked Administrative Survey Index, the indicator series IMC compiled for Johor in 2023, and both are reproduced exactly as the index prints them.\n\nThe index gives coverage of the target population at 83 per cent and the share of records carrying a complete address at 69 per cent. Neither was recalculated here, and the wording the index uses for coverage and for completeness appears in the note to the table beside the figures.\n\nThe remaining rows of that table were compiled from the office's own returns, so the index carries two lines while the office series carries the rest. A figure the index reports for a single district is shown with that district named, and no wider average was inferred from it.\n\nThe annex records the index table and line behind each quoted figure, the date of extraction as well as the number of records behind each share.\n\nMention: \"Linked Administrative Survey Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Linked Administrative Survey Index supplies two coverage figures quoted alongside the office's own series.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:a6f69bba8f106ed747702dbf", "occurrence_id": "04600a4935a120c9e9552323", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Shortfalls against the minimum needs basket were measured for this assessment from the National Minimum Needs Microdata (NMNM), which covers Xinjiang province in 2017, and the figures below are the product of that work rather than of the microdata's tabulations.\n\nHouseholds were scored against the basket and the share unable to afford two or more items computed: 29 per cent of rural households against 14 per cent of urban ones. The microdata report spending on each item separately, so no combined shortfall of this kind exists in them.\n\nHouseholds were then separated by whether anyone in them held a salaried post, and the share with a two-item shortfall falls to 12 per cent in that group against 38 per cent elsewhere, the gap this assessment treats as the reason the transfer is proposed at a flat rate rather than means-tested.\n\nHouseholds with a missing item in the basket were excluded rather than imputed, which removes 6 per cent of the sample, and each share carries the margin around it. The scoring rule is set down in the annex.\n\nMention: \"National Minimum Needs Microdata (NMNM)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The National Minimum Needs Microdata provides the household records used to compare shortfalls by settlement type.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:ae44080c099c2139e397666e", "occurrence_id": "afa1c2eafd7a4b03c81a6213", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Job search behaviour was analysed for this assessment from the Quality-Assured Labour Force Survey (QLFS), which Twitter fielded in Mali, and the figures below come out of that analysis rather than out of any table the survey prints.\n\nRespondents without work were grouped by the contract they last held and the share searching through more than one channel computed: 52 per cent among those last on a casual arrangement against 33 per cent among those last on a written contract. The survey reports search methods for the unemployed as a whole, so the split by previous contract is not something it carries.\n\nYears of experience were then held constant and the difference narrows to 12 percentage points, which is the part this assessment attributes to the contract history itself.\n\nRespondents who had never worked were kept out of the grouping rather than treated as a contract type, removing 9 per cent of the unemployed, and each share carries the respondent count behind it. The grouping rule is in the annex.\n\nMention: \"Quality-Assured Labour Force Survey (QLFS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Quality-Assured Labour Force Survey supplies the records used to compare job search by contract type.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:05bc96687f933fc9d7719d95", "occurrence_id": "e78ac1cfa0a50da3627552c0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Training outcomes in this section are reported from the Pilot Vocational Training Index (PVTI), the indicator series UN DESA maintains for Houston, and the figures are given as the index gives them for its reporting period.\n\nThe index records 4 180 enrolments and a completion share of 58 per cent, with completion lowest in the trades that run more than six months. Neither figure was recalculated here, and the test the index applies before counting an enrolment as completed is printed beside them.\n\nNo enrolment was moved between trades to lift the reported share, and no withdrawal was reclassified as a completion. Where the index marks a trade as having few enrolments, the mark appears with the figure rather than being smoothed away.\n\nBecause the section's proposals about which trades to fund next rest on the index, the annex lists the tables used, the trades included in the denominator and the date of extraction.\n\nMention: \"Pilot Vocational Training Index (PVTI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Pilot Vocational Training Index supplies the enrolment and completion figures reported for the programme.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:12a70aae9c494ea2e8022f00", "occurrence_id": "c32381119ea9ca43075bb0d7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Licensing in informal trade was measured for this assessment from the Comparative Informal Trade Survey, which the UK Cabinet Office fielded in Albania in 2024, and the figures below were derived for the purpose rather than read from the survey's tables.\n\nTraders were grouped by whether they traded on a public stand or inside a premises and the share holding any trade licence computed: 21 per cent of the former against 64 per cent of the latter. The survey publishes licensing rates for all traders together, so the split by location does not exist in its outputs.\n\nThe two groups were then matched on the goods sold and the difference remains 31 percentage points, which is the part this assessment treats as the effect of the premises rule rather than of the trade itself.\n\nTraders who declined the licensing question were excluded rather than counted as unlicensed, 5 per cent of the sample, and each share carries the number of traders behind it. The matching rule is in the annex.\n\nMention: \"Comparative Informal Trade Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Comparative Informal Trade Survey provides the trader records used to compare licensing by market location.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:7c6f882247cbbf259704919b", "occurrence_id": "8499157f142896142cf7494f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The registration figure in this chapter is the office's own, taken from its counter logs, and the Integrated Administrative Records System (IARS), maintained for Nairobi in 2024, is cited to test it.\n\nThe office puts registrations at 92 per cent of the eligible population. The system's tables give 90 per cent for the same period on its own construction. The office counted from transactions processed while the system counted from records closed, and no log is shared between the two.\n\nThe distance between 92 and 90 falls inside the tolerance agreed for the indicator, so the office's figure stands as calculated, and the system is not used to restate it or to supply values for the months the office's series lacks.\n\nThe annex records what each source treats as a completed registration and which populations each counts, since the two populations are not identical, and the comparison is confined to records common to both.\n\nMention: \"Integrated Administrative Records System (IARS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Integrated Administrative Records System checks the office's registration figure, which the agreement leaves unchanged.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:b102e39939ab24a37cf876db", "occurrence_id": "c15041f3e5caa93f9f8db7f5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Internal movement in this section is reported from the Municipal Internal Movement Registry, the register Enerdata keeps for Spain as of 2016, and the figures are carried over as the register prints them.\n\nThe register records 41 300 arrivals in the reporting year and shows 38 per cent of them arriving from two named districts, with the remainder spread across the rest of the country. Nothing was recalculated here, and the register's rule for counting a household rather than an individual is reproduced beside the totals.\n\nNo arrival was assigned to a district the register leaves blank, and no district was merged with another to make the concentration look smaller. Where the register marks an arrival as recorded twice, the duplicate stays excluded and the exclusion is stated.\n\nBecause the section's conclusions about which districts need information campaigns rest on these counts, the annex records the extract date, the tables used together with the arrivals excluded from them.\n\nMention: \"Municipal Internal Movement Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Municipal Internal Movement Registry supplies the movement counts and origin districts quoted in the section.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:73fc618d81c22d43b5b4a53e", "occurrence_id": "849fe1ea931ee0de2b0ca26d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Shortfall against the minimum needs basket was measured for this chapter from the Analytical Minimum Needs Survey (AMNS), Round 7, fielded in Russia in 2015, and the estimates below come from that exercise rather than from the survey's reporting.\n\nThe team scored each household against the basket and derived the share missing two or more items: 24 per cent of households more than 20 kilometres from a market town against 11 per cent of those closer. The survey tabulates item spending by region, so no distance-based shortfall appears in it.\n\nThe comparison supports one claim, that the mobile delivery proposed later in the chapter should begin with the outlying areas. The chapter's price analysis rests on the price series described in the annex, and the survey supplied nothing further to it.\n\nHouseholds with a missing price for any basket item were excluded, 7 per cent of the sample, and each share carries the margin around it. The scoring rule is written up in the annex.\n\nMention: \"Analytical Minimum Needs Survey (AMNS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Analytical Minimum Needs Survey supports the comparison of basket shortfalls between remote and nearer areas.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:89783aee8db1bcd05e0b672b", "occurrence_id": "5fb1469068d93d77dc583086", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two completeness figures in the operations table are taken from the Analytical Administrative Records Index (AARI), Eighth Round, the indicator series VIIRS maintains, and both appear as the index prints them.\n\nThe index puts the share of expected returns received at 88 per cent and the share of those returns that passed its field checks at 71 per cent. Neither was recomputed here, and the note to the table reproduces the index's wording for a received return and for a passed check.\n\nThe remaining rows of the table come from the office's own submission logs, so the index supplies two lines while the logs supply the rest. A figure the index reports for a subset of reporting units is shown with that subset named rather than generalised.\n\nThe annex records the index table and line behind each quoted figure, the date of extraction and the number of returns behind each share, so a reader can rebuild both numbers from the source.\n\nMention: \"Analytical Administrative Records Index (AARI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Analytical Administrative Records Index supplies two completeness figures quoted beside the office's own series.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:df6078ff57e1f056061ebb98", "occurrence_id": "f3d49201773a7d1eb3731088", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Informal sector figures in this section come from the Cross-Sectoral Informal Sector Registry (CISR), the register WhereIsMyTransport maintains for the Arab Republic of Egypt, and they are reported as the register's tables report them.\n\nThe register counts 1.4 million units it classes as informal and records a registration share of 17 per cent among them, with registration highest among units that trade with businesses rather than households. Neither figure was recalculated here, and the register's test for classifying a unit as informal is printed beside them.\n\nNo unit was moved between classes to change the reported share, and no unit counted twice across two municipalities was left duplicated in the total. Where the register marks a municipality as partly enumerated, the mark appears with the figure.\n\nBecause the section's formalisation proposals rest on these two figures, the annex records the extract date, the register tables used plus the units excluded from the count.\n\nMention: \"Cross-Sectoral Informal Sector Registry (CISR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Cross-Sectoral Informal Sector Registry supplies the enterprise counts and registration shares reported for the sector.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:243a7fdea3bde3dafe70d814", "occurrence_id": "c6f3b75295effb73b2a0cdad", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Building coverage in this section is reported from the Baseline Building Footprint Dataset (BBFD), First Phase, the geospatial dataset compiled for Mali in 2015, and the figures are taken from it as published.\n\nThe dataset maps 2.1 million footprints and reports that 68 per cent of the settled area has a footprint assigned, with the remainder left unmapped. Both are carried over unchanged, and the resolution of the imagery and the rule for what counts as a settled area are printed with them.\n\nNo unmapped area was counted as built to close the gap, and no footprint was merged with another to reduce the total. Where the dataset marks an area as cloud-obscured, it stays in the denominator and is shown as unmapped.\n\nThe section's estimates of how many structures need enumeration rest on the dataset, and the annex lists the scenes used, the thresholds behind each share, and the area covered by the extract.\n\nMention: \"Baseline Building Footprint Dataset (BBFD)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Baseline Building Footprint Dataset supplies the footprint counts and coverage share behind the mapping figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:c65412ea5630d0ed9d25dd00", "occurrence_id": "e6b5911e13a1309a1e5107d9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Time use in this section is reported from the Midline Time Use Survey, which the OPT Protection Cluster fielded in Gabon in 2017, and the values below are given as the survey releases them.\n\nThe survey puts mean paid work at 38 hours a week for men and 12 for women, and reports unpaid care work at 26 hours a week for women against 5 for men. All four are carried over as printed, and the survey's own categories for paid and unpaid activity are reproduced with them.\n\nNo activity was reassigned between categories to narrow the difference, and no respondent reporting more than one activity was counted twice in the same category. Where the survey reports a figure for a subset of its sample, the subset is named with the figure.\n\nThe annex gives the survey table and line behind each quoted value, the date of extraction and the diary day each covers, so the four figures can be traced to their source lines.\n\nMention: \"Midline Time Use Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Midline Time Use Survey supplies the reported hours and participation shares behind the section's figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:245fdf36a8e91452772e98a7", "occurrence_id": "9a9f5c6ff7bdfc869f3f117f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The coverage figure in this chapter is the mission's own, computed from its facility listings, and the Rapid Service Register Registry (RSRR), Edition 4, covering Germany in 2017, is cited to test it.\n\nThe mission's calculation gives coverage of 81 per cent for the reporting period while the register's tables give 84 per cent for the same period on their own construction. The mission reached its figure by listing facilities directly and the register reached its own from provider submissions, with no listing shared between them.\n\nThe distance between 81 and 84 sits inside the tolerance agreed for the indicator, so the mission's figure is printed as calculated and the register is not used to adjust it or to fill gaps in the mission's series.\n\nThe annex sets out the definition of coverage each source applies and the facilities each counts, since the two populations differ, and the comparison covers only the facilities common to both.\n\nMention: \"Rapid Service Register Registry (RSRR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Rapid Service Register Registry checks the mission's coverage figure, which the agreement leaves as calculated.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:f6adb6dc41c273a296225318", "occurrence_id": "439d50cea9c4370cb3f35424", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Payment receipt was examined for this chapter from the Recurrent Targeting and Payments Collection, compiled by Federal Police for the Lao PDR in 2018, and the comparisons below were produced for the chapter rather than published by the collection.\n\nThe team computed the share of enrolled households paid on schedule: 71 per cent for the health-linked component, 64 per cent for the education-linked component and 58 per cent for the food component. The collection reports payment counts per component, so no comparable share across the three exists in it.\n\nThat comparison supports one claim, that the payment calendar should be unified before any component is expanded. The chapter's own targeting analysis rests on the registry described in the annex, and the collection supplied nothing to it.\n\nHouseholds enrolled part-way through the period were excluded, 8 per cent of the enrolled total, and each share carries the household count behind it.\n\nMention: \"Recurrent Targeting and Payments Collection\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Recurrent Targeting and Payments Collection supports the comparison of receipt across three service areas.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:29deab0bfc359acb5b48d405", "occurrence_id": "b85bde4dbb764e311a7cce7d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Request resolution was rebuilt for this review from the Municipal Service Delivery Registry (MSDR), Fifth Wave, the register MODIS maintains, and the figures below are the review's own arithmetic on those records.\n\nEach request was followed from logging to closure and the elapsed time converted to days, giving a median of 9 days for water requests, 17 for waste and 31 for structural requests. The registry publishes volumes by category, so no elapsed-time measure of this kind appears in it.\n\nRequests logged in the second half of the window closed 6 days faster on average, and the review treats that as the effect of the intake change made mid-window rather than of the season.\n\nRequests reopened after closure were counted once at their first closure, and the 3 per cent with no closure date were excluded rather than treated as unresolved. Every median carries the request count behind it, and the tracking rule is in the annex.\n\nMention: \"Municipal Service Delivery Registry (MSDR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Municipal Service Delivery Registry provides the request records from which resolution times were rebuilt.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:55adf25f13862a500aa6d0ff", "occurrence_id": "cb16bd0bd6c68909fdb8e069", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Statistical output figures in this section come from the Linked Statistical Bulletin Registry (LSBR), Collection 9, the register maintained for South Asia as of 2015, and both are given as the register's tables give them.\n\nThe register lists 386 bulletins issued in the collection and records that 62 per cent of them were cited by at least one downstream document. Neither figure was recalculated here, and the register's rule for what counts as a citation is reproduced beside the share.\n\nNo bulletin was dropped from the denominator to raise the cited share, and no citation counted twice in one document was left duplicated. Where the register marks a bulletin as withdrawn, it stays in the count and the mark appears with it.\n\nBecause the section's recommendations on publication frequency rest on these two figures, the annex records the extract date, the register tables behind each, with the bulletins excluded from them alongside.\n\nMention: \"Linked Statistical Bulletin Registry (LSBR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Linked Statistical Bulletin Registry supplies the bulletin counts and citation figures used in the section.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:552808701b86174e02df28f7", "occurrence_id": "0bc126b476bef56955340c23", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Transition timing was rebuilt for this review from the Joint Youth Transition System (JYTS), Eighth Release, the administrative system Standard and Poors operates, and the figures below are the review's arithmetic on those records rather than anything the system publishes.\n\nEach leaver was followed from the end of education to their first recorded employment and the elapsed time converted to months, giving a median of 4 months for leavers with a vocational qualification and 8 months for those with none. The system reports stocks at each stage, so no elapsed-time measure exists in its outputs.\n\nLeavers were then grouped by the size of the area they searched in, and those searching beyond their own area found work 3 months sooner, which is the finding the review carries into its advice on mobility support.\n\nLeavers with no recorded start date were excluded rather than treated as still searching, 6 per cent of the cohort, and every median carries the leaver count behind it. The tracking rule is recorded in the annex.\n\nMention: \"Joint Youth Transition System (JYTS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Joint Youth Transition System provides the records from which transition timings by qualification were rebuilt.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:fd497a097db73fc7c6d3f6b4", "occurrence_id": "ba790f554e55f7149f38f0e2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The reporting-completeness figure in this chapter is the mission's own, taken from its submission logs, and the Civil Operations Index Registry, compiled by GeoTerra Image for Nigeria in 2017, is cited to test it.\n\nThe mission's logs give completeness at 86 per cent for the reporting quarter. The registry's tables give 89 per cent for the same quarter on their own construction. The mission counted returns processed at its own office while the registry counted returns acknowledged by the reporting units, and no log is shared between the two routes.\n\nThe distance between 86 and 89 is smaller than the tolerance agreed for the indicator, so the chapter prints the mission's figure as calculated, and the registry is not used to adjust it or to fill the months its own series lacks.\n\nThe annex sets out how each source defines a complete return and which units each counts, since the two populations are not identical, and the comparison covers only the units common to both.\n\nMention: \"Civil Operations Index Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Civil Operations Index Registry checks the mission's reporting figure, which the agreement leaves unchanged.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:cc2e36c0e007eb59a5275210", "occurrence_id": "f11a2fb07eb7459bd7435a12", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Poverty and vulnerability figures in this section are reported from the Sentinel Poverty and Vulnerability Survey, which PPVR fielded in Armenia in 2024, and both are given as the survey releases them.\n\nThe survey puts the share of households below the national adequacy line at 21 per cent and the share reporting a health shock that forced them to borrow in the last year at 14 per cent. Neither was recalculated here, and the survey's own wording for the adequacy line and for a forced loan is reproduced with the figures.\n\nNo household was moved across the line to lower the reported poverty share, and no loan reported for another reason was counted as a health shock. Where the survey reports a figure for a subset of its sample, the subset is named with the figure rather than generalised.\n\nThe annex records the survey table and line behind each quoted figure, the date of extraction and the sample size behind each share, so the two numbers can be rebuilt from the source without further judgement.\n\nMention: \"Sentinel Poverty and Vulnerability Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Sentinel Poverty and Vulnerability Survey supplies the deprivation and health-shock shares reported for the country.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:06220cbdb94da2225b7e4ff8", "occurrence_id": "529e5039d4d84498ba707bf3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The coverage figure in this section is the office's own, computed from its registry extract, and the Integrated Public Records Index (IPRI), Phase 6, produced by the National Institute of Statistics for Beijing, is cited to test it.\n\nThe office puts coverage at 88 per cent of its target population for the reporting period. The index gives 85 per cent for the same period on its own construction. The office worked from records present in its registry while the index worked from household submissions, and no record passes between the two.\n\nThe distance between 88 and 85 falls inside the tolerance agreed for the indicator, so the office's figure is printed as it stands, and the index neither adjusts it nor supplies values for the periods the office's series lacks.\n\nThe annex records how each source defines coverage and which populations each counts, since the two populations differ in age coverage as well as in source, and the comparison is confined to the population both reach.\n\nMention: \"Integrated Public Records Index (IPRI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Integrated Public Records Index checks the office's coverage figure, which the agreement leaves as calculated.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:ad3122a75809452fcf3ab2f3", "occurrence_id": "67b3ae1339a8df5e4cde1387", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Minimum needs figures in this section are reported from the Comparative Minimum Needs Index (CMNI), First Release, the indicator series UNTAD maintains for 2016, and they appear as the index publishes them.\n\nThe index sets the reference basket at 34 items and reports that 27 per cent of households cannot afford four or more of them. Both are carried over unchanged, together with the item list the index uses and the price basis on which affordability is decided.\n\nNo item was removed from the basket to lower the reported shortfall, and no household reporting a missing price was dropped from the denominator to raise it. Where the index publishes a figure for one region only, the region is named with the figure.\n\nBecause the section's conclusions about which groups enter the programme first rest on the index, the annex lists the tables used, the item behind each share and the date of extraction.\n\nMention: \"Comparative Minimum Needs Index (CMNI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Comparative Minimum Needs Index supplies the basket and shortfall figures reported for the reference period.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:c0d39017203b1e28886b158c", "occurrence_id": "b07ed41e1a2c2439f611a1aa", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Teacher vacancies were examined for this assessment from the Endline Teacher Workforce Collection, which the IMF compiled for the CHT and Sylhet divisions in 2017, and the figures below come from that examination rather than from the collection's tables.\n\nSchools were grouped by enrolment and the share carrying at least one vacant post computed: 46 per cent among schools with more than 500 pupils against 19 per cent among those with fewer than 150. The collection reports vacancies per school without a size cross-classification, so the pattern does not exist in its outputs.\n\nThe two groups were then matched on the remoteness score recorded for each school and the difference remains 19 percentage points, which is the part this assessment attributes to enrolment itself rather than to where the school is.\n\nSchools that opened during the collection window were excluded, 4 per cent of the total, and each share carries the school count behind it. The matching rule and the resulting table are in the annex.\n\nMention: \"Endline Teacher Workforce Collection\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Endline Teacher Workforce Collection provides the school records used to compare vacancy rates by school size.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:d24994ccb4719dcdb55b0ae8", "occurrence_id": "01aea52a4a3ff6db58c31c35", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Bulletin reach was measured for this assessment from the Provincial Statistical Bulletin Survey (PSBS), Fourth Release, which the Uppsala Conflict Data Programme fielded, and the measures below were produced for the assessment rather than published by the survey.\n\nBulletins were grouped by the channel through which they were first released and the share reaching more than 500 readers computed: 31 per cent of those released in print against 72 per cent of those released online. The survey tabulates readership and channel separately, so no crossing of the two appears in it.\n\nSubject matter was then held constant and the difference narrows to 33 percentage points, which is the part this assessment attributes to the channel rather than to the topics each channel happens to carry.\n\nBulletins with no recorded channel were excluded rather than assigned to one, 6 per cent of the release, and each share carries the bulletin count behind it. The grouping rule is detailed in the annex.\n\nMention: \"Provincial Statistical Bulletin Survey (PSBS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Provincial Statistical Bulletin Survey supplies the records used to compare bulletin reach by release channel.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:dad5abf357975a853401c023", "occurrence_id": "870b03018bcf5aea4a37e137", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two rollout figures in the education table are quoted from the Verified Curriculum Reform System (VCRS), the administrative system maintained by the Colombian Ministry of Education for 2021, and both appear as the system's tables give them.\n\nThe system records that 68 per cent of schools had received the revised materials and that 54 per cent of teachers had completed the accompanying orientation. Neither was recomputed for this report, and the system's own tests for what counts as received and as completed are reproduced in the note to the table.\n\nThe remaining rows come from the ministry's regional returns, so the system carries two lines while those returns carry the rest. A figure the system reports for one department only is shown with that department named, and no national figure was inferred from it.\n\nThe annex records the system table and line behind each quoted figure, the date of extraction and, alongside it, the number of schools behind each share.\n\nMention: \"Verified Curriculum Reform System (VCRS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Verified Curriculum Reform System supplies two rollout figures quoted beside the ministry's own reporting.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:3efdeed3eef714f34d458235", "occurrence_id": "c056ba8783c29a6949d86754", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Enterprise survival was rebuilt for this assessment from the Provincial Business Demography Registry (PBDR), the register ECLAC maintains for 2019, and the figures below are the assessment's own rather than the register's.\n\nEach registration was followed from opening to either an active status or a removal and the share still active two years later computed: 61 per cent for enterprises opening with one worker against 78 per cent for those opening with five or more. The register publishes counts of openings and closures, so no survival measure of this kind exists in it.\n\nSector was then held constant and the difference narrows to 11 percentage points, which is the part this assessment attributes to initial size rather than to the trade the enterprise enters.\n\nRegistrations with an ambiguous opening date were excluded, 4 per cent of the window, and each share carries the enterprise count behind it. The follow-up rule and the resulting table are in the annex.\n\nMention: \"Provincial Business Demography Registry (PBDR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Provincial Business Demography Registry supplies the records used to compare survival by firm age.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:b666cf68e71806dad2196eb0", "occurrence_id": "a552e73d4cb19c4aa33c5445", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Programme activity figures in this section are taken from the Pilot Programme Monitoring Registry, the register the Lagos State Government keeps for Slovensku as of 2015, and they are reported as the register's tables report them.\n\nThe registry records 1 240 activities scheduled and disbursement of 73 per cent of the committed amount at the reporting date. Neither figure was recalculated here, and the register's rule for counting an activity as started is printed beside them.\n\nNo activity was moved to a different period to lift the reported disbursement, and no commitment cancelled late in the period was removed from the denominator. Where the register marks an activity as suspended, the mark appears with the number.\n\nBecause the section's assessment of delivery pace rests on these two figures, the annex records the extract date, the register tables behind each as well as the activities excluded from the count.\n\nMention: \"Pilot Programme Monitoring Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Pilot Programme Monitoring Registry supplies the activity and disbursement figures reported for the programme.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:591961270b4e5783e6234636", "occurrence_id": "3512b5adda74fbb4064d5e78", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Soil condition in the irrigation area is reported from the Record-Based Soil Fertility Registry, the register INE maintains for Accra as of 2022, and the figures below are carried over as the register prints them.\n\nThe register records 5 700 sampled plots and reports that 44 per cent of them fall below the reference threshold for organic matter, with the share highest on the plots under continuous cropping. Nothing was recalculated here, and the register's thresholds and its sampling rule are reproduced with the figures.\n\nNo plot was dropped from the denominator to lower the reported share, and no plot sampled twice in the season was left duplicated in the total. Where the register marks a plot as sampled outside its protocol, the mark appears with the result.\n\nThe section's recommendations on which inputs to subsidise rest on these figures, and the annex records the extract date, the tables used together with the plots excluded from them.\n\nMention: \"Record-Based Soil Fertility Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Record-Based Soil Fertility Registry supplies the sample counts and nutrient shares reported for the area.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:cd55316d905fce4670b2b24e", "occurrence_id": "0ab62ea6df38c9c4692a74a9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Vaccination coverage in this chapter is taken from the ministry's service returns, and the Independent Vaccination Campaign Index, compiled by Reinhart and Rogoff for Cambodia in 2023, is mentioned for context rather than used as a source of figures.\n\nThe index reports campaign reach of 91 per cent of the target districts and a cold-chain interruption rate of 6 per cent of sessions, both as its own tables give them, with the index's wording for reach and for interruption reproduced beside the figures.\n\nThe chapter's coverage figures come from the service returns, which count doses administered rather than districts reached, and the two are reported side by side without being combined into a single measure.\n\nThe annex lists the index tables consulted, the date of extraction and the districts the index covers, so the reader can see how far the context figures extend before comparing them with the returns.\n\nMention: \"Independent Vaccination Campaign Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Independent Vaccination Campaign Index is quoted for context while the chapter's coverage comes from service returns.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:6b6bfccba2ead0cf4806e259", "occurrence_id": "08205fbf02ec70891e2a8234", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The season described in this component is reconstructed from the mission's own rain-gauge returns, and the Quarterly Rainfall Anomaly Collection (QRAC), Eighth Version, compiled by the Fiscal Affairs Department for Burundi, is mentioned only as background to that reconstruction.\n\nThe collection reports a seasonal anomaly of 18 per cent below the reference mean and 9 weeks in which more than half its reporting stations recorded no measurable rain. Both are given as the collection prints them, with the reference period it uses for the mean stated beside them.\n\nThe component's own figures come from the mission's gauges, which are read weekly rather than quarterly, and the two series are reported without being merged into one number.\n\nThe annex lists the collection tables consulted, the date of extraction and the stations behind its figures, so the reader can see how far the background figures extend across the area the component describes.\n\nMention: \"Quarterly Rainfall Anomaly Collection (QRAC)\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The Quarterly Rainfall Anomaly Collection is quoted as background for the season described in the component.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:f1d6d1c3fecc5e7f78d60d8d", "occurrence_id": "e228d3aa8c06e0d15cb372f4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Graduation from the programme was examined for this assessment from the Rolling Graduation Programmes Survey (RGPS), which covers Madagascar in 2019, and the measures below were derived for the purpose rather than published by the survey.\n\nHouseholds were grouped by how long they had been enrolled and the share whose consumption cleared the graduation threshold computed: 22 per cent among those enrolled under two years against 57 per cent among those enrolled more than four. The survey reports threshold status for enrolled households as a whole, so no duration pattern of this kind appears in it.\n\nThe pattern is the basis for the recommendation that the programme's own exit rule be lengthened, since households leaving before the third year rarely hold the gain.\n\nHouseholds that left the programme for reasons other than graduation were excluded rather than counted as failures, which removes 9 per cent of enrolled households, and each share carries the household count behind it. The grouping rule is in the annex.\n\nMention: \"Rolling Graduation Programmes Survey (RGPS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Rolling Graduation Programmes Survey supplies the records used to compare graduation by duration of membership.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:8f7b4a2f708cbda096c03e78", "occurrence_id": "6556a98fb66897c4a915561f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Ecosystem carbon figures in this section are reported from the Sentinel Ecosystem Carbon Collection (SECC), the collection RMCAB compiled for 2019, and they are carried over as the collection's tables carry them.\n\nThe collection records 3 400 measured plots and reports that 29 per cent of the mapped extent holds carbon stocks above the reference density, with the remainder below it. Both are given unchanged, and the reference density and the plot-measurement rule the collection applies are printed with the figures.\n\nNo plot was excluded to raise the reported share, and no plot measured twice was left duplicated in the total. Where the collection marks a plot as measured outside its protocol, the mark travels with the result.\n\nBecause the section's conclusions about which areas enter the conservation payment rest on the collection, the annex lists the tables used, the plots behind each share plus the date of extraction.\n\nMention: \"Sentinel Ecosystem Carbon Collection (SECC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Sentinel Ecosystem Carbon Collection supplies the plot counts and carbon shares reported for the reference period.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:2fa90de2f9c0abee0717f5cb", "occurrence_id": "7a5a1f5a25098b46cc853561", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Minimum needs figures for the displaced population are reported from the Local Minimum Needs Registry (LMNR), the register HLP AoR maintains for 2019, and both appear in the register's own terms.\n\nThe register puts the share of registered households below the reference adequacy line at 34 per cent and the share receiving any in-kind assistance at 47 per cent. Neither was recomputed here, and the adequacy line and assistance definitions the register applies are printed with the figures.\n\nNo household was removed from the denominator to lower the reported shortfall, and no assistance recorded as requested rather than delivered was counted as received. Where the register reports a figure for one site only, the site is named with the figure.\n\nBecause the section's targeting proposals rest on these two figures, their register tables, the extract date and the households excluded from each are carried in the annex.\n\nMention: \"Local Minimum Needs Registry (LMNR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Local Minimum Needs Registry supplies the adequacy and assistance shares reported for the registered population.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:466219f22fe54108d22b8c63", "occurrence_id": "5f4371618e234857c2d08879", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two energy figures for Uruguay are quoted from the Midline Fuel and Energy Use Survey (MFEUS), which covers the country in 2024, and both are given as the survey's tables give them.\n\nThe survey puts the share of households whose energy spending exceeds a tenth of total spending at 19 per cent and the share collecting fuel more than 3 kilometres from home at 27 per cent. Neither was recalculated here, and the survey's own wording for energy spending and for collection distance travels with the figures.\n\nNo spending category was moved into or out of the energy total to change the reported burden, and no household that bought rather than collected fuel was counted in the distance measure. Where the survey reports for one settlement class, that class is named with the figure.\n\nTheir survey tables, the extract date and the household counts behind each share are set out in the annex, so neither figure can be read as other than what the survey measured.\n\nMention: \"Midline Fuel and Energy Use Survey (MFEUS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Midline Fuel and Energy Use Survey supplies the energy burden and distance figures quoted for the country.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:e09f807be4a65bcded3610e2", "occurrence_id": "e26a6b7702b8895ff0d249ab", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Facility readiness was measured for this assessment from the Audited Facility Assessment Survey, which the U.S. Census Bureau fielded in Switzerland in 2017, and the figures below come from that measurement rather than from the survey's tables.\n\nFacilities were grouped by whether they operated overnight and the share able to begin a resuscitation on arrival computed: 88 per cent of those that did against 41 per cent of those that did not. The survey tabulates equipment and opening hours separately, so the crossing of the two is not available from it.\n\nBed count was then held constant and the gap holds at 33 percentage points, leaving the overnight roster as what the assessment reads as the operative factor rather than the size of the building.\n\nFacilities refusing the equipment module were excluded rather than scored as unequipped, which removes 6 per cent of the assessed total, and the facility count stands beside each share. Their grouping rules are set out in the annex.\n\nMention: \"Audited Facility Assessment Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Audited Facility Assessment Survey provides the facility records used to compare readiness by overnight operation.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:f773159b1c008c3c977280dd", "occurrence_id": "67536fb00681d871546ee5f0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Inspection activity in this section is reported from the Panel Labour Inspection Index, the indicator series the Ministry of Finance maintains for Sudan as of 2024, and the figures are carried over without restatement.\n\nThe index records 6 240 inspections completed in the reporting period and a share of 28 per cent of registered establishments visited at least once. Both travel with the index's own counting rules, which decide what counts as a completed inspection and which establishments enter the register.\n\nNo visit was counted twice because two inspectors attended, and no establishment closed during the period was struck from the denominator. Where the index marks a region as under-enumerated, the mark is printed with the figure.\n\nThe section's proposals on inspection staffing rest on these figures, so the annex lists the index tables used, the fields behind each number and the date the extract was taken.\n\nMention: \"Panel Labour Inspection Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Panel Labour Inspection Index supplies the inspection counts and establishment coverage reported for the period.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:bedc6986ab493f1543bf28fe", "occurrence_id": "e3511e0b8cdf821b40de9406", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The records task in this component concerns the Provincial Integrated Records System (PIRS), the administrative system IMSS operates in Azerbaijan, and the work described is the handling of its records rather than anything drawn from them.\n\nRecords arrive from provincial offices carrying local category codes, and several offices use the same code for different things. The task is to map those codes onto one classification, to flag records that arrive twice under two offices rather than merging them silently, and to keep a log of every decision made on a record so the decision can be checked afterwards.\n\nNo coverage figure and no timeliness figure was produced in the course of this work. The operational shares quoted elsewhere in the component rest on the mission's own returns, and the system contributed nothing to them.\n\nThe mapped extract, the crosswalk and the decision log are deposited together and entered in the national catalogue, with a scheduled review each year, so that the next update begins from one coded extract rather than from parallel provincial submissions.\n\nMention: \"Provincial Integrated Records System (PIRS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The Provincial Integrated Records System is the system whose provincial records are being mapped and catalogued.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:421584997d46d33117876886", "occurrence_id": "4084e27298d54bcfe60e0f9f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Transfer and enrolment figures for programme households are reported from the Rolling Cash Transfer Registry Survey (RCTRS), Seventh Release, which the UN Population Division fields, and both are given as the survey's tables give them.\n\nThe survey puts the share of enrolled households paid within the scheduled month at 66 per cent and the share of their children aged 6 to 14 enrolled in school at 89 per cent. Neither was recalculated here, and the survey's definitions of a scheduled payment and of enrolment appear beside the figures.\n\nNo payment recorded as scheduled but not made was counted as made, and no child reported as enrolled but not attending was counted as attending. Where the survey reports for a subset of households, that subset is named.\n\nTheir survey tables, the extract date and the household counts behind each share are set out in the annex, so neither figure can be read as something other than what the survey measured.\n\nMention: \"Rolling Cash Transfer Registry Survey (RCTRS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Rolling Cash Transfer Registry Survey supplies the payment-timeliness and enrolment shares reported for households.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:dc51a594b6360d3b5d2d1d83", "occurrence_id": "add8a2b0c86df39de5dc884a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Consumption shortfalls were measured for this chapter from the Provincial Household Income and Expenditure Survey, which D¨ur et al. fielded in the Philippines in 2023, and the comparison below was produced for the chapter rather than published by the survey.\n\nHouseholds were grouped by distance from the nearest market town and the share whose food spending fell below the reference line computed: 29 per cent beyond 20 kilometres against 12 per cent within 5. The survey reports food spending by province, so no distance-based reading of it exists.\n\nThat comparison supports one claim in the chapter, that the transport subsidy should begin with the outlying municipalities. The chapter's price analysis rests on the price series described in the annex, to which the survey supplied nothing.\n\nHouseholds with a missing distance record were excluded rather than assigned a median distance, removing 5 per cent of the sample, and the margin stands beside each share. The grouping rule is kept in the annex.\n\nMention: \"Provincial Household Income and Expenditure Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Provincial Household Income and Expenditure Survey supports the comparison of food shortfalls by market distance.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:d1dc64b63029835d1351251f", "occurrence_id": "9de40951a4c89d6c7070afb4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Place-name consistency was measured for this assessment from the Rapid Place Name Registry (RPNR), Release 8, the register compiled for Mexico in 2016, and the figures below are the assessment's own rather than the register's.\n\nEach programme site was matched to the register and the share whose recorded name differs from the register's official form computed: 28 per cent overall, rising to 46 per cent in the southern states. The register holds the official forms alone, so it reports nothing about names used elsewhere.\n\nSites were then checked against the coordinates held in the register, and 6 per cent sit more than 2 kilometres from the place they are named after, which the assessment reads as a registration error rather than a spelling variant and lists for correction.\n\nPlaces the register marks as discontinued were kept out of the matching, and the site count stands beside each share. Their matching rule, the distance threshold and the correction list are in the annex.\n\nMention: \"Rapid Place Name Registry (RPNR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Rapid Place Name Registry provides the official place records used to measure name and coordinate mismatches.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:6210c3fd106f62db8aacaee4", "occurrence_id": "8801b4a7e56b951a8803efbe", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Grievance handling was rebuilt for this assessment from the Departmental Collective Bargaining System (DCBS), Version 2, the administrative system maintained for Sindh in 2022, and the figures below are the assessment's arithmetic rather than the system's.\n\nEach grievance was followed from filing to resolution and the elapsed time converted to days, giving a median of 21 days where a union representative was named on the filing and 44 days where none was. The system reports counts of filings and closures, so no elapsed-time reading of this kind exists in it.\n\nGrievances were then grouped by subject and the gap holds for pay disputes but disappears for disciplinary ones, which is where the assessment places the value of representation.\n\nGrievances still open at the cut-off were censored at their last recorded day rather than counted as unresolved, and the case count stands beside each median. Their tracking rule is held in the annex.\n\nMention: \"Departmental Collective Bargaining System (DCBS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Departmental Collective Bargaining System provides the grievance records from which resolution times were rebuilt.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:56fc3994f790b45e3b507720", "occurrence_id": "aefe2acc2c568b281b3986df", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Skills mismatch was examined for this chapter from the Metropolitan Skills and Employment Survey (MSES), Wave 3, which DAC fielded in Jordan, and the comparison below comes from that examination rather than from the survey's reporting.\n\nRespondents were grouped by whether their highest qualification was acquired abroad and the share reporting a job requiring a lower qualification computed: 34 per cent of the former against 22 per cent of the latter. The survey tabulates qualifications and current job requirements separately, so the crossing does not exist in its outputs.\n\nThat comparison supports one claim, that the recognition service recommended later in the chapter should be funded first. The chapter's wage analysis rests on the payroll series described in the annex, and the survey supplied nothing to it.\n\nRespondents not working at the time of interview were excluded, 17 per cent of the sample, and the margin stands beside each share. Their grouping rule is laid out in the annex.\n\nMention: \"Metropolitan Skills and Employment Survey (MSES)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Metropolitan Skills and Employment Survey supports the comparison of qualification mismatch by place of study.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:cd2f66edf4777d5fa416d2c1", "occurrence_id": "db53807ee69b58d6984030c8", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Insurance take-up was analysed for this assessment from the Combined Unemployment Insurance Survey (CUIS), which covers Cambodia in 2023, and the figures below were derived for the assessment rather than published by the survey.\n\nWorkers were grouped by weekly earnings and the share enrolled in any insurance scheme computed: 19 per cent in the lowest earnings band against 57 per cent in the highest. The survey reports enrolment for the employed as a whole, so no earnings-graded reading of it exists.\n\nThe two bands were then compared on whether a claim had ever been made, and 4 per cent of the lower band had claimed against 21 per cent of the higher band, which the assessment treats as the reason take-up differs rather than the contribution rate alone.\n\nWorkers outside the labour force were excluded rather than counted as unenrolled, removing 12 per cent of respondents, and the respondent count stands beside each share. Their grouping rules are in the annex.\n\nMention: \"Combined Unemployment Insurance Survey (CUIS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Combined Unemployment Insurance Survey provides the records used to compare scheme enrolment by earnings band.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:6e240e447e010c2eef2804e9", "occurrence_id": "aafd628024e95a25a2a123cb", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Irrigation costs were examined for this assessment from the Annual Irrigation and Inputs Survey (AIIS), Collection 7, fielded in Mexico City in 2020, and the figures below come out of that work rather than out of the survey's tables.\n\nHoldings were grouped by water source and the share of gross output spent on water computed: 14 per cent among holdings drawing from the network against 5 per cent among those with their own well. The survey reports water spending and source separately, so the ratio is not something it publishes.\n\nHolding area was then held constant and the difference narrows to 6 percentage points, which leaves the tariff itself as what the assessment treats as the burden rather than the volume of water applied.\n\nHoldings reporting no irrigation were excluded rather than scored at zero, 9 per cent of the collection, and the holding count stands beside each share. Their grouping rule and the resulting table are in the annex.\n\nMention: \"Annual Irrigation and Inputs Survey (AIIS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Annual Irrigation and Inputs Survey provides the holding records used to compare water costs by source.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:95d32f53a2de5d6a8c6ae085", "occurrence_id": "30d2105a3e21c32713ba7801", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Teacher workforce figures in this section are reported from the Continuous Teacher Workforce System (CTWS), Eighth Phase, the administrative system the Digital South Asia Library maintains for 2017, and they are carried over as the system's tables carry them.\n\nThe system counts 214 000 teachers in post and reports a promotion rate of 7 per cent for the year, with the rate lowest in the primary phase. Neither figure was recalculated here, and the system's own test for what counts as a post is printed beside them.\n\nNo teacher was counted twice for holding two part-time assignments, and no vacancy was filled in the reporting to improve the promotion rate. Where the system marks a phase as partially reported, the mark travels with the figure.\n\nTheir system tables, the extract date and the postings excluded from each count are filed in the annex, so the two headline figures can be rebuilt from the source without further judgement.\n\nMention: \"Continuous Teacher Workforce System (CTWS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Continuous Teacher Workforce System supplies the teacher count and promotion rate reported for the year.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:df5bace3a5b45bfd64c5b399", "occurrence_id": "ed7a61ece16873d1169fe0f7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The season narrative in this component rests on the service's own incident log, and the Multi-Topic Wildfire Burn Dataset, the geospatial dataset FEPB produced for Uruguay in 2016, is referred to only as background for the burned area it describes.\n\nThe dataset maps 118 000 hectares of burned extent and reports that 23 per cent of it falls within the protected-area network, both given as the dataset's grid gives them with its mapping threshold stated. Nothing was derived from these figures, and no table in the component is built from the dataset.\n\nThe component's own figures come from the incident log, which records reported fires rather than modelled extent, and the two are reported without being combined into a single measure of the season.\n\nThe annex lists the layers consulted, the date of extraction and the extent the dataset covers, so the reader can judge how far the background figures reach before comparing them with the log.\n\nMention: \"Multi-Topic Wildfire Burn Dataset\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"source\", \"summary\": \"The Multi-Topic Wildfire Burn Dataset is referred to as background for the burned area described in the component.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:bee25647da232eb2bc6fbf36", "occurrence_id": "000005ca51c04dbf6943631f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Shortfall against the minimum needs basket was measured for this assessment from the Audited Minimum Needs Survey (AMNS), which Goskomstat fielded in the Russian Federation, and the figures below are the measurement's own.\n\nHouseholds were scored against the basket and the share unable to afford three or more items computed: 23 per cent of households in single-person dwellings against 8 per cent of those with three or more adults. The survey tabulates item affordability by household type without combining the items, so no combined shortfall exists in it.\n\nDwellings were then matched on tenure and the gap holds at 9 percentage points, which the assessment treats as an income effect rather than as a difference in what the two household types need.\n\nDwellings with a missing item price were excluded rather than imputed, 6 per cent of the sample, and the household count stands beside each share. Their scoring rule is recorded in the annex.\n\nMention: \"Audited Minimum Needs Survey (AMNS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Audited Minimum Needs Survey provides the records used to derive combined shortfall by household type.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:216ccce92bdcc731eb68f4fd", "occurrence_id": "53e18931fbac58c18114acb0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The delivery figures in this chapter are calculated from the mission's own field returns, and the Verified Service Delivery Collection (VSDC), Fifth Release, compiled by SEEG for Guatemala, is mentioned only as background to those calculations.\n\nThe collection reports that 71 per cent of its reporting units met their submission schedule and that the mean gap between reporting and publication ran 42 days, both given as the collection's tables give them with the collection's own schedule definition stated. Nothing was computed from the collection, and Its records are kept out of the chapter's tables.\n\nThe chapter's own figures rest on returns collected weekly by the mission, which is a different cadence from the collection's reporting, and where the two readings differ the difference is noted rather than reconciled into a single number.\n\nThe annex lists the collection tables consulted, the date of extraction and the units behind the two background figures, so the reader can judge how far they extend before comparing them with the mission's returns.\n\nMention: \"Verified Service Delivery Collection (VSDC)\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The Verified Service Delivery Collection is quoted as background while the chapter's figures come from mission returns.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:2f9d48671e8d49f7588872e2", "occurrence_id": "6c52c7dce67d07a2517527e5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Cover change was measured for this assessment from the Standard Land Cover Dataset (SLCD), Seventh Release, the geospatial dataset Standard and Poors maintains, and the figures below are the assessment's own rather than anything the dataset publishes.\n\nEach cell was compared across the two mapped periods and the share whose class changed computed: 9 per cent of the mapped extent, with tree cover converting to cropland in 62 per cent of those cases. The dataset supplies one classification per period and makes no comparison of its own.\n\nThe changed cells were then matched to the slope recorded for them, and 44 per cent of the conversions to cropland lie on slopes above 15 per cent, which the assessment treats as the reason the downstream sediment reports have worsened.\n\nCells whose class was recorded as uncertain in either period were excluded, 7 per cent of the extent, and each share carries the cell count behind it. Their comparison rule and the resulting table are in the annex.\n\nMention: \"Standard Land Cover Dataset (SLCD)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Standard Land Cover Dataset provides the cover classes used to measure change between the mapped periods.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:9c6211a70065ed1e9627a340", "occurrence_id": "4cfd20881d3d6f6b25e30981", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two figures on mental health caseloads are quoted from the Sentinel Mental Health System (SMHS), Eighth Phase, the administrative system operated in Nepal for 2018, and both are given as the system's tables give them.\n\nThe system records 12 700 contacts logged in the reporting year and puts the share of those contacts arriving without a prior referral at 38 per cent. Neither was recalculated here, and the system's own rule for counting a contact rather than a person is reproduced beside the figures, because the two rules give materially different totals.\n\nThe remaining rows of the table come from the ministry's facility series, so the system carries two lines while that series carries the rest. A figure the system reports for one province is shown with that province named, and a contact recorded against two facilities is counted as the system counts it.\n\nThe system table and line behind each quoted figure, the extract date and the contacts behind each share are stored in the annex, together with the facilities that did not report.\n\nMention: \"Sentinel Mental Health System (SMHS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Sentinel Mental Health System supplies two caseload figures quoted beside the ministry's own series.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:eeaf0609cdfe6acf5b85808e", "occurrence_id": "824da22e26af8d7aae3e406e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Rainfall conditions for the season are reported from the Diagnostic Rainfall Anomaly Collection (DRAC), the collection SEDLAC compiled for 2018, and the figures are carried over as the collection prints them.\n\nThe collection records a seasonal anomaly of 19 per cent below its reference mean and reports that 41 per cent of its reporting stations logged at least six consecutive dry weeks. Both travel with the reference period the collection uses for its mean and with its rule for counting a dry week, since neither figure is interpretable without them.\n\nNo station was dropped from the count to soften the reported anomaly, and no district was merged with a wetter neighbour to make the pattern look uniform. Where the collection marks a station as having a partial record, the mark appears with the figure.\n\nBecause the component's planting-calendar advice rests on these figures, the annex lists the stations used, the reference period behind the mean and the date the extract was taken.\n\nMention: \"Diagnostic Rainfall Anomaly Collection (DRAC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Diagnostic Rainfall Anomaly Collection supplies the anomaly and station figures reported for the season.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:6813c818b394ecf515c8465b", "occurrence_id": "874d1dabfd453e1d717a0222", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Road reach was measured for this assessment from the Rolling Transport Network Dataset, the geospatial dataset Bureau Van Dijk compiled for Azerbaijan in 2017, and the figures below come from that measurement rather than from the dataset itself.\n\nEach settlement was matched to the network and the share with a passable route to the nearest town computed: 74 per cent before the displaced population arrived in the district and 51 per cent after, the change coming from routes the dataset marks as carrying loads beyond their class. The dataset holds the network and its condition codes and says nothing about traffic.\n\nSettlements were then ranked by distance from the district road and the shortfall concentrates in the farthest quartile, which is where the assessment places the first repair.\n\nSettlements reached only by tracks the dataset leaves unclassified were excluded, 8 per cent of the total, and each share carries the settlement count behind it. Their matching rule is listed in the annex.\n\nMention: \"Rolling Transport Network Dataset\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Rolling Transport Network Dataset provides the road geometry used to measure reach after the displacement.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:44b78711cfadd327fc7210aa", "occurrence_id": "f74c0841801ec64d2167b317", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two timeliness figures in the operations annex are quoted from the Analytical Operations Monitoring Survey (AOMS), which covers Accra in 2022, and both appear as the survey prints them.\n\nThe survey puts the share of returns filed by their deadline at 67 per cent and the median number of days a late return ran past that deadline at 11. Neither was recomputed here, and the survey's own deadline calendar and its treatment of a return filed on the deadline date are reproduced with the figures.\n\nThe remaining rows of that annex were compiled from the mission's submission logs, so the survey carries two lines while the logs carry the rest. A figure the survey reports for one reporting round is shown with that round named, and no round was pooled with another to smooth the series.\n\nThe survey table and line behind each quoted figure, the extract date and the returns behind each share are set down in the annex, so either number can be rebuilt from the source.\n\nMention: \"Analytical Operations Monitoring Survey (AOMS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Analytical Operations Monitoring Survey supplies two timeliness figures quoted beside the mission's series.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:0b5829f2fa0dc6bcc6e9e6b5", "occurrence_id": "07d48bbe1ab1c6eb067be8ec", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The classroom-practice figure in this chapter is the ministry's own, taken from its inspector observations, and the Rapid Classroom Practice Survey, which EM-DAT fielded in Australia in 2019, is cited to test it.\n\nThe ministry puts the share of lessons in which the teacher questioned pupils at 46 per cent. The survey's own observation round gives 51 per cent for the same period on its own coding. The ministry coded from inspector notes while the survey coded from recorded lessons, and no lesson appears in both.\n\nThe distance between 46 and 51 falls inside the tolerance agreed for the indicator, so the ministry's figure is printed as it stands, and the survey supplies no adjustment to it and no values for the phases the ministry's series lacks.\n\nThe annex records how each source defines a questioned lesson and which classes each observed, since the two samples differ in phase as well as in method, and the comparison is confined to the classes both reach.\n\nMention: \"Rapid Classroom Practice Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Rapid Classroom Practice Survey checks the ministry's observation figure, which the agreement leaves unchanged.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:dbef104a2be5c4a3f8c9e266", "occurrence_id": "5c78314bf9697293e6ef26d3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Hardship was examined for this assessment from the Government Deprivation and Hardship Survey, which Open Data Watch fielded in Bihar in 2015, and the figures below were derived for the assessment rather than published by the survey.\n\nHouseholds were separated by whether anyone in them had moved for work in the last two years and the share reporting an unfinished meal in the prior week computed: 31 per cent of households with a recent mover against 18 per cent of those without. The survey tabulates food shortfall and migration separately, so the crossing does not exist in its outputs.\n\nLand holding was then held constant and the difference narrows to 7 percentage points, which leaves the lost remittance rather than the lost land as what the assessment reads in the data.\n\nHouseholds declining the migration module were excluded rather than counted as non-movers, 5 per cent of the sample, and the household count stands beside each share. Their grouping rule is in the annex.\n\nMention: \"Government Deprivation and Hardship Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Government Deprivation and Hardship Survey provides the records used to compare hardship by displacement status.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:547b8b5c3fa3a45a1a97bb56", "occurrence_id": "4b2747ad8f9f77d65cd1a8d5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The records task in this component concerns the Panel Public Sector Monitoring Registry, the register PROCOMER maintains for Tajikistan as of 2019, and what is described is the handling of its entries rather than anything drawn from them.\n\nUnits enter the register from three ministries, each numbering its own units independently, so the same office can appear three times under three codes. The task is to map those codes onto one identifier, to flag apparent duplicates for review rather than merging them, and to write each decision into a log that travels with the file so it can be checked afterwards.\n\nNo unit count and no performance figure was produced during this work. The operational shares quoted elsewhere in the component rest on the mission's own returns, and the register contributed nothing to them.\n\nThe mapped register, the identifier crosswalk and the review log are deposited together, entered in the national catalogue and placed under annual refresh, so that a later update begins from one coded register instead of three ministry lists.\n\nMention: \"Panel Public Sector Monitoring Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The Panel Public Sector Monitoring Registry is the register whose unit records are being coded and catalogued.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:89f5847124325e9073b34e37", "occurrence_id": "814d7951ea1b9b136246c30f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Structure the camp figures are tabulated by the Multi-Topic Building Footprint Registry (MBFR), Third Round, the register maintained for Zatari camp in 2019, and they are given as the register's tables give them.\n\nThe register counts 21 400 standing structures and puts the share built from materials other than the standard kit at 34 per cent, with the share highest in the blocks opened first. Neither figure was recalculated here, and the register's rule for counting a structure rather than a household is printed beside them.\n\nNo structure demolished during the round was removed from the count, and no structure reported twice by two block teams was left duplicated. Where the register marks a block as remapped, the mark appears with the figures for that block.\n\nBecause the section's proposals on shelter upgrade rest on these two figures, their register tables, the extract date and the structures excluded from the count are written up in the annex.\n\nMention: \"Multi-Topic Building Footprint Registry (MBFR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Multi-Topic Building Footprint Registry supplies the structure counts and shelter shares reported for the camp.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:be2cf1c13383982ba1dc1db5", "occurrence_id": "f6321c8e9e86ec66b06d74de", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two record-quality figures in the annex are quoted from the Quality-Assured Administrative Survey Index, the indicator series NSB compiled for Dar-es-Salaam in 2023, and both are reproduced as the index prints them.\n\nThe index puts the share of registry records carrying no missing critical field at 72 per cent and the share reconciled against a second source at 29 per cent. Neither was recalculated in this report, and the index's list of critical fields is reproduced with the figures, since the share depends directly on how many fields the list names.\n\nThe remaining rows of the annex were compiled from the office's own registry audit, so the index carries two lines while the audit carries the rest. A figure the index publishes for one registry is shown with that registry named rather than averaged across registries.\n\nThe index table and line behind each quoted figure, the extract date and the records behind each share are recorded in the annex, with the registries the index did not reach.\n\nMention: \"Quality-Assured Administrative Survey Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Quality-Assured Administrative Survey Index supplies two record-quality figures quoted in the annex.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:0a8bee0821f21352f2d0fbcd", "occurrence_id": "317bb42bc613afb836cb25f2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The records task in this component concerns the Stratified Reproductive Health Registry (SRHR), the register IMC maintains for 2020, and the work described is the handling of its records rather than anything drawn from them.\n\nRecords arrive from facilities using two different service codes for the same consultation, and a third code is used by facilities that report on paper. The task is to map all three onto one classification, to flag records whose service date precedes their registration date for review rather than correcting them silently, and to keep a log of every decision so the decision can be audited later.\n\nNo coverage figure and no trend figure was produced during this work. The service shares quoted elsewhere in the component rest on the mission's own facility returns, and the registry contributed nothing to them.\n\nThe mapped extract, the crosswalk and the review log are deposited together and entered in the national catalogue under annual refresh, so the next update begins from one coded extract rather than three code systems.\n\nMention: \"Stratified Reproductive Health Registry (SRHR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The Stratified Reproductive Health Registry is the register whose records are being coded, checked and catalogued.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:c21e2415e37e65d4ac85bc2b", "occurrence_id": "fcd741d9613a5a489e79ccfd", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The disbursement figure in this chapter is the unit's own, taken from payment records, and the Quarterly Programme Monitoring Survey (QPMS), Version 6, covering Buenos Aires in 2024, is cited to test it.\n\nThe unit puts disbursement at 78 per cent of committed funds for the reporting quarter. The survey gives 74 per cent for the same quarter on its own construction. The unit counted payments released while the survey counted payments received by the implementing partner, and no payment record is shared between the two.\n\nThe distance between 78 and 74 sits inside the tolerance agreed for the indicator, so the unit's figure stands as calculated, and the survey neither adjusts it nor supplies values for the quarters the unit's series lacks.\n\nThe annex sets out what each source treats as disbursed and which commitments each counts, since a commitment cancelled mid-quarter is handled differently by the two, and the comparison covers only the commitments both include.\n\nMention: \"Quarterly Programme Monitoring Survey (QPMS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Quarterly Programme Monitoring Survey checks the unit's disbursement figure, which the agreement leaves unchanged.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:920ecb62dcd4693a83d7e34b", "occurrence_id": "2c8001085b90854e797e2e85", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Ecosystem carbon figures for Sudan are reported from the Monthly Ecosystem Carbon Dataset (MECD), the geospatial dataset compiled for 2019, and the figures below are taken from it as published.\n\nThe dataset puts mean carbon stock across the mapped extent at 46 tonnes per hectare and reports that 28 per cent of that extent lies in classes holding more than twice the mean, with the concentration in the southern belt. Both are carried over unchanged, together with the class definitions and the mapping resolution behind them.\n\nNo class was merged to raise the reported share, and no cell left unclassified by the mapping was counted into the high-stock classes. Where the dataset marks a period as having incomplete coverage, the mark travels with the figures for that period.\n\nBecause the section's proposals on which areas enter the conservation payment rest on the dataset, the annex lists the layers used, the thresholds behind each share and the date the extract was taken.\n\nMention: \"Monthly Ecosystem Carbon Dataset (MECD)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Monthly Ecosystem Carbon Dataset supplies the stock and land-cover shares reported for the country.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:d735f3f6b589804a761791b7", "occurrence_id": "693e55c72e80c6a588dcfb22", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two operational figures in the regional table are quoted from the National Regional Operations Registry (NROR), Fourth Edition, the register maintained for Lebanon as of 2023, and both appear as the register prints them.\n\nThe register puts the share of scheduled activities completed on time at 61 per cent and the share of activities reporting a cost variance above a tenth at 17 per cent. Neither was recomputed here, and the register's own tests for on-time completion and for a reportable variance are reproduced with the figures.\n\nThe remaining rows of that table were compiled from the office's own activity logs, so the register carries two lines while the logs carry the rest. A figure the register reports for one governorate is shown with that governorate named, and an activity suspended mid-period stays in the denominator as the register leaves it.\n\nThe register table and line behind each quoted figure, the extract date and the activities behind each share are detailed in the annex, with the governorates the register did not reach.\n\nMention: \"National Regional Operations Registry (NROR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The National Regional Operations Registry supplies two activity figures quoted beside the office's own series.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:384cde5325c156d2ad8aecd8", "occurrence_id": "54948d762c29da153d91fefe", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Operational figures for the ministry's programme are reported from the Ministerial Operations Monitoring Registry, the register UNODC maintains for Kinshasa as of 2018, and they are given as the register's tables give them.\n\nThe register lists 3 620 activities under the programme and reports that 68 per cent of those scheduled for the period reached completion, with completion lowest among activities requiring works at more than one site. Neither figure was recalculated here, and the register's rule for counting an activity as completed is printed beside them.\n\nNo activity was shifted into a later period to raise the reported completion, and no activity cancelled for lack of funds was removed from the denominator. Where the register marks an activity as suspended, the mark appears with the number.\n\nBecause the section's assessment of delivery pace rests on these two figures, their register tables, the extract date and the activities excluded from the count are itemised in the annex.\n\nMention: \"Ministerial Operations Monitoring Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Ministerial Operations Monitoring Registry supplies the activity and completion figures reported for the period.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:fa1bcf36c4b7263f9ab03ad8", "occurrence_id": "5d77cd390e5287b03fe4936f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The share of households in temporary shelter reported in this chapter is the unit's own, estimated from the enumeration frame, and the Municipal Housing Conditions Survey (MHCS), Fifth Round, fielded in the CHT and Sylhet divisions in 2020, is cited to check it.\n\nThe unit's estimate puts temporary shelter at 26 per cent of households. The survey's own count for the same round is 23 per cent. The unit worked from structure counts recorded during enumeration while the survey worked from household interviews, and no household appears in both.\n\nThe distance between 26 and 23 falls inside the tolerance agreed for the indicator, so the unit's estimate is reported unchanged and the survey is not used to adjust it or to fill the rounds the unit's series lacks.\n\nThe annex records how each source defines a temporary structure and which dwellings each counted, since the enumeration frame covers dwellings the survey declined to enter, and the comparison is confined to dwellings both reached.\n\nMention: \"Municipal Housing Conditions Survey (MHCS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Municipal Housing Conditions Survey checks the unit's shelter figure, which the agreement leaves as calculated.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:60c9241accba745467121637", "occurrence_id": "76ea5209852a1406100afaf7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Record completeness was examined for this chapter from the National Integrated Records Registry, the register UNODC maintains for Mozambique as of 2022, and the comparison below was produced for the chapter rather than published by the register.\n\nThe two registers held in the registry were scored on the fields each requires and the share of records complete on every required field computed: 71 per cent for the civil register against 48 per cent for the health register. The registry reports volumes per register, so no completeness comparison of this kind appears in it.\n\nThat comparison supports one claim, that the field-level validation recommended later in the chapter should be installed in the health register first. The chapter's coverage analysis rests on the population estimates described in the annex, to which the registry supplied nothing.\n\nRecords created during the migration between systems were excluded rather than scored as incomplete, 6 per cent of the records held, and the record count stands beside each share.\n\nMention: \"National Integrated Records Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The National Integrated Records Registry supports the comparison of record completeness across two registers.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:8ae9e728338b9c017f208af4", "occurrence_id": "620a82154eab2dc1d0bf7305", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Record coverage in this section is reported from the High-Resolution Administrative Records Registry (HARR), the register maintained for Kenya as of 2023, and the figures appear as the register's tables give them.\n\nThe register puts coverage of the resident population at 84 per cent and the share of records matching a second register on two identifying fields at 66 per cent. Both are carried over unchanged, together with the fields the register accepts for a match and the population base behind the first figure.\n\nNo field was dropped from the matching rule to raise the reported share, and no population group was excluded to smooth it. Where the register marks a county as partially enumerated, the mark travels with the figure rather than being quietly resolved.\n\nBecause the section's conclusions about which registers need reconciliation first rest on the HARR, the annex lists the tables used, the fields included in the match and the date the extract was taken.\n\nMention: \"High-Resolution Administrative Records Registry (HARR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The High-Resolution Administrative Records Registry supplies the coverage and match figures reported for the population.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:6586ee54ee36dc24dc2f83e9", "occurrence_id": "3ee700e5203835247f3b99b5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Work on road segments was examined for this assessment from the Longitudinal Road Segment Census (LRSC), the census the General Statistics Office compiled for 2023, and the figures below come from that examination rather than from the census's tables.\n\nSegments were grouped by surface class and the person-days of maintenance labour recorded per kilometre computed: 128 person-days on gravel segments against 41 on sealed ones. The census tabulates labour and surface separately, so no ratio of the two per kilometre exists in it.\n\nSegments were then matched on the traffic carried and the difference narrows to 52 person-days, which is the part the assessment treats as the maintenance burden of the surface itself rather than of the load passing over it.\n\nSegments resurfaced during the census window were excluded, since their labour refers to capital work, 7 per cent of the counted length, and each figure carries the segment count behind it. Their grouping rules are in the annex.\n\nMention: \"Longitudinal Road Segment Census (LRSC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Longitudinal Road Segment Census provides the segment records used to compare employment by road class.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:2930a3fdb809f3e82f4e2946", "occurrence_id": "48fe328e144388919e2e1a23", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Deprivation was measured for this assessment from the Annual Poverty and Vulnerability Microdata (APVM), Eighth Release, the microdata set Afristat distributes, and the estimates below are the product of that work rather than of the microdata's tabulations.\n\nThe team counted the deprivation items each household lacked and derived the share lacking three or more: 32 per cent of households in small towns against 15 per cent in the capital district. Published tabulations for the microdata give the items one at a time, so a combined count of this kind is not available from them.\n\nHouseholds were then separated by whether anyone in them held casual daily work, and the share lacking three or more items is 41 per cent in that group against 13 per cent elsewhere, the gap the assessment treats as the reason the recommendation targets casual labour first.\n\nHouseholds with an incomplete item list were excluded rather than imputed, which removes 5 per cent of the sample, and the margin stands beside each share. The item count is carried in the annex.\n\nMention: \"Annual Poverty and Vulnerability Microdata (APVM)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Annual Poverty and Vulnerability Microdata supplies the records used to derive combined deprivation by area type.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:b4bed262ba9ec1e177ac2860", "occurrence_id": "ac24c981a7f81ac181fd12a5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Overcrowding was measured for this assessment from the Audited Housing Conditions Survey (AHCS), which covers Bolivia in 2023, and the figures below were derived for the purpose rather than read from the survey's tables.\n\nDwellings were grouped by tenure and the share holding more than two occupants per room computed: 24 per cent of rented dwellings against 9 per cent of owner-occupied ones. The survey reports occupants and tenure in separate tabulations, so the crossing of the two does not exist in its outputs.\n\nDwelling size was then held constant and the difference narrows to 8 percentage points, which is the part the assessment attributes to the number of households sharing a roof rather than to the size of the building.\n\nDwellings recorded as vacant were excluded rather than counted as uncrowded, 4 per cent of the surveyed total, and the dwelling count stands beside each share. Their grouping rule and the resulting table are in the annex.\n\nMention: \"Audited Housing Conditions Survey (AHCS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Audited Housing Conditions Survey provides the dwelling records used to compare overcrowding by tenure.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:f8ab82c796ea6939a3136976", "occurrence_id": "bcdb90a8fa87a3add326efff", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Access to the nearest market was measured for this chapter from the Rural Population Grid Dataset (RPGD), Fourth Edition, the geospatial dataset BPS maintains for Vietnam, and the comparison below was produced for the chapter rather than published with the dataset.\n\nCells were grouped by the population density the dataset assigns them and the share more than 5 kilometres from a market town computed: 38 per cent of low-density cells against 12 per cent of high-density ones. The dataset supplies density and nothing about market locations, so the pairing is the chapter's own construction.\n\nThat comparison supports one claim, that the transport support should begin in the low-density cells. The chapter's price analysis rests on the market series described in the annex, to which the dataset supplied nothing.\n\nCells left unpopulated by the dataset were excluded rather than counted as remote, 9 per cent of the mapped extent, and the cell count stands beside each share. Their grouping rule is kept in the annex.\n\nMention: \"Rural Population Grid Dataset (RPGD)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Rural Population Grid Dataset supports the comparison of access between high and low density cells.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:d1c92114020d8247d58706e9", "occurrence_id": "12e4dec20ecd35a26794724a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Rollout timing was rebuilt for this assessment from the Harmonised Curriculum Reform System (HCRS), Collection 7, the administrative system the WEF operates in Burundi, and the figures below are the assessment's arithmetic on those records.\n\nEach school was followed from notification to first use of the new materials and the elapsed time converted to weeks, giving a median of 6 weeks where a district trainer was assigned and 15 weeks where none was. The system reports counts of schools notified and materials shipped, so no elapsed-time reading of this kind exists in it.\n\nSchools were then grouped by enrolment and the gap holds for large and small schools alike, which places the delay in the training step rather than in the delivery of materials.\n\nSchools notified but closed during the window were excluded rather than counted as late, 3 per cent of the notified total, and every median carries the school count behind it. Their tracking rule is in the annex.\n\nMention: \"Harmonised Curriculum Reform System (HCRS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Harmonised Curriculum Reform System provides the school records from which rollout timing was rebuilt.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:59e0b32c1ba8de27356192f9", "occurrence_id": "3c77ab3030f5c0653660d3c2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two completeness figures in the records annex are quoted from the Coordinated Administrative Records Registry (CARR), the register PRIO/Uppsala maintains for Pakistan, and both are given as the register prints them.\n\nThe register puts the share of records carrying every mandatory field at 79 per cent and the share updated within the last two years at 55 per cent. Neither was recomputed here, and the register's list of mandatory fields and its clock for currency are reproduced beside the figures, since both shares move with those two definitions.\n\nThe remaining rows of that annex were compiled from the office's own audit of its registry, so the register carries two lines while the audit carries the rest. A figure reported for one register type is shown with that type named rather than pooled with others.\n\nThe register table and line behind each quoted figure, the extract date and the records behind each share are held in the annex, with the registers the extract did not reach.\n\nMention: \"Coordinated Administrative Records Registry (CARR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Coordinated Administrative Records Registry supplies two completeness figures quoted beside the office's series.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:5d8de90c3d29f0bc5deb49ab", "occurrence_id": "a4dc3f94524c3a0b504b6701", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two figures on administrative boundaries are quoted from the Analytical Administrative Boundary Dataset (AABD), Fifth Wave, the geospatial dataset compiled for Mauritania in 2018, and both appear as the dataset publishes them.\n\nThe dataset reports that 92 per cent of its units carry a boundary closed to the tolerance it specifies and that 34 per cent of the national extent falls inside a mapped flood zone. Neither was re-derived for this report, and the tolerance and the flood-model version behind the second figure travel with it, because a different model version changes the share noticeably.\n\nThe remaining rows of the annex were compiled from the statistics office's own register of units, so the dataset carries two lines while that register carries the rest.\n\nTheir dataset layer, the extract date and the units excluded from each share are laid out in the annex, together with the units whose boundaries the dataset leaves open and the reason it gives for leaving them.\n\nMention: \"Analytical Administrative Boundary Dataset (AABD)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Analytical Administrative Boundary Dataset supplies two boundary and exposure figures quoted in the annex.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:7f16ba8bfdb8cd48e8b65fb2", "occurrence_id": "41e209f3414302c9074df96e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Flood-risk figures in this section are reported from the Verified Climate and Flood Risk System (VCFRS), First Round, the administrative system MINAGRI maintains for 2015, and they are carried over without restatement.\n\nThe system classifies 27 per cent of the registered area in its highest flood-risk band and records 41 800 structures standing inside that band. Both are given as the system's tables give them, with the classification rule that places an area in the band printed beside the figures.\n\nNo area was moved to a lower band to reduce the reported exposure, and no structure counted in two registers was left duplicated in the total. Where the system marks an area as awaiting field verification, it stays in the band it was assigned and the mark appears with it.\n\nBecause the section's proposals on which areas drain first rest on these two figures, their system tables, the extract date and the areas excluded from the count are filed in the annex.\n\nMention: \"Verified Climate and Flood Risk System (VCFRS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Verified Climate and Flood Risk System supplies the risk class and exposure figures reported for the period.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:54772cee90104923a215272c", "occurrence_id": "c58c8e1a8f485dc5a3539735", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Search duration was rebuilt for this assessment from the Linked Job Search System (LJSS), Phase 3, the administrative system the IEA operates, and the figures below are the assessment's own rather than the system's.\n\nEach claim was followed from registration to the first recorded offer and the elapsed time converted to weeks, giving a median of 9 weeks for clerical occupations and 17 weeks for skilled trades. The system reports counts of open claims at each date, so no duration measure of this kind exists in its outputs.\n\nClaims were then grouped by whether the claimant accepted a vacancy outside their own occupation, and those who did closed their claim 6 weeks sooner, which is where the assessment places the case for broader vacancy matching.\n\nClaims closed for administrative reasons rather than placement were excluded, 8 per cent of the window, and every median carries the claim count behind it. Their follow-up rule is recorded in the annex.\n\nMention: \"Linked Job Search System (LJSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Linked Job Search System provides the claim records from which search duration by occupation was rebuilt.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:feaa64e7d46c207201b2e7f4", "occurrence_id": "f84190e42e9101c04869c9e1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Learning figures in this section are reported from the Panel Learning and Equity System (PLES), Wave 6, the administrative system the Ministère des Finances maintains for 2015, and they are given in the system's own terms.\n\nThe system puts enrolment at 94 per cent of the school-age population and the share of pupils advancing to the next grade without repetition at 81 per cent. Neither was recalculated here, and the system's own rules for counting a pupil as enrolled and as advanced are printed beside the figures.\n\nNo pupil withdrawn during the year was removed from the denominator to raise either share, and no repetition was reclassified as a transfer to another school. Where the system marks a region as reporting late, the mark appears with its figures.\n\nBecause the section's proposals on grade repetition rest on the second figure, their system tables, the extract date and the pupils excluded from the count are itemised in the annex.\n\nMention: \"Panel Learning and Equity System (PLES)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Panel Learning and Equity System supplies the enrolment and progression figures reported for the period.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:ffed9c11005e5a8c4957b17d", "occurrence_id": "7be99b903fc1a2f8128904fe", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Pay differences were analysed for this assessment from the Harmonised Collective Bargaining Survey, which DeLorme fielded in the Philippines in 2021, and the figures below come out of that analysis rather than out of the survey's tables.\n\nWorkers were grouped by whether an agreement covered their workplace and the mean hourly wage computed: 1.84 for covered workers against 1.31 for uncovered ones. The survey reports wages for all workers together and bargaining status separately, so no wage split by coverage appears in it.\n\nOccupation and establishment size were then held constant and the difference narrows to 0.22, which leaves what the assessment treats as the bargaining premium itself rather than the kind of work each group does.\n\nWorkers reporting no wage, including the self-employed, were excluded rather than counted at zero, 11 per cent of respondents, and the worker count stands beside each figure. Their grouping rules are in the annex.\n\nMention: \"Harmonised Collective Bargaining Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Harmonised Collective Bargaining Survey provides the workplace records used to compare pay by bargaining status.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:6e8b93ebba535dd9039ec3a0", "occurrence_id": "f751fff12bb3d609ecf9d7ff", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Vacancy figures in this section are reported from the Annual Hiring and Vacancies System (AHVS), Fourth Version, the administrative system maintained in Gabon for 2023, and both appear as the system's tables give them.\n\nThe system records 22 300 vacancies reported by employers and a median of 39 days to fill one, with fill times longest in technical occupations. Neither figure was recalculated here, and the system's rule for counting a vacancy as filled is printed beside them, since vacancies left open and later cancelled are treated differently by its two reporting streams.\n\nNo vacancy was counted twice for appearing in two months, and no unfilled vacancy was dropped from the median to shorten the reported time. Where the system marks an occupation as thinly reported, the mark travels with the figure.\n\nTheir system tables, the extract date and the vacancies excluded from each figure are stored in the annex, together with the employers that reported nothing for the year.\n\nMention: \"Annual Hiring and Vacancies System (AHVS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Annual Hiring and Vacancies System supplies the vacancy counts and fill times reported for the year.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:2617a1a1cb5ce75b68a0b510", "occurrence_id": "37f3abd6f20d67444cff3493", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The records task in this component concerns the District Occupational Safety Survey (DOSS), Fourth Wave, the survey the Barro‐Lee team fielded, and what is described is the handling of its establishment records rather than anything drawn from them.\n\nEstablishments enter the survey through district labour offices, each using its own sector labels, so the same kind of workshop can be recorded under three different headings. The task is to map those labels onto one classification, to flag establishments appearing under two district lists rather than merging them, and to log every decision so the log travels with the file.\n\nNo incident rate and no coverage figure was produced during this work. The safety shares quoted elsewhere in the component rest on the mission's own inspection returns, and the survey contributed nothing to them.\n\nThe mapped extract, the label crosswalk and the review log are deposited together, entered in the national catalogue and placed under annual refresh, so the next round begins from one coded extract rather than parallel district lists.\n\nMention: \"District Occupational Safety Survey (DOSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The District Occupational Safety Survey is the survey whose establishment records are being coded and catalogued.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:4cacb5555e045abcfbe89a51", "occurrence_id": "c1b60f490f911ddd1a605f16", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Access to services was measured for this chapter from the Diagnostic Output Area Census (DOAC), Third Collection, the census conducted in Yemen in 2021, and the comparison below was produced for the chapter rather than published by the census.\n\nAreas were grouped by the household count the census recorded and the share with no health or education point within 3 kilometres computed: 46 per cent of areas below 80 households against 13 per cent of those above 200. The census tabulates household counts and facilities separately, so the crossing does not exist in its outputs.\n\nThat comparison supports one claim, that the mobile teams recommended later should be assigned to the sparse areas first. The chapter's staffing analysis rests on the ministry returns described in the annex, to which the census supplied nothing.\n\nAreas the census left unenumerated were excluded rather than counted as unserved, 12 per cent of the frame, and the area count stands beside each share.\n\nMention: \"Diagnostic Output Area Census (DOAC)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Diagnostic Output Area Census supports the comparison of service access between dense and sparse areas.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:fcef64aaef6495ee40716dbb", "occurrence_id": "0f86caebe941b6f5370b34c2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Reporting delays were examined for this assessment from the Periodic Operations Index Survey, which the U.S. Census Bureau fielded in Panama in 2024, and the figures below are the assessment's calculations rather than the survey's.\n\nOffices were grouped by the number of staff returning data and the share returning after the deadline computed: 44 per cent of offices with fewer than five staff against 18 per cent of those with more than twenty. The survey reports timeliness for all offices together, so no grading by office size appears in it.\n\nThe two groups were then compared on whether the return was filed electronically, and 71 per cent of the prompt group filed electronically against 32 per cent of the late group, which is where the assessment places the case for the electronic channel.\n\nOffices that did not report at all were excluded rather than counted as late, 5 per cent of the surveyed total, and the office count stands beside each share. Their grouping rules are in the annex.\n\nMention: \"Periodic Operations Index Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Periodic Operations Index Survey provides the records used to compare reporting delays by office size.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:ff6a1ce4ea512fbf79c0bb9d", "occurrence_id": "c74f1eddc2eac68203a2a45a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The ownership figure in this chapter is the unit's own, estimated from the cadastral extract, and the Small-Area Asset Ownership Survey (SAOS), Phase 9, compiled by the New York Fed for 2020, is cited to check it.\n\nThe unit's estimate puts owner occupation at 58 per cent of households. The survey's own figure for the same period is 54 per cent. The unit worked from registered titles while the survey worked from household reports of tenure, and no household appears in both readings.\n\nThe distance between 58 and 54 sits inside the tolerance agreed for the indicator, so the unit's estimate is printed unchanged, and the survey neither adjusts it nor supplies values for the areas the unit's series lacks.\n\nThe annex records what each source treats as ownership and which households each reached, since an inherited title with no transfer recorded is handled differently by the two, and the comparison covers only households both describe alike.\n\nMention: \"Small-Area Asset Ownership Survey (SAOS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Small-Area Asset Ownership Survey checks the unit's ownership figure, which the agreement leaves as calculated.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:86e3f6c55240ac9af25f5b78", "occurrence_id": "5fa8647d358bd01e2a2abf15", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The statistical-output figure in this chapter is the office's own count of its releases, and the Diagnostic Statistical Bulletin Registry (DSBR), the register the Department of Defense maintains for Jordan, is cited to test it.\n\nThe office counts 148 releases in the reporting year. The registry lists 141 for the same year on its own counting. The office counted every release it issued while the registry counts only releases carrying its own identifier, and no release list is shared between the two.\n\nThe distance between 148 and 141 is 5 per cent of the office's figure, inside the tolerance set for the indicator, so the office's count is reported as it stands and the registry is used neither to restate it nor to extend it to other years.\n\nThe annex records what each source treats as a release and which series each covers, since the two populations differ, and the comparison is confined to the releases both count as releases.\n\nMention: \"Diagnostic Statistical Bulletin Registry (DSBR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Diagnostic Statistical Bulletin Registry checks the office's output figure, which the agreement leaves unchanged.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:db723e1d8fab6e6736ca46b3", "occurrence_id": "a994e33975216fca8407e409", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Activity completion was examined for this assessment from the Endline Regional Operations Registry (EROR), Version 5, the register maintained for the CHT and Sylhet divisions in 2018, and the figures below are the assessment's calculations rather than the register's.\n\nActivities were grouped by type and the share finished within their scheduled window computed: 72 per cent of procurement activities against 41 per cent of construction works. The register reports completions and activity types in separate tables, so the crossing of the two is not available from it.\n\nActivities were then matched on their contract value and the difference narrows to 19 percentage points, leaving the works themselves rather than the value at stake as what the assessment reads in the data.\n\nActivities cancelled before their scheduled start were excluded rather than counted as incomplete, 6 per cent of the registered total, and the activity count stands beside each share. Their grouping rules are listed in the annex.\n\nMention: \"Endline Regional Operations Registry (EROR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Endline Regional Operations Registry provides the activity records used to compare completion by activity type.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:2e6288f6823ce221664acdc5", "occurrence_id": "9dbeaa2e092ebf9fbc5b9c0c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two remittance figures in the labour annex are quoted from the Standard Migration and Remittance Survey (SMRS), Seventh Round, which ECMWF fields, and both are given as the survey's tables give them.\n\nThe survey puts the share of households receiving a remittance at 19 per cent and the mean share of household income that remittance represents at 11 per cent. Neither was recalculated here, and the survey's own reference period for a receipt and its treatment of in-kind transfers are reproduced with the figures, since counting goods as well as cash moves the second share by several points.\n\nThe remaining rows of that annex were compiled from the office's household panel, so the survey carries two lines while the panel carries the rest. A figure reported for one wave is shown with that wave named rather than averaged across waves.\n\nThe survey table and line behind each quoted figure, the extract date and the households behind each share are set down in the annex, with the waves the extract did not include.\n\nMention: \"Standard Migration and Remittance Survey (SMRS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Standard Migration and Remittance Survey supplies two remittance figures quoted beside the office's series.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:e3becfde61734b6eab1cfd91", "occurrence_id": "e91b9f809f615f03a3b2373b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Operational figures for the city's programme are reported from the National Regional Operations Survey (NROS), Eighth Version, the survey the IEA maintains for Rio de Janeiro, and they are given in the survey's own terms.\n\nThe survey records 5 410 activities under way and puts the share completed within the reporting period at 64 per cent, with completion lowest among activities spanning more than one district. Neither figure was recalculated here, and the survey's test for treating an activity as completed is printed beside them.\n\nNo activity was moved into a later period to lift the reported completion, and no activity suspended for funding reasons was removed from the denominator. Where the survey marks a district as reporting late, the mark travels with its figures.\n\nBecause the section's assessment of delivery pace rests on these two figures, their survey tables, the extract date and the activities excluded from the count are itemised in the annex.\n\nMention: \"National Regional Operations Survey (NROS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The National Regional Operations Survey supplies the activity and completion figures reported for the city.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:7a6075c014e663c08fb7c28f", "occurrence_id": "850c3e077532b13a7ea9c076", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The coverage figure in this section is the office's own, computed from its registry extract, and the Panel Public Records Registry (PPRR), Round 8, maintained for Mozambique in 2022, is cited to test it.\n\nThe office puts coverage at 86 per cent of the population it is mandated to record. The registry gives 82 per cent for the same period on its own construction. The office counted records present in its own system while the registry counted records confirmed by a second source, and no record passes between the two exercises.\n\nThe distance between 86 and 82 falls inside the tolerance agreed for the indicator, so the office's figure is printed as calculated, and the registry supplies no adjustment to it and no values for the periods the office's series lacks.\n\nThe annex records how each source defines coverage and which populations each counts, since the registry's population excludes records it could not confirm, and the comparison is confined to records both reach.\n\nMention: \"Panel Public Records Registry (PPRR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Panel Public Records Registry checks the office's coverage figure, which the agreement leaves unchanged.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:e34bb18ff8129aeae69fdfe0", "occurrence_id": "b27cae4293f1bbc0f3b3d174", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Trading location and licensing were examined for this chapter using the Representative Informal Trade Survey, which Maddison fielded in Niger in 2021, and the comparison below was made for the chapter rather than published by the survey.\n\nTraders were separated by whether their stand lay inside or outside a designated market and the share holding any licence computed: 63 per cent inside the designated areas against 24 per cent outside them. The survey reports licensing and trading location in separate tabulations, so the crossing does not exist in its outputs.\n\nThat comparison supports one claim, that the registration campaign should begin with traders outside the designated areas. The chapter's revenue analysis rests on the treasury series described in the annex, to which the survey supplied nothing.\n\nTraders who declined the location question were excluded rather than assigned to either group, 5 per cent of the sample, and the trader count stands beside each share.\n\nMention: \"Representative Informal Trade Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Representative Informal Trade Survey supports the comparison of trade location and licensing.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:7d6c073b79c6ff4461c1f771", "occurrence_id": "37cf1259ebdf18f195abd8b1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Request handling was measured for this assessment from the Pilot Service Register Index (PSRI), Eighth Round, the indicator series SPMS compiles for Nepal, and the figures below are the assessment's own rather than the index's.\n\nRequests were grouped by the channel through which they arrived and the share resolved at first contact computed: 58 per cent of those arriving in person against 34 per cent of those arriving by letter. The index publishes a single resolution score, so no channel split of this kind appears in it.\n\nRequests were then grouped by subject and the ordering holds for every subject examined, which the assessment treats as a reason to move routine enquiries to the faster channel rather than to add staff.\n\nRequests with no recorded channel were excluded rather than assigned to one, 3 per cent of the window, and each share carries the request count behind it. Their grouping rule is written up in the annex.\n\nMention: \"Pilot Service Register Index (PSRI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Pilot Service Register Index provides the records used to compare request handling by channel.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:1827fc17c39db2a44ef3432e", "occurrence_id": "9fb9797bc58aba73e1db9698", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Vegetation condition for the rangeland is reported from the Standardised Vegetation Condition Registry (SVCR), Release 3, the register the Lagos State Government maintains, and the figures are carried over as the register gives them.\n\nThe register records 9 200 observations in the reporting period and puts the share showing cover below its reference level at 37 per cent, with the share highest in the northern transects. Neither was recalculated here, and the reference level and the observation rule the register applies are printed with the figures.\n\nNo observation was dropped from the count to soften the reported decline, and no transect was merged with a wetter neighbour to make the pattern look uniform. Where the register marks an observation as taken outside its protocol, the mark appears with the result.\n\nBecause the component's proposals on grazing rest periods rest on these two figures, their register tables, the extract date and the observations excluded from the count are listed in the annex.\n\nMention: \"Standardised Vegetation Condition Registry (SVCR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Standardised Vegetation Condition Registry supplies the observation counts and decline shares reported for the period.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:0fe1f8e06cffc92bdb2cbbbb", "occurrence_id": "897919c990594ac53676e6b4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Reporting timeliness was rebuilt for this assessment from the High-Frequency Integrated Records Index (HIRI), Seventh Phase, the indicator series IPSOS maintains, and the series below is the assessment's own rather than the index's.\n\nEach unit's submission was dated against its deadline and the share arriving on time computed: 57 per cent in the first four periods of the window and 73 per cent in the last four. The index prints one composite score per period, so a timeliness series of this kind is not something it carries.\n\nUnits were then grouped by the size of their reporting staff, and units with a dedicated reporter file on time 82 per cent of the time against 44 per cent for those without, which is where the assessment places the improvement.\n\nSubmissions with an unreadable receipt date were excluded, 3 per cent of the window, and each share carries the submission count behind it. Their dating rule is detailed in the annex.\n\nMention: \"High-Frequency Integrated Records Index (HIRI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The High-Frequency Integrated Records Index provides the records used to compare reporting timeliness by unit.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:43dd051ba1ffbaa92811f737", "occurrence_id": "ea6d6517c239b7be74e8904c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The maternal-health figures in this chapter are calculated from the ministry's service data, and the Coordinated Demographic and Health Report (CDHR), Collection 6, issued by NISR for the United States, is mentioned only as background to those calculations.\n\nThe report puts facility delivery at 96 per cent of births and the share of births attended by a skilled provider at 94 per cent, both given as the report prints them with its own reference year stated. Nothing was computed from the report, and no chapter table rests on its pages.\n\nThe chapter's own figures are drawn from facility records rather than from household interviews, and where the two readings differ the difference is noted rather than reconciled, since a facility record and a household report count different things.\n\nThe annex lists the report sections consulted, the date of extraction and the population the report covers, so the reader can judge how far the background figures reach before comparing them with the service data.\n\nMention: \"Coordinated Demographic and Health Report (CDHR)\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The Coordinated Demographic and Health Report is quoted as background while the chapter's figures come from service data.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:a73ddca48574fed80fe9ad46", "occurrence_id": "ed23a0611425d81323adfd5c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Child deprivation was measured for this assessment from the Regional Child Wellbeing Survey (RCWS), which covers Lagos State in 2020, and the figures below come from that measurement rather than from the survey's tables.\n\nWards were ranked by the deprivation score the survey's items produce and the share of children lacking two or more essentials computed: 34 per cent in the lowest-ranked wards against 8 per cent in the highest. The survey reports each item for the sample as a whole, so no ward-level ranking of this kind exists in it.\n\nWards were then matched on the distance to the nearest clinic, and the share lacking two or more essentials is 12 percentage points higher where that distance exceeds 2 kilometres, which the assessment treats as the reason the first sites are proposed there.\n\nWards with fewer than thirty surveyed children were excluded rather than reported as a share, 9 per cent of the surveyed wards, and the child count stands beside each figure. Their ranking rule is in the annex.\n\nMention: \"Regional Child Wellbeing Survey (RCWS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Regional Child Wellbeing Survey provides the household records used to compare deprivation by ward.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:4d3be5746e5ba03a077c8d51", "occurrence_id": "74b4849cc12c8e69b9c92d10", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Shortfall against the reference basket was measured for this assessment from the Statistical Poverty and Vulnerability Microdata (SPVM), Fourth Release, compiled in Gabon for 2020, and the estimates below are the product of that work rather than of the microdata's tabulations.\n\nHouseholds were grouped by size and the share unable to afford three or more basket items computed: 27 per cent of households of two or three adults against 9 per cent of those with five or more. The microdata report affordability for each item separately, so no combined shortfall of this kind is available from them.\n\nTotal spending was then held constant and the difference narrows to 11 percentage points, which is the part the assessment attributes to the fixed costs a small household cannot spread rather than to lower earnings alone.\n\nHouseholds with a missing item price were excluded rather than imputed, 6 per cent of the sample, and the margin stands beside each share. The scoring rule is carried in the annex.\n\nMention: \"Statistical Poverty and Vulnerability Microdata (SPVM)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Statistical Poverty and Vulnerability Microdata supplies the records used to derive shortfall by household size.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:3f1af6116cbf5a79bfb7bf84", "occurrence_id": "c6a60666e682c720a261f897", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Record completeness was examined for this chapter against the Government Administrative Records Registry (GARR), Version 9, the register maintained for Johor in 2019, and the comparison below was produced for the chapter rather than published by the register.\n\nThe registries held in the GARR were scored on the fields each requires and the share of records complete on every required field computed: 74 per cent for the birth register against 51 per cent for the address register. The register reports volumes per registry, so no completeness comparison of this kind appears in it.\n\nThat comparison supports one claim, that field-level validation should be installed in the address register first. The chapter's coverage analysis rests on the population estimates described in the annex, to which the register supplied nothing.\n\nRecords created during the system migration were excluded rather than scored as incomplete, 7 per cent of the records held, and the record count stands beside each share.\n\nMention: \"Government Administrative Records Registry (GARR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Government Administrative Records Registry supports the comparison of record completeness across registries.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:8053f50d3455c5da12085662", "occurrence_id": "6024c91959e5c7a9467da51f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Publication delay was rebuilt for this assessment from the Recurrent Statistical Bulletin Registry (RSBR), the register maintained for Uzbekistan in 2018, and the figures below are the assessment's arithmetic on those records rather than the register's.\n\nEach release was followed from the end of its reference period to publication and the elapsed time converted to weeks, giving a median of 11 weeks for price series and 26 weeks for national accounts. The register reports release dates and series names, so no delay measure of this kind exists in it.\n\nSeries were then grouped by whether they required seasonal adjustment, and the adjusted series published 9 weeks later on average, which is where the assessment places the case for automating that step.\n\nReleases marked as cancelled were excluded rather than treated as infinitely late, 4 per cent of the window, and every median carries the release count behind it. Their tracking rule is kept in the annex.\n\nMention: \"Recurrent Statistical Bulletin Registry (RSBR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Recurrent Statistical Bulletin Registry provides the release records used to compare publication delays by series.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:8d30b617ffc7143565407d5e", "occurrence_id": "6dbe229b9fdcd625f679e409", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The inspection picture in this section is drawn from the ministry's own inspector returns, and the Harmonised Labour Inspection Registry, the register the EBRD maintains for the West Bank and Gaza in 2021, is referred to only as background to that picture.\n\nThe registry reports 8 700 inspections recorded and a share of 31 per cent of establishments visited at least once, both given as the registry's tables give them with its counting rules stated. Nothing was derived from these figures, and no table in the section is built from the registry.\n\nThe section's own figures come from the inspector returns, which record follow-up visits the registry logs as a single inspection, and the two are reported without being merged into one measure of coverage.\n\nThe annex lists the registry tables consulted, the date of extraction and the establishments behind its figures, so the reader can judge how far the background figures reach before comparing them with the returns.\n\nMention: \"Harmonised Labour Inspection Registry\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"source\", \"summary\": \"The Harmonised Labour Inspection Registry is quoted as background for the inspection picture described in the section.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:888a238870525ebcfd912828", "occurrence_id": "04640859354e7dc26e92bca4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Linkage between registers was measured for this assessment from the Multi-Topic Integrated Records Registry, the register Preqin maintains for Dar-es-Salaam as of 2015, and the measures below are the assessment's own rather than the register's.\n\nRecords were matched across the two registers held in the registry and the share matching on two identifying fields computed: 69 per cent overall, falling to 44 per cent for records registered outside the capital. The registry stores the registers and makes no comparison between them.\n\nMatched records were then checked for agreement on the recorded date of birth, and 6 per cent disagree by more than a year, which the assessment treats as a transcription problem rather than as evidence of duplicate registration.\n\nRecords carrying no identifying field were excluded rather than counted as unmatched, 4 per cent of the records held, and each share carries the record count behind it. Their matching rule is in the annex.\n\nMention: \"Multi-Topic Integrated Records Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Multi-Topic Integrated Records Registry provides the records used to compare record linkage across registers.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:885bbe0d2a732f7a9141e3d9", "occurrence_id": "0e4f1033491755d150dc2da2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Wage differences were analysed for this assessment from the Multi-Topic Wage Structure Survey, which UNDYB fielded in Dar-es-Salaam in 2017, and the figures below come out of that analysis rather than out of the survey's tables.\n\nJobs were grouped by contract type and the mean hourly wage computed: 1.42 for permanent posts against 0.98 for fixed-term ones. The survey reports wages by occupation and contracts by employment status separately, so no wage split by contract appears in it.\n\nOccupation, sector and establishment size were then held constant and the difference narrows to 0.21, which leaves what the assessment treats as the contract penalty itself rather than the kind of job each group holds.\n\nJobs reporting no hourly base were excluded rather than converted from monthly pay, 8 per cent of the sample, and the job count stands beside each figure. Their grouping rules are in the annex.\n\nMention: \"Multi-Topic Wage Structure Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Multi-Topic Wage Structure Survey supplies the job records used to compare pay by contract type.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:08ea6bd3cbbf1e28cc92e9f7", "occurrence_id": "8dc27b5e64878670eed27905", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Three figures in the annex table are quoted from the Departmental Migration and Remittance Index (DMRI), Third Collection, the indicator series UNHCR & WFP maintain, and all three appear as the index prints them.\n\nThe index puts the share of mobile households enrolled in school at 61 per cent, the share reaching a health facility within 5 kilometres at 48 per cent, and the share holding any work permit at 22 per cent. None was recalculated here, and the index's own definitions of mobility and of access travel with the figures.\n\nThe remaining rows of that table were compiled from the office's own household listings, so the index carries three lines while the listings carry the rest. A figure the index reports for one collection round is shown with that round named rather than pooled.\n\nThe index table and line behind each quoted figure, the extract date and the households behind each share are held in the annex, with the rounds the extract omitted.\n\nMention: \"Departmental Migration and Remittance Index (DMRI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Departmental Migration and Remittance Index supplies three service figures quoted in the annex table.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:f7321430bf5c25e313a013be", "occurrence_id": "d91dcdc6aedd88c2d024c25b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two coverage figures in the transfer table are quoted from the Biennial Cash Transfer Registry Survey, which DLA Piper conducted in Mozambique in 2015, and both are reproduced as the survey prints them.\n\nThe survey puts enrolment among eligible households at 57 per cent and the share of enrolled households paid within the scheduled month at 69 per cent. Neither was worked out again here, and the survey's own eligibility rule and its definition of a scheduled payment are reproduced with them, since eligibility differs between the survey's two sampling frames.\n\nThe remaining rows of the table were compiled from the programme's payment records, so the survey carries two lines while those records carry the rest. A figure reported for one province is shown with that province named.\n\nThe survey table and line behind each quoted figure, the extract date and the households behind each share are laid out in the annex, together with the districts the survey did not reach.\n\nMention: \"Biennial Cash Transfer Registry Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Biennial Cash Transfer Registry Survey supplies two coverage figures quoted beside the programme's series.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:7a982af561d04c88d0625554", "occurrence_id": "37d8de0b0eaa6f1036688d3e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The records task in this component concerns the Biennial Wage Structure Registry, the register the MRC keeps for Mexico City as of 2022, and what is described is the handling of its records rather than anything drawn from them.\n\nEstablishments are submitted by sector associations using their own occupational titles, so the same job can arrive under four names. The task is to map those titles onto one classification, to flag establishments that appear in two association returns rather than merging them quietly, and to write each decision into a log that travels with the file.\n\nNo wage figure and no dispersion figure was produced during this work. The wage shares quoted elsewhere in the component rest on the mission's own establishment survey, and the register contributed nothing to them.\n\nThe mapped extract, the title crosswalk and the decision log are deposited together, entered in the national catalogue and placed under annual refresh, so the next round begins from one coded extract rather than four association returns.\n\nMention: \"Biennial Wage Structure Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The Biennial Wage Structure Registry is the register whose establishment records are being coded and catalogued.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:a2d345a3f977831583e20b05", "occurrence_id": "957f7deed06bdb4073ebd174", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The coping figure in this chapter is the unit's own, derived from its household panel, and the Quality-Assured Shock Coping Survey, which DLLS fielded in Moscow in 2019, is cited to check it.\n\nThe unit's panel puts the share of households selling a productive asset after a shock at 18 per cent. The survey's own count for the same period is 22 per cent. The panel followed the same households before and after a reported shock while the survey asked once about the last shock recalled, and no household appears in both.\n\nThe distance between 18 and 22 sits inside the tolerance agreed for the indicator, so the panel figure is reported as calculated, and the survey neither adjusts it nor supplies values for the waves the panel lacks.\n\nThe annex records how each source defines a productive asset and a shock, since a recall-based answer counts events the panel may have missed, and the comparison is confined to households both describe alike.\n\nMention: \"Quality-Assured Shock Coping Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Quality-Assured Shock Coping Survey checks the unit's coping figure, which the agreement leaves as calculated.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:801602ae0c2eab59fa98b28a", "occurrence_id": "8aa5f380265164cddcbc585d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Registration in the informal sector was measured for this assessment from the Local Informal Sector Survey (LISS), Fourth Edition, the survey the Turkish Statistical Institute fields, and the figures below are the measurement's own rather than the survey's.\n\nUnits were grouped by trading location and the share holding any registration computed: 58 per cent of units trading from a fixed premises against 19 per cent of those moving between locations. The survey tabulates registration and location separately, so the crossing of the two is not available from it.\n\nUnits were then matched on the goods they sell and the difference remains 27 percentage points, which the assessment treats as the effect of the premises rule rather than of the trade carried on.\n\nUnits reporting more than one location were excluded rather than assigned to one, 6 per cent of the sample, and the unit count stands beside each share. Their grouping rule is filed in the annex.\n\nMention: \"Local Informal Sector Survey (LISS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Local Informal Sector Survey provides the unit records used to compare registration by trading location.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:b980c0c3cb178557ee914c49", "occurrence_id": "998f1d0593e3f13d75ef2228", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Market price the figures for the season are reported by the Baseline Agri-Food Prices Survey (BAPS), Round 2, the survey US Census conducts, and they are given as the survey's tables give them.\n\nThe survey records prices collected at 640 markets and puts the mean retail price of the reference cereal kilogram at 0.71, with the highest prices in the markets reached only by unpaved road. Neither figure was recalculated here, and the survey's reference quality and its collection day are printed beside them.\n\nNo market was dropped from the mean to lower the reported price, and no price collected outside the survey's day was admitted to the calculation. Where the survey marks a market as reporting only part of the week, the mark appears with its prices.\n\nBecause the component's proposals on market infrastructure rest on these figures, their survey tables, the extract date and the markets excluded from the mean are recorded in the annex.\n\nMention: \"Baseline Agri-Food Prices Survey (BAPS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Baseline Agri-Food Prices Survey supplies the market counts and price figures reported for the season.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:0c4befb0ca75dbbb61235a82", "occurrence_id": "6746774cae750ba06f90ce0d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The input-cost figure in this chapter is the unit's own, built from market monitors, and the Government Farm Economy Report (GFER), the report ITU issued for 2019, is cited to test it.\n\nThe unit puts fertiliser cost at 22 per cent of variable costs for a reference hectare. The report's own calculation gives 25 per cent for the same period on its own construction. The unit worked from prices collected at markets while the report worked from farm accounts compiled by the statistics office, and no account appears in both.\n\nThe distance between 22 and 25 sits inside the tolerance agreed for the indicator, so the unit's figure is printed as calculated, and the report neither adjusts it nor supplies values for the seasons the unit's series lacks.\n\nThe annex records which costs each source counts as variable and which farms each covers, since the report excludes hired labour that the unit includes, and the comparison is confined to the cost items both define alike.\n\nMention: \"Government Farm Economy Report (GFER)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Government Farm Economy Report checks the unit's input-cost figure, which the agreement leaves unchanged.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:ca53075f878725f1a33e70b0", "occurrence_id": "4f2a927fbcfa189ab0952569", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The reporting-timeliness figure in this section is the office's own, taken from its submission log, and the Local Operations Monitoring System, the administrative system ProGres maintains for Slovensku in 2020, is cited to test it.\n\nThe office puts returns filed on time at 76 per cent for the reporting quarter. The system gives 80 per cent for the same quarter on its own construction. The office counted returns as they were processed while the system counted them as they were received, and no return appears in both counts.\n\nThe distance between 76 and 80 falls inside the tolerance agreed for the indicator, so the office's figure is reported unchanged, and the system is used neither to restate it nor to fill the quarters its own series lacks.\n\nThe annex records how each source treats a return received on the deadline and which units each counts, since the system's population includes units the office has struck, and the comparison covers only units common to both.\n\nMention: \"Local Operations Monitoring System\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Local Operations Monitoring System checks the office's timeliness figure, which the agreement keeps.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:3f8f7ba3ad7d23f1862f728d", "occurrence_id": "6fae1dcc24be95ef7a48a914", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The records task in this component concerns the Statistical Occupational Safety Index, the indicator series the Central Bank of Peru maintains for Sulaymaniya as of 2022, and what is described is the handling of its establishment records rather than anything drawn from them.\n\nEstablishments arrive on district schedules carrying district hazard labels, and the same activity can be labelled differently in two districts. The task is to map those labels onto one classification, to flag establishments recorded by two districts rather than merging them silently, and to write each decision into a log that travels with the file so it can be checked afterwards.\n\nNo incident rate and no coverage figure was produced during this work. The safety shares quoted elsewhere in the component rest on the mission's own inspection returns, and the index contributed nothing to them.\n\nThe mapped extract, the label crosswalk and the review log are deposited together, entered in the national catalogue and placed under annual refresh, so the next round begins from one coded extract rather than parallel district labels.\n\nMention: \"Statistical Occupational Safety Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The Statistical Occupational Safety Index is the register whose establishment records are being coded and catalogued.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:f3ef9e8b3713b89c1f493758", "occurrence_id": "e02744a4bcebf8db243279a9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Vacancy figures in this section are reported from the Subnational Hiring and Vacancies System (SHVS), Sixth Version, the administrative system the UCDA operates, and they are carried over without restatement.\n\nThe system records 14 900 vacancies open at the reporting date and puts the share filled within 90 days at 52 per cent, with the share lowest for occupations requiring a licence. Neither figure was recalculated here, and the system's rule for treating a vacancy as filled is printed beside them, since vacancies cancelled by the employer are handled differently by its two streams.\n\nNo vacancy was counted twice for appearing in consecutive months, and no unfilled vacancy was removed from the denominator to shorten the reported time. Where the system marks an occupation as thinly reported, the mark travels with the figure.\n\nBecause the section's proposals on skills matching rest on these two figures, their system tables, the extraction date and the vacancies excluded from each are stored in the annex.\n\nMention: \"Subnational Hiring and Vacancies System (SHVS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Subnational Hiring and Vacancies System supplies the vacancy counts and fill shares reported for the period.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:6641ce8fe9f2a907ec6d9739", "occurrence_id": "06f22ecc4deefbaa7374b824", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Informality was examined for this chapter against the Baseline Labour Force Index (BLFI), Edition 9, which the Heritage Foundation compiles for Kyrgyzstan, and the comparison below was produced for the chapter rather than published by the index.\n\nWorkers were grouped by their highest education band and the share in informal work computed: 71 per cent of those with below upper-secondary against 23 per cent of those with tertiary education. The index reports informality for the workforce as a whole, so no grading by education band exists in it.\n\nThat comparison supports one claim, that the transition support recommended later in the chapter should target the lower bands first. The chapter's wage analysis rests on the payroll series described in the annex, to which the index supplied nothing.\n\nWorkers not in employment were excluded rather than counted as formal, 19 per cent of respondents, and the worker count stands beside each share. Their grouping rule is listed in the annex.\n\nMention: \"Baseline Labour Force Index (BLFI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Baseline Labour Force Index supports the comparison of informality by education band.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:70e5e7c69d6e6698f0d81897", "occurrence_id": "7ebcd87efe56f11ff69a8741", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The unpaid-work figures in this chapter come from the office's own diary panel, and the Departmental Time Use Survey (DTUS), which covers Indonesia in 2022, is quoted only as context for them.\n\nThe survey reports 4.1 hours of unpaid care work on a mean day for women and 1.4 hours for men, both given as the survey prints them with its own list of counted activities reproduced beside the figures. Nothing was computed from the survey, and No chapter table rests on its responses.\n\nThe chapter's own figures come from diaries kept across fourteen consecutive days rather than from a single recalled day, and the two are reported without being merged, since a recalled day and a fortnight of diaries describe different behaviour.\n\nThe annex lists the survey tables consulted, the extraction date and the sample the survey reached, so the reader can judge how far the context figures reach before setting them against the panel.\n\nMention: \"Departmental Time Use Survey (DTUS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Departmental Time Use Survey is quoted for context while the chapter's figures come from its own diary panel.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:5993d9af6994daaced3996ab", "occurrence_id": "7b910ed65efe0844a11b54f2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Learning scores were examined for this assessment from the Rolling Literacy and Numeracy Survey, which the Asian Development Bank fielded in China in 2015, and the figures below come from that work rather than from the survey's tables.\n\nSchools were grouped by location and the share of pupils reaching the reading minimum computed: 78 per cent of schools in urban districts against 54 per cent of those in rural ones. The survey reports scores for the sampled pupils as a whole, so no location split of this kind appears in it.\n\nSchools were then matched on the pupil-teacher ratio recorded for them, and the gap narrows to 13 percentage points, which is the part the assessment treats as the school's own effect rather than the staffing it receives.\n\nSchools with fewer than twenty assessed pupils were excluded rather than reported as a share, 8 per cent of the surveyed total, and the pupil count stands beside each figure. Their matching rule is in the annex.\n\nMention: \"Rolling Literacy and Numeracy Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Rolling Literacy and Numeracy Survey provides the assessment records used to compare scores by school location.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:598c62af22b8e4fbc84d3aa3", "occurrence_id": "3818d9464c914d0d2d09de6d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Staffing was compared across service groups for this chapter using the Departmental Public Sector Monitoring Survey, which IMSS fielded in Jordan in 2024, and the comparison below was made for the chapter rather than published by the survey.\n\nThe two groups were separated by service type and the share of posts filled computed: 84 per cent in administration against 67 per cent in front-line delivery. The survey reports filled and vacant posts for the sector as a whole, so the split by service type does not exist in its outputs.\n\nThat comparison supports one claim, that the recruitment round should be directed at front-line posts rather than spread across the sector. The chapter's budget analysis rests on the treasury series described in the annex, to which the survey supplied nothing.\n\nPosts vacant for less than three months were excluded rather than counted as filled, 6 per cent of the posts recorded, and the post count stands beside each share.\n\nMention: \"Departmental Public Sector Monitoring Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Departmental Public Sector Monitoring Survey supports the comparison of staffing between two service groups.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:3509da8061e181c058c40b17", "occurrence_id": "6821265bddaa5fea8fa27610", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Elevation and household condition were examined for this assessment from the Periodic Terrain and Elevation Census, which Burgiss conducted in Germany in 2016, and the figures below are the assessment's own rather than the census's.\n\nEnumeration units were matched to their mean elevation and the share of households receiving a means-tested benefit computed: 26 per cent in the units above 400 metres against 13 per cent in those below 100. The census tabulates elevation and household characteristics separately, so no such contrast exists in it.\n\nUnits were then matched on distance to the nearest service centre, and the difference narrows to 7 percentage points, which leaves the cost of heating rather than remoteness as what the assessment reads in the pattern.\n\nUnits whose elevation could not be resolved were excluded, 4 per cent of the total, and each share carries the household count behind it. Their matching rule and the resulting table are in the annex.\n\nMention: \"Periodic Terrain and Elevation Census\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Periodic Terrain and Elevation Census supplies the records used to compare elevation and household wealth.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:a0e5f2e6299e7ed86eca8116", "occurrence_id": "31773cb6fa742399a17b51b0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The harvest picture in this component is drawn from the fisheries office's own landing records, and the Quality-Assured Fisheries and Aquaculture Dataset, the geospatial dataset SEEG produced for Panama in 2020, is referred to only as background to it.\n\nThe dataset maps 1 830 square kilometres of assigned aquaculture area and reports that 22 per cent of it lies within a seasonally closed zone, both given as the dataset's layers give them with the closure rule stated. Nothing was derived from these figures, and no table in the component is built from the dataset.\n\nThe component's own figures come from landing records, which report what was landed rather than what was licensed, and the two are reported without being combined into a single measure of the sector.\n\nThe annex lists the layers consulted, the extraction date and the coastline the dataset covers, so the reader can judge how far the background figures reach before comparing them with the landing records.\n\nMention: \"Quality-Assured Fisheries and Aquaculture Dataset\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Quality-Assured Fisheries and Aquaculture Dataset is quoted as background for the harvest picture in the component.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:7d53345b51dd9e066f67ef04", "occurrence_id": "6c8c36cbf643a61441c8a21a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The incident-rate figure in this section is the office's own, calculated from its inspection files, and the Multi-Topic Occupational Safety Survey (MOSS), Phase 9, compiled by LinkedInEconomicGraph for 2023, is cited to test it.\n\nThe office puts recorded incidents at 4.2 per thousand workers. The survey's own estimate for the same period is 5.1 per thousand. The office counted incidents that reached an inspection file while the survey counted incidents reported by workers in interview, and no establishment appears in both sources.\n\nThe distance between 4.2 and 5.1 sits inside the tolerance agreed for the indicator, so the office's figure is printed as calculated, and the survey neither adjusts it nor supplies values for the sectors the office's series lacks.\n\nThe annex records which incidents each source counts and which establishments each reached, since an incident settled without inspection is invisible to the first and visible to the second, and the comparison is confined to establishments both describe.\n\nMention: \"Multi-Topic Occupational Safety Survey (MOSS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Multi-Topic Occupational Safety Survey checks the office's incident figure, which the agreement leaves unchanged.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:625408136687cfde3ecbd4ae", "occurrence_id": "96ed5bed6de08b70e0568201", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Coping and shortfall figures covering the county are drawn from the Local Shock Coping Survey, which the ATO conducted in Nairobi County in 2023, and both are given as the survey releases them.\n\nThe survey puts the share of households reducing meals after a shock at 29 per cent and the share of displaced households doing so at 46 per cent. Neither was recalculated here, and the survey's own wording for a shock and for a reduced meal appears beside the figures.\n\nNo shock reported more than once was counted twice, and no household that reduced meals for a reason other than a shock was counted in the share. Where the survey reports for one sub-county, that sub-county is named with the figure rather than generalised to the county.\n\nBecause the section's proposals on food support rest on the second figure, their survey tables, the extraction date and the households behind the share are set down in the annex, together with the sub-counties the survey did not reach.\n\nMention: \"Local Shock Coping Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Local Shock Coping Survey supplies the coping and shortfall shares reported for the county.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:7e174d3e8795a317677777ff", "occurrence_id": "2336617e4aebb6075657e0bd", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two release figures in the annex are quoted from the Combined Statistical Bulletin Survey (CSBS), Version 5, the survey SEP maintains, and both appear as the survey prints them.\n\nThe survey puts the share of scheduled releases issued on their announced date at 63 per cent and the share of those releases carrying a methodological note at 41 per cent. Neither was recalculated here, and the survey's own definition of a scheduled release and of a methodological note travels with the figures, since a technical annex counts as a note under one reading and not under the other.\n\nThe remaining rows of that annex were compiled from the office's own publication log, so the survey carries two lines while the log carries the rest. A figure reported for one release series is shown with that series named.\n\nThe survey table and line behind each quoted figure, the extraction date and the releases behind each share are written up in the annex, with the series the survey did not reach.\n\nMention: \"Combined Statistical Bulletin Survey (CSBS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Combined Statistical Bulletin Survey supplies two release figures quoted beside the office's own count.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:350a06ba119201f101b1df1f", "occurrence_id": "be88a85cce0ba3e0bc079f1d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Service access was compared across boundary versions for this chapter using the Decennial Administrative Boundary Report (DABR), Eighth Wave, the report CEDLAS issues, and the comparison below was produced for the chapter rather than published in the report.\n\nThe two versions were matched and the share of households whose unit changed between them computed: 12 per cent of households overall, rising to 31 per cent in the rapidly built-up units. The report describes the changes and does not count households against them.\n\nThat comparison supports one claim, that the service targets should be restated on the newer boundaries before the next round. The chapter's coverage analysis rests on the population grid described in the annex, to which the report supplied nothing.\n\nUnits whose change could not be matched one to one were excluded rather than split proportionally, 5 per cent of the units, and the household count stands beside each share.\n\nMention: \"Decennial Administrative Boundary Report (DABR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Decennial Administrative Boundary Report supports the comparison of service access across boundary versions.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:53336ae34850963882e250d9", "occurrence_id": "ff72ebba08d2ed8e0e0b62f2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The population figure for the mapped area is the unit's own, built from facility utilisation, and the Quarterly Population Grid System, the administrative system UNESCO-IBE maintains for the Republic of Zambia in 2018, is cited to check it.\n\nThe unit puts the resident population at 214 000. The system's grid gives 226 000 for the same area on its own construction. The unit worked from utilisation records while the system worked from its grid weights, and the two share no inputs at any stage.\n\nThe distance between 214 000 and 226 000 is 6 per cent of the unit's figure, inside the tolerance set for the estimate, so the unit's number is reported unchanged and the system supplies no adjustment to it.\n\nThe annex records how each source defines residence and which cells each includes, since the two grids are not identical, and the comparison is confined to the cells they share rather than extended to the wider district.\n\nMention: \"Quarterly Population Grid System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Quarterly Population Grid System checks the unit's population figure, which the agreement leaves as calculated.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:8bd30b0461f4c9608523c4a8", "occurrence_id": "d290935eb15887645c468544", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The safety picture in this section is drawn from the office's inspection files, and the High-Frequency Occupational Safety Index (HOSI), the indicator series compiled for Tunisia in 2020, is mentioned only as background to that picture.\n\nThe index reports an incident rate of 3.9 per thousand workers and a share of 64 per cent of establishments reporting a safety officer, both given as the index's tables give them with its counting rules stated. Nothing was derived from these figures, and no table in the section is built from the index.\n\nThe section's own figures come from inspection files, which record only establishments an inspector reached, and the two are reported without being merged into a single measure of workplace safety.\n\nThe annex lists the index tables consulted, the extraction date and the establishments behind its figures, so the reader can judge how far the background figures reach before comparing them with the inspection files.\n\nMention: \"High-Frequency Occupational Safety Index (HOSI)\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"source\", \"summary\": \"The High-Frequency Occupational Safety Index is quoted as background for the safety picture in the section.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:34d4bb7d5d1a6842291de988", "occurrence_id": "e123d5ca1195985a947c49fe", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Hardship among farm households was measured for this assessment from the Panel Deprivation and Hardship Survey (PDHS), which covers Poland in 2015, and the figures below are the assessment's own rather than the survey's.\n\nHouseholds were grouped by the area they cultivate and the share reporting an unpaid utility bill in the last year computed: 24 per cent of those farming under two hectares against 7 per cent of those farming more than ten. The survey reports bill arrears and land area in separate tabulations, so the crossing does not exist in its outputs.\n\nHouseholds were then matched on the number of earners in them, and the difference remains 11 percentage points, which the assessment treats as the effect of the plot rather than of the household's labour.\n\nHouseholds reporting no agricultural activity were excluded rather than placed in the smallest class, 9 per cent of the sample, and the household count stands beside each share. Their matching rule is in the annex.\n\nMention: \"Panel Deprivation and Hardship Survey (PDHS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Panel Deprivation and Hardship Survey provides the farm records used to compare hardship by plot size.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:19ae27c6a2892a0ed0217a17", "occurrence_id": "cd0e749180e7183514a3c6d1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Claim outcomes were analysed for this assessment from the District Unemployment Insurance Survey (DUIS), Sixth Phase, the survey Regspec/AcessLab fields, and the figures below come out of that analysis rather than out of the survey's tables.\n\nClaimants were grouped by the contract they last held and the share exhausting their entitlement computed: 38 per cent of those last on a casual arrangement against 17 per cent of those last on a permanent post. The survey reports exhaustion for all claimants together, so no split by prior contract appears in it.\n\nThe two groups were then matched on age band and the difference remains 15 percentage points, which the assessment treats as the effect of the contribution record rather than of the age at which the claim was made.\n\nClaimants whose previous contract was not recorded were excluded rather than grouped together, 7 per cent of claims, and the claim count stands beside each share. Their matching rules are recorded in the annex.\n\nMention: \"District Unemployment Insurance Survey (DUIS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The District Unemployment Insurance Survey supplies the records used to compare claim outcomes by prior contract.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:7374ad4dc7c1b5bd4745e45a", "occurrence_id": "c7b64ad53ca04bf77fbae55b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The vegetation figures in this component are calculated from the service's own field cuts, and the Endline Vegetation Condition Registry (EVCR), the register compiled for Sri Lanka in 2019, is mentioned only to place those figures in context.\n\nThe registry reports cover below its reference level on 31 per cent of its monitoring transects and a decline of 4 per cent against its previous round, both given as the registry prints them with its reference level stated. Nothing was computed from the registry, and no component table rests on its transects.\n\nThe component's own figures come from cuts measured on the ground each season, which are read differently from the registry's photographic scoring, and where the two readings differ the difference is noted rather than reconciled.\n\nThe annex lists the registry tables consulted, the extraction date and the transects behind the two context figures, so the reader can see how far they reach before setting them against the field cuts.\n\nMention: \"Endline Vegetation Condition Registry (EVCR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Endline Vegetation Condition Registry is quoted for context while the component's figures come from field cuts.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:b30b4be45ab382563ab6e56f", "occurrence_id": "f28bf27afb2c1cbd5c22f898", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Employment among people reporting a difficulty was measured for this assessment from the Clustered Disability and Support Survey, which SABESP fielded in Morocco in 2023, and the figures below are the product of that measurement rather than of the survey's tables.\n\nRespondents were separated by whether they received a support payment and the share currently working computed: 47 per cent among recipients against 26 per cent among non-recipients. The survey reports employment and receipt for the sample as a whole, so a contrast of this kind does not exist in its outputs.\n\nThe two groups were then matched on the severity band recorded for the difficulty, and the difference remains 14 percentage points, which the assessment treats as the payment's effect rather than a difference in the severity of what the two groups carry.\n\nRespondents who declined the difficulty questions were excluded, 4 per cent of the sample, and the respondent count stands beside each share. Their matching rule is detailed in the annex.\n\nMention: \"Clustered Disability and Support Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Clustered Disability and Support Survey provides the records used to compare employment by support receipt.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:4bb5c6e90bea7728be1f6965", "occurrence_id": "6313c007f7401adfce432e7b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The deprivation figure in this chapter is the unit's own, computed from its household module, and the Provincial Multidimensional Poverty Survey (PMPS), compiled by the Ministry of Health for 2020, is cited to test it.\n\nThe unit's calculation puts the share of households deprived in three or more dimensions at 19 per cent. The survey's own figure for the same period is 16 per cent. The unit counted a dimension as deprived when any item in it was missing while the survey requires every item, and neither definition was changed to bring the two together.\n\nThe distance between 19 and 16 sits inside the tolerance agreed for the indicator, so the unit's figure is printed as calculated, and the survey neither adjusts it nor supplies values for the provinces the unit's series lacks.\n\nThe annex records the dimensions each source counts and the households each reached, since the two weighting schemes differ, and the comparison is confined to households both score on the same set of dimensions.\n\nMention: \"Provincial Multidimensional Poverty Survey (PMPS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Provincial Multidimensional Poverty Survey checks the unit's deprivation figure, which the agreement keeps.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:d84d05b1d0acb639750e4679", "occurrence_id": "ccc783521387aebb9652212e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Overcrowding was compared between area types for this chapter using the Small-Area Housing Conditions Survey, which Google fielded in Turkey in 2020, and the comparison below was made for the chapter rather than published by the survey.\n\nAreas were separated by the density class the survey assigns them and the share of households with more than two occupants per room computed: 29 per cent in the highest class against 6 per cent in the lowest. The survey reports occupants and density class in separate tabulations, so the crossing does not exist in it.\n\nThat comparison supports one claim, that the first housing upgrade should go to the highest-density class. The chapter's cost analysis rests on the unit-cost series described in the annex, to which the survey supplied nothing.\n\nAreas with fewer than twenty-five surveyed households were excluded rather than reported as a share, 8 per cent of the surveyed areas, and the household count stands beside each figure.\n\nMention: \"Small-Area Housing Conditions Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Small-Area Housing Conditions Survey supports the comparison of overcrowding between two area types.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:207c20b99d46d8a35d944e9c", "occurrence_id": "24e0db3f1707227aa59c6d67", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Meal delivery and pupil health absences were examined for this assessment from the Follow-Up School Feeding Collection (FSFC), Third Wave, compiled by WhereIsMyTransport for 2020, and the figures below are the examination's own rather than the collection's.\n\nSchools were grouped by enrolment and the share delivering meals on every scheduled day computed: 79 per cent among schools with more than 400 pupils against 52 per cent among those with fewer than 100. The collection reports deliveries per school-day and makes no comparison by school size.\n\nThe same records were then read for absence caused by illness, which is the health measure the collection carries alongside feeding, and schools delivering on every scheduled day report 9 absence-days per enrolled pupil for illness against 15 days at the schools delivering least often. Feeding records alone would not have shown that pairing, since the collection holds the two series in separate tables.\n\nSchools closed for any part of the observation window were left out of both calculations rather than counted as failing to deliver, which removes 6 per cent of the collection. Denominators, the grouping rule and the extraction date sit with the tables in the annex.\n\nMention: \"Follow-Up School Feeding Collection (FSFC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Follow-Up School Feeding Collection provides the school records used to compare meal delivery and illness absence.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:51fe5fed2f1523d8eac78a8d", "occurrence_id": "caa6ec572148e4c676c5c535", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Household density, shelter condition and income were mapped for this assessment from the Quarterly Cadastral Frame Census, the census IOM compiled for Niger in 2022, and the figures below are the assessment's own rather than the census's.\n\nEach enumeration area was matched to its settlement and the share of households in structures classed as temporary computed: 44 per cent in the areas nearest the town's outer edge and 12 per cent closer in. The census tabulates structure type and location separately, so nothing like that contrast appears in its outputs.\n\nThe same areas were then read for household income, which is how the assessment connects the shelter condition to poverty, and 58 per cent of households in the outer areas report an income below the national adequacy line against 26 per cent closer in. Density completes the picture, 31 households per hectare against 9, and it is that combination of poverty and crowding the assessment treats as the reason service delivery fails there.\n\nAreas with an incomplete structure count were left out, 5 per cent of the total. The matching rule, the resulting table and the household counts behind each share are carried with the annex tables.\n\nMention: \"Quarterly Cadastral Frame Census\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Quarterly Cadastral Frame Census supplies records used to compare shelter condition, density and household income.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:5e9049eb26d6578ffdb4248e", "occurrence_id": "6c031fe45125d8b6d399c03d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Asset ownership and employment figures in this section are quoted from the Independent Asset Ownership Survey (IAOS), Fourth Phase, conducted by Heston, Summers, and Aten for 2019, and they are given without restatement.\n\nThe survey puts the share of households owning the dwelling they occupy at 61 per cent, the share holding a title document in the name of a woman at 27 per cent, and the share of working-age household members in paid employment at 54 per cent. None was recomputed here, and the survey's own definitions of ownership, of a registered document and of paid employment are reproduced with them.\n\nNo household reporting an in-progress transfer was counted as owning, and no adult doing unpaid farm work for the household was counted as employed. Where the survey reports a figure for one stratum, the stratum is named with the figure rather than extended to the whole sample.\n\nThe section's proposals on registration reform rest on the second figure and its labour-market reading on the third, so the annex gives the survey table and line behind each, the extraction date, and the households behind each share.\n\nMention: \"Independent Asset Ownership Survey (IAOS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Independent Asset Ownership Survey supplies the ownership, title and employment shares quoted in the section.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:6f667b2c28681ea1ea2a94ce", "occurrence_id": "cafe94ecd72be40692134745", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Housing conditions in this section are reported from the Continuous Housing Conditions Survey (CHCS), which covers Saint Lucia in 2015, and the four figures below are given exactly as the survey releases them.\n\nThe survey puts the share of households in structures classed as temporary at 23 per cent and the share without a written tenure document at 41 per cent. It puts the share of household members in paid work at 66 per cent, and the share of households that had moved at least once because of a storm at 12 per cent, which is the displacement measure the survey carries alongside its housing items. None was recalculated here, and the survey's own definitions of a temporary structure, of paid work and of a storm-related move are reproduced with them.\n\nNo category was merged to simplify the table, and no household reporting two reasons for a move was counted twice. Where the survey reports a figure for a subset of its sample, that subset is named with the figure.\n\nTheir survey table and line, the extraction date and the sample size behind each share are carried in the annex notes rather than paraphrased here.\n\nMention: \"Continuous Housing Conditions Survey (CHCS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Continuous Housing Conditions Survey supplies shelter, tenure, work and storm-displacement shares for the country.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:7c91d545dba79804591f47ef", "occurrence_id": "48a72eb66c3d85bce9ebb235", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The documentation figure in this chapter is the mission's own, drawn from its protection filings by displaced households, and the Midline Access and Documentation Census (MADC), Phase 6, conducted in Ghana in 2016, is cited to test it.\n\nThe mission's filings put the share of displaced households holding any identity document at 57 per cent. The census gives 62 per cent for the same period on its own counting. The mission counted households that presented a document during registration while the census asked displaced households directly in the interview, and no household appears in both exercises.\n\nThe distance between 57 and 62 falls inside the margin the census prints around its own estimate, so the mission's figure is reported as calculated and the census is not used to adjust it or to fill the rounds the mission's series lacks.\n\nExcluded from the comparison are households the census reached outside a displacement site, since the filings describe only those who came forward. Which documents each source accepts and where each interviewed are stated with the annex tables.\n\nMention: \"Midline Access and Documentation Census (MADC)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Midline Access and Documentation Census checks the mission's documentation figure for displaced households.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:815a13ed98172507e6ae26b0", "occurrence_id": "1768daa5265a34f190d61bce", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The labour-market picture in this chapter is built from the office's household panel, and the Local Migration and Remittance Survey (LMRS), Round 8, conducted in Houston in 2016, is mentioned only to place that picture in context.\n\nThe survey reports that 21 per cent of households received a remittance in the reference year, that the mean remittance equalled 9 per cent of household income, and that 14 per cent of households contained an adult who had taken work in another city during the year. All three are given as the survey prints them, with its own reference year stated, and nothing was computed from the survey here.\n\nNo chapter table rests on its responses. The employment figures the chapter argues from come from the panel, which follows the same households across waves rather than sampling once, and where the two readings of work and remittance differ the difference is noted rather than reconciled.\n\nThe annex lists the survey tables consulted, the extraction date and the sample the survey reached, so the reader can see how far the context figures extend.\n\nMention: \"Local Migration and Remittance Survey (LMRS)\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The Local Migration and Remittance Survey is mentioned for context while the chapter's employment figures come from the panel.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:94cfbddc754c7ac74558eae4", "occurrence_id": "26b946d64ff25a4a5b6bce5d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two staffing figures in the education annex are quoted from the Municipal Teacher Workforce Survey (MTWS), Second Round, which covers Chile in 2024, and both are reproduced as the survey prints them.\n\nThe survey puts the share of classes taught by a teacher holding the qualification the curriculum requires at 79 per cent and the share of teachers reporting a second job at 13 per cent. Neither was recalculated here, and the survey's own wording for a required qualification and for a second job appears with the figures.\n\nThe remaining rows of that annex were compiled from the ministry's school returns, so the survey carries two lines while the returns carry the rest. A figure the survey reports for one municipality is shown with that municipality named rather than averaged with others.\n\nThe survey is also the only place in this chapter that reports the household income band of the teachers it reached, and that reading is what the section quotes when it discusses whether low pay pushes teachers into second jobs. Excluded schools, the extraction date and the counts behind each share are kept in the annex notes.\n\nMention: \"Municipal Teacher Workforce Survey (MTWS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Municipal Teacher Workforce Survey supplies staffing and teacher household-income figures quoted beside ministry returns.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:b2698c3e0f3b20c2fd154c97", "occurrence_id": "f4cb1ecbdeb3e427b7e997b3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Crop condition was measured for this assessment from the Inter-Agency Remote Sensing Microdata (IRSM), Eighth Round, the microdata set compiled for Accra in 2020, and the figures below come from that work rather than from the microdata's published tables.\n\nParcels were grouped by area and the share whose vegetation index stayed below the reference line through the season computed: 33 per cent of parcels under half a hectare against 14 per cent of those above two hectares. The microdata report the index per parcel without an area cross-classification, so the contrast is not available from them.\n\nThe same parcels were then matched to the household income band recorded for the holding, which is how the assessment connects field size to poverty, and 46 per cent of small-plot households sit in the lowest income band against 9 per cent of the larger ones. That pairing is why the assessment treats the poor condition of small plots as an inability to invest rather than as a difference in soil.\n\nParcels with a cloud-obscured observation in any critical week were left out, 8 per cent of the set. The matching rule, the parcel counts behind each share and the extraction date are carried with the annex tables.\n\nMention: \"Inter-Agency Remote Sensing Microdata (IRSM)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Inter-Agency Remote Sensing Microdata provides parcel records used to compare crop condition by plot size and income.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:b49b8277b70d87ef82cb30e4", "occurrence_id": "1a312b2a570de64c6ec9fb80", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Deprivation was compared across parishes for this chapter using the Departmental Living Standards Survey, which УВКБ ООН, REACH fielded in Ecuador in 2016, and the comparison below was made for the chapter rather than published by the survey.\n\nParishes were separated by the location recorded for them, specifically their altitude and their road distance from the departmental capital, and the share of households without a piped supply computed: 52 per cent of those above 2 000 metres against 19 per cent of those below, and 44 per cent of those more than 25 kilometres from the capital against 16 per cent of those nearer. The survey reports water access by department, so no reading of it by location exists.\n\nThat comparison supports one claim, that the pipeline work should begin in the high and distant parishes. The chapter's cost analysis rests on the unit-cost series described in the annex, to which the survey supplied nothing.\n\nParishes with fewer than twenty surveyed households were left out rather than reported as a share, which removes 7 per cent of the surveyed total. The distance and altitude cut-points are stated with the annex tables.\n\nMention: \"Departmental Living Standards Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Departmental Living Standards Survey supports the comparison of water access by altitude and road distance.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:b57347cf6058f0e60e7127d0", "occurrence_id": "c27afbab096cb6c91700aca2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Public works wages were analysed for this assessment from the Baseline Public Works Survey, which INE fielded in Delhi in 2015, and the figures below are this assessment's calculations rather than values the survey publishes.\n\nWorkers were grouped by assignment and the mean daily wage computed: 340 for those on earthworks against 415 for those on structure work. The survey reports wages for programme workers as a whole, so no split by assignment exists in its tables.\n\nThe same records carry the years of schooling each worker completed, which is the education measure the assessment uses next, and workers with more than ten years of schooling earn 62 more per day on the same assignment than those with fewer. Days worked in the prior month were then held constant and the assignment difference narrows to 44, which leaves the structure work itself as what the assessment reads as the larger return.\n\nWorkers reporting work outside the programme were left out, since their wage refers to other employment, which removes 7 per cent of the sample. The grouping rules, the schooling cut-point and the counts behind each figure are stated with the annex tables.\n\nMention: \"Baseline Public Works Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Baseline Public Works Survey provides worker records used to compare wages by assignment and years of schooling.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:b9895733f9a157993d9b1276", "occurrence_id": "1f3a5026b5906db72b2fb38f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The housing-deprivation figure in this chapter is the unit's own, calculated from its survey module, and the Standard Housing Conditions Index, compiled by TURKSTAT for Ukraine in 2019, is cited to test it.\n\nThe unit's calculation puts the share of households lacking basic services and living below the national income line at 21 per cent. The index gives 18 per cent for the same period on its own construction. The unit counted a household as deprived when it lacked at least one service while the index requires all three, and neither definition was changed to bring the two together.\n\nThe distance between 21 and 18 is smaller than the interval the index prints around its own estimate, so the unit's figure is printed as calculated and the index supplies no adjustment to it and no values for the years the unit's series lacks.\n\nThe annex records the services each source counts, the income line each applies and the households each reached, since the two weightings are not equivalent. A comparison is given only for households both score on the same set of items.\n\nMention: \"Standard Housing Conditions Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Standard Housing Conditions Index checks the unit's combined service and income deprivation figure.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:ce959613812355fd609faa22", "occurrence_id": "a0f6e69323a8f69a022255b7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Disability, work and support figures for the city are taken from the Urban Disability and Support Survey (UDSS), which covers Nepal in 2020, and all three are given as the survey releases them.\n\nThe survey puts the share of households with at least one member reporting a long-standing difficulty at 18 per cent, the share of those households receiving any support payment at 26 per cent, and the share of adults with such a difficulty currently in paid work at 31 per cent against 59 per cent of adults without one. None was recomputed here, and the survey's own wording for a long-standing difficulty, for a support payment and for paid work is reproduced with them.\n\nNo difficulty was reclassified to enlarge the first share, no in-kind assistance was counted as a payment to raise the second, and no adult doing unpaid household duties was counted as working. Where the survey reports a figure for a subset of its sample, the subset is named with the figure.\n\nTheir survey table and line, the extraction date and the sample size behind each share are held in the annex notes rather than restated here.\n\nMention: \"Urban Disability and Support Survey (UDSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Urban Disability and Support Survey supplies the difficulty, employment and support shares reported for the city.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:e203009168903783ac5de1d0", "occurrence_id": "5570d2264bbddc27eb0a6ecc", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Programme completion was worked out for this assessment from the Metropolitan Tertiary Education Collection (MTEC), Second Release, compiled by the Penn World table, and the measures below are this assessment's own rather than the collection's.\n\nStudents were grouped by entry route and the share finishing within the published duration computed: 64 per cent among those entering by examination and 49 per cent among those entering by transfer. The collection reports enrolment and completions as separate counts, so no completion measure of this kind is available from it.\n\nThe collection also records whether a health condition interrupted study, which is the health measure the assessment reads next, and 22 per cent of transfer students reported such an interruption against 9 per cent of examination entrants. Holding that interruption constant narrows the completion difference to 9 percentage points, which is the part the assessment attributes to the route itself rather than to illness.\n\nStudents with no recorded end status were left out, 8 per cent of the cohort. The grouping rule, the derivation and the counts behind each share are stated with the annex tables.\n\nMention: \"Metropolitan Tertiary Education Collection (MTEC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Metropolitan Tertiary Education Collection provides records used to compare completion by entry route and health status.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:e560fddc88ecee36dfb41a1c", "occurrence_id": "8161ecadf06290a155f5a99a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Workplace incidents were measured for this assessment from the Local Occupational Safety Survey (LOSS), Seventh Version, fielded in the Tigray Region in 2021, and the figures below are the product of that measurement rather than of the survey's reporting.\n\nEstablishments were grouped by the share of their workforce displaced from elsewhere in the last year and the rate of recorded incidents per 100 workers computed: 7.4 in establishments where more than half had been displaced against 3.1 in those where fewer than a tenth had. The survey tabulates incidents and worker origin separately, so no such rate appears in it.\n\nEstablishment size and branch were then held constant and the difference remains 3.2 per 100. The survey records no prior safety training for the newly displaced workers in most establishments, which is the explanation the assessment carries forward rather than the kind of work performed.\n\nEstablishments reporting no workers at the time of the survey were left out, 5 per cent of the sample. The grouping rule, the worker counts behind each rate and the extraction date are stated with the annex tables.\n\nMention: \"Local Occupational Safety Survey (LOSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Local Occupational Safety Survey provides establishment records used to compare incidents by displaced-worker share.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:e6fda0475e5d3216b8ef9056", "occurrence_id": "d64f0e43f390f3cd939e2c2a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The emergency-care figures in this chapter are calculated from the ministry's facility data, and the Coordinated Emergency Care Survey (CECS), Eighth Round, compiled by Afristat for 2023, is quoted only to place those figures in context.\n\nThe survey puts the share of facilities admitting emergency cases within 30 minutes at 49 per cent, the share of those cases involving a household pushed into debt by the visit at 17 per cent, and the share of those households falling below the national income line within a year at 8 per cent. All three are given as the survey prints them, with its own definition of an emergency admission stated, and nothing was derived from the survey here.\n\nNo chapter table rests on its responses. The health figures the chapter argues from come from facility records, which capture admissions rather than the household's own account of payment and income, and where the two readings differ the difference is noted rather than reconciled.\n\nThe annex lists the survey tables consulted, the extraction date and the facilities behind the context figures, which is all the survey is used for here.\n\nMention: \"Coordinated Emergency Care Survey (CECS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Coordinated Emergency Care Survey is quoted for context while the chapter's health and income figures come from facility data.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:e7052a07406608ecff4d1e1e", "occurrence_id": "9c9eb91645443ab67c1aa9f6", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Disability, income and work figures in this section are reported from the Quality-Assured Disability and Support Survey (QDSS), Release 7, compiled by MEF for 2022, and all three are carried over as the survey prints them.\n\nThe survey puts the share of adults reporting at least one functional difficulty at 15 per cent, the share of those adults receiving any support at 28 per cent, and the share of them in paid work at 33 per cent against 64 per cent of adults without a difficulty. None was recomputed for this report, and the survey's own difficulty questions, its definition of a support payment and its test for paid work appear beside the figures.\n\nNo difficulty reported as temporary was counted toward the first share, and no in-kind service was counted as a payment to raise the second. Where the survey reports for one age band, the band is named with the figure rather than extended to all adults.\n\nThe section's proposals on support coverage rest on the second and third figures, so their survey tables, the extraction date and the respondents behind each share are laid out in the annex.\n\nMention: \"Quality-Assured Disability and Support Survey (QDSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Quality-Assured Disability and Support Survey supplies the difficulty, work and support shares reported for the period.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:ec45eb9a7775da97845c53b1", "occurrence_id": "084c1e4029518ae50a35573a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The delivery table in this section is compiled from the programme's own needs assessment, and the Annual Service Register Registry, the register Relief International keeps for Aceh province as of 2023, is quoted only to place that table in context.\n\nThe register puts the share of villages with a functioning water point at 68 per cent, the share with a health post staffed on weekdays at 41 per cent, and the share of households within 1 kilometre of either at 73 per cent. All three are reproduced as the register prints them, with its counting rules stated, and none was recalculated here.\n\nTheir purpose is to let the reader see the same indicators from a second vantage point. The programme's targets rest on the needs assessment, which counts households rather than villages, and no register figure enters them or adjusts them.\n\nVillages are counted as the register counts them, and no village is credited with a service it does not hold. The register table and line behind each quoted figure and the extraction date are filed in the annex, since the register is background here and nothing more.\n\nMention: \"Annual Service Register Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Annual Service Register Registry is quoted for context while the programme's targets rest on the needs assessment.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:00f2036cc0c903dec3e1f908", "occurrence_id": "b7db0577f7a65361a461326f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two claim figures in the insurance annex come from the Continuous Unemployment Insurance Survey (CUIS), Second Round, which Oxford fields, and they are reproduced here without amendment.\n\nMedian claim duration stands at 14 weeks, and 22 per cent of claimants exhaust their entitlement before finding work. The survey's definitions of a completed claim and of exhaustion appear in the note to the table, since both quantities move depending on whether a claim paused for a training course still counts as running.\n\nThe rest of that annex is drawn from the ministry's administrative series, so the survey accounts for two lines and the ministry series for the remainder. Where the survey reports a figure for one cohort only, the cohort is named rather than the figure presented as a whole-sample result, and a restatement the survey issues for an earlier round is shown against the round it replaces.\n\nExcluded are claims filed by workers who had already left the labour force, which the survey counts separately. Their extraction date, the survey table and the claim counts standing behind both figures are recorded with the annex notes.\n\nMention: \"Continuous Unemployment Insurance Survey (CUIS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Continuous Unemployment Insurance Survey supplies two claim figures quoted beside the ministry's administrative series.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:64ff68d9c9e428f25e24c9c3", "occurrence_id": "a7fe22302238451fb5e622e2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Movement intentions in this section are reported from the Statistical Movement Intentions Survey, which Copernicus fielded in Buenos Aires in 2021, and the three figures below are carried over as the survey releases them.\n\nThe survey puts the share of households with at least one member planning to move to another city within a year at 17 per cent, the share of those who name the cost of housing as the reason at 44 per cent, and the share of households that had already moved once in the previous two years at 9 per cent. None was recomputed here, and the survey's own wording for a planned move and for a completed move appears beside them.\n\nNo intention was counted twice because two household members reported one, and no move within the same neighbourhood was treated as displacement. Where the survey reports for one commune, the commune is named rather than the figure shown as a city-wide result.\n\nThe section's argument about displacement pressure rests on the first and third figures while the second is quoted only to explain them, so the survey table and line, the extraction date and the households behind each share are listed with the annex notes.\n\nMention: \"Statistical Movement Intentions Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Statistical Movement Intentions Survey supplies planned and completed movement shares for the city's households.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:8db9933cfe3f38364e58ff85", "occurrence_id": "31ca79ce712f00e671cefaed", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The poverty figure in this chapter is the office's own, calculated from its household panel, and the Quarterly Inequality Survey (QIS), Version 4, conducted in Nairobi in 2017, is cited to test it.\n\nThe office's calculation puts the share of households below the metropolitan adequacy line at 26 per cent. The survey gives 31 per cent for the same period on its own construction. The office applied the national line to equivalised household income while the survey uses a metropolitan basket and does not equivalise, and neither definition was altered to bring the two together.\n\nThat gap of 5 percentage points sits inside the interval the survey publishes around its own estimate, so the office's figure is printed as calculated and the survey supplies no adjustment to it and no values for the periods the panel lacks.\n\nHouseholds the survey reached only in its informal-sector frame are excluded from the comparison, since the panel carries no equivalent stratum. The basket each source applies, the equivalence rule each uses and the households each reached are recorded with the annex tables.\n\nMention: \"Quarterly Inequality Survey (QIS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Quarterly Inequality Survey checks the office's adequacy-line poverty figure without adjusting it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:92fbec8222199e7b5e64077d", "occurrence_id": "2eff1163a18f3c7f12ea9b35", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The incident-rate figure in this chapter is the committee's own, drawn from its member returns, and the National Occupational Safety Index (NOSI), Seventh Round, compiled by MSCI for 2017, is cited to test it.\n\nThe committee's returns put recorded incidents at 4.2 per 1 000 insured workers. The index puts them at 3.8 per 1 000 covered workers for the same year on its own construction. The committee counts every reported incident while the index counts only those leading to more than three days away from work, and no definition was changed to close the difference.\n\nThe distance between 4.2 and 3.8 is smaller than the interval the index prints around its own estimate, so the committee's figure stands as calculated and the index is not used to adjust it or to extend the committee's series into years it does not cover.\n\nExcluded from the comparison are the members whose returns arrive without a worker count, 4 per cent of filings, since no rate can be formed for them. Which events each source counts and how each denominator is built are set out with the annex tables.\n\nMention: \"National Occupational Safety Index (NOSI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The National Occupational Safety Index checks the committee's incident rate for insured workers.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:1809ccd184ba5e46fefef9ce", "occurrence_id": "f6c29fb34e5ae45e296a62f9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Enterprise and employment figures for the district are taken from the Sentinel Business Demography Survey, which IVAP fielded in Bihar in 2017, and all four are given exactly as the survey releases them.\n\nThe survey puts the share of registered units active at the time of enumeration at 62 per cent, the share of those units employing anyone outside the owning household at 38 per cent, the mean number of workers per active unit at 4.1, and the share of units opened within the previous two years at 19 per cent. None was recalculated here, and the survey's own definitions of an active unit, of outside employment and of a new opening are reproduced with the figures.\n\nNo dormant registration was counted as an active unit, no household member working without pay was counted as employed, and no unit reporting two locations was entered twice. Where the survey reports for one size class, that class is named with the figure.\n\nThe annex carries the survey table and line for each figure, the extraction date and the number of units standing behind each share.\n\nMention: \"Sentinel Business Demography Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Sentinel Business Demography Survey supplies enterprise activity and employment shares for the district.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:17c71e49bb2090bdbd46d7d6", "occurrence_id": "c616748a4c56235ace645387", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The registration records for the camp were catalogued for the labour file by the Decennial Skills and Employment System, the administrative system the International Monetary Fund maintains for Zatari camp as of 2022, and the work described here is handling rather than measurement.\n\nEach registration was given the file's own occupation code, checked against the entry made at arrival and, where the two disagreed, held for review rather than overwritten. Registrations with no occupation recorded were left uncoded rather than assigned a residual category, and 1 840 of them were carried into the file in that condition. The system's own registration date was kept with every entry so that later rounds could be matched without re-asking the household.\n\nThe labour file is built from the ministry's permit series, and this cataloguing supports that work in one respect only, by keeping camp registrations under the same occupation codes so that permit holders who also registered in the camp can be found in both. Nothing in the file's employment estimates is calculated from the registrations themselves.\n\nThe coding sheet, the count of entries held for review and the date the transfer was deposited are recorded in the annex log.\n\nMention: \"Decennial Skills and Employment System\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The Decennial Skills and Employment System supplies coded camp registrations for the labour file's crosswalk.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:e010a6f2a375dbb7d3b57f97", "occurrence_id": "060213f179e406c49b41e4de", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Transport and clinic-access figures covering the whole country derive from the Combined Transport Network System (CTNS), the administrative register maintained for Lebanon as of 2016, and the four values below are given as the register holds them.\n\nThe register puts the length of all-season road open to civilian traffic at 2 140 kilometres, the share of the population living within 5 kilometres of a clinic admitting walk-in cases at 71 per cent, the number of border crossings operating on the eastern route at 3, and the mean travel time from a secondary settlement to the nearest such clinic at 38 minutes. None was recalculated here, and the register's own definitions of an open road, of an admitting clinic and of travel time appear with them.\n\nNo road closed for part of the year was counted as open, no facility staffed only for emergencies was counted as a clinic, and no settlement was credited with access to a facility beyond the crossing that was shut when the register was compiled.\n\nTheir register table and reference, the extraction date and the settlements behind each value are stored in the annex notes.\n\nMention: \"Combined Transport Network System (CTNS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Combined Transport Network System supplies road length and clinic-distance values for the country.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:af06ed34023ce4a359cb6c0f", "occurrence_id": "df295e1bd8a040d87a097985", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Whether the cooperative's own coverage figure holds was tested against the Urban Market Access Registry, the register NCAER keeps for Italy as of 2018, and the registry is the principal source for that test.\n\nThe cooperative reports that 88 of its 126 member groups sell at a market within 25 kilometres of their holding. The registry, which records each trading site and the holding locations registered against it, puts the equivalent share at 66 per cent, because it counts only sites open on a weekday the group can reach by its own route. The cooperative counts any site within the radius on a straight line while the registry follows the road the register holds.\n\nThe registry is therefore what this section relies on for the market-access claim, and the cooperative's larger count is reported beside it so the reader can see how much of the difference is the road rather than the distance. Where the registry lists a site whose opening day is unrecorded, that site is excluded and the count of exclusions is given.\n\nThe radius each source applies, the route each follows and the trading sites each counted are set out with the annex tables.\n\nMention: \"Urban Market Access Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Urban Market Access Registry tests the cooperative's claim on how many member groups reach a market.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:5ec9fccac8880484d2a493cd", "occurrence_id": "17eda3511b1d85aa7cf8714f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Curriculum uptake was rebuilt for this assessment from the Midline Curriculum Reform System, the administrative system FRS operates in Cyprus in 2019, and the measures below are this assessment's own rather than the system's.\n\nSchools were grouped by the date they entered the reform and the share of their classes teaching the revised syllabus computed: 74 per cent among schools enrolled for more than a year against 41 per cent among those enrolled for one term. The system records the enrolment event and the syllabus code per class without ever joining them into a rate, so no such share exists in its outputs.\n\nThe same records carry the training sessions each teacher attended, and holding attendance constant narrows the difference to 12 percentage points, which leaves the length of enrolment itself as the smaller part of the gap than the assessment had assumed.\n\nClasses with no syllabus code recorded were left out rather than assumed unchanged, which removes 6 per cent of classes. The grouping rule, the derivation and the class counts behind each share are stated with the annex tables.\n\nMention: \"Midline Curriculum Reform System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Midline Curriculum Reform System provides class records used to compute syllabus uptake by length of enrolment.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:989bec151fe38674fa4e7fad", "occurrence_id": "033820842039b0825fae3fe5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Time-use figures for the country are quoted from the Joint Time Use System (JTUS), Fourth Edition, the administrative collection covering Somalia in 2023, and they are reproduced without amendment.\n\nThe collection puts the mean hours per week that adults spend in paid work at 38 for men and 21 for women, the share of households in which no adult reports any paid work at 27 per cent, and the mean hours per week spent on collecting water and fuel at 11. None was recomputed here, and the collection's own categories for paid work, for unpaid domestic work and for water and fuel collection are reproduced with the figures.\n\nNo hour spent tending household livestock was moved into the paid column, and no adult working through an intermediary without a payment in the reference week was counted as employed. Where the collection reports for one region, the region is named with the figure rather than pooled with others.\n\nThe collection's table and line for each figure, the extraction date and the adults behind each estimate are listed in the annex notes.\n\nMention: \"Joint Time Use System (JTUS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Joint Time Use System supplies paid-work and unpaid-work hours for adults in the country.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:71925e2a3c1687e1081bf5a6", "occurrence_id": "d24a07432a333268456eb737", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Fuel poverty in this chapter is measured against the National Fuel and Energy Use Survey, the survey the International Trade Centre conducted in France in 2019, and that survey is the principal source for the claim.\n\nThe survey puts the share of households spending more than 10 per cent of their income on heating at 12 per cent, and the share reporting they left a room unheated through the winter at 19 per cent. Neither figure was recalculated here, and the survey's own definitions of household income and of an unheated room are reproduced with them.\n\nWhat the chapter adds is the comparison between the two, since 7 per cent of households appear in neither measure while reporting a bill they could not pay on time, and reading them together rather than separately is what makes the survey the basis for the section instead of one of its references.\n\nHouseholds whose income was reported as a band rather than an amount are excluded from the first share, 4 per cent of responses, because no ratio can be formed for them. The survey table, the extraction date and the households behind each share are carried with the annex notes.\n\nMention: \"National Fuel and Energy Use Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The National Fuel and Energy Use Survey is the basis for the chapter's fuel-poverty reading of income and heating.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:67ef3542833f60f07827ee76", "occurrence_id": "36f77e07b7d6ac184a46f952", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Enterprise size and household income were joined for this assessment from the Cross-Sectoral Enterprise and Jobs Survey (CEJS), conducted in Argentina in 2024, and the figures below are the assessment's own rather than the survey's.\n\nUnits were grouped by the number of workers they employ and the share whose owners report household income below the national adequacy line computed: 47 per cent of units with no outside workers, 29 per cent of those with one to four, and 8 per cent of those with more than twenty. The survey publishes employment and income items in separate tables, so no reading of them by unit size is available from it.\n\nIncome was then held at the same band for workers and owners and the gap between the two narrows to 6 percentage points, which tells the assessment that most of the distance is the size of the unit rather than the difference between wage and self-employment income.\n\nOwners reporting income from a source outside the unit were left out, 9 per cent of responses, since that income does not track the enterprise. The grouping rule, the income band used and the counts behind each share are set out with the annex tables.\n\nMention: \"Cross-Sectoral Enterprise and Jobs Survey (CEJS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Cross-Sectoral Enterprise and Jobs Survey provides unit records used to compare owner poverty by enterprise size.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:5ad853be1e60b751003b8f42", "occurrence_id": "ee7a9cddf3b7c69aa8e93a13", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Exam performance was reworked for this review from the Analytical Exam Performance System, the administrative system the UK Cabinet Office maintains in Vietnam in 2021, and the measures below are this review's calculations rather than the system's.\n\nSchools were grouped by the share of pupils eligible for schooling support and the mean subject score computed: 61 for schools above 40 per cent eligibility against 74 for those below 10 per cent. The system reports each pupil's score and each school's eligibility count without combining them, so no score by eligibility band appears in it.\n\nClass size was then taken from the same records and held constant, and the difference between those two groups narrows to 8 points, which leaves resourcing as the smaller part of the distance than the review's opening draft had assumed.\n\nSchools with fewer than twenty pupils entered for the examination were left out rather than shown as a mean, which removes 4 per cent of schools. The eligibility cut-points, the derivation and the pupil counts behind each mean are stated with the annex tables.\n\nMention: \"Analytical Exam Performance System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Analytical Exam Performance System provides exam and eligibility records used to compare scores by support band.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:006271fc35567ce95188080e", "occurrence_id": "e31d305502734b0ea13fe371", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Waiting times were measured for this assessment from the Harmonised Clinic Quality Survey (HCQS), Edition 6, which covers Peru in 2024, and the figures below are the assessment's own rather than the survey's.\n\nClinics were grouped by the number of clinicians on the roster and the share of consultations beginning within 30 minutes of the appointed hour computed: 71 per cent of clinics with six or more clinicians against 38 per cent of those with two or fewer. The survey reports waiting time per consultation without any grouping by staffing, so the comparison is not available from it.\n\nThat comparison supports one claim, that the additional posts should go to the under-staffed clinics rather than be spread across all of them. The budget figures the chapter works from come from the ministry's payroll series, to which the survey contributed nothing.\n\nClinics closed on more than a fifth of their scheduled days were left out, 6 per cent of those surveyed. The staffing cut-points and the counts behind each share are given with the annex tables.\n\nMention: \"Harmonised Clinic Quality Survey (HCQS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Harmonised Clinic Quality Survey supports one staffing claim by comparing clinic waiting times by roster size.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:dfdfdbe07cdc7f9ec638993d", "occurrence_id": "0309dae3e7791540e7fe5f0c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Referral times were computed for this assessment from the Monthly Maternal and Newborn Survey (MMNS), Wave 5, compiled by the US Census Bureau, and the measures below are this assessment's own rather than the survey's.\n\nFacilities were matched to their location and grouped by the road distance from the nearest referral hospital, and the share of births at a facility reaching a specialist within 2 hours computed: 22 per cent of facilities more than 60 kilometres from a referral hospital against 68 per cent of those within 20 kilometres. The survey records the distance for each facility and the outcome for each birth without ever reporting the two together.\n\nComplication rates were then read from the same records, and facilities at the longer distance report 11 complications per 100 births against 6 at the nearer facilities, which is the pairing the assessment treats as the case for the mobile referral service rather than the raw distance alone.\n\nFacilities with no recorded coordinates were left out, 3 per cent of those enumerated, since no distance could be formed for them. The distance bands, the derivation and the birth counts behind each share are stated with the annex tables.\n\nMention: \"Monthly Maternal and Newborn Survey (MMNS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Monthly Maternal and Newborn Survey provides facility records used to compare referral reach by distance.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:abcc34cad5c01a19ebab715c", "occurrence_id": "1ca0995c786b44f5fb1aca63", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two coverage figures in the statistical appendix are quoted from the High-Resolution Statistical Bulletin Registry (HSBR), Sixth Collection, compiled by ALCS, and both are reproduced as the registry records them.\n\nThe registry puts the share of its bulletins revised at least once after publication at 14 per cent and the median interval between first publication and that revision at 9 weeks. Neither was recomputed here, and the registry's own rule for what counts as a revision rather than a reprint appears with them.\n\nThe appendix itself is compiled from the producing offices' returns, so the registry accounts for two lines of it while the returns account for the remainder. A figure the registry gives for one series is shown with that series named rather than averaged across the collection.\n\nTheir registry entry, the extraction date and the bulletins standing behind each figure are set down in the annex log.\n\nMention: \"High-Resolution Statistical Bulletin Registry (HSBR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The High-Resolution Statistical Bulletin Registry supplies two revision figures quoted inside the appendix.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:47c78ca0626464f688b19b76", "occurrence_id": "d4723c35c9cac95b07ed7da5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Work histories were rebuilt for this assessment from the Sentinel Migration and Remittance Survey, which IMSS fielded in Tamil Nadu in 2019, and the measures below are this assessment's own rather than the survey's.\n\nHouseholds were grouped by whether any member had moved for work in the previous five years and the share of adults in regular paid employment computed: 58 per cent among households with such a move against 43 per cent among those without. The survey reports migration events and current employment in separate tables, so no reading of employment by migration history exists in it.\n\nRemittance receipt was then held constant and the difference narrows to 9 percentage points, which leaves the move itself doing more of the work than the money arriving afterwards, and that is the reading the assessment carries into its recommendation on skills support.\n\nHouseholds whose move was for schooling rather than work were left out, 7 per cent of the sample, since their employment status refers to a different decision. The grouping rule, the derivation and the household counts behind each share are stated with the annex tables.\n\nMention: \"Sentinel Migration and Remittance Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Sentinel Migration and Remittance Survey provides household records used to compare employment by work-move history.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:554aeea494c98ab5ff18bd46", "occurrence_id": "26f5cad67001b7155b099a96", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two service-coverage lines in the district annex are quoted from the Local Service Register Registry (LSRR), Version 8, the register maintained by DANE. for Johor, and both are reproduced without amendment.\n\nThe register puts the share of premises served by piped water at least 12 hours a day at 81 per cent and the share of streets with a collected waste service at 64 per cent. Neither was recalculated here, and the register's own counting rule for a served premises, which treats a standpipe as service only where a collection schedule is posted, appears beside the figures.\n\nThe district's own figures on these services come from the utility's billing records, so the register accounts for two lines of the annex while the billing records account for the rest. Where the register reports a figure for one sub-district, that sub-district is named rather than the figure shown for the district as a whole.\n\nThe register entry and line behind each figure and the extraction date are recorded with the annex notes.\n\nMention: \"Local Service Register Registry (LSRR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Local Service Register Registry supplies two utility coverage lines quoted inside the district annex.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:b6850c05f09773c5ccca3079", "occurrence_id": "633dc2bcd32cb5d3ce3bc544", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Training outcomes were analysed for this assessment from the Municipal Vocational Training Survey, which INDEC fielded in Aceh province in 2024, and the figures below are this assessment's calculations rather than values the survey publishes.\n\nGraduates were grouped by the trade they trained in and the share holding related work twelve months after completion computed: 52 per cent among those trained in vehicle repair against 31 per cent among those trained in general office work. The survey reports enrolment and current status without any cross-tabulation by trade, so no such share appears in it.\n\nPrior schooling was then taken from the same records and held constant, and the distance between the two trades narrows to 12 percentage points, which the assessment reads as evidence that the trade itself rather than the intake explains most of the gap.\n\nGraduates reporting further study at the time of interview were left out rather than counted as out of work, which removes 8 per cent of graduates. The grouping rule, the derivation and the graduate counts behind each share are carried with the annex tables.\n\nMention: \"Municipal Vocational Training Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Municipal Vocational Training Survey provides graduate records used to compare related work by trade.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:4358736b516e1250b7a92414", "occurrence_id": "914164810a9879cd7400202c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Lights and household income were matched for this assessment from the Sentinel Nighttime Lights Census, the census GTIS compiled for Bangladesh in 2019, and the measures below are this assessment's own rather than the census's.\n\nEach enumeration area was matched to its grid cell and the share of households reporting an income below the national adequacy line computed: 19 per cent of areas in the brightest third of the cell distribution against 52 per cent of those in the darkest third. The census reports radiance per cell and income per household in separate outputs, so no reading of poverty by brightness exists in either.\n\nElectrification was then read from the same records, and 91 per cent of households in the brighter areas report a connected supply against 44 per cent in the darker ones, which is the pairing the assessment uses to argue that the lights measure tracks access rather than preference.\n\nCells with cloud obscuration through the whole observation window were left out, 6 per cent of the area. The matching rule, the brightness bands and the household counts behind each share are stated with the annex tables.\n\nMention: \"Sentinel Nighttime Lights Census\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Sentinel Nighttime Lights Census provides grid cells matched to households to compare poverty by brightness.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:52d733c59103a7f11c6ac25d", "occurrence_id": "e511affe647a3df9f277d5ea", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The teacher figures in this chapter are drawn from the ministry's administrative returns, and the Endline Teacher Workforce Survey, conducted by MoWECP and BPS in Kyrgyzstan in 2020, is referred to only as background to them.\n\nThe survey reports that 68 per cent of teachers hold the qualification the post requires, that the mean teaching load stands at 19 hours a week, and that 22 per cent of teachers hold a second post. All three are given as the survey prints them, with its own definition of a qualified teacher stated, and none was used to build a chapter figure.\n\nNo table in the chapter rests on its responses. The figures the chapter argues from come from the ministry's returns, which count every post rather than sampling teachers, and where the two readings of workload differ the difference is noted rather than reconciled.\n\nThe annex lists the survey tables consulted and the extraction date, so the reader can see the extent of the background the survey provides here.\n\nMention: \"Endline Teacher Workforce Survey\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The Endline Teacher Workforce Survey is cited as background while the chapter's figures come from ministry returns.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:63e0d592e65b23a1bdb99776", "occurrence_id": "9dc204e02f29087da8270023", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Coverage was estimated for this assessment from the Harmonised Vaccination Campaign Dataset, the geospatial dataset CEDLAS-World Bank compiled for Chad in 2016, and the figures below are this assessment's own rather than the dataset's.\n\nSettlements were matched to their catchment polygon and the share of children completing the schedule computed: 73 per cent of settlements within 10 kilometres of a campaign post against 46 per cent of those beyond 30 kilometres. The dataset reports doses administered per post and the location of each settlement without producing any coverage figure by distance.\n\nNomadic rounds were then read from the same records, and settlements reached by a round in the same season reach 68 per cent regardless of distance, which is what the assessment treats as the operative variable rather than the distance itself.\n\nSettlements whose polygon could not be matched to a post, 4 per cent of those listed, were left out rather than assigned to the nearest one. The matching rule, the distance bands and the child counts behind each share are stated with the annex tables.\n\nMention: \"Harmonised Vaccination Campaign Dataset\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Harmonised Vaccination Campaign Dataset provides settlement polygons used to estimate coverage by distance.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:dd39f37123c976a6cb756ccb", "occurrence_id": "445ae79fcaea5a977ceaf57d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Water stress was measured for this chapter from the Regional Water Stress Collection (RWSC), Release 8, compiled by the PRS Group, and the measures below are this chapter's own rather than the collection's.\n\nBasins were grouped by the anomaly in the wet season's rainfall against the thirty-year reference and the share of their cropland left unharvested computed: 29 per cent of basins in the driest anomaly band against 7 per cent of those in the wettest. The collection reports the rainfall anomaly per basin and harvested area per district, so nothing of this kind appears in it.\n\nThat comparison supports one claim, that the supplementary seeding should be triggered on the rainfall anomaly rather than on reported losses. The chapter's cost figures rest on the ministry's procurement series, from which the collection supplied nothing.\n\nBasins whose gauges report fewer than twenty years of record were left out of the anomaly bands, 12 per cent of those listed. The anomaly bands, the reference period and the harvested areas behind each share are given with the annex tables.\n\nMention: \"Regional Water Stress Collection (RWSC)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Regional Water Stress Collection supports one trigger claim by comparing unharvested cropland by rainfall anomaly.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:0241653d3cee3fb2a88d82d4", "occurrence_id": "48f9f8c0386661ab3c5353ec", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Household hardship and schooling the numbers held for the city come from the Municipal Deprivation and Hardship Survey, which the German Socioeconomic Panel conducted in Buenos Aires in 2016, and all four are given as the survey releases them.\n\nThe survey puts the share of households that could not meet an unexpected expense at 34 per cent, the share behind on rent or utilities at 21 per cent, the share of children aged fifteen not enrolled in school at 6 per cent, and the share of households receiving any public transfer at 28 per cent. None was recomputed here, and the survey's own wording for an unexpected expense, for arrears and for enrolment appears with the figures.\n\nNo household reporting an expense it had paid by borrowing was counted as meeting it, and no child enrolled but not attending was counted as in school. Where the survey reports for one commune, the commune is named with the figure.\n\nThe survey table and line behind each figure, the extraction date and the households behind each share are recorded with the annex notes.\n\nMention: \"Municipal Deprivation and Hardship Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Municipal Deprivation and Hardship Survey supplies hardship, arrears, enrolment and transfer shares for the city.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:d7af1cf276378d7910e5077a", "occurrence_id": "6521f5143eae49dc195d3904", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Bargaining and work-condition figures for the country are taken from the Pilot Collective Bargaining Survey, which KTB fielded in Burundi in 2017, and the three figures below are carried over unchanged.\n\nThe survey puts the share of formal workplaces covered by a collective agreement at 23 per cent, the share of those agreements containing a clause on working hours at 61 per cent, and the share of covered workers whose pay was revised under a bargaining round in the previous year at 17 per cent. None was recomputed here, and the survey's own definitions of a covered workplace, of a clause and of a bargaining round are reproduced with them.\n\nNo workplace with an agreement under dispute was counted as covered, and no worker whose pay changed for an unrelated reason was counted under the third figure. Where the survey reports for one sector, the sector is named with the figure rather than pooled across the sample.\n\nThe survey tables and lines behind each figure, the extraction date and the workplaces behind each share are written up in the annex notes.\n\nMention: \"Pilot Collective Bargaining Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Pilot Collective Bargaining Survey supplies coverage, clause and pay-revision shares for formal workplaces.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:e57215823445a1578164dc07", "occurrence_id": "77e9bedf6c74da75419cb64a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Child deprivation was computed for this assessment from the Sentinel Child Wellbeing Survey (SCWS), Sixth Collection, compiled by the Bureau of Statistics, and the measures below are this assessment's own rather than the collection's.\n\nHouseholds were grouped by the income band recorded for them and the share of children lacking two or more of the collection's deprivation items computed: 38 per cent of children in the lowest band against 9 per cent of those in the highest. The collection reports each item as a separate prevalence, so no combined count by income band is available from it.\n\nHousehold size was then held constant and the difference narrows to 21 percentage points, which leaves the income band doing more of the work than the number of children in the household, and that is the reading the assessment carries into its recommendation on transfer value.\n\nChildren in households that did not answer the income module were left out rather than assigned to a band, 11 per cent of households. The item count used, the income bands and the child counts behind each share are stated with the annex tables.\n\nMention: \"Sentinel Child Wellbeing Survey (SCWS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Sentinel Child Wellbeing Survey provides household records used to count child deprivation by income band.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:ead4eee82bc0c57aae551136", "occurrence_id": "7d70de0f00a054de34a8ac23", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Elevation, work and income were joined for this assessment from the Multi-Topic Settlement and Elevation Census (MSEC), Third Collection, compiled by Michelin, and the figures below are this assessment's own rather than the census's.\n\nSettlements were assigned to an elevation band and the share of adults in paid work outside the settlement computed: 41 per cent of settlements above 1 500 metres against 57 per cent of those below 500 metres, while the share of households below the national adequacy line runs 39 per cent in the first group and 22 per cent in the second. The census holds elevation, employment and income as separate outputs, so no such joint reading of the three exists in it.\n\nRoad distance to the nearest market town was then held constant and both gaps close by roughly half, which tells the assessment that access rather than altitude explains most of the difference in work and income.\n\nSettlements with no elevation recorded were left out rather than placed at the band midpoint, 5 per cent of those listed. The bands used, the derivation and the settlement counts behind each share are stated with the annex tables.\n\nMention: \"Multi-Topic Settlement and Elevation Census (MSEC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Multi-Topic Settlement and Elevation Census provides settlement records linking altitude to work and household income.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:b37d7352df2f5e129d58dab2", "occurrence_id": "9a494d4644357515e825e654", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Hazard exposure was derived for this assessment from the Provincial Landslide Hazard Dataset (PLHD), the geospatial dataset covering Argentina in 2022, and the measures below are this assessment's own rather than the dataset's.\n\nHouseholds were matched to their slope class and the share reporting damage to a dwelling from a slip in the wet season computed: 14 per cent of households on slopes above 30 degrees against 3 per cent of those below 10 degrees. The dataset grades slope and hazard class per cell and reports no damage outcome, so the pairing is not available from it.\n\nRainfall anomaly for the season was then read from the same records, and households in the wettest anomaly band report damage at roughly twice the rate of those in the driest band at the same slope class, which is what the assessment treats as the trigger for the relocation programme rather than slope alone.\n\nHouseholds whose dwelling could not be matched to a cell, 4 per cent of those surveyed, were left out rather than assigned to the nearest cell. The slope bands, the anomaly bands and the household counts behind each share are carried with the annex tables.\n\nMention: \"Provincial Landslide Hazard Dataset (PLHD)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Provincial Landslide Hazard Dataset supplies slope cells matched to dwellings to compare slip damage by slope.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:587c523832c98d19a58f1bff", "occurrence_id": "20c08f907038cda694feac72", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The clinic-attendance figures in this chapter come from the ministry's facility returns, and the Representative Health and Household Survey (RHHS), Round 6, conducted for 2021 by Rogun PMG, is quoted only as context for them.\n\nThe survey puts the share of households reporting an unmet need for care at 18 per cent, the share of those who did not attend at all at 23 per cent, and the mean number of attendances per adult in the reference year at 1.9. All three are given as the survey prints them, with its own definition of unmet need stated, and no chapter figure was built from them.\n\nTheir purpose is to show what the population's own account of attendance looks like beside the facility counts. The chapter's argument rests on the returns, which record attendances rather than the reasons people give for not coming, and the difference between the two readings is noted rather than reconciled.\n\nThe survey tables consulted, the extraction date and the households behind the context figures appear in the annex, which is all the survey is used for here.\n\nMention: \"Representative Health and Household Survey (RHHS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Representative Health and Household Survey is quoted as context beside the ministry's attendance returns.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:5f99dcd8bde800bd99e14b93", "occurrence_id": "50d940405ed1e191cecff241", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Flood extent for the province is reported from the Local Flood Extent Dataset, the geospatial dataset CGAP compiled for China in 2023, and the four values below are carried over as the dataset releases them.\n\nThe dataset puts the area under water at the peak of the event at 1 240 square kilometres, the number of settlements intersected by the extent polygon at 63, the length of road covered for more than three days at 180 kilometres, and the share of that area falling inside the floodplain mapped a decade earlier at 78 per cent. None was recomputed here, and the dataset's own mapping rule, which treats open water as extent only where it persists across two observations, appears with them.\n\nNo area mapped as wetland in the base layer was counted as flood extent, and no settlement was credited with flooding unless its centroid falls inside the polygon. Where the dataset reports for one district, the district is named with the value.\n\nThe dataset's table and reference for each value, the extraction date, and the observations behind them are recorded in the annex notes.\n\nMention: \"Local Flood Extent Dataset\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Local Flood Extent Dataset supplies peak extent, settlement and road-cover values for the event.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:621071383c4e6e291fb5adc8", "occurrence_id": "0acf0c18b85f0a5321fd3d3f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two timeliness lines in the statistical annex are quoted from the Stratified Statistical Bulletin Index (SSBI), Wave 3, compiled by GBV IMS for 2019, and both are reproduced as the index records them.\n\nThe index puts the median interval between data collection and bulletin publication at 11 weeks and the share of bulletins published within their own announced schedule at 57 per cent. Neither was recalculated here, and the index's own rule for what counts as a collection month and what counts as a published bulletin appears beside them.\n\nThe annex's other lines are compiled from the offices' own production logs, so the index accounts for two rows while the logs account for the remainder. A figure the index gives for one series is shown with that series named rather than averaged across the wave.\n\nThe index entry and line behind each figure and the extraction date are recorded with the annex notes.\n\nMention: \"Stratified Statistical Bulletin Index (SSBI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Stratified Statistical Bulletin Index supplies two publication-timeliness lines quoted in the annex.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:45a34de2bdc445fa96be660a", "occurrence_id": "6646ba93a52c6f8d6188969c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Crowding was computed for this assessment from the District Housing Conditions Survey (DHCS), conducted in Mozambique in 2024, and the figures below are the assessment's own rather than the survey's.\n\nDwellings were grouped by the number of households sharing them and the share of residents sleeping in a room that also serves for cooking computed: 62 per cent of dwellings shared by two households against 24 per cent of those held by a single household. The survey reports occupants per dwelling and room use as separate prevalences, so nothing joins the two, and the survey's own definition of a dwelling, which treats a separate entrance as a separate dwelling, was kept rather than the plan's wider one.\n\nThat comparison supports one claim, that the rehousing plan should treat a shared dwelling as a single unit rather than each household inside it separately. The unit-cost figures the plan works from come from the works office, to which the survey supplied nothing at all.\n\nDwellings with a room count left blank were excluded rather than assumed to have a single room, which removes 6 per cent of those visited. The grouping rule and the household counts behind each share are given with the annex tables.\n\nMention: \"District Housing Conditions Survey (DHCS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The District Housing Conditions Survey supports one rehousing claim by comparing room use in shared dwellings.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:f32053c66244e5a7f2617336", "occurrence_id": "1542620638a2412aaa94544c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two transfer figures in the graduation annex are quoted from the Regional Graduation Programmes Report, which Groningen issued for Yemen in 2018, and both are reproduced as the report prints them.\n\nThe report puts the share of enrolled households receiving the full transfer value in the last quarter at 74 per cent and the share living more than 10 kilometres from the payment point at 19 per cent. Neither was recalculated here, and the report's own definition of a full transfer, which counts a partial payment as unmet, appears with them.\n\nThe annex itself is compiled from the implementing partners' payment records, so the report accounts for two lines while those records carry the rest. Where the report gives a figure for one governorate, that governorate is named rather than the figure shown for the programme as a whole.\n\nThe report's page and table for each figure, the extraction date and the households behind each share are recorded with the annex notes.\n\nMention: \"Regional Graduation Programmes Report\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Regional Graduation Programmes Report supplies transfer-delivery and payment-distance lines quoted in the annex.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:ae6078b8f1e1dc5cd10d20e8", "occurrence_id": "2bca180bcef189c713c2739d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Ownership and work were compared for this assessment from the Clustered Asset Ownership Registry, the database maintained by the International Water Association for Nepal in 2022, and the measures below are this assessment's own rather than the registry's.\n\nRegistered households were grouped by whether the title is recorded in a woman's name and the share of adults in that household reporting paid work outside the holding computed: 34 per cent of households with such a title against 26 per cent of those without. The registry records the title holder and the work status as separate fields and never reports them together.\n\nHousehold income band was then held constant and the difference narrows to 3 percentage points, which leaves income rather than the title itself as what the assessment reads as the operative factor in who works outside the holding.\n\nHouseholds whose registration carries no title field at all were left out rather than counted as untitled, 7 per cent of entries. The grouping rule, the income bands and the counts behind each share are stated with the annex tables.\n\nMention: \"Clustered Asset Ownership Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Clustered Asset Ownership Registry provides title records used to compare outside work by title holder.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:bd3e982e3c6abfaed9f5a378", "occurrence_id": "c6e2a27ff88f5afa20d89115", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The admission figures in this chapter are compiled from the regional hospital board's returns, and the Statistical Emergency Care Survey (SECS), conducted in Central Asia in 2024, is mentioned only as background to them.\n\nThe survey reports that 46 per cent of emergency attendances arrived by private transport, that the mean wait before a clinician sees a case stands at 52 minutes, and that 12 per cent of attendances leave without being seen. All three are given as the survey prints them, with its own counting rule for an attendance stated, and nothing in the chapter is calculated from them.\n\nNo chapter table rests on its responses. The figures the chapter argues from come from the board's returns, which record admissions rather than attendances that ended without one, and where the two readings differ the difference is noted rather than reconciled.\n\nThe annex lists the survey tables consulted and the extraction date, so the reader can see the limits of the background the survey provides here.\n\nMention: \"Statistical Emergency Care Survey (SECS)\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The Statistical Emergency Care Survey is background while the chapter's admission figures come from board returns.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:cb05d89819344aa2c16183ff", "occurrence_id": "cb7dcd41b23f48e8a1cf9a38", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two price lines in the market annex are quoted from the Sentinel Agri-Food Prices Survey (SAPS), which the US Census conducted for 2022, and both are reproduced without amendment.\n\nThe survey puts the mean retail price of maize at 34 per kilogram and the share of market towns quoting a price on more than three days a week at 58 per cent. Neither was recomputed here, and the survey's own definition of a quoted price, which requires a sale actually made rather than an asking figure, appears beside them. The distance between the farm-gate and retail prices the survey records for the same commodity is also given as the survey prints it, since the marketing margin it implies is discussed later in the chapter without being recalculated.\n\nThe annex is otherwise compiled from the ministry's market monitoring series, so the survey accounts for two lines of it while the monitoring series accounts for the remainder. Where the survey reports for one commodity, that commodity is named rather than averaged into a basket, and no price from a town with fewer than three quoted sales was carried into the annex at all.\n\nThe survey table and line behind each figure, the extraction date and the towns behind each share are detailed in the annex notes.\n\nMention: \"Sentinel Agri-Food Prices Survey (SAPS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Sentinel Agri-Food Prices Survey supplies price lines quoted beside the ministry's monitoring series.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:18e29dcfa5289051a0d0282e", "occurrence_id": "d4900c1135d1180e037627b9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The population figure for the capital district in this chapter is the office's own, taken from its register-based estimate, and the Enumerative Population Grid Dataset (EPGD), Round 9, compiled by Global Financial Data for 2015, is cited to test it.\n\nThe office's estimate puts the district at 1.42 million residents. The grid dataset puts it at 1.51 million for the same year on its own allocation, which distributes a national total across grid cells by built-up area. The office counts residents at their registered address while the dataset counts people assigned to a cell, and no adjustment was made to align the two.\n\nThe gap of 90 000 falls within the band the dataset publishes for its allocated cells, so the office's figure stands as printed and the grid supplies no values for the districts the register does not cover.\n\nCells that straddle the district boundary are excluded from the comparison, since neither source assigns their population to one side. The allocation rule each source uses and the boundary each applies are recorded with the annex tables.\n\nMention: \"Enumerative Population Grid Dataset (EPGD)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Enumerative Population Grid Dataset checks the office's register-based district population against its grid allocation.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:4c4b6dd6fc54a5de4c600813", "occurrence_id": "cfcb76ed6e3113def1dd6617", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The claim records were catalogued for the insurance file by the Representative Unemployment Insurance Registry (RUIR), Edition 6, the database maintained by the Central Bank of Peru, and what was done to them is handling rather than measurement.\n\nEach record was matched to the file's own sector code, checked against the employer's return and, where the two disagreed, flagged for the officer rather than recoded automatically. Records with no employer identifier were left uncoded rather than assigned to a residual sector, and 2 310 of them entered the file in that condition. The registry's own claim date was carried with every record so that later editions could be matched without re-requesting the employer's details.\n\nThe insurance file's series is built from the ministry's monthly returns, and this cataloguing supports that work in one respect only, by keeping registry claims under the same sector codes so that a claim and a return for the same employer can be found in both. No estimate in the file is calculated from the registry's records themselves.\n\nThe coding sheet, the count of records flagged and the date the transfer was deposited are carried in the annex log.\n\nMention: \"Representative Unemployment Insurance Registry (RUIR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The Representative Unemployment Insurance Registry supplies claim records coded into the insurance file's sector crosswalk.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:3f0835163ba31834de650009", "occurrence_id": "77225a3cd5c5c4ee524c23b9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Water and rainfall country figures are assembled from the Standardised Water and Climate System (SWCS), the administrative system maintained for Republic of Zambia in 2018, and the four values below are given as the system holds them.\n\nThe system puts the number of functioning boreholes on its inventory at 3 840, the share of its rain gauges reporting in every month of the wet season at 62 per cent, the mean depth to water at the monitoring points at 24 metres, and the share of its catchments where abstraction exceeds the annual recharge estimate at 11 per cent. None was recalculated here, and the system's own definitions of a functioning borehole and of a reporting gauge appear with the values.\n\nNo borehole reported as under repair was counted as functioning, and no gauge with a gap of more than one month was counted as reporting. Where the system reports for one catchment, that catchment is named with the value rather than pooled with the others.\n\nThe system's inventory reference for each value, the extraction date as well as the points behind them are kept in the annex notes.\n\nMention: \"Standardised Water and Climate System (SWCS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Standardised Water and Climate System supplies borehole, gauge and catchment values for the country.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:9603f4f93421965caffb31db", "occurrence_id": "ca51d9d9d72b03bf312fcfa0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Neighbourhood density was rebuilt for this assessment from the Combined Population Grid Dataset (CPGD), the geospatial dataset compiled by the Penn World table for Fortaleza, and the measures below are this assessment's own rather than the dataset's.\n\nGrid cells were assigned to their administrative neighbourhood and the mean number of residents per hectare computed: 218 in the cells the dataset assigns to the northern neighbourhoods against 96 in those assigned to the southern ones. The dataset publishes population per cell without any neighbourhood assignment, so no reading of density by neighbourhood exists in it.\n\nBlock length was then taken from the same grid, and cells with shorter blocks and denser cover report 41 more residents per hectare than cells of the same neighbourhood with longer blocks, which is the pairing the assessment uses to argue that the density difference follows the street pattern rather than the neighbourhood name.\n\nCells whose population was allocated rather than enumerated were left out, 6 per cent of the grid. The assignment rule, the density derivation and the cell counts behind each figure are stated with the annex tables.\n\nMention: \"Combined Population Grid Dataset (CPGD)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Combined Population Grid Dataset provides grid cells used to compute residential density by neighbourhood.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:553a83a5c39926d7e03ce776", "occurrence_id": "8ca6b533ae8daaf923192d9d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Distress was measured for this assessment from the Multi-Topic Mental Health Survey, which Open Data Watch conducted in the Dominican Rep. in 2023, and the figures below are this assessment's own rather than the survey's.\n\nAdults were grouped by the household income band recorded for them and the share reporting frequent anxiety symptoms computed: 27 per cent of adults in the lowest band against 11 per cent of those in the highest. The survey reports symptoms as a single prevalence and income as a separate item, so no reading of symptoms by income band is available from it.\n\nWhether the household had missed a rent or utility payment in the previous six months was then held constant, and the difference between the two income bands narrows to 5 percentage points, which tells the assessment that the strain of falling behind on bills, rather than income level on its own, is what the survey's health measure is picking up.\n\nAdults who did not complete the symptom module were left out rather than counted as asymptomatic, 9 per cent of respondents. The income bands, the symptom threshold and the counts behind each share are given with the annex tables.\n\nMention: \"Multi-Topic Mental Health Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Multi-Topic Mental Health Survey provides adult records used to compare anxiety symptoms by income band.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:09133029b04f04cc080337f6", "occurrence_id": "f35fac9740a5980065f57575", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Household needs and farm figures for the province are taken from the Clustered Minimum Needs Survey (CMNS), conducted in Aceh province in 2016, and all four are carried over as the survey releases them.\n\nThe survey puts the share of households unable to meet a minimum food basket at 29 per cent, the share selling a productive asset to cover the gap at 14 per cent, the mean holding operated by each household at 0.8 hectares, and the share of that holding planted to food crops at 71 per cent. None was recomputed here, and the survey's own basket, its test for a productive asset and its definition of an operated holding appear with the figures.\n\nNo household receiving food assistance was counted as meeting the basket, and no land held but not planted in the season was counted as operated. Where the survey reports for one cluster of camps, that cluster is named with the figure.\n\nThe survey table and line behind each figure, the extraction date and the households behind each share are recorded with the annex notes.\n\nMention: \"Clustered Minimum Needs Survey (CMNS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Clustered Minimum Needs Survey supplies food-basket, asset-sale and holding figures for the province.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:9954ce9e49e891e9e1cf02de", "occurrence_id": "88962ac84b90936b7f5c2f1b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The coping figures in this chapter are calculated from the programme's own household panel, and the Pilot Shock Coping Survey, which Economist conducted in Mexico in 2020, is quoted only to place those figures in context.\n\nThe survey puts the share of households that cut spending on food after a shock at 37 per cent, the share borrowing from relatives at 22 per cent, and the share drawing down savings at 18 per cent. All three are given as the survey prints them, with its own definition of a shock stated, and none was used to build a chapter figure.\n\nTheir purpose is to show how a separately sampled population describes the same behaviour the panel records. The chapter's argument rests on the panel, which observes the household's expenditure rather than its account of it, and where the two readings of coping differ the difference is noted rather than reconciled.\n\nThe annex lists the survey tables consulted, the extraction date and the households behind the context figures, which is the whole use made of the survey here.\n\nMention: \"Pilot Shock Coping Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Pilot Shock Coping Survey is quoted for context beside the panel's own coping figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:2b78d536d9ace88cc3d87892", "occurrence_id": "5d4fa95dd93c19e1c91c6c12", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Site conditions were compared for this assessment from the Baseline Settlement Location Report (BSLR), issued by the CCCM Cluster for 2017, and the measures below are this assessment's own rather than the report's.\n\nSites were grouped by the share of their residents displaced more than once and the share of households reporting income below the national adequacy line computed: 58 per cent of sites where more than a third had moved repeatedly against 34 per cent of those where fewer than a tenth had. The report lists displacement history for some sites and household income for others, so the two never appear together in it.\n\nDistance to the nearest market town was then read from the same locations, and sites beyond 25 kilometres report poverty at 9 percentage points above sites of the same displacement history nearer in, which is the combination the assessment treats as the case for locating the livelihood programme at the outlying sites.\n\nSites with fewer than fifty households enumerated were left out rather than reported as a share, 7 per cent of those listed. The grouping rule, the distance bands and the household counts behind each share are stated with the annex tables.\n\nMention: \"Baseline Settlement Location Report (BSLR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Baseline Settlement Location Report provides site records comparing repeat displacement and household poverty.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:f9f6e9fb2f37d8d5c17413ac", "occurrence_id": "bf46dda786a7519ad3b91dcc", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The registration figures in this chapter come from the agency's own arrival log, and the Stratified Arrival and Registration Survey, which Lederman y Saenz conducted in Moldova in 2017, is quoted only as context for them.\n\nThe survey reports that 41 per cent of arrivals registered within seven days of crossing, that 26 per cent never registered at all, and that the mean stay before moving on again stands at 34 days. All three are given as the survey prints them, with its own definition of an arrival stated, and no chapter figure was derived from them.\n\nTheir purpose is to show what the log's coverage looks like from outside it. The chapter's argument rests on the log, which counts only those who present themselves, and the share the survey reports as never registering is the part the log cannot see at all.\n\nThe survey tables consulted, the extraction date and the arrivals behind the context figures are held in the annex, and the survey is used for nothing else here.\n\nMention: \"Stratified Arrival and Registration Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Stratified Arrival and Registration Survey is quoted as context for the coverage of the agency's arrival log.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:631b069a00bebafe2ddfae79", "occurrence_id": "96488bcd107b88c530c7e98a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Hiring and vacancy figures spanning the country are taken from the Clustered Hiring and Vacancies Survey (CHVS), Wave 2, which covers Chile in 2018, and the three figures below are given exactly as the survey releases them.\n\nThe survey puts the share of establishments with at least one vacancy open more than four weeks at 19 per cent, the mean number of applicants per filled post at 6.2, and the share of hires made without any advertised post at 44 per cent. None was recalculated here, and the survey's own definitions of an open vacancy, of an applicant and of an unadvertised hire are reproduced with the figures.\n\nNo post frozen pending a restructuring was counted as vacant, and no enquiry that did not lead to an application was counted as an applicant. Where the survey reports for one size class of establishment, that class is named with the figure rather than pooled.\n\nThe survey table and line for each figure, the extraction date and the establishments behind each share are laid out in the annex notes.\n\nMention: \"Clustered Hiring and Vacancies Survey (CHVS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Clustered Hiring and Vacancies Survey supplies vacancy duration, applicant and unadvertised-hire shares.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:0713309a2b267ba3602673cb", "occurrence_id": "103cb18d58ab26a8bf4b969e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Unpaid work was computed for this assessment from the Endline Time Use Survey (ETUS), Wave 3, compiled by the Uganda Bureau of Statistics, and the measures below are this assessment's own rather than the survey's.\n\nAdults were grouped by whether they report any paid work in the reference week and the mean hours spent on unpaid care computed: 31 hours among adults with no paid work against 12 among those with full-time paid work. The survey reports each activity's hours as separate totals without any grouping by employment status, so no such comparison appears in it.\n\nThe number of children under six in the household was then held constant and the gap between the two groups narrows to 7 hours, which leaves the presence of paid work doing more of the separation than the care burden itself, and that is the reading the assessment carries into its proposal on childcare places.\n\nAdults who did not complete the activity diary were left out rather than assigned zero hours, 13 per cent of respondents. The grouping rule, the diary threshold and the counts behind each mean are given with the annex tables.\n\nMention: \"Endline Time Use Survey (ETUS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Endline Time Use Survey provides diary records used to compare unpaid care hours by paid-work status.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:879854c4bc7bd16d03bf1fb9", "occurrence_id": "762266aaced073cd73d4fdbe", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Consumption and yields were joined for this assessment from the High-Frequency Living Standards Survey (HLSS), Release 5, compiled by FFMS, and the figures below are this assessment's own rather than the release's.\n\nHouseholds were grouped by the yield their main plot produced against the district's median and the share whose food spending fell below the minimum basket computed: 44 per cent of households below half the median yield against 12 per cent of those above it. The release reports consumption and agricultural items in separate tables, so no reading of poverty by yield is available from it.\n\nWhether the household also received a transfer was then held constant, and the difference between the two yield groups narrows to 18 percentage points, which the assessment reads as evidence that the transfer cushions but does not close the distance a poor harvest opens.\n\nHouseholds reporting no plot at all were left out rather than placed in the lowest yield group, 9 per cent of the sample. The yield bands, the basket used and the counts behind each share are stated with the annex tables.\n\nMention: \"High-Frequency Living Standards Survey (HLSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The High-Frequency Living Standards Survey links plot yields to food-basket adequacy for the assessment.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:39e14e048cdd5c7c1f48bec8", "occurrence_id": "ee66450fa4c224dfa2ef5c4c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Work and schooling were compared for this assessment from the Cross-Sectoral Rural Livelihoods Survey, which Fed fielded in Lagos State in 2022, and the measures below are this assessment's own rather than the survey's.\n\nAdults were grouped by the highest qualification they completed and the share reporting steady work in the previous month computed: 69 per cent of those completing secondary schooling against 41 per cent of those leaving before it. The survey reports attainment and current activity as separate distributions, so no reading of work by qualification exists in its tables.\n\nWhether the adult lived more than 10 kilometres from a market town was then held constant, and the difference between the two groups narrows to 14 percentage points, which tells the assessment that distance from work opportunities accounts for a substantial part of what looks like a qualification effect.\n\nAdults reporting full-time study were left out rather than counted as out of work, 6 per cent of respondents. The qualification bands, the distance cut-point and the counts behind each share are recorded with the annex tables.\n\nMention: \"Cross-Sectoral Rural Livelihoods Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Cross-Sectoral Rural Livelihoods Survey provides adult records comparing steady work by schooling and distance.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:44511395c71dba00f77855df", "occurrence_id": "ed9e384f4a7412de86ed9374", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two erosion lines in the coastal annex are quoted from the Household Coastal Erosion Index (HCEI), compiled for Gabon in 2018, and both are reproduced as the index prints them.\n\nThe index puts the share of households in the coastal band reporting shoreline loss behind their dwelling at 31 per cent and the mean annual retreat recorded along the surveyed stretch at 1.8 metres. Neither was recalculated here, and the index's own rule for counting a household as exposed, which requires the loss to fall within the compound boundary, appears with the figures.\n\nThe annex is otherwise compiled from the ports authority's survey series, so the index accounts for two lines while that series carries the rest. A figure the index gives for one settlement is shown with the settlement named rather than averaged along the coast.\n\nThe index entry behind each figure, the extraction date and the surveyed points behind each value are recorded with the annex notes.\n\nMention: \"Household Coastal Erosion Index (HCEI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Household Coastal Erosion Index supplies two shoreline-loss lines quoted beside the ports authority series.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:0cd438496556f1d0747372f5", "occurrence_id": "09033a024494fe4ba01034ec", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Housing cost burden was derived for this assessment from the High-Frequency Housing Conditions Survey (HHCS), compiled by BOOST for Uruguay, and the figures below are this assessment's own rather than the survey's.\n\nRenting households were grouped by the share of income their rent consumes and the share reporting they skipped a meal for money at least once in the month computed: 22 per cent of households above 40 per cent of income against 6 per cent of those below 20 per cent. The survey reports rent and food insufficiency as separate items without any such cross-classification.\n\nThat comparison supports one claim, that the rent supplement should be means-tested on the rent-to-income ratio rather than on income alone. The fiscal figures the proposal works from come from the treasury's budget note, to which the survey supplied nothing.\n\nHouseholds not reporting an income amount were left out rather than assumed to be below the line, 8 per cent of renters. The ratio bands and the counts behind each share are given with the annex tables.\n\nMention: \"High-Frequency Housing Conditions Survey (HHCS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The High-Frequency Housing Conditions Survey supports one rent-support claim by comparing rent burden and skipped meals.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:6fe085465898096a75fd21f6", "occurrence_id": "89c9ca6d0ccee1dc694ded78", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Public works participation figures are reported from the Nationwide Public Works System (NPWS), the administrative system OECD-PISA maintains for 2023, and the four values below are given as the system holds them.\n\nThe system puts the number of work-days offered under the programme at 41 million, the share of registered workers who took at least one assignment at 63 per cent, the mean number of days worked by those who took one at 27, and the share of districts whose allocation was fully used by the end of the year at 48 per cent. None was recalculated here, and the system's own definitions of an offered work-day and of a used allocation appear with the values.\n\nNo day cancelled for weather was counted as offered, and no worker registered in two districts was counted twice. Where the system reports for one province, that province is named with the value rather than pooled into the national figure.\n\nThe system's return reference for each value, plus the extraction date and the districts behind them are filed in the annex notes.\n\nMention: \"Nationwide Public Works System (NPWS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Nationwide Public Works System supplies work-day, participation and allocation-use values for the programme.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:e3235d8d8178e9bf69e4fe8f", "occurrence_id": "616f7accf73a729cbd78d0ad", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Completion of apprenticeships was rebuilt for this assessment from the High-Frequency Apprenticeship Survey (HAS), Fifth Collection, compiled by IEA, and the measures below are this assessment's own rather than the collection's.\n\nApprentices were grouped by the size of the establishment training them and the share reaching the final assessment computed: 71 per cent in establishments with more than fifty workers against 43 per cent in those with fewer than ten. The collection reports enrolment and final outcomes as separate counts, so no completion rate by establishment size appears in it.\n\nThe trade trained in was then held constant and the difference narrows to 12 percentage points, which leaves the size of the establishment as the smaller part of the distance than the trade itself, and that is the reading the assessment carries into its proposal on mentoring support for small firms.\n\nApprentices whose record shows a transfer between establishments were left out rather than counted twice, 7 per cent of records. The grouping rule, the derivation and the counts behind each share are stated with the annex tables.\n\nMention: \"High-Frequency Apprenticeship Survey (HAS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The High-Frequency Apprenticeship Survey provides records used to compute completion by establishment size.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:7b17804d41acda8ab36476fe", "occurrence_id": "8a5b7990eb385b58991de718", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The inflation figure in this chapter is the statistical office's own, taken from its published series, and the Annual Statistical Bulletin Report, issued by the Central Bank for Ethiopia in 2015, is cited to test it.\n\nThe office's series puts the annual increase in the consumer basket at 14 per cent. The bulletin reports 16 per cent for the same year on its own compilation. The office weights the basket by its household expenditure survey while the bulletin uses the weights carried over from the previous round, and neither weighting was altered to bring the two together.\n\nThe distance between 14 and 16 sits inside the band the bulletin itself notes around its compiled figure, so the office's series is printed as published and the bulletin supplies no revision to it and no values for the months the series lacks.\n\nExcluded from the comparison are the two components the bulletin reports only as a combined item, since no separate movement can be taken from them. The basket weights each source applies and the months each covers are recorded with the annex tables.\n\nMention: \"Annual Statistical Bulletin Report\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Annual Statistical Bulletin Report checks the office's consumer-basket inflation figure without revising it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:e7902716b35e28a9732a5b71", "occurrence_id": "4a6a312022de014fa444020d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Support and schooling were compared for this assessment from the National Disability and Support Registry (NDSR), Collection 8, the database maintained by FAPRI, and the measures below are this assessment's own rather than the registry's.\n\nRegistered children were grouped by the household income band on their record and the share enrolled in a mainstream school computed: 38 per cent of children in the lowest band against 66 per cent of those in the highest. The registry holds the income band for enrolment purposes and the enrolment status as operational fields without reporting the two together.\n\nWhether the record notes a mobility difficulty was then held constant, and the difference between the two income bands narrows to 15 percentage points, which the assessment reads as evidence that cost, rather than the difficulty itself, keeps most of these children out of mainstream classrooms.\n\nChildren whose record carries no income band were left out rather than assigned to the lowest, 11 per cent of registrations. The bands used, the derivation and the counts behind each share are given with the annex tables.\n\nMention: \"National Disability and Support Registry (NDSR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The National Disability and Support Registry provides registrations used to compare mainstream enrolment by income band.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:3bbc39798303b882a09ea3cd", "occurrence_id": "0c992b9b79310401d0e830ca", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Care access was measured for this assessment from the Combined Reproductive Health Survey (CRHS), Eighth Phase, compiled by CONDUSEF, and the figures below are this assessment's own rather than the survey's.\n\nHouseholds were grouped by whether they had been displaced within the previous two years and the share of women reporting an antenatal visit in each of the expected contacts computed: 54 per cent among displaced households against 73 per cent among those that had not moved. The survey reports displacement and maternal contacts in separate sections, so no reading of care by displacement history is available from it.\n\nDistance to the nearest facility was then taken from the same records and held constant, and the difference between the two groups narrows to 8 percentage points, which leaves the interruption of registration and follow-up, rather than the distance itself, as what the assessment reads as the operative cause.\n\nHouseholds with no recorded move date were left out rather than counted as not displaced, 6 per cent of the sample. The grouping rule, the distance bands and the counts behind each share are stated with the annex tables.\n\nMention: \"Combined Reproductive Health Survey (CRHS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Combined Reproductive Health Survey compares antenatal contact coverage for displaced and non-displaced households.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:e37832fb2b3afcc3615d0961", "occurrence_id": "785099a27e34d8a9f6fd6662", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The household income figure in this chapter is the team's own, computed from its survey module, and the Joint Socioeconomic System (JSS), Phase 9, the administrative collection maintained by NISR, is cited to test it.\n\nThe team's module puts the share of households below the national adequacy line at 24 per cent. The system's own estimate for the same period is 27 per cent, built from tax and transfer records rather than from a reported income figure. The team asked households for a monthly figure while the system reconstructs income from administrative payments, and no reconciliation was applied to either.\n\nThe 3 percentage points between them fall inside the sampling interval the team's module carries, so the team's figure is reported as computed and the system is not used to adjust it or to extend the module's coverage.\n\nHouseholds with no tax identifier are excluded from the comparison, since the system has no record for them and the team's module cannot be matched to one. The income definition each source applies is recorded with the annex tables.\n\nMention: \"Joint Socioeconomic System (JSS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Joint Socioeconomic System checks the team's adequacy-line income figure against administrative income records.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:9066a8f3aa765baaada99bcb", "occurrence_id": "976cec27ff1e3b2d9d2f7a2b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The seasonal employment figure in this chapter is the union's own, taken from its member returns, and the Stratified Unemployment Insurance Survey (SUIS), First Round, compiled by UNTAD for 2018, is cited to test it.\n\nThe union's returns put the share of agricultural members without paid work in the lean month at 38 per cent. The survey puts it at 42 per cent for the same period on its own sampling. The union counts members who drew no benefit in the month while the survey counts anyone reporting no paid hours, and neither definition was changed to close the gap.\n\nThe distance between 38 and 42 is smaller than the interval the survey reports for its rural stratum, so the union's figure stands as reported and the survey supplies no adjustment to it and no values for the months the returns do not cover.\n\nExcluded are members the survey places in non-agricultural work, since the union's figure refers only to those whose main occupation is farming. The event each source counts as work and the strata each covers are set out with the annex tables.\n\nMention: \"Stratified Unemployment Insurance Survey (SUIS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Stratified Unemployment Insurance Survey checks the union's lean-season work figure for farm members.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:f8726768bd0f464203cb2350", "occurrence_id": "6a223487fcabd16a0ddbb9a5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two maternal lines in the health annex are quoted from the Government Demographic and Health Registry (GDHR), Edition 6, the register REACH keeps for Sindh, and both are reproduced as the registry holds them.\n\nThe registry puts the share of births attended by a trained clinician at 76 per cent and the share of facilities posting a delivery register at 63 per cent. Neither was recalculated here, and the registry's own rule for counting an attended birth, which requires a clinician recorded against the entry, appears beside the figures.\n\nThe annex's other lines are compiled from the province's monthly health returns, so the registry accounts for two lines while the returns carry the rest. A figure the registry gives for one district is shown with that district named rather than averaged across the register.\n\nThe registry entry and line behind each figure and the extraction date are recorded with the annex notes.\n\nMention: \"Government Demographic and Health Registry (GDHR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Government Demographic and Health Registry supplies two maternal-care lines quoted in the health annex.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:3d25d8ef529ff0640ab04f92", "occurrence_id": "86ce9f1b055773f00fe346ad", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Whether the mission's poverty figure for displaced households holds was tested against the Continuous Refugee and Migrant System (CRMS), Second Wave, the administrative register FEPB maintains, and the register is the principal source for that test.\n\nThe mission's own assessment puts the share of registered households living below the national adequacy line at 47 per cent. The register, which records each household's declared income at registration and its benefit status since, puts the equivalent share at 55 per cent, because it counts households that have stopped collecting benefits as still reporting their declared income. The mission surveyed a sample during one month while the register carries every registration since the programme began.\n\nThe register is therefore what this section relies on for the poverty claim, and the mission's lower figure is reported beside it so the reader can see how much of the difference is the coverage of lapsed households rather than a different population. Where a household appears in both, the register's declared income is used and the mission's figure is not averaged with it.\n\nThe benefit rule each source applies and the households each counts are set out with the annex tables.\n\nMention: \"Continuous Refugee and Migrant System (CRMS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Continuous Refugee and Migrant System is the principal source testing the mission's poverty figure for displaced households.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:0576c3f96dc2cb56d75b169c", "occurrence_id": "1d7b817b8cd25c4f9549a19c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The settlement figures in this chapter are compiled from the office's own gazetteer, and the Endline Output Area Registry, maintained by Heston, Summers, and Aten in Mongolia in 2024, is referred to only as background to them.\n\nThe registry reports that 1 940 output areas are defined across the country, that the mean population assigned to each stands at 1 180 residents, and that 8 per cent of areas have no settled population at all. All three are given as the registry prints them, with its own rule for defining an area stated, and no chapter figure was built from them.\n\nNo table in the chapter rests on its entries. The figures the chapter argues from come from the gazetteer, which lists settlements by name rather than allocating residents to boundaries, and the two ways of dividing the country are described side by side rather than merged.\n\nThe annex lists the registry pages consulted and the extraction date, so the reader can see the extent of the background the registry provides here.\n\nMention: \"Endline Output Area Registry\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The Endline Output Area Registry is background while the chapter's settlement figures come from the gazetteer.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:39a76aec711ec4419445f074", "occurrence_id": "2b2830f2ba2096c0aa06cc88", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Facility readiness was computed for this assessment from the Statistical Service Delivery and Readiness Survey (SSDRS), Second Phase, compiled by CASEN, and the measures below are this assessment's own rather than the survey's.\n\nFacilities were grouped by whether they report a supply chain without a stock-out in the previous quarter and the share able to manage a complicated birth computed: 58 per cent of facilities without a stock-out against 31 per cent of those reporting one. The survey reports stock status and each readiness item separately, so no readiness figure crossed with supply is available from it.\n\nThe number of clinicians rostered was then held constant and the difference narrows to 11 percentage points, which leaves the supply interruption doing more of the work than staffing, and that is the reading the assessment carries into its proposal on buffer stocks.\n\nFacilities closed during the survey window were left out rather than counted as unready, 4 per cent of those listed. The grouping rule, the readiness items used and the counts behind each share are stated with the annex tables.\n\nMention: \"Statistical Service Delivery and Readiness Survey (SSDRS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Statistical Service Delivery and Readiness Survey links stock-outs to the ability to manage a complicated birth.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:a13385aaa7f75f64cf92cfbd", "occurrence_id": "e35c4e0f82e7d654a4e0a822", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Household operations figures for the programme are taken from the Household Operations Index Registry, the database UNHCR Global Appeal maintains for Jordan in 2019, and the four values below are carried over as the registry holds them.\n\nThe registry puts the share of households with a file complete enough to be processed without a follow-up visit at 72 per cent, the mean number of days from application to decision at 41, the share of decisions requiring a second verification at 18 per cent, and the share of households whose file was last touched more than a year before the count at 23 per cent. None was recalculated here, and the registry's own definitions of a complete file and of a verification round appear with the values.\n\nNo file awaiting an applicant document was counted as complete, and no decision reopened on appeal was counted as a second verification. Where the registry reports for one office, that office is named with the value rather than pooled across the programme.\n\nThe registry table and line behind each value, the extraction date and the files behind each share are recorded in the annex notes.\n\nMention: \"Household Operations Index Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Household Operations Index Registry supplies file completeness, decision time and verification values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:999c15748ec176a2c9b08e32", "occurrence_id": "0a8dc05a5811849049e17288", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The caseload figure in this chapter is the secretariat's own, drawn from its case-tracking sheet, and the Quarterly Integrated Records Registry, compiled by IPEA for South Asia in 2018, is cited to test it.\n\nThe secretariat's sheet puts open cases at the end of the quarter at 12 400. The registry gives 13 100 for the same quarter on its own counting. The secretariat excludes cases transferred between offices while the registry keeps the receiving office's entry, and no adjustment was made to either count.\n\nThe difference of 700 cases is smaller than the reconciliation allowance the registry itself publishes between its quarterly extracts, so the secretariat's figure is printed as tracked and the registry supplies no correction to it and no values for the quarters the sheet does not cover.\n\nCases recorded in the registry with no office attached are excluded from the comparison, since they cannot be attributed to either count. The transfer rule each source applies is recorded with the annex tables.\n\nMention: \"Quarterly Integrated Records Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Quarterly Integrated Records Registry checks the secretariat's open-case count for the quarter.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:3ee5613db2847e1b185d471a", "occurrence_id": "ee507ff84f71907de3f1d283", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Service coverage figures are reported from the Household Service Register Census (HSRC), Fifth Wave, which covers the Arab Republic of Egypt in 2020, and the three figures below are given exactly as the census releases them.\n\nThe census puts the share of households within 500 metres of a collected waste point at 84 per cent, the share with a contract for a piped supply at their dwelling at 69 per cent, and the share reporting no service of any kind within 2 kilometres at 7 per cent. None was recomputed here, and the census's own definitions of a collection point and of a contracted supply are reproduced with the figures.\n\nNo point serving a commercial premises only was counted for a household, and no dwelling connected to a tanked supply was counted as piped. Where the census reports for one governorate, that governorate is named with the figure rather than shown for the country.\n\nThe census table and line behind each figure, the extraction date and the households behind each share are stored in the annex notes.\n\nMention: \"Household Service Register Census (HSRC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Household Service Register Census supplies waste, piped-supply and no-service coverage shares.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:bc2d990d032ea788f0ce816c", "occurrence_id": "0a2f07dc8dc283b2f6c1d17a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Seed use figures for the crop year are reported from the Panel Seed Use Registry (PSUR), Fifth Release, the database maintained by UNTAD, and all four are given as the registry holds them.\n\nThe registry puts the share of registered plantings using certified seed at 46 per cent, the mean quantity sown per hectare at 14 kilograms, the share of holdings saving seed from the previous harvest at 38 per cent, and the share of registered suppliers holding at least one lot below germination standard at 6 per cent. None was recalculated here, and the registry's own definitions of a registered planting and of a saved-seed holding appear beside the figures.\n\nNo planting sown from a lot withdrawn from the register was counted as certified, and no holding reporting both saved and certified seed was counted twice. Where the registry reports for one crop, the crop is named rather than merged into a cereals aggregate.\n\nThe registry's table and line for each figure, the extraction date and the plantings behind each share are recorded with the annex notes.\n\nMention: \"Panel Seed Use Registry (PSUR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Panel Seed Use Registry supplies certified-seed use, sowing rate, seed-saving and lot-quality shares.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:763c78242a77ec34be94c5c9", "occurrence_id": "7a2f7a7fcd31aad65f3d3807", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Consumption growth was computed for this assessment from the Longitudinal Household Consumption Census (LHCC), Edition 4, compiled by the Penn World table, and the figures below are this assessment's own rather than the census's.\n\nHouseholds followed across both editions were grouped by whether their reported spending fell in real terms and the share whose food share of spending rose computed: 71 per cent of households whose spending fell against 22 per cent of those whose spending held. The census publishes each edition's spending distribution without following households between them, so no such movement appears in it.\n\nThat comparison supports one claim, that the transfer should be indexed to food prices rather than to the general basket. The fiscal cost the proposal quotes comes from the treasury's note, to which the census supplied nothing at all.\n\nHouseholds appearing in only one edition were left out rather than treated as unchanged, 18 per cent of records. The real-terms deflator and the counts behind each share are given with the annex tables.\n\nMention: \"Longitudinal Household Consumption Census (LHCC)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Longitudinal Household Consumption Census supports one indexation claim by following households between editions.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:fd43367bf2a5bfdf8b1cc88e", "occurrence_id": "72224ba3ba92d93ba766002e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Catchment definitions for the country are reported from the Standardised Service Catchments Dataset (SSCD), Collection 7, the geospatial dataset covering Egypt in 2016, and the four values below are given as the dataset holds them.\n\nThe dataset puts the number of defined catchments at 2 480, the median population within each catchment boundary at 9 400 residents, the share of households falling outside every defined catchment at 6 per cent, and the mean distance from a household to the boundary of its assigned catchment at 1.6 kilometres. None was recalculated here, and the dataset's own rule for assigning a household to a catchment, which uses the shortest road route rather than straight-line distance, appears with the values.\n\nNo household assigned to a facility closed at the time of collection was counted as covered, and no catchment whose polygon failed the topology check was included in the median. Where the dataset reports for one governorate, that governorate is named with the value.\n\nThe dataset's layer reference for each value, along with the extraction date and the households behind each share are listed in the annex notes.\n\nMention: \"Standardised Service Catchments Dataset (SSCD)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Standardised Service Catchments Dataset supplies catchment counts, boundary populations and assignment distances.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:10758278bff0230105cc38c1", "occurrence_id": "edecf0562f02d533301fd741", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Scarcity and movement were joined for this assessment from the Panel Water and Climate Collection (PWCC), Round 4, compiled by US Customs, and the measures below are this assessment's own rather than the collection's.\n\nHouseholds were grouped by the dry-season shortfall in their source against the annual requirement and the share reporting a member who left the holding for more than three months computed: 29 per cent of households with a shortfall above 40 per cent against 9 per cent of those with a shortfall under 10 per cent. The collection reports the water balance per source and movement per household in separate tables, so the two never appear together in it.\n\nRainfall anomaly for the season was then read from the same records and held constant, and the difference between the two shortfall groups narrows to 12 percentage points, which leaves the condition of the household's own source rather than the season's rain as what the assessment reads as driving the departure.\n\nHouseholds whose source could not be matched to a location, 5 per cent of the panel, were left out rather than assigned to the nearest point. The shortfall bands, the anomaly bands and the counts behind each share are stated with the annex tables.\n\nMention: \"Panel Water and Climate Collection (PWCC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Panel Water and Climate Collection links source shortfall to seasonal departure from the holding.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:8392a094f6a758ea2562f3bf", "occurrence_id": "e84826cf62f73cd29471625c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Pond output was mapped for this assessment from the Decennial Aquaculture Output Dataset, the geospatial dataset the National Statistics Office compiled for Mexico in 2022, and the measures below are this assessment's own rather than the dataset's.\n\nPonds were matched to their parcel and grouped by the mean water temperature recorded for the cell, and the yield per hectare computed: 4.2 tonnes for ponds in the warmest third of cells against 2.8 tonnes for those in the coolest. The dataset reports the pond polygon and the cell's temperature series without producing any yield comparison between them.\n\nDistance to the nearest processing point was then taken from the same polygons, and ponds within 15 kilometres of one report 0.6 tonnes more per hectare at the same temperature band, which the assessment reads as the loss of perishable stock rather than as a difference in husbandry.\n\nPonds whose polygon could not be matched to a parcel, 4 per cent of those mapped, were left out rather than assigned to the nearest cell. The temperature bands, the distance cut-point and the pond counts behind each figure are given with the annex tables.\n\nMention: \"Decennial Aquaculture Output Dataset\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Decennial Aquaculture Output Dataset matches pond polygons to cells to compare yield by temperature.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:2d8875af8e701ace25146e04", "occurrence_id": "9d25a1d243d597de770931f3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: School infrastructure and staffing were compared for this assessment from the Household School Infrastructure Survey (HSIS), which UNESCO conducts in Mongolia, and the figures below are this assessment's own rather than the survey's.\n\nSchools were grouped by whether they report heating through the whole winter and the share of posts filled at the start of the term computed: 82 per cent of heated schools against 64 per cent of those without full heating. The survey records infrastructure items and staffing counts separately, so no reading of staffing by building condition exists in it.\n\nThe distance from the nearest town was then held constant and the difference narrows to 9 percentage points, which leaves the condition of the building as the smaller part of the staffing gap than the assessment's opening draft assumed, and the remaining part is what the review attributes to the difficulty of posting staff to unheated buildings in the first place.\n\nSchools with no enrolment recorded were left out rather than counted as unfilled, 5 per cent of schools. The infrastructure items used, the distance bands and the post counts behind each share are stated with the annex tables.\n\nMention: \"Household School Infrastructure Survey (HSIS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Household School Infrastructure Survey compares teacher post fill rates by school heating condition.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:dac80f4d6f9e2f2b49f7f1ce", "occurrence_id": "7d7609d68a66caf8965639bc", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Accessibility was derived for this assessment from the Record-Based Transport Network Dataset, the geospatial dataset NCHS/CDC/WHO compiled for Germany in 2020, and the measures below are this assessment's own rather than the dataset's.\n\nDwellings were matched to the network and the share whose nearest stop lies within 800 metres of the door computed: 78 per cent of dwellings in the inner ring against 31 per cent of those in the outer districts. The dataset publishes the network and the dwelling locations without computing any such accessibility share.\n\nService frequency on the serving line was then held constant and the difference narrows to 19 percentage points, which leaves the geometry of the network rather than how often it runs as the larger part of the distance between the two areas.\n\nDwellings whose address could not be geocoded, 3 per cent of those listed, were left out rather than placed at the district centroid. The matching rule, the buffer distance and the counts behind each share are given with the annex tables.\n\nMention: \"Record-Based Transport Network Dataset\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Record-Based Transport Network Dataset supports stop-accessibility shares computed by dwelling location.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:9f245811c1e31666bd0953de", "occurrence_id": "be204ba7be80e0b35214399c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Administrative coverage figures at the programme level are taken from the District Administrative Records Registry (DARR), First Collection, compiled by the World Bank, and the four values below are carried over as the registry holds them.\n\nThe registry puts the share of districts filing their returns within the reporting window at 74 per cent, the mean number of records per district return at 1 260, the share of returns that reconcile to the previous period without an adjustment at 61 per cent, and the share of districts whose register carries a unique identifier for every person at 49 per cent. None was recalculated here, and the registry's own definitions of a filed return and of a reconciling adjustment appear with the values.\n\nNo return filed late and later withdrawn was counted as filed, and no district using two identifiers for the same person was counted as unique. Where the registry reports for one level of government, that level is named with the value rather than pooled.\n\nThe registry table and reference for each value, the extraction date and the returns behind each share are set down in the annex notes.\n\nMention: \"District Administrative Records Registry (DARR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The District Administrative Records Registry supplies filing, reconciliation and identifier coverage values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:561a60c29b742b8c727836eb", "occurrence_id": "f86b94d2a9bac3d483b5afa5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The hardship figures in this chapter are drawn from the office's panel, and the Departmental Deprivation and Hardship Survey (DDHS), Fourth Release, compiled by Economist journal, is quoted only to place them in context.\n\nThe release reports that 31 per cent of households could not replace worn clothing when needed, that 19 per cent went without a heating fuel in the cold month, and that 12 per cent reported a debt arrears episode in the previous year. All three are given as the release prints them, with its own wording for each item stated, and nothing in the chapter was computed from them.\n\nTheir purpose is to show how a differently framed questionnaire records the same deprivations the panel tracks. The chapter's figures come from the panel, which observes the same households over successive periods rather than sampling once, and where the two readings differ the difference is noted rather than reconciled.\n\nThe annex lists the release tables consulted and the extraction date, which is the whole of the use made of the release here.\n\nMention: \"Departmental Deprivation and Hardship Survey (DDHS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Departmental Deprivation and Hardship Survey is quoted for context beside the office's panel hardship figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:c8dd1b139a53216f30164303", "occurrence_id": "568d85961c429b33a82d1c8e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The registry entries were reconciled into the programme's file by the Diagnostic Public Records Registry (DPRR), the database the Banque de France maintains for Lagos State, and what was done to them is handling rather than measurement.\n\nEach entry was given the file's own district code, matched against the ward list and, where the entry named a ward absent from that list, held for the officer rather than assigned to the nearest match. Entries with no district stated were left uncoded rather than allocated by address, and 1 620 of them were carried into the file in that condition. The registry's own identifier was kept on every row so that a later collection can be matched without re-keying the entry.\n\nThe programme's coverage estimates come from its own enumeration, and this reconciliation supports that work in one respect only, by letting a registry entry and an enumeration record for the same premises be found in both files. No estimate is calculated from the registry's entries themselves.\n\nThe coding sheet, the count of entries held and the date the transfer was deposited are written up in the annex log.\n\nMention: \"Diagnostic Public Records Registry (DPRR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The Diagnostic Public Records Registry supplies entries coded into the programme file's district crosswalk.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:48a7921ebd5d362b0683c0db", "occurrence_id": "2840faeaed9cd1e69ca467fc", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two records-quality lines in the annex are quoted from the Periodic Public Records Index, compiled by INVALSI for Sulaymaniya in 2022, and both are reproduced as the index prints them.\n\nThe index puts the share of offices whose records can be retrieved within one working day at 57 per cent and the median age of the oldest unclosed file at 3 years. Neither was recalculated here, and the index's own rule for counting a retrieval, which requires the file to be produced complete rather than located, appears with the figures.\n\nThe annex's remaining lines come from the inspectorate's own visits, so the index accounts for two lines while the visit reports carry the rest. A figure the index gives for one office class is shown with that class named rather than averaged across the service.\n\nThe index entry behind each figure, the extraction date and the offices behind each share are recorded with the annex notes.\n\nMention: \"Periodic Public Records Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Periodic Public Records Index supplies two records-retrieval lines quoted in the annex.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:79518d8a5bf3d587aa375992", "occurrence_id": "601c9bc960da101ce45ab605", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two immunisation lines in the city annex come from the Ministerial Immunisation Coverage Registry, the register World Bank/IFC keeps for Kinshasa as of 2023, and they appear here exactly as published.\n\nThe registry records third-dose coverage at 71 per cent of surviving infants and the share of health posts reporting a cold-chain failure in the last quarter at 14 per cent. No figure was re-derived for this report, and the registry's counting rule for a surviving infant is quoted with them rather than replaced.\n\nEverything else in that annex is compiled from the ministry's weekly disease bulletins, so the registry stands behind two lines and the bulletins behind the remainder. Where the registry gives coverage for one health zone, the zone is named and the figure is not averaged over the city.\n\nUnder the table the annex carries the registry reference, the date the extract was taken and the number of posts behind each share.\n\nMention: \"Ministerial Immunisation Coverage Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Ministerial Immunisation Coverage Registry supplies two coverage lines quoted inside the city annex.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:a9c7979d37dc79428512d014", "occurrence_id": "dd02850b474f7be5e2d562be", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Operation logs were brought into the camp's file from the Diagnostic Operations Monitoring Registry (DOMR), Release 9, the database kept for Zatari camp in 2017, and the work done on them was cataloguing, not analysis.\n\nEvery log line was tagged with the file's activity code and set against the entry made by the team on the ground; where the two disagreed the line was parked for the coordinator instead of being corrected in place. Lines carrying no activity at all stayed untagged rather than being filed under a general heading, and 940 of them entered the file that way. The registry's timestamp was preserved on each line so that a later release can be matched without asking the teams to re-send anything.\n\nThe camp's monthly statistics are built from the teams' own returns, and this cataloguing assists that work only in the narrow sense that a log line and a return for the same day can now be found in both places. No statistic in the monthly file is derived from the registry's lines.\n\nThe annex log holds the coding sheet, the number of lines parked and the date the deposit was made.\n\nMention: \"Diagnostic Operations Monitoring Registry (DOMR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The Diagnostic Operations Monitoring Registry supplies log lines coded into the camp file's activity crosswalk.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:888b9493f9e7f21b8f6f4d7d", "occurrence_id": "321f4434882cc4aa54ff536a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Whether the office's own deforestation figure survives was checked against the Departmental Land Cover Change System (DLCCS), First Version, the administrative system UCDA maintains, and that system is the principal source for the check.\n\nThe office reports 8 200 hectares converted to cropland in the department over the observation period. The system, which compares each parcel's classified cover between two acquisitions, puts conversion at 9 750 hectares, because it counts a parcel whose cover changed once during the period even if it was replanted afterwards. The office's figure comes from field inspections at a single point in time.\n\nThe system is what the section relies on for the conversion claim, and the office's smaller figure is shown beside it so the reader can see how much of the gap is the replanted parcels rather than a disagreement about where conversion happened. Parcels the two sources classify differently are listed individually rather than netted off.\n\nThe classification each source applies and the acquisitions each uses are recorded with the annex tables.\n\nMention: \"Departmental Land Cover Change System (DLCCS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Departmental Land Cover Change System is the principal source checking the office's cropland conversion figure.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:598e24be2be8d184a99b3677", "occurrence_id": "cc8b8fcf420d409294f6bf98", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Support, work and movement were read together for this assessment from the Pilot Disability and Support Survey (PDSS), Fifth Version, compiled by SEP.\n\nAdults with a functional difficulty were grouped by whether they had moved residence in the previous two years and the share reporting paid work in the reference month computed: 29 per cent of those who had moved against 44 per cent of those who had not. What the survey prints is a difficulty prevalence and an employment rate, each on its own; the crossing of difficulty with a move is not in it.\n\nReceipt of a support payment was then equalised between the two groups, and 6 percentage points of the 15-point distance remain, which tells the assessment that the payment offsets part but not most of what a move costs a disabled adult in work terms.\n\nAdults whose move date is missing are not folded into either group. They are reported separately and account for 7 per cent of respondents, and the bands, denominators and extraction date are recorded under the table.\n\nMention: \"Pilot Disability and Support Survey (PDSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Pilot Disability and Support Survey crosses difficulty, support receipt and residential moves for work outcomes.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:54bdad52338b7083c8d02562", "occurrence_id": "ef903d7213c3d129afcf4d54", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two brightness lines in the appendix are quoted from the Coordinated Nighttime Lights Index, the indicator OECD-PISA compiled for Australia in 2023, and both are given without restatement.\n\nThe index puts the share of the population living in cells above its own radiance threshold at 88 per cent, the share of cells showing a decline between the two observations at 4 per cent, and the mean radiance of cells classed as settlements at 26 nanowatts per square centimetre. Nothing here recomputes any of the three, and the threshold and the radiance unit are the index's own rather than ones this report adopted.\n\nThe appendix's other lines are drawn from the statistics office's settlement series, so the index accounts for two rows and the settlement series for the rest. A value the index reports for one state is printed with that state named rather than pooled across the country, and no cell excluded by the index's own cloud mask was reinstated for this report.\n\nThe annex records the index layer, the extraction date, with the cells behind each share alongside.\n\nMention: \"Coordinated Nighttime Lights Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Coordinated Nighttime Lights Index supplies radiance lines quoted inside the appendix.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:355e2e657d39b368719cdedc", "occurrence_id": "e8a18cff6e2d44fe1d595965", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two return-intention lines in the displacement annex are quoted from the Urban Return Intention Report (URIR), Collection 3, issued by UN‐DESA, and both are reproduced as the report prints them.\n\nThe report puts the share of displaced households that intend to return to their origin neighbourhood within a year at 37 per cent and the share naming damage to their dwelling as the reason they have not done so at 44 per cent. Neither figure was reworked for this annex, and the report's own wording for an intention and for a dwellable dwelling appears with them.\n\nThe annex's remaining lines come from the agency's movement tracking matrix, so the report stands behind two lines while the matrix carries the others. Where the report gives a share for one origin district, that district is named rather than the share shown as a general result.\n\nThe report's table and page for each line, the extraction date and the households behind each share are set out in the annex notes.\n\nMention: \"Urban Return Intention Report (URIR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Urban Return Intention Report supplies return-intention and dwelling-damage lines in the displacement annex.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:34ad47e99c81e1bdd9ebfee5", "occurrence_id": "561ce556b673f5399b846ec3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two forest-product lines in the livelihoods annex come from the Monthly Forest Products Registry, the register HCP keeps for the Amhara region as of 2015, and they are given as the register prints them.\n\nThe register puts the share of rural households collecting fuelwood from communal land at 63 per cent and the mean value of forest products sold by those households at 9 per cent of their annual cash income. Neither was rebuilt here, and the register's definition of a collecting household, which requires collection in the reference month, is reproduced with the figures.\n\nThe annex's other lines are compiled from the agricultural office's crop returns, so the register accounts for two lines while the crop returns carry the rest. A figure the register gives for one woreda is shown with that woreda named rather than merged into a regional average.\n\nThe register reference behind each line, the extraction date and the households behind each share are carried in the annex notes.\n\nMention: \"Monthly Forest Products Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Monthly Forest Products Registry supplies fuelwood collection and forest-income lines for the region.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:acb86eb96a0eaf73b178c5f7", "occurrence_id": "e2c622208145a0032a37ebf2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two register lines in the provincial annex are quoted from the Provincial Service Register Registry (PSRR), the register kept for the Republic of Congo in 2024, and both are carried over unchanged.\n\nThe register puts the share of villages appearing in it with a current service entry at 76 per cent and the share whose last entry is more than two years old at 15 per cent. Neither was recalculated here, and the register's rule for a current entry, which requires a visit recorded within the reporting year, appears with the figures.\n\nThe rest of the annex lines come from the provincial office's field reports, so the register carries two lines and the field reports carry the rest. A figure given for one department is printed with that department named rather than pooled across the province.\n\nThe register entry behind each line, the extraction date and the villages behind each share are detailed in the annex notes.\n\nMention: \"Provincial Service Register Registry (PSRR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Provincial Service Register Registry supplies two register-currency lines quoted in the provincial annex.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:17090e6231f2d1eb27a9cfad", "occurrence_id": "d5572b0054fc126e3fd983c2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two graduation lines in the rural annex are quoted from the Rural Graduation Programmes Index (RGPI), Collection 9, compiled for Sindh in 2017, and both are reproduced as the index holds them.\n\nThe index puts the share of enrolled households holding an asset above its own graduation threshold at 52 per cent, the share of those households more than 5 kilometres from an all-season road at 28 per cent, and the share receiving a transfer in the reference quarter at 61 per cent. No figure was recalculated for this annex, and the threshold, the distance band and the transfer definition are the index's own rather than ones this report imposed.\n\nThe rest of the annex is compiled from the programme's own payment records, so the index carries two lines and the payment records carry the remainder. Where the index gives a figure for one tehsil, that tehsil is named rather than averaged across the division, and a household counted by the index in two programmes is counted once, as the index itself counts it.\n\nThe index entry behind each line, the extraction date and the households behind each share are recorded with the annex notes.\n\nMention: \"Rural Graduation Programmes Index (RGPI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Rural Graduation Programmes Index supplies asset, road-distance and transfer lines in the rural annex.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:c87f19d2c344eb2f5c887ba3", "occurrence_id": "70b7164ec4a188865ffcfeae", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two delivery lines in the district annex are quoted from the Combined Service Delivery Registry (CSDR), the register SEP maintains for Johor, and both appear here as the registry records them.\n\nThe registry puts the share of applications answered within its own published deadline at 58 per cent, the share of premises visited and found closed at 6 per cent, and the mean number of follow-up visits before a file is closed at 1.7. Nothing was recomputed for this annex, and the deadline each service declares is the registry's own rather than one assumed here.\n\nThe district's own performance figures come from its complaint log, so the registry accounts for two lines of the annex while the log accounts for the balance. Where the registry reports for one service, that service is named and its figure is not blended with others, and an application withdrawn by the applicant is excluded from the first share as the registry excludes it.\n\nThe registry line behind each figure, the extraction date and the applications behind each share are recorded with the annex notes.\n\nMention: \"Combined Service Delivery Registry (CSDR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Combined Service Delivery Registry supplies deadline, closure and follow-up lines quoted in the district annex.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:282218f5941dde96ccd668ac", "occurrence_id": "341d2a15afb0ebcc33fa6356", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two health lines in the state annex are quoted from the Experimental Demographic and Health Registry (EDHR), Edition 8, the register kept for Florida in 2024, and both are carried over as the register prints them.\n\nThe register puts the share of adults reporting a check-up in the previous year at 64 per cent, the share without any regular place of care at 17 per cent, and the mean number of nights admitted per thousand residents at 82. None was re-derived for this annex, and the register's own wording for a regular place of care, which excludes a walk-in clinic visited once, is reproduced with the figures.\n\nThe annex's other lines are compiled from the hospital discharge series, so the register accounts for two lines while that series carries the rest. A figure the register reports for one county is printed with the county named rather than averaged across the state.\n\nThe register table behind each line, the extraction date and the respondents behind each share are carried in the annex notes.\n\nMention: \"Experimental Demographic and Health Registry (EDHR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Experimental Demographic and Health Registry supplies check-up, regular-care and admission lines for the state.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:b5958f2c061f989b71969b63", "occurrence_id": "460219abf5be8fe0d821d419", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Adequacy of household spending was rebuilt for this review from the Multi-Topic Household Consumption Survey, which Jacobs et al conducted in Buenos Aires in 2019, and none of what follows appears in the survey's own tables.\n\nHouseholds were ranked by the share of income committed to rent and utilities and then grouped at the two ends of that ranking. Among the top fifth, 46 per cent fall below the national adequacy line once housing is subtracted from available income, against 8 per cent among the bottom fifth. The survey publishes the housing share and the consumption aggregate separately, so the subtraction that produces this contrast is the review's.\n\nWhether the household receives a rent subsidy was then equalised across the two groups and the distance falls to 19 percentage points, which leaves the subsidy doing part of the work and the composition of household spending doing the rest.\n\nHouseholds that declined the income module are excluded rather than imputed, and they are 10 per cent of responses. Bands, denominators and the extraction date are recorded under the table.\n\nMention: \"Multi-Topic Household Consumption Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Multi-Topic Household Consumption Survey supports a housing-adjusted adequacy measure built by this review.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:b06bd522397c711a7edc0831", "occurrence_id": "6456751a05107eed98e55098", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Whether the bureau's statelessness figure stands was tested against the Periodic Statelessness Survey, which Lane Financial conducted in Croatia in 2021, and that survey is the principal source for the test.\n\nThe bureau's register puts the number of residents without a recognised citizenship at 1 930. The survey, which reaches people who never appear at a register office, puts the equivalent count at 2 780, because it counts households whose members hold no document from any state rather than only those whose status has been determined. The bureau's figure counts determinations completed while the survey counts what respondents report.\n\nThe survey is what this section relies on for the size of the affected population, and the bureau's smaller count is shown beside it so the reader can see that the difference is the share never determined rather than a disagreement about who qualifies. Where one household appears in both, the survey's report is used and the two are not averaged.\n\nThe determination rule each source applies and the households each reaches are set out with the annex tables.\n\nMention: \"Periodic Statelessness Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Periodic Statelessness Survey is the principal source testing the bureau's count of undocumented residents.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:d322c59f6b090c55da075fe0", "occurrence_id": "4158f4a653d379a0050a8322", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The processing-time figure in this chapter is the ombudsman's own, taken from its complaint log, and the Independent Public Sector Monitoring Index (IPSMI), Collection 9, compiled by People’s Bank of China, is cited to test it.\n\nThe ombudsman's log puts the median time to a substantive reply at 21 working days. The index gives 26 days for the same period on its own construction. The log starts its clock when a complaint is opened while the index starts it when the responsible office acknowledges the case, and neither clock was reset to match the other.\n\nThat 5-day distance is smaller than the interval the index reports for its administrative panel, so the ombudsman's figure is printed as logged and the index supplies no correction to it and no values for the categories the log does not separate.\n\nExcluded from the comparison are complaints closed as out of scope, since the index does not count them and the log records them with a reply time of zero. Which events start each clock and which categories each covers are recorded with the annex tables.\n\nMention: \"Independent Public Sector Monitoring Index (IPSMI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Independent Public Sector Monitoring Index checks the ombudsman's median reply-time figure.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:ba815df4a39cdce1dfffff64", "occurrence_id": "51b3488ac8df93be4ce5840e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Programme monitoring city-level figures come from the Triennial Programme Monitoring Survey, which RMCAB conducted in Houston in 2023, and the four values below are carried over exactly as published.\n\nThe survey puts the share of eligible households aware of the programme at 57 per cent, the share of those who applied and were enrolled at 71 per cent, the mean months between enrolment and first benefit at 5, and the share of enrolled households still receiving a benefit at the time of interview at 64 per cent. No value was recalculated here, and the survey's own definitions of eligibility, of enrolment and of a current recipient appear with them.\n\nNo household screened out for a missing document was counted among applicants, and no household whose benefit was suspended pending review was counted as current. Where the survey reports for one ward, the ward is named with the value rather than pooled into the city figure.\n\nThe survey table and line for each value: the extraction date and the households behind each share are kept in the annex notes.\n\nMention: \"Triennial Programme Monitoring Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Triennial Programme Monitoring Survey supplies awareness, enrolment, delay and retention values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:da475ef4ca008e53d72693a7", "occurrence_id": "6c91db6552142eabe05aa0e5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Service monitoring figures for the state are drawn from the Joint Public Sector Monitoring Registry (JPSMR), the database maintained for Bihar in 2024, and the four values below are reproduced without amendment.\n\nThe registry puts the share of facilities submitting their return on time at 69 per cent, the share of inspected sites whose findings were entered into the register within a month at 44 per cent, the mean number of staff recorded against each facility at 7.3, and the share of entries revised after the inspection report was issued at 9 per cent. None of these was recomputed here, and the registry's own counting rules for a timely return and for a revision accompany the values.\n\nNo facility whose return arrived after the register closed was counted as timely, and no entry corrected in the same sitting as the inspection was counted as revised. Where the registry reports for one district, the district is named with the value rather than shown for the state.\n\nThe annex holds the registry reference for each value, the date of extraction, and the facilities behind each share.\n\nMention: \"Joint Public Sector Monitoring Registry (JPSMR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Joint Public Sector Monitoring Registry supplies return-timeliness, staffing and revision values for the state.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:92700ee11a7ad3e3bcf90539", "occurrence_id": "7d1b78106d283163c551ee6a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Whether the federation's own catch figure survives was tested against the Independent Aquaculture Output Survey (IAOS), Round 5, compiled by Social Watch, and the survey is the principal source for that test.\n\nThe federation reports 214 000 tonnes harvested by its member cooperatives over the season. The survey, which samples ponds rather than cooperative rolls, arrives at 268 000 tonnes, because it counts harvest sold directly from the pond and never entered into the cooperative's books. The federation's figure is a membership tally while the survey's is a production estimate built from sampled ponds.\n\nThe survey therefore carries the section's production claim, and the federation's lower tonnage is printed beside it so the reader can see that the distance is unrecorded direct sales rather than a different definition of a harvest. Catches both sources record are taken from the survey and are not averaged with the cooperative tally.\n\nHouseholds below the national adequacy line make up 34 per cent of the survey's sampled pond owners, which is what the section quotes when it explains why unrecorded sales matter to household income. The counting rule each source applies is recorded with the annex tables.\n\nMention: \"Independent Aquaculture Output Survey (IAOS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Independent Aquaculture Output Survey is the principal source testing the federation's harvest tonnage.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:9ca64af63b16be0ae44f8c6d", "occurrence_id": "2281661046b9da1d97f30554", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: How long job search takes was worked out for this assessment from the Follow-Up Job Search Survey (FJSS), First Wave, conducted in Indonesia in 2023, and the figures that follow come from that work rather than from the survey's tables.\n\nRegistrants were grouped by the qualification recorded for them and the mean weeks between registration and a first offer computed: 11 weeks among registrants with a tertiary qualification against 24 weeks among those with none. The survey publishes the two distributions separately and never crosses them.\n\nThe number of applications logged in the first month was then equalised between the groups, and 4 weeks of the 13-week distance remain, which tells the assessment that search intensity explains a part of the qualification gap while the remainder sits with the kinds of vacancy each group can reach.\n\nRegistrants who took work outside the employment office's register are excluded rather than counted as hired, and they are 8 per cent of the sample. The bands, the denominators and the date of extraction are recorded under the table.\n\nMention: \"Follow-Up Job Search Survey (FJSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Follow-Up Job Search Survey provides registrant records used to compute search duration by qualification.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:7f10fb504c37a9b3badc2bdc", "occurrence_id": "14c6c6ec280e79551d86c120", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Programme monitoring figures for the country are held in the Harmonised Programme Monitoring System (HPMS), Sixth Wave, the administrative register maintained for Malawi in 2019, and the four values below are transferred here without change.\n\nThe register puts the share of planned activities completed by the end of the reporting year at 63 per cent, the share of budget released to implementing units on schedule at 48 per cent, the mean delay in months between an activity's planned and recorded completion at 3.2, and the share of activity records carrying a verification note at 77 per cent. None of these was recomputed for this report, and the register's own definitions of completion and of verification accompany the values.\n\nAn activity deferred into the following year was not counted as completed, and a release held at provincial level was not counted as released. Where the register reports for one ministry, that ministry is named with the value rather than merged into the national figure.\n\nThe register reference for each value, the date of extraction as well as the activity counts behind each share are recorded with the annex notes.\n\nMention: \"Harmonised Programme Monitoring System (HPMS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Harmonised Programme Monitoring System supplies completion, release, delay and verification values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:5e271c22e5f0a064980b9139", "occurrence_id": "67c231ca90acd3774456fefe", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two bargaining lines in the labour annex are quoted from the High-Resolution Collective Bargaining Survey (HCBS), which covers Sulaymaniya in 2016, and both are carried over exactly as published.\n\nThe survey puts the share of establishments with a recognised bargaining unit at 27 per cent, the share of agreements reviewed within the year at 39 per cent, and the mean number of workers covered per agreement at 46. No figure was re-derived for this annex, and the survey's own test for a recognised unit, which requires a filed agreement rather than a negotiated one, is reproduced with them.\n\nThe annex's other lines come from the labour office's inspection records, so the survey stands behind two lines while the inspection records carry the rest. A figure given for one sector is printed with that sector named rather than averaged across the survey.\n\nThe survey table behind each line, the extraction date and the establishments behind each share are set out in the annex notes.\n\nMention: \"High-Resolution Collective Bargaining Survey (HCBS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The High-Resolution Collective Bargaining Survey supplies bargaining-unit and agreement-review lines for the annex.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:72e3ee6e232ea9e7958bdadd", "occurrence_id": "c4d51c622e0ac25cee947cd0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The adaptation spending figure in this chapter is the ministry's own, taken from its disbursement ledger, and the Triennial Adaptation Finance Registry (TAFR), Phase 2, compiled for Rwanda in 2022, is cited to test it.\n\nThe ministry's ledger puts spending on water-related adaptation at 84 million in the reporting year. The registry puts it at 96 million, because it counts a tranche committed and not yet drawn as adaptation spending while the ledger counts only cash actually paid out. Neither basis was converted to the other for the comparison.\n\nThe 12 million distance sits inside the band the registry publishes for its own commitments series, so the ministry's figure is printed as disbursed and the registry provides no revision to it and no values for the years the ledger does not separate by purpose.\n\nExcluded are projects the registry cannot locate to a catchment, 6 per cent of entries, and those it tags cross-cutting, since neither source assigns them a water share and any split would be this chapter's invention. The registry's catchment mapping is what allows the water tag at all, and the commitment rule each source applies is recorded with the annex tables.\n\nMention: \"Triennial Adaptation Finance Registry (TAFR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Triennial Adaptation Finance Registry checks the ministry's water-related adaptation disbursement figure.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:ca6211339da6f2b70d189570", "occurrence_id": "25ffd867a778d33ac02ba958", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Hiring inside households was measured for this assessment from the Household Hiring and Vacancies Survey (HHVS), Wave 5, compiled by US-EPA, and what follows was produced here rather than taken from the survey's tables.\n\nHouseholds were grouped by whether they operate a business away from home and the share reporting an unfilled post at the time of interview computed: 21 per cent of those operating away from home against 7 per cent of those working only within it. The survey records each household's activity and its vacancies in different sections, so the two are never brought together in it.\n\nWeeks the post had been open were then read from the same records, and half of the unfilled posts in the first group had been open longer than eight weeks against a third in the second, which the assessment reads as a search problem rather than a shortage of willing workers.\n\nHouseholds reporting a post frozen rather than vacant are excluded from the first share, and they are 5 per cent of respondents. Bands, denominators and the extraction date are recorded under the table.\n\nMention: \"Household Hiring and Vacancies Survey (HHVS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Household Hiring and Vacancies Survey supports a measure of unfilled household business posts.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:3f75c20763a27029ce97f04b", "occurrence_id": "429475cc8f9907940f5db62c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two enterprise lines in the employment annex are quoted from the Analytical Business Demography Microdata (ABDM), a microdata set covering Chile in 2023, and both are reproduced without amendment.\n\nThe set puts the share of surviving units employing the same number of workers as the previous year at 58 per cent and the share that reduced their workforce while remaining open at 12 per cent. Neither was recalculated here, and the set's own rule for treating a unit as surviving, which requires at least one worker in both periods, appears with the figures.\n\nThe other annex entries are assembled from the social security contribution series, so the microdata account for two lines while that series accounts for the remaining lines. A figure given for one size class is printed with the class named rather than blended into an all-size average.\n\nThe set's table and line behind each figure, the extraction date and the units behind each share are recorded with the annex notes.\n\nMention: \"Analytical Business Demography Microdata (ABDM)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Analytical Business Demography Microdata supplies two workforce-change lines quoted in the annex.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:eb29330d4cdd165a4067866c", "occurrence_id": "93f2ffe1e756551c3650d04c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Unit survival was computed for this review from the Panel Business Demography Survey, which MoF follows in Bhopal in 2021, and the numbers below are the review's rather than the survey's.\n\nUnits were split by whether the owner had opened a second register-based enterprise and the share still trading eighteen months later computed: 74 per cent among owners with a second registration against 55 per cent among sole registrants. The survey reports each registration event and each trading status without linking them through the owner, which is the link this review had to build.\n\nThat comparison supports one claim, that the credit line should accept an owner's second registration as evidence of track record. The loan-loss figures the proposal quotes come from the bank's own portfolio, from which the survey supplied nothing.\n\nOwners whose second enterprise was registered after the follow-up interview are excluded rather than treated as sole registrants, and they are 6 per cent of owners. The linking rule and the counts behind each share are given under the table.\n\nMention: \"Panel Business Demography Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Panel Business Demography Survey supports one credit claim by comparing survival of serial registrants.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:5695af5cc6a40634216a8471", "occurrence_id": "91f0d660bfb48e7cfe69dabc", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Parcels were brought into the monitoring file from the Rural Land Cover Change Dataset (RLCCD), Second Wave, the geospatial dataset UNHCR RBWCA compiled, and what happened to them was filing rather than measurement.\n\nEach parcel polygon was given the file's cover class, checked against the previous wave's classification and, where the two disagree on a change of class, marked for review instead of overwritten. Parcels that arrived without a class were left unclassified rather than mapped to the modal class of their district, and 2 140 of them entered the file that way. The dataset's own acquisition date was kept on every polygon so a later wave can be matched without re-requesting the imagery.\n\nThe file's cover statistics come from its own interpretation of the imagery, and this filing assists that work only in the sense that a parcel now carries the same class label in both places, so a change reported in one can be looked up in the other. No statistic in the file is derived from the dataset's classifications.\n\nThe coding sheet, the number of parcels marked and the date of deposit are held in the annex log.\n\nMention: \"Rural Land Cover Change Dataset (RLCCD)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The Rural Land Cover Change Dataset supplies parcel polygons classified into the monitoring file's cover scheme.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:8c94830b339469e9e8ba3457", "occurrence_id": "99027498176c405a6735c94b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Workload per office was rebuilt for this assessment from the Representative Regional Operations Index, compiled by Fed in Mozambique in 2020, and the measures below are the assessment's rather than the index's.\n\nOffices were grouped by the population assigned to them and the mean minutes a visitor spends at a counter computed: 38 minutes in offices serving more than 60 000 residents against 16 minutes in those serving fewer than 15 000. The index publishes population per office and a service-time item per office in separate tables and never divides one by the other.\n\nStaff per counter was then equalised between the groups, and 12 minutes of the 22-minute distance remain, which leaves the volume each office must process as the larger part of the queue rather than how many people stand behind it.\n\nOffices open fewer than three days a week are excluded rather than averaged in, and they are 4 per cent of offices. The bands used and the counts behind each mean are recorded under the table.\n\nMention: \"Representative Regional Operations Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Representative Regional Operations Index supports a counter-wait measure built against office population.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:78a6b3f05a59187e9b5f0be1", "occurrence_id": "7efa5987ac4e485955b41c85", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Poverty and vulnerability figures for each part of the country are drawn from the Biennial Poverty and Vulnerability Report, which INVALSI issued for Sri Lanka in 2021, and the four values below are given exactly as the report releases them.\n\nThe report puts the national poverty rate at 21 per cent, the share of households vulnerable to falling below the line after a single health payment at 34 per cent, the depth of poverty among the poor at 4 per cent and the share of households receiving any social assistance at 29 per cent. None was recomputed here, and the report's own basket and its vulnerability threshold are reproduced with the figures.\n\nNo household whose income was imputed by the report's own procedure was re-weighted for this chapter, and no regional figure was extended to the national total. Where the report gives a rate for one district, the district is named with the rate.\n\nThe report's table and page for each value, plus the extraction date and the households behind each share are laid out in the annex notes.\n\nMention: \"Biennial Poverty and Vulnerability Report\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Biennial Poverty and Vulnerability Report supplies rate, vulnerability, depth and assistance values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:ee0204aecc322c2834919b3a", "occurrence_id": "2bab6d7af17589b59e573c61", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The clinic throughput figures in this chapter come from the ministry's facility returns, and the High-Resolution Immunisation Coverage Survey (HICS), compiled by Water Aid for Botswana, is quoted only to place them in context.\n\nThe survey puts third-dose coverage at 82 per cent, the share of caregivers who named distance as their main obstacle at 18 per cent, and the share of sessions postponed in the reference quarter at 7 per cent. All three are reproduced as the survey issues them, with its own definition of a postponed session stated, and no chapter figure was built from them.\n\nTheir use here is to show how a household survey describes the same service the returns measure from the counter. The chapter's argument rests on the returns, which record sessions held rather than children missed, and the difference between the two views is noted rather than merged.\n\nThe annex lists the survey tables consulted and the date of extraction, which is the extent of the survey's use in this chapter.\n\nMention: \"High-Resolution Immunisation Coverage Survey (HICS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The High-Resolution Immunisation Coverage Survey is context while the chapter's throughput figures come from returns.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:91d31c06f657a934c97f8d08", "occurrence_id": "169d5c930cf93f9e9a00829e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two boundary lines in the statistics annex are quoted from the Rolling Administrative Boundary Registry (RABR), Phase 8, the register GAD maintains, and both are carried over unchanged.\n\nThe registry puts the number of valid administrative units at its current level at 4 310 and the share of units whose boundary geometry was last revised more than five years ago at 22 per cent. Neither was recalculated for this annex, and the registry's own rule for a valid unit, which requires a published code rather than a proposed one, is reproduced with them.\n\nThe annex's other lines come from the electoral commission's constituency file, so the registry carries two lines while that file carries the remainder. A figure given for one level of administration is printed with that level named rather than merged with the others.\n\nThe registry reference behind each line, the extraction date and the units behind each share are recorded with the annex notes.\n\nMention: \"Rolling Administrative Boundary Registry (RABR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Rolling Administrative Boundary Registry supplies unit-count and boundary-currency lines in the annex.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:3a6afbc3695344435a489905", "occurrence_id": "b8d9411d942a0ed44ecefa68", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Records completeness was derived for this assessment from the Multi-Topic Administrative Records Index, the indicator compiled by Twitter for Papua New Guinea in 2016, and the figures below are the assessment's rather than the index's.\n\nFacilities were grouped by the number of record types they report monthly and the share whose totals reconcile with the provincial office computed: 81 per cent of facilities reporting six or more types against 43 per cent of those reporting three or fewer. The index scores each record type separately and offers no reconciliation measure at all.\n\nDistance to the provincial office was then read from the same records and equalised, and the distance between the two groups narrows to 11 percentage points, which leaves the breadth of a facility's reporting habit, rather than how far it sits from the office, as what the assessment treats as the operative factor.\n\nFacilities that closed during the reporting period are excluded rather than counted as non-reporting, and they are 3 per cent of facilities. The grouping rule and the counts behind each share are recorded under the table.\n\nMention: \"Multi-Topic Administrative Records Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Multi-Topic Administrative Records Index supports a reconciliation measure by breadth of reporting.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:a22a54527e40e35043a3f6f9", "occurrence_id": "908994521f72ca567d1696a0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Informal employment was measured for this assessment from the Nationwide Informal Sector System (NISS), Round 3, the administrative collection National Institute of maintains, and the measures below are the assessment's rather than the collection's.\n\nWorkers were grouped by whether their activity holds a municipal trading permit and the share reporting any cash sickness benefit computed: 26 per cent of permit holders against 7 per cent of those trading without one. The collection tabulates permits and benefit receipt separately, so no such cross-reading appears in it.\n\nYears in the trade were then equalised between the two groups and the difference narrows to 8 percentage points, which leaves the permit itself as the larger part of the distance rather than how long a worker has been in the trade.\n\nWorkers who hold a permit but have not traded in the reference month are excluded rather than counted as covered, and they are 9 per cent of respondents. The bands, the denominators and the date of extraction are recorded under the table.\n\nMention: \"Nationwide Informal Sector System (NISS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Nationwide Informal Sector System links trading permits to cash sickness benefit cover.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:b08f4f45837fea38da18c139", "occurrence_id": "d1e715b2b7280f4e8b0250ae", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Return to work was computed for this review from the Follow-Up Unemployment Insurance Survey (FUIS), Version 3, which Kenya ICT Action Network conducts in Bangladesh, and none of what follows appears in the survey's own reporting.\n\nClaimants were grouped by the reason recorded for the separation from their last job and the share taking up work within thirteen weeks computed: 62 per cent among those separated by a closure against 38 per cent among those separated for performance. The survey publishes separations and exits from the register as two separate distributions.\n\nWeeks of benefit drawn were then equalised between the groups and 11 percentage points of the 24-point distance remain, which the review reads as a difference in how transferable a worker's last occupation is rather than in how long the register kept them in view.\n\nClaimants whose case was closed for a missing report are excluded rather than counted as re-employed, and they are 7 per cent of cases. The grouping rule and the counts behind each share are recorded under the table.\n\nMention: \"Follow-Up Unemployment Insurance Survey (FUIS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Follow-Up Unemployment Insurance Survey compares return-to-work speed by reason for separation.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:626bc7c560ea522803a1c7c2", "occurrence_id": "7552344e77148fa20cfc2857", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Dispersion of income was rebuilt for this review from the Harmonised Inequality Microdata, the microdata set Fed compiled for Afghanistan in 2015, and the figures that follow are the review's construction.\n\nQuintiles were formed on equivalised household income and the share of total income held by the top quintile computed: 47 per cent nationally, against 34 per cent among households whose main occupation is farming and 58 per cent among those whose main occupation is trade. The set publishes incomes at the household level and never reports a quintile share, by design.\n\nWhether the household sits in an urban location was then equalised within each occupation group and the distance between farming and trade narrows to 19 percentage points, which leaves the occupation itself rather than the town-country split as what the review treats as the larger part of the dispersion.\n\nHouseholds reporting no income in the reference period are kept in the distribution rather than dropped, since removing them would raise the top share by 3 points and that choice would be the review's rather than the set's. The equivalence rule and the extraction date sit with the annex tables.\n\nMention: \"Harmonised Inequality Microdata\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Harmonised Inequality Microdata supports quintile-share readings built by this review by occupation.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:6d93ef8d1de81b31a46f64cb", "occurrence_id": "ef16a5c3c374190fe4ff7bd1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Brightness and dryness were compared for this assessment using the Provincial Nighttime Lights Dataset, the geospatial dataset the Colombian Ministry of Education compiled for Papua New Guinea in 2022, and the comparison is the assessment's rather than the dataset's.\n\nCells were grouped by the share of their dry-season nights below the detection floor and the mean radiance recorded where a settlement is present computed: 21 nanowatts per square centimetre in cells with fewer than 20 such nights against 8 in cells with more than 60. The dataset reports radiance and the detection count per cell without combining them into any such reading.\n\nThat comparison supports one claim, that the electrification baseline should be built on cells with fewer than 20 undetected dry-season nights rather than on the full grid, and cells flagged with a burn event in the same period, which report 3 nanowatts less at the same detection count, are set aside for the same reason. The cost figures the plan uses come from the utility's capital programme, to which the dataset supplied nothing.\n\nCells whose class flags them as permanent water are excluded rather than read as unlit settlements, 5 per cent of the grid. The grouping rule and the counts behind each mean are recorded under the table.\n\nMention: \"Provincial Nighttime Lights Dataset\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Provincial Nighttime Lights Dataset supports one baseline claim by comparing radiance and undetected nights.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:5098e1e12c6b66e73f693ab0", "occurrence_id": "e32f743cd9597c8d510ce931", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Settlement records were catalogued into the elevation file from the Local Settlement and Elevation Census (LSEC), Version 9, the census conducted across South Asia in 2023, and the work done on them was filing rather than measurement.\n\nEach settlement was given the file's own feature code and matched against the elevation recorded for its coordinates; where the census and the file's existing entry differ by more than 50 metres the record was parked for review rather than corrected in place. Settlements whose coordinates arrive without an elevation value were left uncoded rather than filled from the neighbouring polygon, and 1 320 of them entered the file in that condition. The census's own enumeration date was preserved on every record so a later version can be matched without re-asking the field team.\n\nThe file's elevation statistics come from its own survey of the high-resolution surface, and this filing assists that work only in the narrow sense that a settlement now carries the same code in both places, so a record in one can be looked up in the other. No figure in the file is computed from the census's records.\n\nThe coding sheet, the count of parked records and the date of deposit are filed in the annex log.\n\nMention: \"Local Settlement and Elevation Census (LSEC)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The Local Settlement and Elevation Census supplies settlement records coded into the elevation file.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:560f9a8c40fa34665b46dea8", "occurrence_id": "4e624db58ec10382caf39915", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Cost burden was measured for this assessment from the Monthly Housing Conditions Survey, which GTAP runs in Romania in 2020, and the readings below are the assessment's own.\n\nHouseholds were grouped by tenure and the share whose housing cost leaves them under the national adequacy line after rent or upkeep computed: 31 per cent of renters against 9 per cent of owner-occupiers with no loan. The survey reports tenure and each cost item separately and never nets cost against the adequacy line.\n\nWhether the household receives a housing allowance was then read from the same records, and renters receiving one sit 12 points above other renters but still 19 points above owner-occupiers, which the assessment reads as the allowance offsetting part of a structural gap rather than closing it.\n\nHouseholds not reporting a monthly cost figure are excluded rather than treated as cost-free, and they are 8 per cent of respondents. The line used, the tenure groups and the counts behind each share sit with the annex tables.\n\nMention: \"Monthly Housing Conditions Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Monthly Housing Conditions Survey compares adequacy after housing cost by tenure and allowance receipt.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:db8cdbb84796c4516b649ec7", "occurrence_id": "3089e534ddc6200bdcc8633a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Debt and savings figures for the country are taken from the Quality-Assured Debt and Savings Survey, which Germanwatch e.V. conducted in Burundi in 2020, and the four values below are given exactly as the survey releases them.\n\nThe survey puts the share of households carrying a cash debt at 57 per cent, the median debt as a multiple of monthly income at 1.4, the share of households with any formal savings account at 19 per cent, and the share of households that had moved and carried a debt with them at 8 per cent. None was recomputed here, and the survey's own definitions of a cash debt, of a formal account and of a carried debt are reproduced with the values.\n\nNo debt owed to another household member was counted as a cash debt, and no livestock held for sale was counted as savings. Where the survey reports for one province, the province is named with the value rather than pooled into the national figure.\n\nThe survey table and line for each value, along with the extraction date and the households behind each share are recorded in the annex notes.\n\nMention: \"Quality-Assured Debt and Savings Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Quality-Assured Debt and Savings Survey supplies debt, savings and carried-debt shares for the country.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:86be91a6988458539797365e", "occurrence_id": "f51fb68315429c4ade83d725", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two non-state schooling lines in the education annex come from the Subnational Non-State Education System, the register World Bank/IFC maintains for Uzbekistan as of 2024, and both are reproduced without amendment.\n\nThe register puts the share of pupils enrolled in a non-state provider at 11 per cent and the share of those providers holding a current operating licence at 68 per cent. Nothing here was re-derived, and the register's own test for a current licence, which requires a renewal filed within the year, appears with the figures.\n\nThe annex's other lines are taken from the ministry's school census, so the register stands behind two lines while the census carries the rest. A figure given for one region is printed with that region named rather than averaged across the country.\n\nThe register reference behind each line, the extraction date and the providers behind each share are recorded with the annex notes.\n\nMention: \"Subnational Non-State Education System\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Subnational Non-State Education System supplies enrolment share and licence currency for the annex.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:e038ab937b33fd41d2bebc1b", "occurrence_id": "1bede1866918affe9709fa07", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Targeting and payment figures for the programme are held in the Coordinated Targeting and Payments System (CTPS), the administrative register Colombian Ministry of Education operates for 2017, and the four values below are transferred unchanged.\n\nThe register puts the share of enrolled households whose score falls below the programme threshold at 92 per cent, the share of payments reaching the intended recipient's own account at 84 per cent, the mean months between enrolment and first payment at 4, and the share of households dropped from the roll for a failed re-verification at 6 per cent. None was recomputed here, and the register's definitions of a score, of a recipient account and of a failed re-verification accompany the values.\n\nNo payment returned by the bank as undeliverable was counted as reaching its recipient, and no household re-scored upward was counted as wrongly dropped. Where the register reports for one operational region, that region is named with the value rather than pooled.\n\nThe register reference for each value, the date of extraction and the households behind each share are stored in the annex notes.\n\nMention: \"Coordinated Targeting and Payments System (CTPS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Coordinated Targeting and Payments System supplies targeting accuracy, payment reach and delay values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:ed310e0bf2a07e6ee73ea7dd", "occurrence_id": "d23c3aba2d2ecf063a62f6e0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Entry and exit rates were worked out for this assessment from the High-Frequency Business Demography Survey (HBDS), Round 2, which covers Azerbaijan in 2015, and the rates below are the assessment's own arithmetic on the survey's records.\n\nQuarterly, units were classified by whether they had begun trading within the quarter and whether an existing unit ceased, and the entry rate computed at 3.1 per cent of the active stock against an exit rate of 2.2 per cent. The survey publishes a stock count for each round and the dates of opening and closing, but no rate derived from them.\n\nSector was then separated out and services record an entry rate of 4.4 per cent against 1.2 per cent in manufacturing, while exits run the other way at 1.1 and 3.6 per cent, which is the asymmetry the assessment treats as the reason the services stock grows without any net subsidy.\n\nUnits whose round-to-round status cannot be resolved are excluded from both rates rather than assumed unchanged, and they are 5 per cent of units. The rate definitions and the counts behind each figure sit with the annex tables.\n\nMention: \"High-Frequency Business Demography Survey (HBDS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The High-Frequency Business Demography Survey supports entry and exit rates derived by this assessment.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:6ce203e3e5d92fe09f890e2d", "occurrence_id": "093c1b92c2c1bb53eea3713d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Whether the council's own displacement figure holds was tested against the Periodic Internal Movement Survey, which Barro and Lee conducted in the Philippines in 2022, and the survey is the principal source for the test.\n\nThe council's counting sheet puts households displaced by the storm season at 148 000. The survey, which samples households whether or not they present at a site, puts the equivalent count at 196 000, because it counts households that moved in with relatives and never registered at a site. The council's figure counts those who appear at a designated site.\n\nThe survey therefore carries the section's displacement claim, and the council's smaller count is printed beside it so the reader can see that the distance is the households absorbed by relatives rather than a disagreement about what counts as displaced. Where a household appears in both, the survey's report is used and the two are not averaged.\n\nIncome enters the test because the survey finds 41 per cent of the additional households below the national adequacy line before the storm against 27 per cent among those the council counts, and that difference is what the section uses to argue the sites under-count the poorest. The counting rule each source applies is recorded with the annex tables.\n\nMention: \"Periodic Internal Movement Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Periodic Internal Movement Survey is the principal source testing the council's storm displacement count.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:86b4e3fc7ca7cb79e95e7b1a", "occurrence_id": "81bb903bd0ed6d6b585d9590", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The inspection figures in this chapter are drawn from the labour office's own returns, and the Inter-Agency Labour Inspection Survey (ILIS), Seventh Phase, compiled by ILO for 2018, is referred to only as background to them.\n\nThe survey reports that 12 per cent of establishments were visited at least once in the reference year, that the mean interval between visits at a visited establishment stands at 3 years, and that 34 per cent of visits follow a worker complaint rather than a plan. All three are given as the survey issues them, with its own definition of a visit stated, and no chapter figure rests on them.\n\nNone of the chapter's tables rest on its responses. The figures the chapter argues from come from the office's returns, which record visits actually logged rather than establishments recalling one, and where the two accounts of visit frequency differ the difference is noted rather than reconciled.\n\nThe annex lists the survey tables consulted and the date of extraction, which is the whole of the use made of the survey here.\n\nMention: \"Inter-Agency Labour Inspection Survey (ILIS)\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The Inter-Agency Labour Inspection Survey is background while the chapter's inspection figures come from returns.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:4d8ab1801db89c058eafc195", "occurrence_id": "ee356408ef8c33c6dffbedef", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Referral capacity was assessed for this review from the Municipal Facility Assessment Survey (MFAS), conducted in Nicaragua in 2016, and the measures below are the review's rather than the survey's.\n\nFacilities were matched to their location and grouped by the road distance to the nearest hospital with a surgical unit, and the share keeping a surgeon on site at night computed: 12 per cent of facilities beyond 40 kilometres against 46 per cent of those within 10 kilometres. The survey holds the distance for each facility and the staffing items for each facility but never reports one against the other.\n\nBed count was then equalised between the two groups and the distance narrows to 15 percentage points, which leaves how far the facility sits from a surgical unit as the larger part of the staffing gap than the size of the facility itself.\n\nFacilities with no coordinates recorded are excluded rather than placed at the municipal seat, and they are 4 per cent of facilities. The bands, the derivation and the counts behind each share sit with the annex tables.\n\nMention: \"Municipal Facility Assessment Survey (MFAS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Municipal Facility Assessment Survey compares night surgical cover by distance to a surgical hospital.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:bb70814e11ba0bdb619ba6f5", "occurrence_id": "3b278064cf90700ec581441d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The registry figures in this chapter are compiled from the agency's own service register, and the Enumerative Service Register Survey (ESRS), Eighth Phase, conducted by UNHCR, IOM, is mentioned only as background to them.\n\nThe survey reports that 79 per cent of registered households could name the service point nearest to them, that 22 per cent had used a service point in the previous month, and that the mean number of households behind each point stands at 410. All three are given as the survey issues them, with its own definition of a service point stated, and nothing in the chapter is computed from them.\n\nNo chapter table rests on its responses. The figures the chapter argues from come from the register, which counts contacts made rather than whether households know a point exists, and the two accounts are kept apart rather than merged.\n\nThe annex lists the survey tables consulted and the date of extraction, and the survey is used for nothing beyond that background.\n\nMention: \"Enumerative Service Register Survey (ESRS)\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The Enumerative Service Register Survey is background while the chapter's figures come from the service register.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:35da80a7106e5c05df55c1ef", "occurrence_id": "4ea2657851da1c78a8c53aec", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Transition after school was measured for this assessment from the District Youth Transition Survey, which the World Food Programme follows in Malaysia in 2023, and the figures below are the assessment's own.\n\nLeavers were grouped by whether they completed the final year of secondary school and the share in education or work nine months later computed: 77 per cent of completers against 44 per cent of non-completers. The survey reports completion status and current activity as separate items and does not cross them.\n\nWhether the young person reported a long-standing health problem was then read from the same records, and non-completers reporting such a problem sit 13 points below other non-completers, which the assessment treats as evidence that health support, not only re-enrolment, belongs in the transition programme.\n\nLeavers in full-time military service are excluded rather than counted as employed, and they are 3 per cent of leavers. The grouping rule and the counts behind each share are recorded under the table.\n\nMention: \"District Youth Transition Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The District Youth Transition Survey links school completion and health problems to transition outcomes.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:3bffd25a8aeef4dc11049988", "occurrence_id": "2b8eeecd4b1b8028cb31e381", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Apprenticeship figures for 2022 are reported from the Integrated Apprenticeship Survey (IAS), which OECD conducts, and the four values below are carried over as the survey issues them.\n\nThe survey puts the share of employers offering at least one apprenticeship place at 16 per cent, the mean duration of a completed programme at 2.1 years, the share of apprentices holding a paid contract throughout at 71 per cent, and the share of employers reporting a vacancy they could not fill with a former apprentice at 28 per cent. None was recomputed here, and the survey's own definitions of an offering employer and of a completed programme appear with the values.\n\nNo employer whose only place was withdrawn before the start was counted as offering, and no apprentice whose contract was converted to permanent work was counted as a remaining vacancy. Where the survey reports for one size class, the class is named with the value rather than pooled.\n\nThe survey table and line for each value: the extraction date and the establishments behind each share are listed in the annex notes.\n\nMention: \"Integrated Apprenticeship Survey (IAS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Integrated Apprenticeship Survey supplies offering, duration, contract and vacancy-fill values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:153878f252624b9f212475a1", "occurrence_id": "e0f6707b82e361b8f2191723", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Working hours in informal activity were analysed for this review from the Cross-Sectoral Informal Sector Survey (CISS), Phase 7, which covers Sri Lanka in 2018, and the readings below are the review's rather than the survey's.\n\nWorkers were grouped by whether they work at a fixed structure or move and the mean hours worked in the reference week computed: 58 hours among mobile workers against 47 hours among those with a fixed stall. The survey publishes each activity's hours separately and offers no comparison of this kind.\n\nThe number of dependents was then equalised between the two groups and the distance narrows to 6 hours, which leaves the mobility of the activity rather than household pressure as what the review reads as the larger driver of the longer week.\n\nWorkers whose main activity is seasonal and who were idle in the reference week are excluded rather than counted as working zero hours, and they are 11 per cent of respondents. The grouping rule and the counts behind each mean sit with the annex tables.\n\nMention: \"Cross-Sectoral Informal Sector Survey (CISS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Cross-Sectoral Informal Sector Survey compares working hours between mobile and fixed informal workers.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:bba9cb1b923be043ebb8b77a", "occurrence_id": "e541bac76c3de14d53715fc6", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Placement after training was computed for this assessment from the Harmonised Vocational Training Survey, which World Bank-IFC runs in Aceh province in 2023, and the figures below are the assessment's own.\n\nGraduates were grouped by whether their course included a workplace placement and the share in paid work six months after completion computed: 63 per cent of those with a placement against 39 per cent of those without. The survey reports course design and current status in separate tables and never joins them.\n\nThe trade trained for was then equalised across the two groups and the distance narrows to 9 percentage points, which leaves the placement itself as the larger part of the effect rather than the choice of trade.\n\nGraduates who enrolled in further study are excluded rather than counted as jobless, and they are 8 per cent of graduates. The grouping rule and the counts behind each share are recorded under the table.\n\nMention: \"Harmonised Vocational Training Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Harmonised Vocational Training Survey compares paid work after training by workplace placement.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:e09352b2b2cfd1da9d305bf0", "occurrence_id": "0817f8764e32d29f6d830f1b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two monitoring lines in the delivery annex are quoted from the Statistical Programme Monitoring Index, compiled by People’s Bank of China for Argentina in 2024, and both are given without restatement.\n\nThe index puts the share of programme milestones met by their planned date at 66 per cent and the share of reporting units whose data were complete for every period at 54 per cent. Neither was recalculated here, and the index's own rule for counting a milestone as met, which requires evidence filed rather than a claim by the unit, appears with the figures.\n\nThe annex's other lines come from the programme's own quarterly narrative, so the index carries two lines while the narrative carries the remainder. A figure given for one component is printed with that component named rather than averaged across the programme.\n\nThe index entry behind each line, the extraction date and the units behind each share are recorded with the annex notes.\n\nMention: \"Statistical Programme Monitoring Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Statistical Programme Monitoring Index supplies milestone and completeness lines quoted in the annex.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:1c0c14459c8364e8f221e669", "occurrence_id": "6e65953d959a3ca2c850fdc6", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Emergency care figures measured across the operation are read from the Endline Emergency Care Survey (EECS), which Frontex conducted for 2017, and the four values below are given as the survey issues them.\n\nThe survey puts the share of incidents receiving first aid at the point of encounter at 74 per cent, the mean minutes between an incident being logged and a clinician reaching it at 22, the share of vessels or sites carrying a resuscitation kit at 61 per cent, and the share of incidents transferred to shore care within two hours at 48 per cent. None was recomputed here, and the survey's own definitions of an encounter, of a logged incident and of a transfer appear with the values.\n\nNo incident resolved without medical contact was counted in the response time, and no kit past its expiry date was counted as carried. Where the survey reports for one location of the operation, that location is named with the value rather than pooled.\n\nThe survey table and line for each value, the extraction date, and the incidents behind each share are set down in the annex notes.\n\nMention: \"Endline Emergency Care Survey (EECS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Endline Emergency Care Survey supplies first-aid, response-time, kit and transfer values for the operation.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:28b7fb846dc8dfaa6cd35c5a", "occurrence_id": "804eb64522c1fab5a6f75a32", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Trader records were catalogued into the livelihoods file from the Joint Informal Trade Registry (JITR), Wave 6, the database IRC maintains, and the work done on them was cataloguing rather than measurement.\n\nEach trader record was given the file's own trade code, checked against the market roll and, where the record names a stall number absent from that roll, left for the field officer rather than matched to the nearest stall. Records with no market stated were left uncoded rather than assigned by the trader's residence, and 1 470 of them entered the file in that condition. The registry's own registration date was kept on every record so a later wave can be matched without re-interviewing the trader.\n\nThe file is the primary place where informal trading is counted for this report, and the registry is what supplies its records: each trader in the file originates from a registry entry, and the coding step is what lets the file's totals be read by trade rather than by market alone. Nothing in the file is derived from the registry's own summary tables.\n\nThe coding sheet, the count of records left for the officer and the date of deposit are written up in the annex log.\n\nMention: \"Joint Informal Trade Registry (JITR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The Joint Informal Trade Registry supplies the coded trader records that underpin the livelihoods file.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:ddcaf2fdbb3c5d6a9f7619a2", "occurrence_id": "a7a45f3341615cf0857f54cc", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two monitoring lines in the programme annex are quoted from the Continuous Programme Monitoring Index (CPMI), Wave 9, the index Baier and Bergstrand compile for Azerbaijan, and both are carried over unchanged.\n\nThe index puts the share of disbursement linked to a verified activity at 88 per cent and the share of activities whose verification was carried out by a party other than the implementing unit at 37 per cent. Neither was recomputed for this annex, and the index's own definition of a verified activity, which requires a signed completion note, is reproduced with them.\n\nThe annex's other entries are drawn from the fund's own payment ledger, so the index carries two lines while the ledger carries the rest. A figure given for one tranche is printed with that tranche named rather than pooled across the programme.\n\nThe index reference behind each line, the extraction date and the activities behind each share are recorded with the annex notes.\n\nMention: \"Continuous Programme Monitoring Index (CPMI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Continuous Programme Monitoring Index supplies verified-disbursement and independent-verification lines.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:3d923515109c29aa919ec53f", "occurrence_id": "c87434b8679c5d820072d88c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Asset holdings and farm operation were compared for this assessment using the High-Frequency Asset Ownership Survey, which the Asian Disaster Reduction Center conducted in the West Bank and Gaza in 2016, and the comparison is the assessment's rather than the survey's.\n\nHouseholds were grouped by whether they hold title to the land they farm and the share reporting an income below the national adequacy line computed: 41 per cent of households without title against 18 per cent of those holding it. The survey reports land tenure and consumption in separate sections and never crosses them.\n\nIrrigated area was then read from the same records and equalised, and the distance between the two groups narrows to 12 percentage points, which leaves the area under water as a substantial part of what looks like a tenure effect, and the assessment reports both readings rather than choosing between them.\n\nHouseholds farming land they neither own nor rent are excluded rather than grouped with the untitled, and they are 6 per cent of households. The tenure categories and the counts behind each share sit with the annex tables.\n\nMention: \"High-Frequency Asset Ownership Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The High-Frequency Asset Ownership Survey links land title to adequacy of income among farming households.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:4e53374d57660027e5213142", "occurrence_id": "b69a99e669e72c731085c8d8", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Office records were catalogued into the operations file from the Decennial Regional Operations Index (DROI), Edition 6, the indicator compiled for Serbia in 2022, and the work done on them was cataloguing rather than measurement.\n\nEach office record was given the file's own function code and matched to its administrative level; where a record names a function the file does not carry, it was parked for the statistician rather than mapped to the nearest existing heading. Records arriving without an administrative level were left uncoded rather than assumed to sit at district level, and 310 of them entered the file that way. The index's own reporting period was kept on every record so a later edition can be matched without re-collecting the returns.\n\nThe operations file is where this report's office-level statistics are held, and the index is what supplies its records: every office in the file originates from an index entry, and the coding step is what allows the file's totals to be read by function rather than by region alone. No figure in the file is computed from the index's published aggregates.\n\nThe coding sheet, the count of parked records and the date of deposit are recorded in the annex log.\n\nMention: \"Decennial Regional Operations Index (DROI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The Decennial Regional Operations Index supplies the coded office records behind the operations file.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:dad8f582eea09f0e8b5c79dc", "occurrence_id": "1a3633b2ec33e8cb0256b00f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The transfer value in this chapter is the programme's own, taken from its payment runs, and the Pilot Graduation Programmes Survey (PGPS), Collection 4, conducted in Burkina Faso in 2016, is cited to test it.\n\nThe programme's runs put the mean annual transfer per enrolled household at 210. The survey puts the mean actually received at 186, because it counts only payments households could evidence and excludes transfers reduced at the collection point. The programme's figure is gross of those deductions.\n\nThe gap of 24 sits inside the tolerance the programme applies to its own reconciliation between committed and evidenced payments, so the programme's figure is printed as committed and the survey supplies no revision to it and no values for the periods its own fieldwork missed.\n\nExcluded from the comparison are households enrolled after the last payment run, since neither source can report a receipt for them. What each source treats as received and what each excludes are recorded with the annex tables.\n\nMention: \"Pilot Graduation Programmes Survey (PGPS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Pilot Graduation Programmes Survey checks the programme's mean transfer value against evidenced receipts.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:d830d23b11200327c1b8bdb0", "occurrence_id": "157f679fe7dc223e61a0202e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Continuity of care was measured for this review from the Rolling Primary Care Survey (RPCS), Phase 2, which covers the CHT and Sylhet divisions in 2018, and the measures below are the review's rather than the survey's.\n\nPatients with a chronic condition were grouped by whether their facility keeps a paper register and the share returning for their next scheduled visit computed: 68 per cent at facilities with a register against 49 per cent at those without. The survey records the register item and the visit history separately and produces no such comparison.\n\nDistance to the facility was then read from the same records and equalised, and the distance narrows to 8 percentage points, which leaves the register rather than the journey as what the review treats as the operative difference between the two groups of facilities.\n\nPatients who moved out of the survey area during the follow-up are excluded rather than counted as non-attenders, and they are 5 per cent of patients. The grouping rule and the counts behind each share are recorded under the table.\n\nMention: \"Rolling Primary Care Survey (RPCS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Rolling Primary Care Survey compares chronic-care return rates by whether a facility keeps a register.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:3da830f4a53ba2460715789f", "occurrence_id": "f39cc8d8f3d9e51f560f0e48", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two built-up area lines in the urban annex are quoted from the Combined Built-Up Surface Report, issued by Amadeus for Eastern Europe in 2020, and both are reproduced as the report prints them.\n\nThe report puts the built-up surface per inhabitant at 78 square metres across the cities it maps and the share of new built-up area falling inside the existing settlement edge at 61 per cent. Neither was recalculated here, and the report's own rule for what counts as built-up, which requires an impervious surface, is reproduced with the figures.\n\nThe annex's other lines are compiled from the cadastral office's parcel counts, so the report carries two lines while the parcel counts carry the rest. A figure given for one country is printed with that country named rather than averaged across the region.\n\nThe report's page and table behind each line, the extraction date and the cities behind each share are recorded with the annex notes.\n\nMention: \"Combined Built-Up Surface Report\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Combined Built-Up Surface Report supplies built-up surface per inhabitant and infill shares for the annex.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:c1fab71092c1d3beea8bd831", "occurrence_id": "ce4109d69a81dc0ed619b18f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The storm-surge figure in this chapter is the harbour office's own, taken from its gauge record, and the Recurrent Storm Surge Registry (RSSR), Third Collection, compiled for Mauritania in 2017, is cited to test it.\n\nThe office's gauge record puts the highest surge in the reporting year at 1.9 metres. The registry puts it at 2.2 metres for the same event, because it reports the still-water level at the tide gauge nearest the event while the office adjusts its reading to the quay datum. No conversion was applied to either figure for the comparison.\n\nThe 0.3-metre distance is smaller than the datum difference the registry itself notes between the two reference levels, so the office's figure is printed as recorded and the registry supplies no revision to it and no values for the years its own collection does not reach.\n\nExcluded are events the registry lists without a peak height, 7 of its entries, since no comparison can be formed for them. Which reference level each source uses is recorded with the annex tables.\n\nMention: \"Recurrent Storm Surge Registry (RSSR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Recurrent Storm Surge Registry checks the harbour office's peak surge height for the reporting year.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:b9ca9a41aa0fb2782fa08793", "occurrence_id": "09c2256a9c380b602db2639b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Curriculum reform figures for the department are reported from the Departmental Curriculum Reform Survey (DCRS), which covers Russia in 2023, and the four values below are given exactly as the survey releases them.\n\nThe survey puts the share of schools teaching the revised curriculum at 72 per cent, the share of teachers reporting they received the new materials before the term began at 45 per cent, the share of pupils in a class containing a child who had moved to the school during the year at 19 per cent, and the share of schools receiving additional support for pupils from households below the regional income line at 33 per cent. None was recomputed here, and the survey's own definitions of a revised class, of a moved child and of the income line are reproduced with the values.\n\nNo school that had received the materials but not begun teaching was counted as teaching the curriculum, and no child who moved between two classes in the same school was counted as new. Where the survey reports for one district, the district is named with the value.\n\nThe survey table and line for each value, the extraction date as well as the schools behind each share are detailed in the annex notes.\n\nMention: \"Departmental Curriculum Reform Survey (DCRS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Departmental Curriculum Reform Survey supplies curriculum, materials, mobility and support shares.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:ae112624a9e2cb287e93ee21", "occurrence_id": "094300b323eabcf5e7a74215", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two operations lines in the utility annex are quoted from the Continuous Operations Monitoring Registry, the register Water Aid maintains for Italy as of 2019, and both are given without restatement.\n\nThe register puts the share of supply zones whose pressure is logged continuously at 69 per cent and the median hours of unserved demand per zone per year at 14. Neither was re-derived for this annex, and the register's own rule for counting unserved demand, which requires a recorded interruption rather than an estimate, appears with the figures.\n\nThe remaining annex lines come from the regulator's annual return, so the register carries two lines while the return carries the rest. A figure given for one utility is printed with that utility named rather than averaged across the sector.\n\nThe register reference behind each line, the extraction date and the zones behind each share are recorded with the annex notes.\n\nMention: \"Continuous Operations Monitoring Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Continuous Operations Monitoring Registry supplies pressure-logging and unserved-demand lines.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:104dcc087dd63190898301e3", "occurrence_id": "a32a50b38d5cb6cfec33bb12", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two records lines in the health annex are quoted from the Periodic Integrated Records Index (PIRI), Release 7, compiled by Hungarian NHIF for 2021, and both are carried over unchanged.\n\nThe index puts the share of provider records matched to a person register at 93 per cent, the share of records requiring a manual correction before use at 8 per cent, and the median age of a claim at the point of matching at 11 days. Neither of the first two was recomputed here, and the index's own definition of a match, which requires an identifier rather than a name and date, is reproduced with them.\n\nThe annex's other lines come from the fund's own claims file, so the index carries two lines while the claims file carries the rest. A figure given for one provider class is printed with that class named rather than pooled across providers, and records the index itself sets aside as unmatchable are not restored to the denominator for this report.\n\nThe index entry behind each line, the extraction date and the records behind each share are recorded with the annex notes.\n\nMention: \"Periodic Integrated Records Index (PIRI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Periodic Integrated Records Index supplies matching, correction and claim-age lines for the health annex.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:b602fd2c4f3d8578e7383711", "occurrence_id": "07998759cbc06fc7e2a81d5d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Market reach was measured for this assessment from the Diagnostic Market Access Dataset, the geospatial dataset Fed compiled for Sub-Saharan Africa in 2018, and the measures below are the assessment's own rather than the dataset's.\n\nRural localities were matched to the road network and grouped by the travel time to the nearest market of national size, and the share of households selling a surplus crop computed: 63 per cent of localities within 2 hours of travel against 34 per cent of those beyond 6 hours. The dataset holds the network and the market locations and publishes no travel-time reading of either.\n\nWhether the locality sits in a district whose rainfall fell below its own thirty-year norm in the same year was then equalised, and the distance between the two travel groups narrows to 18 percentage points, which leaves the road rather than the season's rain as what the assessment treats as the binding constraint on selling.\n\nLocalities with no mapped track at all are excluded rather than treated as infinitely remote, and they are 4 per cent of localities. The travel bands, the rainfall norm and the counts behind each share sit with the annex tables.\n\nMention: \"Diagnostic Market Access Dataset\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Diagnostic Market Access Dataset links travel time to market with crop sales and the season's rainfall.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:5df8f298115c44137f5bac93", "occurrence_id": "57de639793c87e5a1f0de1ed", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Consumption dispersion was built for this review from the Subnational Inequality Survey (SIS), Collection 5, conducted in Guatemala in 2018, and the figures below are the review's rather than the survey's.\n\nDepartments were ranked by the median household consumption recorded for them and grouped at the two ends of that ranking, and the share of households below the national adequacy line computed: 61 per cent in the lowest-ranked group against 14 per cent in the highest. The survey reports consumption and each item's prevalence without publishing any departmental ranking of its own.\n\nThat comparison supports one claim, that the targeting roll should be re-weighted toward the lowest-ranked departments rather than adjusted by population. The budget figures the proposal uses come from the finance ministry's allocation note, from which the survey supplied nothing.\n\nDepartments with fewer than thirty completed interviews are excluded rather than reported, and they are 9 per cent of departments. The ranking rule and the counts behind each share are recorded under the table.\n\nMention: \"Subnational Inequality Survey (SIS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Subnational Inequality Survey supports one targeting claim by ranking departments on consumption.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:32f8032aa3b36167341b6e43", "occurrence_id": "db88619a9790bdc3239a866e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two registry lines in the statistical annex are quoted from the Rolling Administrative Survey Registry (RASR), Wave 6, the database UNIDO maintains, and both are given exactly as the registry holds them.\n\nThe registry puts the share of surveys in its inventory with a published microdata release at 41 per cent and the median interval between fieldwork and that release at 20 months. Neither was recalculated for this annex, and the registry's own rule for counting a release, which requires an accessible file rather than a request procedure, appears with the figures.\n\nThe annex's remaining lines are compiled from the statistical office's own publication log, so the registry carries two lines while the log holds the other lines. A figure given for one survey family is printed with that family named rather than averaged across the inventory.\n\nThe registry reference behind each line, the extraction date and the surveys behind each share are recorded with the annex notes.\n\nMention: \"Rolling Administrative Survey Registry (RASR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Rolling Administrative Survey Registry supplies microdata-release shares and publication delay lines.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:f0b842824cac5843526c4946", "occurrence_id": "c1ba5a35bc7afae73ce4e913", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Operations the offices figures are compiled from the Nationwide Regional Operations System (NROS), Third Phase, the administrative register maintained by Development Strategy Center, and the four values below are transferred here unchanged.\n\nThe register puts the share of offices with a functioning internet connection at 77 per cent, the mean number of cases a clerk closes in a day at 12, the share of offices whose registers were inspected within the year at 55 per cent, and the share of inspected offices whose register agreed with the returns filed at 68 per cent. None was recomputed here, and the register's own definitions of a functioning connection and of an agreeing register accompany the values.\n\nAn office closed for more than a month for refurbishment was excluded from the caseload mean, and an inspection conducted by the office itself was not counted as an inspection. Where the register reports for one region, that region is named with the value rather than pooled.\n\nThe register reference for each value, plus the extraction date and the offices behind each share are carried in the annex notes.\n\nMention: \"Nationwide Regional Operations System (NROS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Nationwide Regional Operations System supplies connectivity, caseload and inspection agreement values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:1a5a1c53b35c82ef1a14687e", "occurrence_id": "440133fd63e6f1aa38d01c42", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Poverty dynamics were rebuilt for this review from the Rapid Socioeconomic Survey, a survey conducted in Thailand in 2015 by The World Bank.\n\nHouseholds were grouped by whether a member changed occupation during the previous three years and the share falling below the national adequacy line in the survey year computed: 29 per cent of households with such a change against 13 per cent of those without. The survey publishes an occupational history and a consumption aggregate but never reports the two together.\n\nWhether the household also experienced a health payment larger than one month's income was then read from the same records, and the distance between the two groups narrows to 7 percentage points, which leaves the health shock, rather than the job change itself, as what the review treats as the mechanism moving households below the line.\n\nHouseholds whose occupation history is incomplete are excluded rather than treated as unchanged, and they are 12 per cent of households. The grouping rule and the counts behind each share sit with the annex tables.\n\nMention: \"Rapid Socioeconomic Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Rapid Socioeconomic Survey links occupation change and health payments to falling below the poverty line.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:3564bf3493f0701bed57a98f", "occurrence_id": "be7a704312ab8da7a5112428", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The water-board figures in this chapter are calculated from the board's own abstraction returns, and the Analytical Irrigation and Inputs Dataset (AIID), Edition 4, compiled by WorldPop, is quoted only to place those figures in context.\n\nThe dataset puts the share of equipped irrigation area within its mapped extent at 63 per cent, the share of those areas using a recorded input package at 47 per cent, and the mean field size within a mapped scheme at 1.8 hectares. All three are given as the dataset issues them, with its own definition of an equipped area stated, and no chapter figure was built from them.\n\nTheir use here is to show what the physical layout of the schemes looks like beside the volumes the board meters. The chapter's argument rests on the returns, which record water taken rather than the area it serves, and the two are reported separately rather than combined.\n\nThe annex lists the dataset layers consulted and the date of extraction, which is the full extent of the dataset's use in this chapter.\n\nMention: \"Analytical Irrigation and Inputs Dataset (AIID)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Analytical Irrigation and Inputs Dataset is quoted as context beside the water board's abstraction returns.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:f157f1d29996d9c35d44ecd6", "occurrence_id": "a84f683ddf87bc0a6a11950b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Debt stress was computed for this assessment from the Continuous Debt and Savings Survey (CDSS), Collection 4, compiled by NCAER, and the readings below are the assessment's rather than the collection's.\n\nHouseholds were grouped by the multiple of monthly income their total debt represents and the share reporting they met a repayment by cutting food spending computed: 38 per cent of households above two times monthly income against 11 per cent of those below half a month's income. The collection reports debt and each coping item separately and publishes no such crossing.\n\nWhether the debt is owed to a formal lender was then equalised between the two groups and the distance narrows to 15 percentage points, which leaves the size of the debt rather than the identity of the lender as what the assessment treats as the binding condition.\n\nHouseholds reporting no debt at all are kept in the lowest group rather than removed, since excluding them would raise both shares by 4 points and that choice would be the assessment's rather than the collection's. The bands and the counts behind each share sit with the annex tables.\n\nMention: \"Continuous Debt and Savings Survey (CDSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Continuous Debt and Savings Survey compares repayment by cutting food spending against debt size.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:9a7f094ef12888f92658fffd", "occurrence_id": "bf2ab16a236a61a3bd0b99ec", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Drought and response figures for the whole clusters are sourced from the Clustered Drought and Resilience Registry (CDRR), Seventh Round, the register maintained for Malawi in 2017, and the four values below are given as the registry holds them.\n\nThe registry puts the share of households in a declared cluster whose maize plot failed at 46 per cent, the share of those households that sold livestock in response at 22 per cent, the mean number of months a household reports a food gap at 4, and the share receiving any external assistance at 31 per cent. None was recomputed here, and the registry's own definitions of a failed plot and of a food gap appear with the values.\n\nNo plot watered from a scheme was counted as failed, and no animal sold for a school fee was counted as a drought response. Where the registry reports for one cluster, that cluster is named with the value rather than pooled.\n\nThe registry table and line for each value, along with the extraction date and the households behind each share are kept in the annex notes.\n\nMention: \"Clustered Drought and Resilience Registry (CDRR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Clustered Drought and Resilience Registry supplies plot failure, livestock sale, food gap and aid values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:497fc5df4fb1990c40a35e2d", "occurrence_id": "a6f67fdc3683de63d7ef05f9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two terrain lines in the appendix are quoted from the Subnational Elevation Model Dataset, the geospatial dataset NSS compiled for Serbia in 2020, and both are reproduced without amendment.\n\nThe dataset puts the mean elevation of its covered area at 337 metres above its own vertical datum, the share of the population living above 500 metres at 14 per cent, and the share of its surface where the slope exceeds 15 degrees at 21 per cent. Neither of the first two was recalculated here, and the datum the dataset uses is named with the figures rather than converted to another reference.\n\nThe appendix's other lines are compiled from the statistical office's settlement register, so the dataset carries two lines while the register carries the rest. A figure given for one district is printed with that district named rather than averaged across the country, and cells the dataset masks as no-data were not filled from the surrounding terrain for this report.\n\nThe dataset layer behind each line, the extraction date and the settlements behind each share are recorded with the annex notes.\n\nMention: \"Subnational Elevation Model Dataset\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Subnational Elevation Model Dataset supplies elevation, altitude-population and slope lines in the appendix.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:196aa05d4bddc11b02580a60", "occurrence_id": "cf93825803d640b727885931", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Whether the municipality's own count of standing structures survives was tested against the Nationwide Building Footprint Dataset (NBFD), the geospatial dataset compiled for Jordan in 2020, and the dataset is the principal source for the test.\n\nThe municipality's inspection roll lists 41 200 standing structures within the planned area. The dataset, which traces a polygon for every roof it can resolve, traces 46 800, because it counts a structure on a parcel even where the municipality has no inspection record for it. The municipality's roll is a record of inspected buildings rather than of buildings.\n\nThe dataset therefore carries the section's count of standing structures, and the municipality's lower figure is printed beside it so the reader can see that the distance is uninspected buildings rather than a disagreement about what a structure is. Where a polygon matches an inspection record, the municipality's attributes are used and the two counts are not averaged.\n\nStructures under construction at the time of acquisition are excluded from both counts rather than counted once by each source, 3 per cent of the dataset's polygons. The tracing rule and the inspection rule each source applies are recorded with the annex tables.\n\nMention: \"Nationwide Building Footprint Dataset (NBFD)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Nationwide Building Footprint Dataset is the principal source testing the municipality's structure count.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:a91d48e93b9bdf97adfb89d7", "occurrence_id": "233aabad2df589bac6babb5b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Reporting delays were rebuilt for this assessment from the Recurrent Programme Monitoring System (RPMS), the administrative register kept for Burkina-Faso in 2017, and the measures below are the assessment's rather than the register's.\n\nSites were grouped by the number of reporting lines required of them and the median weeks between a period closing and the site's return arriving computed: 3 weeks for sites filing three lines or fewer against 9 weeks for those filing more than eight. The register stamps each return with the period it covers and the date it arrived but never relates either to the reporting burden.\n\nDistance to the coordinating office was then read from the same records and equalised, and the distance narrows to 3 weeks, which leaves the number of lines demanded of a site as the larger part of the delay than how far it sits from the office.\n\nSites that filed nothing for the period are excluded rather than given a delay of the full period, and they are 6 per cent of sites. The grouping rule and the counts behind each median sit with the annex tables.\n\nMention: \"Recurrent Programme Monitoring System (RPMS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Recurrent Programme Monitoring System supports a return-delay measure related to reporting burden.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:733db1172bcdf4b8a1a48b7a", "occurrence_id": "2b66e40860f59e9d8c225a6f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Caseload per office was computed for this review from the Coordinated Regional Operations Census (CROC), Phase 9, conducted for 2017 by IFC, and the figures below are the review's own.\n\nOffices were grouped by the number of staff recorded against them and the mean cases each office closed per staff member in the period computed: 26 cases in offices with fewer than five staff against 41 in those with ten or more. The census counts staff and cases for each office and publishes no productivity figure.\n\nYears since the office opened were then equalised between the groups and the distance narrows to 5 cases, which leaves experience rather than size as the smaller part of the difference than the review's first draft assumed.\n\nOffices open for less than half the period are excluded rather than scaled to a full period, and they are 4 per cent of offices. The grouping rule and the counts behind each mean are recorded under the table.\n\nMention: \"Coordinated Regional Operations Census (CROC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Coordinated Regional Operations Census supports a cases-per-staff measure built by this review.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:aff471c00f6169badd695da2", "occurrence_id": "b91a95fb3ea7fff1306fe9c5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Youth transition the continent figures are tabulated by the National Youth Transition Report (NYTR), Round 5, issued by FPS for Africa, and the four values below are given as the report releases them.\n\nThe report puts the share of young people not in education, employment or training at 24 per cent, the mean years of schooling completed at 8, the share of school leavers who found paid work within a year at 31 per cent, and the share of employers reporting they would hire a leaver without prior experience at 42 per cent. None was recomputed here, and the report's own definitions of a leaver and of paid work appear with the values.\n\nNo young person doing unpaid work in a family enterprise was counted as employed, and no period of training after the age of eighteen was included in the schooling mean. Where the report gives a figure for one region, that region is named with the figure.\n\nThe report's table and page for each value, the extraction date and the respondents behind each share are held in the annex notes.\n\nMention: \"National Youth Transition Report (NYTR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The National Youth Transition Report supplies NEET, schooling, hiring and employer-willingness values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:c23d523afc41ff07ae9c8f35", "occurrence_id": "da017016e63ebaed955c009f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Fees and learning were compared for this assessment using the Harmonised Non-State Education Collection (HNEC), Edition 4, which covers Dar-es-Salaam in 2023, and the comparison is the assessment's rather than the collection's.\n\nProviders were grouped by the monthly fee they charge and the share of their pupils reaching the end-of-year benchmark computed: 58 per cent of providers charging above 20 per month against 34 per cent of those charging below 5. The collection reports fees for providers and benchmark results for pupils in separate tables and never relates the two.\n\nPupil-teacher ratio was then read from the same records and equalised, and the distance narrows to 11 percentage points, which leaves the ratio as a substantial part of what looks like a fee effect, and the assessment reports both readings.\n\nProviders that opened during the school year are excluded rather than included without a result, and they are 7 per cent of providers. The fee bands and the counts behind each share sit with the annex tables.\n\nMention: \"Harmonised Non-State Education Collection (HNEC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Harmonised Non-State Education Collection compares benchmark attainment by school fee level.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:964f5da1f14378d1f7de1990", "occurrence_id": "87664016a0db5ce3105e4b45", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two campaign lines in the health annex are quoted from the Sentinel Vaccination Campaign Microdata, the microdata set JFO compiled for Malawi in 2021, and both are reproduced as the set records them.\n\nThe set puts the share of children reached at a fixed post at 64 per cent and the share reached only by a mobile round at 19 per cent. Neither was recalculated here, and the set's own rule for counting a child as reached, which requires a recorded dose rather than a household report, appears with the figures.\n\nThe annex's other lines are compiled from the ministry's campaign returns, so the microdata carry two lines while the returns carry the rest. A figure given for one district is printed with that district named rather than averaged across the campaign, and households the set could not revisit after a missed round are left in the denominator rather than dropped.\n\nThe set's table behind each line, the extraction date and the children behind each share are recorded with the annex notes.\n\nMention: \"Sentinel Vaccination Campaign Microdata\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Sentinel Vaccination Campaign Microdata supplies fixed-post and mobile-round reach lines for the annex.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:acaaff6483510e04ca7748ca", "occurrence_id": "9a11550fff1b38c5ba769cca", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Bulletin production figures are reported from the Decennial Statistical Bulletin Index (DSBI), the index MEC compiled for 2022, and the four values below are transferred here unchanged.\n\nThe index puts the share of scheduled bulletins issued in the year at 83 per cent, the median pages per bulletin at 42, the share of bulletins issuing a revision after publication at 11 per cent, and the share whose source tables are published with the bulletin at 57 per cent. None was recomputed here, and the index's own definitions of a scheduled bulletin and of a revision accompany the values.\n\nA bulletin withdrawn before issue was not counted as issued, and a correction confined to the website was not counted as a revision. Where the index reports for one statistical domain, that domain is named with the value rather than pooled.\n\nThe index reference for each value: the extraction date and the bulletins behind each share are laid out in the annex notes.\n\nMention: \"Decennial Statistical Bulletin Index (DSBI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Decennial Statistical Bulletin Index supplies issuance, size, revision and transparency values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:15ea522e65edb355045ae39a", "occurrence_id": "99eb64b68f15ad09aabb887a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Energy burden was measured for this review from the Periodic Fuel and Energy Use Survey (PFEUS), Wave 8, compiled by IRF, and the figures below are the review's rather than the survey's.\n\nHouseholds were grouped by the fuel they name as primary and the share whose energy spending leaves them below the national adequacy line computed: 34 per cent of households relying on charcoal against 12 per cent of those on a grid connection. The survey reports primary fuel and each spending item without netting spending against the line.\n\nWhether the household reports an irregular supply was then equalised between the two groups and the distance narrows to 13 percentage points, which leaves the fuel itself rather than the reliability of supply as what the review treats as the larger part of the burden.\n\nHouseholds reporting no energy spending in the reference period are excluded rather than treated as free of burden, and they are 3 per cent of households. The fuel groups and the counts behind each share are recorded under the table.\n\nMention: \"Periodic Fuel and Energy Use Survey (PFEUS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Periodic Fuel and Energy Use Survey compares energy-related poverty by the household's primary fuel.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:3587978384ab859cb72c82f9", "occurrence_id": "12e2b16375a4b3e56d641e35", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two adaptation finance lines in the annex are quoted from the District Adaptation Finance System, the register People’s Bank of China maintains for Europe and Central Asia as of 2024, and both are given without restatement.\n\nThe register puts the share of district budgets carrying a dedicated adaptation line at 27 per cent and the share of that spending reaching works started within the year at 61 per cent. Neither was re-derived for this annex, and the register's own rule for counting adaptation spending, which requires a purpose stated in the budget document, is reproduced with them.\n\nThe other lines in the annex are copied from the regional office's project file, so the register carries two lines while the project file carries the lines the table does not. A figure given for one country is printed with that country named rather than averaged across the region.\n\nThe register reference behind each line, the extraction date and the districts behind each share are recorded with the annex notes.\n\nMention: \"District Adaptation Finance System\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The District Adaptation Finance System supplies budget-line and spending-reach lines for the annex.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:f8a78bb2fe9457b0ed2f82d3", "occurrence_id": "badb3aa773d686bfe1dc36fd", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The unit-cost figure in this chapter is the agency's own, taken from its framework rates, and the Integrated Regional Operations System (IROS), compiled by Oxford Economics for 2023, is cited to test it.\n\nThe agency's framework puts the cost of a standard office fit-out at 1 180 per square metre. The system's construction series gives 1 340 for the same period on its own basis, because it includes supervision and the provisional sums the framework excludes. Neither figure was adjusted for the comparison.\n\nThe 160 distance is smaller than the spread the system reports between its low and high tender bands, so the agency's figure is printed as framed and the system supplies no revision to it and no rates for the categories the framework does not price.\n\nExcluded from the comparison are the regions the system prices in a different currency band, 4 of its 26 entries, since no conversion was made for this chapter. Which cost elements each source includes is recorded with the annex tables.\n\nMention: \"Integrated Regional Operations System (IROS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Integrated Regional Operations System checks the agency's framework unit cost for a standard fit-out.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:d6b8f0cebbe50c7126e30204", "occurrence_id": "8c1f47aa62cb4f06f189c4f9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Support receipt and income were compared for this review using the Rural Disability and Support Microdata (RDSM), Version 6, a microdata set covering Accra in 2015, and the comparison is the review's rather than the set's.\n\nHouseholds with a member reporting a functional difficulty were grouped by whether they receive any support payment and the share below the national adequacy line computed: 52 per cent of those receiving nothing against 33 per cent of those receiving one. The set publishes difficulty, payment receipt and consumption without crossing them.\n\nThat comparison supports one claim, that the payment's coverage rather than its value is the first thing to change. The cost figures in the proposal come from the treasury's own parameter set, from which the set supplied nothing.\n\nHouseholds whose payment status is recorded as pending are excluded rather than counted as receiving, and they are 6 per cent of households. The categories and the counts behind each share are recorded under the table.\n\nMention: \"Rural Disability and Support Microdata (RDSM)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Rural Disability and Support Microdata supports one coverage claim by comparing poverty with support receipt.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:74823228ba5c991fee962b80", "occurrence_id": "819a1268ccb092e1b2c482da", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Waiting for a specialist was computed for this assessment from the Provincial Facility Assessment Survey (PFAS), Second Version, conducted in Moscow in 2016, and the measures below are the assessment's rather than the survey's.\n\nFacilities were grouped by the number of specialist sessions they run each week and the median days a patient waits for an appointment computed: 9 days at facilities running five sessions or more against 26 days at those running one or fewer. The survey records sessions per facility and a waiting-time item per facility but publishes no relationship between them.\n\nBed capacity was then equalised between the two groups and the distance narrows to 8 days, which leaves the number of sessions rather than the size of the facility as what the assessment treats as the binding constraint on waiting.\n\nFacilities that did not offer the specialty at all are excluded rather than given the longest observed wait, and they are 8 per cent of facilities. The grouping rule and the counts behind each median sit with the annex tables.\n\nMention: \"Provincial Facility Assessment Survey (PFAS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Provincial Facility Assessment Survey relates specialist waiting time to the number of weekly sessions.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:ab7e59802ff39b5f28b71f85", "occurrence_id": "cefd50af14f4a11ac5cbf1d0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Slope hazard and cropping were joined for this assessment from the Regional Landslide Hazard Dataset (RLHD), First Round, the geospatial dataset compiled for Botswana in 2021, and the figures below are the assessment's own.\n\nFields were matched to their hazard class and grouped by the rainfall anomaly of the season, and the share of fields losing more than half their crop computed: 27 per cent of fields in the highest hazard class during a wet anomaly against 6 per cent of the same class during a dry one. The dataset grades hazard per polygon and the anomaly per period without reporting any crop outcome.\n\nDistance to the nearest all-weather road was then equalised and the distance narrows to 14 percentage points, which leaves the season's rainfall rather than the difficulty of evacuation as what the assessment treats as the operative factor in the loss.\n\nFields whose polygon could not be matched to a parcel, 4 per cent of those mapped, were excluded rather than assigned to the nearest class. The hazard classes, the anomaly bands and the counts behind each share are recorded with the annex tables.\n\nMention: \"Regional Landslide Hazard Dataset (RLHD)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Regional Landslide Hazard Dataset links hazard class and rainfall anomaly to crop loss.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:f76b6b9fa192465c2ff6e1e2", "occurrence_id": "cabb859095244427448ca9f5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Soil condition and input use were compared for this assessment using the Endline Soil Fertility Survey, which World Bank – UNDP conducted in Nairobi County in 2019, and the comparison is the assessment's rather than the survey's.\n\nFields were grouped by the organic matter measured in their topsoil and the share whose owner applied fertiliser in the season computed: 71 per cent of fields above the county median against 39 per cent of those below it. The survey reports the soil measurement and the input items in separate tables and never relates them.\n\nDistance to the nearest input dealer was then read from the same records and equalised, and the distance narrows to 14 percentage points, which leaves access to the dealer as a substantial part of what looks like a difference in soil knowledge.\n\nFields grazed rather than cropped in the season are excluded rather than grouped with the unimproved, and they are 9 per cent of fields. The measurement bands and the counts behind each share are recorded under the table.\n\nMention: \"Endline Soil Fertility Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Endline Soil Fertility Survey relates topsoil organic matter to fertiliser use by field owners.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:2c80010cee5702128aa3d8b0", "occurrence_id": "e60e63c0c5c97193733d30ea", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two service lines in the district annex are quoted from the High-Frequency Service Delivery Registry (HSDR), Eighth Phase, the register CAS maintains, and both are carried over unchanged.\n\nThe register puts the share of service points reporting uninterrupted operation through the month at 72 per cent and the mean number of days a point stays closed once it stops at 6. Neither was recalculated for this annex, and the register's own rule for counting a closure, which requires a field report rather than an absence of data, appears with the figures.\n\nThe annex's remaining entries follow the department's own facility list, so the register carries two lines while the list carries the balance. A figure given for one service type is printed with that type named rather than averaged across services.\n\nThe register reference behind each line, the extraction date and the service points behind each share are recorded with the annex notes.\n\nMention: \"High-Frequency Service Delivery Registry (HSDR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The High-Frequency Service Delivery Registry supplies uptime and closure-duration lines for the district annex.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:c99c0edd6648c149d32a70e7", "occurrence_id": "5d558f1a4ffd8208dc25d310", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Access and documentation the figures for the country are reported by the Representative Access and Documentation Survey (RADS), which covers Ethiopia in 2022, and the four values below are given exactly as the survey releases them.\n\nThe survey puts the share of displaced households holding any identity document for each member at 54 per cent, the share of their children enrolled in a school at the time of interview at 61 per cent, the mean distance a household walks to the nearest documentation office at 12 kilometres, and the share reporting a fee as the reason a document was not obtained at 27 per cent. None was recomputed here, and the survey's own definitions of a document and of enrolment are reproduced with the values.\n\nNo household holding a receipt for an application was counted as holding a document, and no child enrolled but not attending was counted as in school. Where the survey reports for one camp or settlement, that site is named with the value.\n\nThe survey table and line for each value, the extraction date, and the households behind each share are filed in the annex notes.\n\nMention: \"Representative Access and Documentation Survey (RADS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Representative Access and Documentation Survey supplies document, enrolment, distance and fee barriers.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:afe69584fec29e4c1dc1ef12", "occurrence_id": "78e04b6545bce6fe14722f30", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Price records were catalogued into the market file from the Audited Agri-Food Prices Registry (AAPR), Fourth Wave, the database Lane and Milesi-Ferretti maintains, and the work done on them was cataloguing rather than measurement.\n\nEach price record was given the file's own commodity code and checked against the market's own price book; where a record names a grade the book does not carry, it was parked for the editor rather than mapped to the nearest grade. Records with no market stated were left uncoded rather than assigned by the enumerator's origin, and 2 260 of them entered the file in that condition. The registry's own audit stamp was kept on every record so a later wave can be matched without re-auditing the quotation.\n\nThe market file is where this report's price series are held, and the registry is what supplies its records: every quotation in the file originates from a registry entry, and the coding step is what lets the file's series be read by commodity rather than by market alone. No series in the file is computed from the registry's published aggregates.\n\nThe coding sheet, the count of parked records and the date of deposit are recorded in the annex log.\n\nMention: \"Audited Agri-Food Prices Registry (AAPR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The Audited Agri-Food Prices Registry supplies the coded quotations that underpin the market file's series.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:e882d1b640d23d1db79a7984", "occurrence_id": "8ae84e2e3341658aff215404", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Whether the office's own population figure for the grid cells holds was tested against the Subnational Population Grid Census (SPGC), Seventh Round, the census compiled by the National Treasury, and the census is the principal source for the test.\n\nThe office's estimate puts 3.1 million residents inside the cells covered by the new road plan. The census puts 3.6 million for the same cells, because it allocates population by building footprint while the office allocates by administrative registers at the ward level. Neither allocation was converted for the comparison.\n\nThe census therefore carries the section's figure for who lives inside the plan area, and the office's lower count is printed beside it so the reader can see that the distance is how the two distribute a similar total across cell boundaries rather than a disagreement about the total itself. Where a cell is split by the plan boundary, the census's apportionment is used and the two are not averaged.\n\nCells the census reports as unpopulated are excluded from the comparison rather than counted as zero by both sources, 4 per cent of cells. The allocation rule each source applies is recorded with the annex tables.\n\nMention: \"Subnational Population Grid Census (SPGC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Subnational Population Grid Census is the principal source testing the office's cell population estimate.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:bfe429c8f9fa11aec9ad91a4", "occurrence_id": "2460914e3756f8c5f9e96225", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Output-area records were catalogued into the analysis file from the Triennial Output Area Dataset (TOAD), the geospatial dataset compiled for Buenos Aires in 2015, and the work done on them was cataloguing rather than measurement.\n\nEach area was given the file's own concordance code and its boundary checked against the file's previous vintage; where a boundary had been redrawn across two earlier areas, the record was parked for the geographer rather than split by population. Areas the dataset carries without a population figure were left unweighted rather than given the municipal mean, and 318 of them entered the file in that condition. The dataset's own vintage date was kept on every record so a later edition can be matched without redrawing the concordance.\n\nThe file's neighbourhood figures come from the census's tabulations, from which the dataset supplied nothing; it is the concordance that lets those tabulations be read against the older boundaries the programme's baseline used.\n\nThe concordance sheet, the count of parked areas and the date of deposit are stored in the annex log.\n\nMention: \"Triennial Output Area Dataset (TOAD)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The Triennial Output Area Dataset supplies the concordance records catalogued into the analysis file.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:6462d00ee6250d3fa4713380", "occurrence_id": "428b402e8bd2d4c81cd27cf9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: School feeding figures covering the municipality are drawn from the Multi-Topic School Feeding System, the register IMF maintains for Beijing as of 2021, and the four values below are transferred here unchanged.\n\nThe register puts the share of schools serving a meal each day at 88 per cent, the mean cost per pupil served per day at 3.4, the share of meals whose composition was checked by the office within the year at 46 per cent, and the share of schools reporting a shortfall of cooking staff at 17 per cent. None was recomputed here, and the register's own definitions of a served meal and of a check accompany the values.\n\nA school serving only milk was not counted as serving a meal, and a check carried out by the caterer itself was not counted as an office check. Where the register reports for one district, that district is named with the value rather than pooled.\n\nThe register reference for each value, the extraction date as well as the schools behind each share are listed in the annex notes.\n\nMention: \"Multi-Topic School Feeding System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Multi-Topic School Feeding System supplies coverage, cost, inspection and staffing values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:0346c5264cad88a8c365af40", "occurrence_id": "50320c384b32f9fd0f71dd42", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The readiness score in this chapter is the health department's own, computed from its inspection file, and the Regional Service Delivery and Readiness Report (RSDRR), First Phase, issued for Houston in 2018, is cited to test it.\n\nThe department's file puts the share of facilities meeting every readiness item at 62 per cent. The report puts 49 per cent for the same facilities, because it requires a signed corrective plan for each item as well as the item itself, and the department records only whether the item was met at the visit. Neither standard was applied to the other's data.\n\nThe 13-point distance is smaller than the year-on-year movement the department's own series shows for this measure, so the department's figure is printed as computed and the report supplies no revision to it and no items for the facilities its fieldwork did not reach.\n\nExcluded are facilities the report lists without a readiness assessment, 5 of its entries, since no comparison can be formed for them. Which items each source requires is recorded with the annex tables.\n\nMention: \"Regional Service Delivery and Readiness Report (RSDRR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Regional Service Delivery and Readiness Report checks the department's share of fully ready facilities.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:aa07b077052284b251ccb1b9", "occurrence_id": "63be340789d035a612384d31", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Record requests were catalogued into the transparency file from the Follow-Up Public Records Report (FPRR), Release 6, the report compiled for Egypt in 2022, and the work done on them was cataloguing rather than measurement.\n\nEach request was given the file's own subject code and its outcome matched to the file's headings; where a request was answered in parts across several dates, it was parked for the editor rather than counted once for each part. Requests with no outcome recorded were left open rather than assumed refused, and 640 of them entered the file in that condition. The report's own filing date was kept on every request so a later release can be matched without re-reading the correspondence.\n\nThe transparency file is where this report's request statistics are held, and the report is what supplies its records: every request in the file originates from a report entry, and the coding step is what lets the file be read by subject as well as by agency. Nothing in the file is drawn from the report's own summary of response rates.\n\nThe coding sheet, the count of parked requests and the date of deposit are set down in the annex log.\n\nMention: \"Follow-Up Public Records Report (FPRR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The Follow-Up Public Records Report supplies the coded request records behind the transparency file.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:52f032d6b5774d5139251824", "occurrence_id": "6c89951cf0ef484b0a1fa4ef", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The wage figures in this chapter are computed from the ministry's payroll returns, and the Clustered Wage Structure Survey (CWSS), Seventh Round, compiled by IDMC, is referred to only to set them in context.\n\nThe survey puts the share of employees covered by a collective agreement at 34 per cent, the ratio between the upper and lower quartile of the wage distribution at 2.1, and the share of establishments reporting a wage above the statutory floor for entry-level work at 71 per cent. All three are given as the survey issues them, with its own definition of an establishment stated, and no chapter figure was built from them.\n\nTheir use here is to show what the distribution looks like beside the mean the chapter reports. The chapter's argument rests on the payroll returns, which record what was paid rather than what an agreement stipulates, and the two are kept apart rather than merged.\n\nThe annex lists the survey tables consulted and the date of extraction, which is the whole of the survey's use in this chapter.\n\nMention: \"Clustered Wage Structure Survey (CWSS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Clustered Wage Structure Survey is quoted as context beside the ministry's payroll wage figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:404cdcc412ebda2c98b0c042", "occurrence_id": "994744a903a70870ac2c4fbc", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Supply chain figures covering the whole region derive from the High-Resolution Supply Chain System (HSCS), Collection 4, the register ActivityInfo maintains for Sub-Saharan Africa, and the four values below are given as the register holds them.\n\nThe register puts the share of distribution points whose stock of a first-line commodity was adequate on the day of reporting at 67 per cent, the mean days a stock-out of that commodity lasts at 9, the share of points whose location was confirmed by a coordinate rather than a name at 81 per cent, and the share of reports arriving within 48 hours of the visit at 74 per cent. None was recomputed here, and the register's own definitions of an adequate stock and of a confirmed location appear with the values.\n\nA point closed for a public holiday was not counted as stocked out, and a location supplied as a village name alone was not counted as confirmed. Where the register reports for one corridor, that corridor is named with the value rather than pooled.\n\nThe register table and line for each value, plus the extraction date and the distribution points behind each share are written up in the annex notes.\n\nMention: \"High-Resolution Supply Chain System (HSCS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The High-Resolution Supply Chain System supplies stock adequacy, stock-out length, location and reporting values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:cf20b3e8c21edbaebd446364", "occurrence_id": "9bf6529a913a2b26b94b7d66", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Hardship after displacement was measured for this review from the Longitudinal Deprivation and Hardship Survey (LDHS), Seventh Phase, which Bloomberg follows for 2024, and the figures below are the review's own.\n\nHouseholds that had moved within the previous two years were grouped by whether they occupy a structure of their own and the share reporting an unmet health need in the reference month computed: 44 per cent of those in borrowed or shared accommodation against 26 per cent of those in their own structure. The survey records tenure and health need for the same household and never publishes the crossing.\n\nIncome adequacy was then read from the same records and equalised between the two groups and the distance narrows to 11 percentage points, which leaves income as a substantial part of what the first reading attributes to the housing alone.\n\nHouseholds that moved more than once in the period are excluded rather than counted at their latest destination, and they are 7 per cent of displaced households. The tenure categories and the counts behind each share sit with the annex tables.\n\nMention: \"Longitudinal Deprivation and Hardship Survey (LDHS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Longitudinal Deprivation and Hardship Survey relates housing tenure to unmet health need among displaced households.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:5181e3adca77e61ff4bb7dc2", "occurrence_id": "ad7fa1e8a02c0fd12f842e19", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Earnings and origin were compared for this assessment using the Small-Area Informal Trade Survey (SITS), Version 3, compiled by IPEA, and the comparison is the assessment's rather than the survey's.\n\nTraders were grouped by whether they moved to the area within the previous five years and the mean daily earnings from their principal activity computed: 14 among recent arrivals against 21 among traders established longer. The survey reports an arrival date and daily earnings for each trader and offers no such comparison.\n\nWhether the trader holds a stall permit was then read from the same records and equalised, and the distance narrows to 3, which leaves the permit as a substantial part of what looks like an origin effect, and the assessment reports both readings.\n\nTraders who work only on market days are excluded from the mean rather than counted as earning nothing on the other days, and they are 13 per cent of traders. The grouping rule and the counts behind each mean are recorded under the table.\n\nMention: \"Small-Area Informal Trade Survey (SITS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Small-Area Informal Trade Survey compares informal trader earnings by length of residence.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:e0df94cd7b79ff7ecab2a5bb", "occurrence_id": "f6f56fb9fc7a7b28b1bd1afe", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Completion of training was computed for this review from the Provincial Vocational Training Survey (PVTS), Second Edition, conducted in Delhi in 2021, and the measures below are the review's rather than the survey's.\n\nEnrollees were grouped by whether the course runs in the day or in the evening and the share completing the programme computed: 52 per cent of day students against 71 per cent of evening students. The survey records the schedule and the completion status for the same person and publishes no table relating them.\n\nWhether the enrollee reports caring for a dependent was then read from the same records and the distance narrows to 9 points, which leaves the schedule as the larger part of the difference than the caring obligation the review first assumed.\n\nEnrollees still on their course at the time of interview are excluded rather than counted as dropouts, and they are 10 per cent of enrollees. The grouping rule and the counts behind each share are recorded with the annex tables.\n\nMention: \"Provincial Vocational Training Survey (PVTS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Provincial Vocational Training Survey compares training completion between day and evening cohorts.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:b59b9ead6142efd8ca09dc61", "occurrence_id": "d4f7d3f1b5b2868f7e3dcd3d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Settlement stability was analysed for this assessment from the Ministerial Resettlement Survey (MRS), Eighth Release, the survey the National Statistics Office conducts, and the readings below are the assessment's rather than the office's.\n\nResettled households were grouped by the year of their arrival at the site and the share reporting they intend to remain at the end of two years computed: 68 per cent of households arriving in the first year of the programme against 41 per cent of those arriving in the most recent year. The office reports arrival year and intention as separate items and never crosses them.\n\nWhether the household holds a plot title was then read from the same records and equalised, and the distance narrows to 12 percentage points, which leaves the title as a substantial part of what the first reading attributes to the passage of time alone.\n\nHouseholds whose intention was recorded as undecided are excluded rather than counted as leavers, and they are 9 per cent of households. The arrival groups and the counts behind each share sit with the annex tables.\n\nMention: \"Ministerial Resettlement Survey (MRS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Ministerial Resettlement Survey relates arrival year and plot title to the intention to remain at a site.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:5fef963614c1a8373181e887", "occurrence_id": "77aa3c4e76187c3c9658515c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two lit-area lines in the appendix are quoted from the Experimental Nighttime Lights Dataset (ENLD), Seventh Version, the geospatial dataset PRIO/Uppsala publishes, and both are reproduced without amendment.\n\nThe dataset puts the mean radiance over its settled cells at 22 nanowatts per square centimetre per steradian, the share of its cells showing no measurable light at all at 46 per cent, and the share whose brightness rose between its own two observation dates at 9 per cent. Neither of the first two was recalculated here, and the dataset's own rule for calling a cell settled, which requires a neighbouring lit cell, is reproduced with the figures.\n\nThe appendix's other lines are compiled from the grid census, so the dataset carries two lines while the census carries the rest. A figure given for one band of the radiance scale is printed with that band named rather than averaged over the extent, and cells flagged as cloud-affected were not replaced with the previous observation for this report.\n\nThe dataset layer behind each line, the extraction date and the cells behind each share are recorded with the annex notes.\n\nMention: \"Experimental Nighttime Lights Dataset (ENLD)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Experimental Nighttime Lights Dataset supplies radiance, unlit-cell and brightening lines for the appendix.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:67131addf773aa20d7fc2ef1", "occurrence_id": "282722608c87f5151827d548", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Survival of new enterprises was computed for this assessment from the Quality-Assured Business Demography Survey, which YouGov conducted in Sulaymaniya in 2017, and the figures below are the assessment's own.\n\nUnits were grouped by whether the owner had previously worked for another employer and the share still trading two years after registration computed: 57 per cent of owners with prior employment against 38 per cent of those who had not. The survey records each owner's work history and the unit's current status without publishing any such crossing.\n\nWhether the unit took a bank loan in its first year was then read from the same records and equalised, and the distance narrows to 10 points, which leaves the loan as a substantial part of what the first reading attributes to the owner's history.\n\nUnits that changed owner within the two years are excluded rather than followed under the new owner, and they are 6 per cent of units. The grouping rule and the counts behind each share sit with the annex tables.\n\nMention: \"Quality-Assured Business Demography Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Quality-Assured Business Demography Survey relates the owner's prior employment to enterprise survival.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:605ba4157425c6a3662e977a", "occurrence_id": "29bca19a8b8c28512e2bbe94", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two delivery lines in the municipal annex are quoted from the Small-Area Service Delivery Index, compiled by New Cronos for Dar-es-Salaam in 2018, and both are given exactly as the index issues them.\n\nThe index puts the share of small areas within walking distance of a functioning water point at 63 per cent and the share whose nearest waste collection point is met on the advertised day at 41 per cent. Neither was re-derived for this annex, and the index's own walking rule, which measures along its footpath layer rather than in a straight line, appears with the figures.\n\nThe rest of the annex lines come from the authority's own complaint log, so the index carries two lines while the log carries the rest. A figure given for one ward is printed with that ward named rather than averaged across the city.\n\nThe index entry behind each line, the extraction date and the areas behind each share are recorded with the annex notes.\n\nMention: \"Small-Area Service Delivery Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Small-Area Service Delivery Index supplies water-point access and collection reliability lines.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:06831f4f0b38ead910e2b262", "occurrence_id": "12ee38be1ae966ccf8888284", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Include and exclude errors were built for this review from the Provincial Targeting and Payments Survey (PTPS), Release 5, conducted for 2018 by Digital South Asia Library, and the figures below are the review's rather than the survey's.\n\nHouseholds were classified against the programme's own eligibility rule and grouped by whether they hold a card, and the share of ineligible households holding one computed at 14 per cent against a share of eligible households holding none at 22 per cent. The survey reports card possession and the eligibility items without applying any rule of its own to them.\n\nThat comparison supports one claim, that the roll needs re-verification before the next payment round rather than expansion. The cost of re-verification in the proposal comes from the programme's unit-rate sheet, from which the survey supplied nothing.\n\nHouseholds whose eligibility cannot be determined from the survey's items are excluded rather than counted on either side, and they are 5 per cent of households. The rule applied and the counts behind each share are recorded under the table.\n\nMention: \"Provincial Targeting and Payments Survey (PTPS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Provincial Targeting and Payments Survey supports one targeting claim built from include and exclude errors.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:baff3f12d1937150d951f5ad", "occurrence_id": "80f55e489a015d6fab1a6251", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Work and income adequacy were compared for this assessment using the Monthly Living Standards Survey, which SECEX follows across South Asia in 2015, and the comparison is the assessment's rather than the survey's.\n\nHouseholds were grouped by whether any member works more than 40 hours a week without a written contract and the share below the national adequacy line computed: 39 per cent of households in that position against 17 per cent of those with a contract. The survey publishes hours, contract status and consumption for the same household and never reports the three together.\n\nWhether the household also depends on a single earner was then read from the same records and equalised, and the distance narrows to 12 points, which leaves the absence of a contract as a substantial part of what looks at first like a low-wage effect.\n\nHouseholds with no member reporting work in the reference month are excluded rather than grouped with the uncontracted, and they are 14 per cent of households. The grouping rule and the counts behind each share are recorded with the annex tables.\n\nMention: \"Monthly Living Standards Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Monthly Living Standards Survey links uncontracted long hours to household income adequacy.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:4ced489058f5054ab2b75535", "occurrence_id": "a2981c18ac7f1bb6a3ecdb25", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Zero-dose children were identified for this review from the Nationwide Vaccination Campaign Survey (NVCS), Fifth Wave, compiled by IFC, and the figures below are the review's rather than the wave's.\n\nChildren were grouped by the distance band of their nearest fixed post and the share with no recorded dose at any age computed: 15 per cent beyond 10 kilometres against 4 per cent within 2 kilometres. The wave publishes a distance item and a dose history and never reports the two against each other.\n\nWhether the household reports the post's fees as a barrier was then read from the same records and equalised, and the distance narrows to 6 points, which leaves the fee as a substantial part of what the first reading attributes to distance.\n\nChildren whose card was lost and who could not be verified from the post's register are excluded rather than counted as unvaccinated, and they are 7 per cent of children. The bands and the counts behind each share sit with the annex tables.\n\nMention: \"Nationwide Vaccination Campaign Survey (NVCS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Nationwide Vaccination Campaign Survey relates distance to a fixed post with zero-dose status.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:07b2c55e5d9ad4d4b207e6a7", "occurrence_id": "eb0de72b2bf54988e3540f3a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The coverage figure in this chapter is the office's own, taken from its client register, and the Endline Administrative Records Registry (EARR), compiled by CDR for Riga, is cited to test it.\n\nThe office's register puts 91 400 residents holding an active entitlement. The registry puts 84 900 for the same register period, because it removes records whose entitlement lapsed and whose holder has since died or moved, a step the office applies only at the annual sweep. No adjustment was made to either count for the comparison.\n\nThe 6 500 distance is smaller than the movement the office's own series shows between two sweeps, so the office's figure is printed as registered and the registry supplies no revision to it and no records for the categories it does not hold.\n\nExcluded are records the registry marks as duplicate identifiers, 1 120 of them, since neither count can be trusted for a duplicate. Which lapsed records each source keeps is recorded with the annex tables.\n\nMention: \"Endline Administrative Records Registry (EARR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Endline Administrative Records Registry checks the office's count of residents holding an entitlement.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:ec36f7ea535a9866511b5b4c", "occurrence_id": "ed31ddca69b73434dedd3884", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Operations the numbers held for the schemes come from the Cross-Sectoral Regional Operations Registry (CROR), Seventh Wave, the register International Water Association maintains for Aceh province, and the four values below are given as the registry holds them.\n\nThe registry puts the share of schemes whose operator filed a return for the period at 79 per cent, the mean hours per day that distribution runs at a metered scheme at 14, the share of schemes whose tariffs cover their recorded operating cost at 31 per cent, and the share reporting a main break repaired within 24 hours at 58 per cent. None was recomputed here, and the registry's own definitions of a filed return and of a covered cost appear with the values.\n\nA scheme operating without a meter was not counted in the distribution hours, and a repair completed by the community rather than the operator was not counted as repaired. Where the registry reports for one scheme, that scheme is named with the value rather than pooled.\n\nThe registry table and line for each value, along with the extraction date and the schemes behind each share are carried in the annex notes.\n\nMention: \"Cross-Sectoral Regional Operations Registry (CROR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Cross-Sectoral Regional Operations Registry supplies return, supply, cost-recovery and repair values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:9c1764f4f306fa8d9df03446", "occurrence_id": "108a01ffb70454a3bc1aa4cb", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Shock response country figures are assembled from the Verified Shock Coping Survey, which SEEG conducted in Peru in 2018, and the four values below are carried over as the survey issues them.\n\nThe survey puts the share of households reporting an income shock in the previous year at 37 per cent, the share meeting the loss by withdrawing a child from school at 9 per cent, the share holding any savings at the time of the shock at 44 per cent, and the share receiving public assistance afterwards at 21 per cent. None was recalculated here, and the survey's own definitions of an income shock and of a withdrawal are reproduced with the values.\n\nA household that borrowed from a relative was not counted as receiving assistance, and a child who changed school rather than leaving was not counted as withdrawn. Where the survey reports for one department, that department is named with the value.\n\nThe survey table and line for each value, the extraction date and the households behind each share are detailed in the annex notes.\n\nMention: \"Verified Shock Coping Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Verified Shock Coping Survey supplies shock, school withdrawal, savings and assistance values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:4fca9a8a4dd0df10ce6ff0e5", "occurrence_id": "789d614e317f13a53744be67", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two poverty lines in the country annex are quoted from the Recurrent Poverty and Vulnerability Report, issued by CONAPO for Mauritania in 2018, and both are reproduced as the report prints them.\n\nThe report puts the share of households below its own multidimensional line at 47 per cent and the share vulnerable to falling below it after a single bad season at 22 per cent. Neither was recalculated here, and the report's own weighting of the dimensions behind its line is stated with the figures rather than reweighted for this annex.\n\nThe annex's other lines are compiled from the household economy survey, so the report carries two lines while that survey carries the rest. A figure given for one livelihood zone is printed with that zone named rather than averaged across the country.\n\nThe report's page and table behind each line, the extraction date and the households behind each share are recorded with the annex notes.\n\nMention: \"Recurrent Poverty and Vulnerability Report\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Recurrent Poverty and Vulnerability Report supplies headcount and vulnerability lines for the annex.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:625693e7e552d69fc5aab942", "occurrence_id": "386cc7cc7c534c357ada45f2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Graduation from the programme was measured for this assessment from the Record-Based Socioeconomic Registry (RSR), Fourth Release, the register maintained for Nicaragua in 2015, and the figures below are the assessment's own.\n\nEnrolled households were grouped by the number of visits a worker paid them during enrolment and the share whose income cleared the programme's exit line at the following review computed: 46 per cent of households receiving four visits or more against 21 per cent of those receiving one or none. The registry holds the visit log and the review result for the same household and publishes no relationship between them.\n\nWhether the household also holds a productive asset was then read from the same records and equalised, and the distance narrows to 11 points, which leaves the asset as a substantial part of what the first reading credits to the visits.\n\nHouseholds reviewed before their second year were excluded rather than counted as failures, and they are 8 per cent of enrolled households. The grouping rule and the counts behind each share sit with the annex tables.\n\nMention: \"Record-Based Socioeconomic Registry (RSR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Record-Based Socioeconomic Registry relates worker visits to households clearing the exit line.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:8c05ffcca97ecf541e3713a2", "occurrence_id": "e3e9e03bc79834d3e759d8dc", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The distribution figures in this chapter are computed from the office's own tax register, and the Panel Inequality Survey conducted by ATO in Niger in 2018 is referred to only as background to them.\n\nThe survey puts the Gini coefficient for its consumption measure at 0.39, the share of working adults in an activity with no written agreement at 78 per cent, and the ratio between the top and bottom decile of its own expenditure measure at 4.6. All three are given as the survey issues them, with its own consumption aggregate stated, and no chapter figure rests on them.\n\nNo chapter table is built from its panels. The figures the chapter argues from come from the register, which records declared income rather than what a household reports spending, and where the two accounts of inequality differ the difference is noted rather than reconciled.\n\nThe annex lists the survey tables consulted and the date of extraction, and the survey is used for nothing beyond that background.\n\nMention: \"Panel Inequality Survey\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The Panel Inequality Survey is background while the chapter's distribution figures come from the tax register.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:27e4ab3a8245b87b544eac2a", "occurrence_id": "94a8a8aa628937b7e67d31bd", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Movement figures spanning the department are taken from the Harmonised Mixed Movement Registry, the register the Ministry of Trade and Industry maintains for Bolivia as of 2019, and the four values below are transferred here unchanged.\n\nThe registry puts the number of movements recorded at a checkpoint in the period at 214 000, the share of those movements made by a person holding no travel document at 18 per cent, the mean duration a recorded stay lasts at 40 days, and the share of movements crossing more than once in the period at 26 per cent. None was recomputed here, and the registry's own definitions of a movement and of a recorded stay accompany the values.\n\nA movement recorded twice at the same checkpoint on the same day was counted once, and a stay interrupted by removal was not counted as completed. Where the registry reports for one crossing point, that point is named with the value rather than pooled.\n\nThe registry reference for each value: the extraction date and the movements behind each share are carried in the annex notes.\n\nMention: \"Harmonised Mixed Movement Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Harmonised Mixed Movement Registry supplies movement counts, documentation, stay length and repeat shares.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:7a66e72d3141686547b69740", "occurrence_id": "fac930995b7eeb27237d8345", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The poverty figure in this chapter is the office's own, computed from its own household panel, and the Longitudinal Multidimensional Poverty Survey (LMPS), compiled by IFMIS for the Republic of Congo, is cited to test it.\n\nThe office's panel puts 52 per cent of households below its multidimensional line. The survey puts 58 per cent for the same period, because it counts a household as deprived on housing where only one indicator fails while the office requires two. Neither rule was applied to the other's data.\n\nThe 6-point distance is smaller than the movement the office's panel shows between its own two waves, so the office's figure is printed as computed and the survey supplies no revision to it and no measurements for the dimensions it does not carry.\n\nExcluded are the regions the survey fielded in but the office's panel did not reach, 3 of its 12, since no comparison can be formed for them. Which indicators each source counts is recorded with the annex tables.\n\nMention: \"Longitudinal Multidimensional Poverty Survey (LMPS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Longitudinal Multidimensional Poverty Survey checks the office's multidimensional poverty headcount.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:33ebc14482978c8f90106e51", "occurrence_id": "a93c76dd17b202887f585c3f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Emission figures at the region level are taken from the Ministerial Emission Inventory Dataset, the geospatial dataset the Fraser Institute compiled for Australia in 2020, and the four values below are given exactly as the dataset issues them.\n\nThe dataset puts total carbon dioxide equivalent for its covered area at 402 megatonnes, the share arising from stationary energy at 38 per cent, the share from agriculture at 14 per cent, and the share of its grid cells whose estimate carries an uncertainty above 20 per cent at 23 per cent. None was recomputed here, and the dataset's own boundary of what the inventory covers is stated with the values.\n\nNo emission arising outside the covered area was attributed to it, and no cell flagged as interpolated was counted as a measurement. Where the dataset reports for one sector, that sector is named with the value rather than pooled.\n\nThe dataset layer and cell behind each value, the extraction date and the cells behind each share are kept in the annex notes.\n\nMention: \"Ministerial Emission Inventory Dataset\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Ministerial Emission Inventory Dataset supplies total emissions, sector shares and cell uncertainty.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:fadc01ba9e263de575ab53e4", "occurrence_id": "5851e64a1575acb67133b2c2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two child lines in the household annex are quoted from the Midline Child Wellbeing Survey, compiled by FEPB in China in 2015, and both are given without restatement.\n\nThe survey puts the share of children in a household reporting an income below the national adequacy line at 13 per cent and the share of those children whose household also depends on farming at 44 per cent. Neither was recalculated here, and the survey's own definition of the household's income, which counts subsistence production at local farm-gate prices, appears with the figures.\n\nThe other annex entries are assembled from the education ministry's enrolment file, so the survey carries two lines while the file carries the rest. A figure given for one province is printed with that province named rather than averaged across the country.\n\nThe survey table behind each line, the extraction date and the children behind each share are recorded with the annex notes.\n\nMention: \"Midline Child Wellbeing Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Midline Child Wellbeing Survey supplies child poverty and farming-dependence lines for the annex.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:76883e363c02407573f97ff4", "occurrence_id": "c880d012325a13413160b6b5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Migration and remittance figures for the metropolitan area are reported from the Metropolitan Migration and Remittance Survey (MMRS), Version 9, the survey UN COMTRADE follows in Ukraine, and the four values below are given as the survey issues them.\n\nThe survey puts the share of households with a member working in the capital for at least three months a year at 27 per cent, the mean monthly remittance such a household receives at 180, the share of receiving households spending any of it on school fees at 36 per cent, and the share of migrants holding a qualification above upper secondary at 41 per cent. None was recomputed here, and the survey's own definitions of a receiving household and of a remittance appear with the values.\n\nA transfer made in kind was not counted as a remittance, and a migrant studying full-time was not counted as working. Where the survey reports for one city, that city is named with the value rather than pooled.\n\nThe survey table and line for each value, the extraction date, and the households behind each share are held in the annex notes.\n\nMention: \"Metropolitan Migration and Remittance Survey (MMRS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Metropolitan Migration and Remittance Survey supplies migration, remittance, fee-use and qualification values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:0410ffea66fb68a85fc5f451", "occurrence_id": "412d29861812ac4cb2f1309f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two referral lines in the health annex are quoted from the Endline Referral Pathways Index (ERPI), Second Version, compiled for Accra in 2022, and both are reproduced without amendment.\n\nThe index puts the share of primary facilities with a named referral destination for each of five common conditions at 71 per cent and the mean hours a patient waits at the receiving facility once referred at 5. Neither was re-derived here, and the index's own rule for counting a pathway as named, which requires a written protocol rather than a custom, is reproduced with them.\n\nThe annex's other lines are compiled from the facilities' own registers, so the index carries two lines while the registers carry the rest. A figure given for one condition is printed with that condition named rather than averaged across the five.\n\nThe index entry behind each line, the extraction date and the facilities behind each share are recorded with the annex notes.\n\nMention: \"Endline Referral Pathways Index (ERPI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Endline Referral Pathways Index supplies named-pathway coverage and receiving-facility wait lines.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:3a4d816eefe126106c966a4b", "occurrence_id": "31615cd71d62d1e37317e7a7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The trader count in this chapter is the association's own, taken from its membership book, and the Ministerial Informal Trade Survey conducted by REACH across Africa and South Asia in 2021 is cited to test it.\n\nThe association's book lists 18 400 traders holding a stall licence. The survey puts 24 100 traders working the same markets, because it counts traders working without a licence, whom the association does not enrol. Neither figure was adjusted for the comparison.\n\nThe 5 700 distance is smaller than the turnover the association's own book shows between two renewals, so the association's figure is printed as enrolled and the survey supplies no revision to it and no listings for the markets its teams did not enter.\n\nExcluded are the markets the survey visited once rather than on a trading day, 6 of its 41, since a count taken then cannot be compared with a book. Which traders each source counts is recorded with the annex tables.\n\nMention: \"Ministerial Informal Trade Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Ministerial Informal Trade Survey checks the association's licensed trader count against market listings.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:196564814ed5475fbbd0d6bf", "occurrence_id": "6d054500731f9c2cf0d008bc", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Vacancy fill rates were computed for this review from the Rolling Hiring and Vacancies Census (RHVC), Release 3, the census Asian Development Bank conducts, and the figures below are the review's own.\n\nEstablishments were grouped by the number of vacancies open at the time of enumeration and the share reporting a post unfilled for more than three months computed: 48 per cent of establishments with four vacancies or more against 22 per cent of those with one. The census counts open posts and records a duration for each and publishes no relationship between the two.\n\nWhether the establishment offers training was then read from the same records and equalised, and the distance narrows to 13 points, which leaves the absence of training as a substantial part of what the first reading attributes to the number of posts alone.\n\nEstablishments that withdrew a vacancy before the enumeration are excluded rather than counted as having filled it, and they are 5 per cent of establishments. The grouping rule and the counts behind each share sit with the annex tables.\n\nMention: \"Rolling Hiring and Vacancies Census (RHVC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Rolling Hiring and Vacancies Census relates the number of open posts to long unfilled vacancies.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:135f61a4d0ed2c806b54ec35", "occurrence_id": "4d658c1152de67160714f066", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The support figures in this chapter are computed from the agency's own payment file, and the Standardised Disability and Support Registry (SDSR), Collection 6, compiled by the CCCM Cluster, is quoted only to set them in context.\n\nThe registry puts the share of registered households with a member reporting a functional difficulty at 19 per cent, the share of those households receiving any support at 34 per cent, and the mean months between registration and first payment at 5. All three are given as the registry issues them, with its own difficulty screen stated, and no chapter figure was built from them.\n\nTheir use here is to show what the registered population looks like beside the caseload the chapter reports. The chapter's argument rests on the payment file, which records what was paid to whom rather than who is registered, and the two are kept apart rather than merged.\n\nThe annex lists the registry tables consulted and the date of extraction, which is the whole of the registry's use in this chapter.\n\nMention: \"Standardised Disability and Support Registry (SDSR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Standardised Disability and Support Registry is quoted as context beside the agency's payment caseload.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:0caeb7b59100231fd9d079e6", "occurrence_id": "faa72eb901458ee850dcdf55", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two workforce lines in the health annex are quoted from the Comparative Public Sector Monitoring Registry, the register MOH maintains for Florida as of 2020, and both are carried over unchanged.\n\nThe register puts the share of public health posts vacant at the census point at 12 per cent and the median months a post stays vacant at 7. Neither was recalculated for this annex, and the register's own rule for counting a post as vacant, which requires it to be funded and unfilled on the same date, appears with the figures.\n\nThe annex's other lines are taken from the ministry's payroll extract, so the register carries two lines while the extract supplies the balance. A figure given for one professional group is printed with that group named rather than averaged across the workforce.\n\nThe register reference behind each line, the extraction date and the posts behind each share are recorded with the annex notes.\n\nMention: \"Comparative Public Sector Monitoring Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Comparative Public Sector Monitoring Registry supplies vacancy rates and median vacancy duration lines.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:f7838bb09c666cf189374812", "occurrence_id": "f97a7f14bb954d30a0372e8a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Return intention figures are reported from the Quarterly Return Intention System (QRIS), Round 3, the register NISR maintains, and the four values below are given as the register holds them.\n\nThe register puts the share of displaced households expressing an intention to return within a year at 38 per cent, the share naming the state of services at the place of origin as the reason for hesitation at 47 per cent, the mean months a household has spent in displacement at 26, and the share holding a documented claim to a dwelling at origin at 29 per cent. None was recomputed here, and the register's own definitions of an expressed intention and of a documented claim appear with the values.\n\nA household that returned and came back was not counted as intending to return, and a claim held by a relative rather than the household was not counted as held. Where the register reports for one place of origin, that place is named with the value rather than pooled.\n\nThe register table and line for each value, the extraction date as well as the households behind each share are laid out in the annex notes.\n\nMention: \"Quarterly Return Intention System (QRIS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Quarterly Return Intention System supplies return intention, hesitation reasons, duration and claim values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:fbdb0b1f7d4ace11831107c8", "occurrence_id": "6de5cd35322e14c1e410cb5b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The efficiency figures in this chapter are drawn from the authority's own performance returns, and the Harmonised Operations Index Registry compiled by Hungarian NHIF for Uzbekistan in 2016 is mentioned only as background to them.\n\nThe registry puts the mean cost recovery ratio across the bodies it covers at 0.7, the share of bodies publishing an annual account at 64 per cent, and the staff per thousand connections it records at 4. All three are given as the registry issues them, with its own coverage of bodies stated, and no chapter figure rests on them.\n\nNo table in the chapter uses its entries. The figures the chapter argues from come from the returns, which report one authority's own accounts rather than a panel of comparable bodies, and where the two accounts of cost recovery differ the difference is noted rather than reconciled.\n\nThe annex lists the registry tables consulted and the date of extraction, and the registry is used for nothing beyond that background.\n\nMention: \"Harmonised Operations Index Registry\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The Harmonised Operations Index Registry is background while the chapter's efficiency figures come from returns.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:080a4ba7f554f2f2de5d20f8", "occurrence_id": "bbb38a5858f01c1d58e1f1a9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The reach figure in this chapter is the team's own, taken from its site list, and the Household Access and Documentation System (HADS), the register kept for the Amhara region in 2018, is cited to test it.\n\nThe team's list puts 96 000 displaced people living within an hour's walk of a documentation point. The register puts 71 000 for the same sites, because it measures the walk along its road layer while the team draws a straight line from each settlement. No conversion was applied to either figure for the comparison.\n\nThe 25 000 distance is smaller than the population of the three largest sites the team counts, so the team's figure is printed as measured and the register supplies no revision to it and no settlements for the woredas its collection did not reach.\n\nExcluded are settlements the register places without coordinates, 14 of them, since no walking distance can be formed for them. How each source measures the walk is recorded with the annex tables.\n\nMention: \"Household Access and Documentation System (HADS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Household Access and Documentation System checks how many displaced people live within reach of a point.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:68ffe130d26d904d70f01233", "occurrence_id": "b321643758a782e96fbaf473", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Transition from school was measured for this assessment from the Representative Youth Transition System (RYTS), Sixth Round, the register WDI maintains, and the figures below are the assessment's own.\n\nLeavers were grouped by the level they last attended and the share enrolled in further study or training within a year computed: 72 per cent of upper-secondary leavers against 34 per cent of those leaving lower secondary. The register records last level attended and current activity for the same person and never publishes the crossing.\n\nWhether the household sits in the lowest wealth quintile was then read from the same records and equalised, and the distance narrows to 19 points, which leaves household wealth as a substantial part of what the first reading attributes to the level alone.\n\nLeavers in unpaid family work are excluded rather than counted as neither studying nor working, and they are 11 per cent of leavers. The grouping rule and the counts behind each share sit with the annex tables.\n\nMention: \"Representative Youth Transition System (RYTS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Representative Youth Transition System relates last school level to further study within a year.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:96bfd5b704f6dcbb79a49fce", "occurrence_id": "ee70d58888c35152685ccbf9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Plot records were catalogued into the water-use file from the Standardised Irrigated Area Registry (SIAR), the database IFPRI maintains for Xinjiang province, and the work done on them was cataloguing rather than measurement.\n\nEach plot was given the file's own crop code and its service area matched to the file's scheme list; where a plot lies outside every listed area, it was parked for the engineer rather than assigned to the nearest scheme. Plots with no water source stated were left uncoded rather than assumed to draw from the scheme serving their village, and 1 890 of them entered the file in that condition. The registry's own survey date was kept on every plot so a later edition can be matched without re-mapping the boundary.\n\nThe water-use file is where this report's irrigation totals are held, and the registry is what supplies its plots: every plot in the file originates from a registry entry, and the coding step is what lets the file's totals be read by crop as well as by scheme. Nothing in the file is drawn from the registry's published area figures.\n\nThe coding sheet, the count of parked plots and the date of deposit are filed in the annex log.\n\nMention: \"Standardised Irrigated Area Registry (SIAR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The Standardised Irrigated Area Registry supplies the coded plot records behind the water-use file.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:e8e5b0846a35b4071d0c56bd", "occurrence_id": "26861a1b36b6c4f96051f6fe", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two farm lines in the peri-urban annex are quoted from the Subnational Agricultural Household Survey, compiled by S&P in Kinshasa in 2024, and both are reproduced as the survey prints them.\n\nThe survey puts the share of households cultivating a plot within the city boundary at 31 per cent, the share of those plots watered from a stream or well at 58 per cent, and the mean cultivated area per cultivating household at 0.4 hectares. Neither of the first two was recalculated here, and the survey's own minimum plot size for counting a holding, a quarter of a hectare, appears with the figures.\n\nThe annex's other lines are compiled from the office's market stall census, so the survey carries two lines while the census carries the rest. A figure given for one commune is printed with that commune named rather than averaged across the city, and households the survey lists as cultivating without stating an area were left out of the mean rather than given the median.\n\nThe survey table behind each line, the extraction date and the households behind each share are recorded with the annex notes.\n\nMention: \"Subnational Agricultural Household Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Subnational Agricultural Household Survey supplies urban plot, watering and plot-size lines.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:729a88a0d049243d47279404", "occurrence_id": "8dff2ab27c888948f94030bd", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two operations lines in the municipal annex are quoted from the Record-Based Operations Monitoring Index, compiled by Bureau van Dijk for Maputo in 2017, and both are given without restatement.\n\nThe index puts the share of registered firms filing their annual return on time at 57 per cent, the mean days the licensing office takes to decide an application at 21, and the share of applications decided without a further request for documents at 63 per cent. Neither of the first two was re-derived for this annex, and the index's own rule for counting a return as on time, which uses the postmark rather than the filing date, is reproduced with them.\n\nThe annex's other entries are drawn from the council's own receipts ledger, so the index carries two lines while the ledger holds the remainder. A figure given for one firm size is printed with that size named rather than averaged across registrations.\n\nThe index entry behind each line, the extraction date and the firms behind each share are recorded with the annex notes.\n\nMention: \"Record-Based Operations Monitoring Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Record-Based Operations Monitoring Index supplies filing timeliness, decision delay and rework lines.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:1b8ae701a5833cff6c7877cb", "occurrence_id": "d812a87bd33ac770f9193f5d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two delivery lines in the public administration annex are quoted from the Government Service Delivery Index, compiled by NCHS/CDC/WHO for Jordan in 2015, and both are carried over unchanged.\n\nThe index puts the share of service outlets open the full advertised week at 82 per cent, the share of visits ending with the requested document issued the same day at 49 per cent, and the mean number of visits a household makes to complete one transaction at 2. Neither of the first two was recalculated here, and the index's own definition of a completed transaction, which requires the document in hand rather than a promise, appears with the figures.\n\nThe annex's other lines are compiled from the ministry's own inspection returns, so the index carries two lines while the returns carry the rest. A figure given for one outlet type is printed with that type named rather than pooled across outlets.\n\nThe index entry behind each line, the extraction date and the visits behind each share are recorded with the annex notes.\n\nMention: \"Government Service Delivery Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Government Service Delivery Index supplies opening hours, same-day issue and repeat-visit lines.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:edba71e45cb224ba19103643", "occurrence_id": "620e05be569e2d173574054e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Non-state education figures for the programme countries are reported from the Inter-Agency Non-State Education Index (INEI), Third Release, compiled by Growth Lab at Harvard, and the four values below are given exactly as the index issues them.\n\nThe index puts the share of schools operating outside the ministry's network at 22 per cent, the share of those schools recognised by the ministry at 38 per cent, the share of pupils in a recognised non-state school eligible for the public grant at 45 per cent, and the mean fee a recognised school charges as a multiple of public per-pupil spending at 0.6. None was recomputed here, and the index's own definitions of operating outside the network and of recognition appear with the values.\n\nA school charging nothing but refusing ministry inspection was not counted as recognised, and a pupil holding a grant but attending an unrecognised school was not counted as eligible. Where the index reports for one country, that country is named with the value rather than pooled.\n\nThe index table and line for each value, plus the extraction date and the schools behind each share are recorded in the annex notes.\n\nMention: \"Inter-Agency Non-State Education Index (INEI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Inter-Agency Non-State Education Index supplies share, recognition, grant eligibility and fee values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:66f72c84bc195ebd097ab961", "occurrence_id": "844a676af1465f1b7b37cde2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The immunisation figure in this chapter is the programme's own, taken from its dose register, and the Local Health and Household Report issued by CAS for Bolivia in 2024 is cited to test it.\n\nThe programme's register puts 89 per cent of infants completing the schedule. The report puts 80 per cent for the same birth cohort, because it asks the caregiver to produce the card and counts a child without one as unverified, while the register counts a dose entered at any facility. The two bases were not applied to each other's data.\n\nThe 9-point distance is smaller than the year-on-year movement the register itself shows for this measure, so the programme's figure is printed as registered and the report supplies no revision to it and no children for the departments its teams did not enter.\n\nExcluded are caregivers the report interviewed but who could not state the child's date of birth, 3 per cent of its sample, since cohort membership cannot be fixed for them. What each source counts as a completed schedule is recorded with the annex tables.\n\nMention: \"Local Health and Household Report\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Local Health and Household Report checks the programme's infant schedule completion figure.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:803d79ed54ef24812ab5288a", "occurrence_id": "083086cc6ca2e57a612213d9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two registry lines in the statistical annex are quoted from the Record-Based Administrative Survey Registry (RASR), Edition 6, the register compiled for Guatemala in 2015, and both are reproduced without amendment.\n\nThe registry puts the share of administrative sources in its inventory undergoing a documented quality review at 34 per cent, the median years since a source's last external audit at 6, and the share whose microdata can be requested rather than only read at aggregate level at 21 per cent. Neither of the first two was recalculated here, and the registry's own definition of a documented review, which requires a written protocol on file, is stated with the figures.\n\nThe annex's other lines are compiled from the office's own legislative schedule, so the registry carries two lines while the schedule carries the rest. A figure given for one source type is printed with that type named rather than averaged across the inventory.\n\nThe registry entry behind each line, the extraction date and the sources behind each share are recorded with the annex notes.\n\nMention: \"Record-Based Administrative Survey Registry (RASR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Record-Based Administrative Survey Registry supplies review coverage, audit age and access lines.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:63f3163ac09b7d2b3e875809", "occurrence_id": "8e54d14e9a62cc79f9d38895", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Consumption growth by position was built for this review from the Follow-Up Inequality Survey, which UNU‐Wider conducted in Zambia in 2020, and the figures below are the review's rather than the survey's.\n\nHouseholds were grouped by their position in the consumption distribution at the first wave and the share whose real consumption rose by the follow-up computed: 41 per cent of households in the bottom quintile at the first wave against 66 per cent of those in the top quintile. The survey publishes both waves and never reports growth by starting position.\n\nWhether the household head changed activity between the waves was then read from the same records and equalised, and the distance narrows to 12 points, which leaves the change of activity as a substantial part of what the first reading attributes to position alone.\n\nHouseholds the survey could not re-interview at the follow-up are excluded rather than assumed unchanged, and they are 17 per cent of first-wave households. The grouping rule and the counts behind each share sit with the annex tables.\n\nMention: \"Follow-Up Inequality Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Follow-Up Inequality Survey relates starting consumption position to later growth.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:02acd2247bf9482c4563a4d5", "occurrence_id": "dc2ebb66c5a501ce31930431", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Whether the authority's own count of resettled households holds was tested against the Diagnostic Resettlement Registry (DRR), First Version, the register Reinhart & Rogoff compiled for 2024, and the registry is the principal source for the test.\n\nThe authority's relocation orders list 12 800 resettled households. The registry lists 14 300 for the same sites, because it records households displaced by the same clearance that were rehoused by a self-settlement payment rather than by an order. The authority's list counts orders rather than households.\n\nThe registry therefore carries the section's count of resettled households, and the authority's lower figure is printed beside it so the reader can see that the distance is self-settled households rather than a disagreement about who was displaced. Where a household appears in both, the authority's rehousing date is used and the two are not averaged.\n\nHouseholds the registry lists at a site whose coordinates it could not fix are excluded from the site-level comparison rather than assigned to the nearest site, 4 per cent of its entries. The rule each source applies is recorded with the annex tables.\n\nMention: \"Diagnostic Resettlement Registry (DRR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Diagnostic Resettlement Registry is the principal source testing the authority's resettlement count.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:0899336954cff77383e87542", "occurrence_id": "1f0142df835955041be27fbf", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Whether the shelter's own occupancy count survives was tested against the Civil Forced Population Survey, which SBP conducted in Fortaleza in 2015, and the survey is the principal source for the test.\n\nThe shelter's roll shows 2 340 residents on the count night. The survey, which interviews households wherever they sleep, puts 2 910 for the same settlements, because it counts households sleeping in a structure adjacent to the shelter that the roll does not list. The roll counts beds assigned rather than people present.\n\nThe survey therefore carries the section's occupancy figure, and the shelter's lower count is printed beside it so the reader can see that the distance is unlisted households rather than double counting. Where a household appears in both, the survey's household composition is used and the two counts are not averaged.\n\nIncome enters the test because the survey finds 57 per cent of the additional households below the municipal adequacy line against 44 per cent among those the roll lists, and the section uses that difference to argue the roll under-states need. Households staying less than a week are excluded from both rather than counted by one only. The counting rule each source applies is recorded with the annex tables.\n\nMention: \"Civil Forced Population Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Civil Forced Population Survey is the principal source testing the shelter's occupancy count.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:196d5247ab71afad93c0efd5", "occurrence_id": "611adf574a1cde2ae262083a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Complaint resolution was measured for this assessment from the Analytical Service Delivery Registry (ASDR), Round 5, the register kept for Dar-es-Salaam in 2018, and the figures below are the assessment's own.\n\nComplaints were grouped by the channel through which they arrived and the median days before the authority closed them computed: 11 days for complaints lodged in person at the counter against 29 days for those arriving by letter. The registry stamps each complaint with its channel and its closing date and publishes no comparison between channels.\n\nWhether the complaint names a street the authority has on file was then read from the same records and equalised, and the distance narrows to 7 days, which leaves the channel itself as the larger part of the delay than how identifiable the complaint is.\n\nComplaints reopened after closing are excluded from the median rather than counted to their latest closing, and they are 6 per cent of complaints. The grouping rule and the counts behind each median sit with the annex tables.\n\nMention: \"Analytical Service Delivery Registry (ASDR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Analytical Service Delivery Registry relates the complaint channel to how quickly it is closed.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:0817f61a3fc878cc8fa13b73", "occurrence_id": "4e98307aff44829bb8c37100", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Delivery gaps were built for this review from the Standard Service Delivery Index (SSDI), the index compiled for Argentina in 2022, and the measures below are the review's rather than the index's.\n\nLocalities were grouped by population band and the share reporting all four basic services delivered at standard computed: 58 per cent of localities above 50 000 residents against 23 per cent of those below 5 000. The index scores each service for each locality and reports no joint figure across the four.\n\nDistance to the nearest service centre was then read from the same records and equalised, and the distance narrows to 14 points, which leaves distance as a substantial part of what the first reading attributes to size alone.\n\nLocalities the index scores on three services or fewer are excluded rather than treated as failing the rest, and they are 8 per cent of localities. The bands and the counts behind each share are recorded under the table.\n\nMention: \"Standard Service Delivery Index (SSDI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Standard Service Delivery Index supports a combined four-service coverage measure by locality size.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:03fe138238c8e9166f8ee1e4", "occurrence_id": "94571a0fd44eb0ec93effc69", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Terrain records were catalogued into the site file from the Civil Elevation Model Dataset (CEMD), Wave 3, the geospatial dataset covering the Amhara region in 2015, and the work done on them was cataloguing rather than measurement.\n\nEach site was given the file's own catchment code and its elevation checked against the dataset's cell; where a site falls on the edge of two cells of different height, it was parked for the geographer rather than given the higher reading. Sites the file lists without coordinates were left unheighted rather than assigned their district's mean, and 212 of them entered the file in that condition. The dataset's own wave date was kept on every site so a later wave can be matched without re-sampling the surface.\n\nThe site file's population figures come from the agency's registration, from which the dataset supplied nothing; it is the terrain coding that lets those figures be read against slope and height for the flood warning the report draws up.\n\nThe coding sheet, the count of parked sites and the date of deposit are stored in the annex log.\n\nMention: \"Civil Elevation Model Dataset (CEMD)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The Civil Elevation Model Dataset supplies the terrain coding catalogued into the agency's site file.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:138894fb8b9414005b7471a7", "occurrence_id": "cb995146a35504b3d5538f9b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The prevalence figure in this chapter is the programme's own, taken from its clinic records, and the Rapid Mental Health Survey conducted by SIPRI in the Republic of Congo in 2016 is cited to test it.\n\nThe programme's records put 6 per cent of attendees presenting a condition of interest. The survey puts 11 per cent of its sampled adults, because it screens everyone it meets while the clinic counts only those who present with a symptom that prompts a recording. Neither basis reworks the other's data.\n\nThe 5-point distance is smaller than the difference the clinic's own series shows between its two busiest seasons, so the programme's figure is printed as recorded and the survey supplies no revision to it and no screens for the departments its teams did not reach.\n\nExcluded are respondents the survey screened but whose symptom module was left incomplete, 2 per cent of its sample, since no condition can be assigned to them. What each source counts as a case is recorded with the annex tables.\n\nMention: \"Rapid Mental Health Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Rapid Mental Health Survey checks the clinic programme's prevalence figure for conditions of interest.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:621caa4397dd0c94d5d0a1fd", "occurrence_id": "35518dedd3053773593b3d21", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The coverage figure in this chapter is the ministry's own, taken from its facility census, and the National Service Delivery Index (NSDI) compiled for South Asia in 2021 is cited to test it.\n\nThe ministry's census puts 91 per cent of the population within 5 kilometres of a functioning facility. The index puts 84 per cent for the same countries, because it measures along its road layer and counts a facility as functioning only where staff are present on the day of the visit. Neither measure was applied to the other's data.\n\nThe 7-point distance is smaller than the spread the ministry's own census shows between its rural and urban divisions, so the ministry's figure is printed as computed and the index supplies no revision to it and no countries for which it holds no road layer.\n\nExcluded are the countries the index covers on a different reference year, 4 of its 19, since the two cannot be aligned for this chapter. How each source measures access is recorded with the annex tables.\n\nMention: \"National Service Delivery Index (NSDI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The National Service Delivery Index checks the ministry's population coverage within 5 kilometres.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:988ace9a0323db38b7a15d97", "occurrence_id": "a2c779c89a6caad1e6651b8a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Built surface per person was derived for this chapter from the Linked Elevation Model Census, the census Copernicus conducted in Cyprus in 2022, and the figures below come from the chapter, not from the census's.\n\nOutput zones were grouped by the mean height of their terrain and the surface of roofed area per resident computed: 41 square metres in zones below 100 metres against 66 square metres in those above 400 metres. The census reports roofed area and a height for each zone and never relates them.\n\nThat comparison supports one claim, that the pressure on coastal land is being measured against a floor that ignores how dispersed building is inland. The land-price series the chapter quotes comes from the registry's transaction file, from which the census supplied nothing.\n\nZones the census reports with no residents are excluded rather than given an infinite area, and they are 5 per cent of zones. The bands and the counts behind each figure are recorded under the table.\n\nMention: \"Linked Elevation Model Census\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Linked Elevation Model Census supports one claim by relating roofed area per person to terrain height.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:03481b2b2588b5a7d4bd5486", "occurrence_id": "740bcb943fc9993a67504285", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Parcel coverage was computed for this review from the Clustered Parcel Footprint System (CPFS), Edition 5, the register Samoa Bureau of Statistics maintains for Jordan, and the readings below are the review's own.\n\nParcels were grouped by whether their footprint touches a public road and the share carrying a structure of more than two storeys computed: 34 per cent of road-touching parcels against 9 per cent of those reached only through a servitude. The register holds a footprint and an access indicator for each parcel and publishes no crossing of the two.\n\nPlot area was then equalised between the groups and the distance narrows to 12 points, which leaves the size of the plot as a substantial part of what the first reading attributes to access.\n\nParcels whose footprint the register marks as uncertain are excluded rather than counted as unbuilt, and they are 7 per cent of parcels. The grouping rule and the counts behind each share sit with the annex tables.\n\nMention: \"Clustered Parcel Footprint System (CPFS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Clustered Parcel Footprint System relates road frontage to the height of the structure on a parcel.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:608bb826697c711f47542f03", "occurrence_id": "282909ab92d6ab5665b8d542", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Job creation was measured for this assessment from the Joint Enterprise and Jobs Survey (JEJS), Round 3, which covers Ile de France in 2024, and the figures below are the assessment's own.\n\nEstablishments were grouped by whether they introduced a new product or process in the previous two years and the net change in employees over the same period computed: a gain of 6 per cent among innovating establishments against a loss of 2 per cent among those that did not. The survey records innovation and headcount for each establishment and never reports the two together.\n\nWhether the establishment exports was then read from the same records and equalised, and the distance narrows to 3 points, which leaves exporting as a substantial part of what the first reading credits to innovation.\n\nEstablishments that closed between the two observations are excluded rather than counted as shedding all their staff, and they are 4 per cent of establishments. The grouping rule and the counts behind each figure are recorded with the annex tables.\n\nMention: \"Joint Enterprise and Jobs Survey (JEJS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Joint Enterprise and Jobs Survey compares employment change between innovating and other establishments.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:ed4f3b52cecdcb869ff02698", "occurrence_id": "5244906f9de0fe7d1b638c8a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Programme monitoring city-level figures come from the Standardised Programme Monitoring Index (SPMI), the index compiled for Addis Ababa in 2019, and the four values below are given exactly as the index issues them.\n\nThe index puts the share of planned activities completed within their scheduled quarter at 62 per cent, the share of expenditure released against a completed activity at 48 per cent, the mean months between an activity's completion and its verification at 3, and the share of verified activities whose evidence is filed electronically at 71 per cent. None was recomputed here, and the index's own definitions of a completed activity and of verification appear with the values.\n\nAn activity completed late was not counted as completed within its quarter, and an activity verified only by the implementing unit was not counted as verified. Where the index reports for one sector, that sector is named with the value rather than pooled.\n\nThe index table and line for each value, along with the extraction date and the activities behind each share are listed in the annex notes.\n\nMention: \"Standardised Programme Monitoring Index (SPMI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Standardised Programme Monitoring Index supplies completion, release, verification and filing values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:a9714a9c7b3363bc82408b84", "occurrence_id": "dbbdfe3df7ff760befff3d02", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Vegetation change was worked out for this chapter from the Ministerial Vegetation Condition Collection (MVCC), Fourth Wave, the collection Reinhart & Rogoff publish, and the figures below are the chapter's own rather than the collection's.\n\nCells were grouped by the dryness of their own long-term mean and the change in greenness between the collection's two observation dates computed: a decline of 0.04 index points in the driest third against a gain of 0.02 in the wettest third. The collection reports an index for each cell and date and publishes no reading of change by dryness band.\n\nThat comparison supports one claim, that the programme's restoration works should be sited where the ground is already marginal rather than where the maps show the largest loss. The cost per hectare the chapter quotes comes from the works office's schedule, from which the collection supplied nothing.\n\nCells flagged as burned in either observation are excluded rather than counted as a decline, and they are 6 per cent of cells. The bands and the counts behind each figure are recorded under the table.\n\nMention: \"Ministerial Vegetation Condition Collection (MVCC)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Ministerial Vegetation Condition Collection supports one siting claim by relating greenness change to dryness.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:53cbff1e19dddaa5107617ba", "occurrence_id": "741daaec6368ef9633ea437e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The poverty figures in this chapter are computed from the office's own household panel, and the Baseline Asset Ownership Survey conducted by BCL in Iraq in 2020 is referred to only as background to them.\n\nThe survey puts the share of households owning their dwelling at 71 per cent, the mean number of asset categories a household holds at 6, and the share holding no productive asset at 28 per cent. All three are given as the survey issues them, with its own list of asset categories stated, and no chapter figure rests on them.\n\nThe chapter's tables draw on none of its responses. The figures the chapter argues from come from the panel, which follows the same households over time rather than describing them once, and where the two accounts of deprivation differ the difference is noted rather than reconciled.\n\nThe annex lists the survey tables consulted and the date of extraction, and the survey is used for nothing beyond that background.\n\nMention: \"Baseline Asset Ownership Survey\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The Baseline Asset Ownership Survey is background while the chapter's poverty figures come from the panel.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:e9f5a3c56f7f0310bb1226f3", "occurrence_id": "b7d25c2c13267c5182e67689", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Bargaining coverage was computed for this review from the Inter-Agency Collective Bargaining Survey (ICBS), which covers Punjab in 2021, and the measures below are the review's rather than the survey's.\n\nEstablishments were grouped by the number of workers they employ and the share whose wages are set by a signed agreement computed: 64 per cent of establishments with 100 workers or more against 12 per cent of those with fewer than 20. The survey records employment size and agreement status for each establishment and never crosses them.\n\nWhether the establishment operates in a designated industrial estate was then read from the same records and equalised, and the distance narrows to 17 points, which leaves the estate as a substantial part of what the first reading attributes to size alone.\n\nEstablishments reporting a union but no agreement in negotiation are excluded rather than counted as covered, and they are 5 per cent of establishments. The grouping rule and the counts behind each share sit with the annex tables.\n\nMention: \"Inter-Agency Collective Bargaining Survey (ICBS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Inter-Agency Collective Bargaining Survey relates establishment size to signed wage agreements.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:a18dfaaaf8b532dbe32c2b46", "occurrence_id": "a7cc722c5d137fb7e7397478", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Attendance and the meal were related for this assessment using the Regional School Feeding Survey, which MoF follows in Sierra Leone in 2019, and the figures below are the assessment's own.\n\nSchools were grouped by the number of days per week they served a meal and the mean share of enrolled pupils present on the last day of the term computed: 88 per cent at schools serving five days a week against 79 per cent at those serving two or fewer. The survey reports the serving schedule and the attendance count for the same school and publishes no relationship between them.\n\nWhether the school sits more than an hour's walk from a road was then read from the same records and equalised, and the distance narrows to 4 points, which leaves remoteness as a substantial part of what the first reading credits to the meal.\n\nSchools closed for the whole of the reference period are excluded rather than counted at zero attendance, and they are 3 per cent of schools. The grouping rule and the counts behind each mean are recorded with the annex tables.\n\nMention: \"Regional School Feeding Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Regional School Feeding Survey relates meal days per week to end-of-term attendance.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:f14f11990afe67cdda8483f5", "occurrence_id": "3a7b06ac4f7477f430c06cfb", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Treatment interruption was built for this chapter from the Quality-Assured Health and Household Survey (QHHS), Sixth Collection, compiled by the Uppsala Conflict Data Programme, and the figures below are the chapter's rather than the collection's.\n\nHouseholds were grouped by whether a health worker was present at their nearest facility in the reference month and the share reporting an interrupted course of treatment computed: 31 per cent of households whose facility was unstaffed at some point in the month against 13 per cent of those whose facility was staffed throughout. The collection records staffing visits and treatment history without publishing the crossing.\n\nThat comparison supports one claim, that continuity of staffing rather than drug supply is the first thing the referral plan should secure. The vehicle-availability figures the plan uses come from the logistics register, from which the collection supplied nothing.\n\nHouseholds whose nearest facility closed permanently during the period are excluded rather than counted as interrupted, and they are 4 per cent of households. The grouping rule and the counts behind each share are recorded under the table.\n\nMention: \"Quality-Assured Health and Household Survey (QHHS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Quality-Assured Health and Household Survey supports one staffing claim through treatment interruption.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:115ee286d03064a6019e74a0", "occurrence_id": "8a430cf4a1d392c8e12a8ad3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The staffing figures in this chapter are compiled from the authority's own establishment list, and the Municipal Public Sector Monitoring Registry maintained by ZamStats for Romania in 2016 is cited only as background against which to read them.\n\nThe registry puts the share of local authorities whose establishment list agrees with their payroll at 76 per cent, the mean difference between the two at 4 per cent of posts, and the share of authorities revising the list at least annually at 58 per cent. All three are given as the registry issues them, with its own definition of an establishment list stated, and no figure in the chapter is computed from them.\n\nThe registry's own agreement test was applied to the authority's list and to no other, and the two accounts of staffing are reported separately rather than merged. The chapter's figures rest on the establishment list, which names each post rather than summarising a headcount, and where the two accounts differ the difference is noted rather than reconciled.\n\nThe annex lists the registry tables consulted and the date of extraction, which is the whole of the use made of the registry here.\n\nMention: \"Municipal Public Sector Monitoring Registry\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"validate\", \"summary\": \"The Municipal Public Sector Monitoring Registry is background to the authority's establishment staffing figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:64051fdeb1604191ff2c2adc", "occurrence_id": "0cc21badb65681a96463f19c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Informal trade figures for each part of the country are drawn from the National Informal Trade Index, the index compiled by Small Arms Survey for Turkey in 2015, and the four values below are transferred here unchanged.\n\nThe index puts the share of working adults whose main activity is unregistered at 32 per cent, the share of unregistered traders holding a municipal stall permit at 21 per cent, the mean monthly receipts of an unregistered trader at 640, and the share reporting a seizure of goods in the previous year at 9 per cent. None was recomputed here, and the index's own definitions of an unregistered activity and of a seizure accompany the values.\n\nA trader registered for tax but not with the municipality was counted as unregistered only where the index does so, and a seizure reported without a receipt was not counted. Where the index reports for one region, that region is named with the value rather than pooled.\n\nThe index table and line for each value, the extraction date and the traders behind each share are set down in the annex notes.\n\nMention: \"National Informal Trade Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The National Informal Trade Index supplies unregistered shares, permits, receipts and seizure values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:50552ae4fdd757724ec786c8", "occurrence_id": "6e678b456e1231a8481153a0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Lit-area figures measured across the country are read from the Quality-Assured Nighttime Lights Registry, the register the Government of India compiled for Burkina Faso in 2023, and the four values below are given as the registry holds them.\n\nThe registry puts the share of the population living in a lit cell at 44 per cent, the mean nights per month a lit cell stays lit at 21, the share of cells newly lit over the registry's own interval at 3 per cent, and the share of its cells whose observation was cloud-free at 67 per cent. None was recomputed here, and the registry's own threshold for calling a cell lit appears with the values.\n\nA cell lit only on market nights was not counted as newly lit, and a cell observed only through cloud was not counted as unlit. Where the registry reports for one region, that region is named with the value rather than pooled.\n\nThe registry layer and cell behind each value, the extraction date and the cells behind each share are written up in the annex notes.\n\nMention: \"Quality-Assured Nighttime Lights Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Quality-Assured Nighttime Lights Registry supplies lit-population, persistence, new-light and cloud values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:f412fd82f8a54b6522034e95", "occurrence_id": "c4c4666d0083801e7f66cac6", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Condition change and grazing were joined for this assessment from the Audited Vegetation Condition Collection (AVCC), Phase 7, the collection BCL compiles, and the figures below are the assessment's own.\n\nPasture polygons were matched to their condition class and grouped by the rainfall of the season, and the share whose condition improved between the collection's two phases computed: 18 per cent of polygons in the degraded class in a wet season against 5 per cent in a dry one. The collection grades each polygon and reports rainfall by period without publishing any such crossing.\n\nStock density per hectare was then read from the same records and equalised, and the distance narrows to 7 points, which leaves the season rather than the herd as what the assessment treats as the larger part of the improvement.\n\nPolygons whose class changed between phases for reasons the collection records as mapping error are excluded rather than counted as improved, and they are 4 per cent of polygons. The classes, the season bands and the counts behind each share sit with the annex tables.\n\nMention: \"Audited Vegetation Condition Collection (AVCC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Audited Vegetation Condition Collection relates pasture condition change to season and stock density.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:cbcb07580fd865dab41f825f", "occurrence_id": "cf30797bb9f233e1523d9881", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two operations lines in the protection annex are quoted from the Comparative Operations Monitoring Registry (COMR), the register OPT Protection Cluster maintains for Brazil, and both are given without restatement.\n\nThe registry puts the share of field offices filing their incident return on time at 68 per cent, the mean days between an incident being reported and its entry appearing in the register at 5, and the share of entries later revised by the reporting office at 12 per cent. Neither of the first two was re-derived for this annex, and the registry's own rule for counting an incident as reported on time, which uses the date the office received it rather than the date it occurred, appears with the figures.\n\nThe remaining annex lines come from the cluster's own case file, so the registry carries two lines while the case file carries what is left over. A figure given for one state is printed with that state named rather than averaged across the country.\n\nThe registry reference behind each line, the extraction date and the offices behind each share are recorded with the annex notes.\n\nMention: \"Comparative Operations Monitoring Registry (COMR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Comparative Operations Monitoring Registry supplies return timeliness, entry delay and revision lines.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:63c1662d240bf276d4b6a792", "occurrence_id": "687bbc38834a235f89cbd415", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Night-time lighting figures for the country are reported from the Subnational Nighttime Lights Dataset (SNLD), Edition 2, the geospatial dataset UNHCR and IOM compiled for Senegal, and the four values below are given exactly as the dataset issues them.\n\nThe dataset puts the share of its settlements showing measurable light at 37 per cent, the mean radiance of a lit settlement at 18 nanowatts per square centimetre per steradian, the share of settlements whose light rose between its own two observations at 6 per cent, and the share of its extent observed without cloud at 72 per cent. None was recomputed here, and the dataset's own threshold for measurable light is stated with the values.\n\nA settlement lit only by a flare at a worksite was not counted as lit, and a settlement seen only through cloud was not counted as unlit. Where the dataset reports for one region, that region is named with the value rather than pooled.\n\nThe dataset layer and table behind each value, the extraction date and the settlements behind each share are carried in the annex notes.\n\nMention: \"Subnational Nighttime Lights Dataset (SNLD)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Subnational Nighttime Lights Dataset supplies lit-settlement share, radiance, brightening and cloud values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:ad6283463e3f5941b690d467", "occurrence_id": "0aa04485df7627468eb5fbbf", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Whether the ministry's own count of women farming holds was tested against the Rapid Gender and Farming Registry, the register Latinobarómetro Corporation compiled for Tunisia in 2024, and the registry is the principal source for the test.\n\nThe ministry's extension records put 21 per cent of holders of a farm registration card at women. The registry puts 39 per cent of plot operators at women, because it interviews the person who works the plot rather than the person whose name is on the card, and a plot registered to a husband but worked by his wife is counted by the registry as hers. The ministry counts cards.\n\nThe registry therefore carries the section's figure for who farms, and the ministry's lower share is printed beside it so the reader can see that the distance is cards held in a household name rather than a disagreement about who works the land. Where a plot appears in both, the registry's operator is used and the two shares are not averaged.\n\nPlots left fallow for the whole season are excluded from both rather than counted by the registry only, 6 per cent of plots. The counting rule each source applies is recorded with the annex tables.\n\nMention: \"Rapid Gender and Farming Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Rapid Gender and Farming Registry is the principal source testing the ministry's count of women farmers.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:b56e5e65c5abf1a043da91d4", "occurrence_id": "b44c265b05575967b9827c1f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Absenteeism was derived for this chapter from the National Public Sector Monitoring System, the administrative register the National Weather Service maintains for Costa Rica in 2018, and the figures below are the chapter's own, not the register's.\n\nPosts were grouped by the distance of the workplace from the nearest public transport stop and the mean days absent per post in the year computed: 9 days for posts within 500 metres of a stop against 15 days for those beyond 3 kilometres. The register records a distance for each workplace and days lost for each post and never reports the two together.\n\nThat comparison supports one claim, that the travel allowance should be extended before the shift pattern is changed. The wage bill figures the proposal uses come from the treasury's vote, from which the register supplied nothing.\n\nPosts vacant for more than half the year are excluded from the mean rather than counted at zero absence, and they are 7 per cent of posts. The bands and the counts behind each mean are recorded under the table.\n\nMention: \"National Public Sector Monitoring System\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The National Public Sector Monitoring System supports one allowance claim through distance and absence.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:7261e349e9874fe171ddf440", "occurrence_id": "fc39beb747f343fab2610cd3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two operations lines in the country annex are quoted from the Ministerial Operations Index Registry, the register BPS maintains for Eritrea as of 2018, and both are carried over unchanged.\n\nThe registry puts the share of ministries publishing their annual statistics at 46 per cent and the median months between a period closing and its statistics appearing at 9. Neither was recalculated for this annex, and the registry's own rule for counting a ministry as publishing, which requires tables rather than a narrative, is reproduced with them.\n\nThe annex's remaining lines are compiled from the planning office's budget review, so the registry carries two lines while the review carries the rest. A figure given for one ministry is printed with that ministry named rather than averaged across government.\n\nThe registry reference behind each line, the extraction date and the ministries behind each share are recorded with the annex notes.\n\nMention: \"Ministerial Operations Index Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Ministerial Operations Index Registry supplies publication coverage and statistics delay lines.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:23fa0cad5abff783c84a717a", "occurrence_id": "f15026ad79e0e3ddc6203d74", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Vocational training the country figures are compiled from the Representative Vocational Training Survey, which Federal Police conducted in Kenya in 2022, and the four values below are transferred here unchanged.\n\nThe survey puts the share of young people aged 18 to 35 holding a vocational certificate at 19 per cent, the mean months a certificate-holder takes to find work related to the trade at 7, the share of employers accepting a certificate in place of experience at 43 per cent, and the share of graduates working outside their trained trade at 38 per cent. None was recomputed here, and the survey's own definitions of a certificate and of related work appear with the values.\n\nA short course of fewer than 40 hours was not counted as a certificate, and work in a different branch of the same trade was not counted as related. Where the survey reports for one county, that county is named with the value rather than pooled.\n\nThe survey table and line for each value: the extraction date and the respondents behind each share are detailed in the annex notes.\n\nMention: \"Representative Vocational Training Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Representative Vocational Training Survey supplies certificate holding, job search, acceptance and mismatch values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:55950511cacb7998f53cb74a", "occurrence_id": "ba8583f5710aa7916b2fa4e1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Land height and income were compared for this assessment using the Analytical Terrain and Elevation Census, the census УВКБ ООН, REACH conducted in Xinjiang province in 2022, and the comparison is the assessment's rather than the census's.\n\nHouseholds were grouped by the height of the land they live on and the share below the regional adequacy line computed: 34 per cent of households above 1 500 metres against 16 per cent of those below 800 metres. The census records an elevation for each dwelling and a consumption aggregate for each household and never reports the two together.\n\nDistance to the nearest market town was then read from the same records and equalised, and the distance narrows to 9 points, which leaves how far the household sits from a town as a substantial part of what looks at first like an altitude effect.\n\nHouseholds the census places at a settlement without an elevation figure are excluded rather than given the district mean, and they are 5 per cent of households. The bands and the counts behind each share sit with the annex tables.\n\nMention: \"Analytical Terrain and Elevation Census\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Analytical Terrain and Elevation Census relates elevation of the dwelling to income adequacy.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:5548aef7533edd48f305d4c3", "occurrence_id": "87a99b6b73aa1e08fda8b17f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Parcel records were catalogued into the shelter file from the Midline Parcel Footprint Survey, the survey WFP carried out in Tajikistan in 2016, and the work done on them was cataloguing rather than measurement.\n\nEach parcel was given the file's own plot code and its structure count checked against the survey's count; where the two disagree by more than one structure, the record was parked for the field officer rather than settled on the survey's figure. Parcels the survey lists without an owner were left unattributed rather than assigned to the village register, and 388 of them entered the file in that condition. The survey's own visit date was kept on every parcel so a later round can be matched without re-walking the block.\n\nThe shelter file's household counts come from the agency's registration, from which the survey supplied nothing; it is the parcel coding that lets those counts be read against plot size for the relocation plan.\n\nThe coding sheet, the count of parked parcels and the date of deposit are carried in the annex log.\n\nMention: \"Midline Parcel Footprint Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The Midline Parcel Footprint Survey supplies the coded parcel records behind the shelter file.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:434ee798d2095b8ea59afcc4", "occurrence_id": "10ccf1abf5032a1eceda1a15", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two capacity lines in the laboratory annex are quoted from the Sentinel Diagnostic Capacity Survey (SDCS), Version 2, compiled for Serbia in 2021, and both are reproduced without amendment.\n\nThe survey puts the share of laboratories able to return a result within 24 hours at 57 per cent, the share whose nearest served district lies beyond 30 kilometres at 22 per cent, and the mean number of tests a laboratory performs in a week at 310. Neither of the first two was recalculated here, and the survey's own rule for counting a result as returned within 24 hours, which runs from receipt of the specimen rather than from the order, is stated with the figures.\n\nThe annex's other lines are compiled from the network's own requisition log, so the survey carries two lines while the log carries the rest. A figure given for one region is printed with that region named rather than averaged across the country.\n\nThe survey table behind each line, the extraction date and the laboratories behind each share are recorded with the annex notes.\n\nMention: \"Sentinel Diagnostic Capacity Survey (SDCS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Sentinel Diagnostic Capacity Survey supplies turnaround, served-distance and workload lines.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:c18e52f3a0e5344085ada160", "occurrence_id": "538ac63a3505f1285fc6be6e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two public sector lines in the annex are quoted from the Harmonised Public Sector Monitoring Index (HPSMI), the index compiled for Iraq in 2017, and both are given exactly as the index issues them.\n\nThe index puts the share of agencies with a published staffing table at 61 per cent and the mean months an agency takes to fill a sanctioned post at 8. Neither was re-derived for this annex, and the index's own definition of a sanctioned post, which requires a budget line as well as an approval, appears with the figures.\n\nThe other lines in the annex are copied from the ministry's own establishment returns, so the index carries two lines while the returns carry the rest. A figure given for one agency is printed with that agency named rather than averaged across the service.\n\nThe index entry behind each line, the extraction date and the agencies behind each share are recorded with the annex notes.\n\nMention: \"Harmonised Public Sector Monitoring Index (HPSMI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Harmonised Public Sector Monitoring Index supplies staffing-table coverage and filling time lines.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:7755200606e020f0423be0fe", "occurrence_id": "3e168c6ecbf88594ceea5940", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Displacement figures for the whole region are sourced from the Diagnostic Displacement Tracking Registry, the register Census maintains for Niger as of 2019, and the four values below are given as the register holds them.\n\nThe register puts the number of displaced persons in the region at 128 000, the share living with a host household rather than at a site at 63 per cent, the mean months since the household last moved at 14, and the share reporting a second move since displacement at 18 per cent. None was recomputed here, and the register's own definitions of a displaced person and of a site appear with the values.\n\nA household that moved within its own village was not counted as displaced, and a site operating without registration was not counted as a site. Where the register reports for one department, that department is named with the value rather than pooled.\n\nThe register table and line for each value, the extraction date, and the persons behind each share are kept in the annex notes.\n\nMention: \"Diagnostic Displacement Tracking Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Diagnostic Displacement Tracking Registry supplies displaced numbers, hosting, duration and repeat-move values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:12808c463bb137651be48279", "occurrence_id": "4bf26c2190e29e915dc3c363", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Administrative records the country figures are tabulated by the Annual Administrative Records Survey, which DOLISA conducted in Korea in 2020, and the four values below are transferred here unchanged.\n\nThe survey puts the share of registers reconciled against a second source in the year at 52 per cent, the mean years since a register's structure was last documented at 4, the share of registers whose updates are logged rather than overwritten at 37 per cent, and the share of custodians reporting access to the documentation as adequate at 44 per cent. None was recomputed here, and the survey's own definitions of a reconciliation and of a logged update accompany the values.\n\nA register reconciled only within its own department was not counted as reconciled against a second source, and a change recorded in a paper file was not counted as logged. Where the survey reports for one custodian class, that class is named with the value rather than pooled.\n\nThe survey table and line for each value, the extraction date as well as the registers behind each share are held in the annex notes.\n\nMention: \"Annual Administrative Records Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Annual Administrative Records Survey supplies reconciliation, documentation, logging and access values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:438e8e82988ada546268e2d8", "occurrence_id": "a341917e864bf362068cf535", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Housing cost burden was measured for this review from the Quarterly Housing Conditions Survey (QHCS), which covers Lebanon in 2016, and the figures below are the review's rather than the survey's.\n\nHouseholds were grouped by the share of their income taken by rent and the share reporting they cut food spending in the last quarter computed: 29 per cent of households paying more than 40 per cent of income against 9 per cent of those paying less than 15 per cent. The survey publishes the rent share and the coping items for the same household and never crosses them.\n\nWhether the household holds a written tenancy agreement was then read from the same records and equalised, and the distance narrows to 12 points, which leaves the absence of an agreement as a substantial part of what the first reading attributes to the rent alone.\n\nHouseholds owning their dwelling are excluded rather than counted as paying no rent, and they are 34 per cent of households. The bands and the counts behind each share sit with the annex tables.\n\nMention: \"Quarterly Housing Conditions Survey (QHCS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Quarterly Housing Conditions Survey relates the rent share of income to cuts in food spending.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:4b8bb3ec3fafc559ddfc90ed", "occurrence_id": "3c57c0f73dd5e34fd66673b4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The statistical capacity figure in this chapter is the office's own, taken from its capability checklist, and the Subnational Administrative Survey Index (SASI), Wave 5, compiled for Ethiopia in 2015, is cited to test it.\n\nThe office's checklist puts 78 per cent of offices at the second capability level or above. The index puts 65 per cent for the same offices, because it requires evidence of a published output as well as the staffing and budget items the checklist counts. Neither standard was applied to the other's data.\n\nThe 13-point distance is smaller than the movement the office's checklist shows between its own two rounds, so the office's figure is printed as scored and the index supplies no revision to it and no scores for the offices it did not visit.\n\nExcluded are offices the index visited but scored on fewer than six of its items, 9 of its entries, since a partial score cannot be placed on either scale. Which items each source counts is recorded with the annex tables.\n\nMention: \"Subnational Administrative Survey Index (SASI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Subnational Administrative Survey Index checks the office's share of offices at the capability level.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:019611eb76ceacbdf91f39b9", "occurrence_id": "9149f476a1d1246a9a027983", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Seed use the figures for the districts are reported by the District Seed Use Survey (DSUS), Seventh Version, the survey the US census runs in Comoros, and the four values below are given exactly as the survey issues them.\n\nThe survey puts the share of households sowing saved seed from their own previous harvest at 74 per cent, the share buying certified seed in the season at 6 per cent, the mean kilograms of seed a household sows per hectare at 12, and the share reporting seed shortage as a reason for not planting at 17 per cent. None was recalculated here, and the survey's own definitions of saved seed and of certified seed appear with the values.\n\nSeed exchanged between neighbours without payment was counted as saved rather than bought, and seed used for food was not counted as sown. Where the survey reports for one island, that island is named with the value rather than pooled.\n\nThe survey table and line for each value, plus the extraction date and the households behind each share are laid out in the annex notes.\n\nMention: \"District Seed Use Survey (DSUS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The District Seed Use Survey supplies saved-seed, certified purchase, sowing rate and shortage values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:c241bc11464ba7f005287fa1", "occurrence_id": "00d2170ddbeeb8465bea66f0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Harvest and adequacy were compared for this assessment using the Cross-Sectoral Food Security Survey, which E.Wong conducted in Azerbaijan in 2023, and the comparison is the assessment's rather than the survey's.\n\nHouseholds were grouped by the share of their crop they sold rather than kept and the share reporting a month with too little food in the previous year computed: 26 per cent of households selling more than half their crop against 47 per cent of those keeping almost all of it. The survey records the sold share and the food-gap months for the same household and never reports the two together.\n\nWhether the household holds irrigation was then read from the same records and equalised, and the distance narrows to 8 points, which leaves irrigation as a substantial part of what the first reading attributes to the decision to sell.\n\nHouseholds with no land of their own are excluded rather than grouped with those keeping everything, and they are 9 per cent of households. The bands and the counts behind each share sit with the annex tables.\n\nMention: \"Cross-Sectoral Food Security Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Cross-Sectoral Food Security Survey relates the share of crop sold to months of inadequate food.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:b0ac7602c6fe11fa77d3f203", "occurrence_id": "d31adf0864d013729883a494", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The claimant count in this chapter is the ministry's own, taken from its benefit file, and the Coordinated Unemployment Insurance Registry (CUIR), Seventh Release, compiled by FAO, is cited to test it.\n\nThe ministry's file puts 214 000 claimants in payment at the reference month. The registry puts 197 000 for the same month, because it removes claimants whose entitlement ended during the month but who were paid for the whole of it. Neither rule was applied to the other's data.\n\nThe 17 000 distance is smaller than the month-to-month movement the ministry's own file shows, so the ministry's figure is printed as paid and the registry supplies no revision to it and no claimants for the schemes it does not cover.\n\nExcluded are claimants the registry lists without a scheme identifier, 1 300 of them, since neither count can be attributed to a scheme for them. Which claimants each source counts is recorded with the annex tables.\n\nMention: \"Coordinated Unemployment Insurance Registry (CUIR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Coordinated Unemployment Insurance Registry checks the ministry's count of claimants in payment.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:1bcfe4dce89be5e8480595ea", "occurrence_id": "c482d628acfe47fc8fbc841d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two bulletin lines in the statistical annex are quoted from the Midline Statistical Bulletin Registry, the register SEP maintains for Slovensku as of 2015, and both are carried over unchanged.\n\nThe registry puts the share of bulletins issued on their announced date at 58 per cent and the median number of revisions a bulletin undergoes after issue at 1. Neither was recalculated for this annex, and the registry's own rule for counting a revision, which requires a change to a published figure rather than to its notes, is reproduced with them.\n\nThe annex's remaining entries follow the office's own production calendar, so the registry carries two lines while the calendar carries the rest. A figure given for one bulletin series is printed with that series named rather than averaged across the office.\n\nThe registry reference behind each line, the extraction date and the bulletins behind each share are recorded with the annex notes.\n\nMention: \"Midline Statistical Bulletin Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Midline Statistical Bulletin Registry supplies on-time issuance and revision counts for the annex.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:be251c02674ec19dfba3a566", "occurrence_id": "379ae3639464adcd3f36ebd8", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Vocational training and earnings figures covering the country are drawn from the Annual Vocational Training Survey, which the Bank of Italy conducted in Mozambique in 2015, and the four values below are given as the survey issues them.\n\nThe survey puts the share of trained adults holding paid work at 58 per cent, the mean monthly earnings of a trained adult at 2 400, the share of trained adults whose household falls below the national adequacy line at 22 per cent, and the share reporting training in a trade with no local employer at 15 per cent. None was recomputed here, and the survey's own definitions of trained and of paid work appear with the values.\n\nAn unpaid apprenticeship was not counted as paid work, and a household receiving a remittance larger than its own earnings was still counted on its own earnings. Where the survey reports for one province, that province is named with the value rather than pooled.\n\nThe survey table and line for each value, along with the extraction date and the adults behind each share are filed in the annex notes.\n\nMention: \"Annual Vocational Training Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Annual Vocational Training Survey supplies trained employment, earnings, poverty and trade-mismatch values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:6aa76cff3a670896301835e3", "occurrence_id": "90f190d69ad4c18bb33bb66e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Nutrition and schooling were related for this assessment from the Regional Nutrition Surveillance Index (RNSI), the index compiled by Lagos State Government for 2022, and the figures below are the assessment's own.\n\nAreas were grouped by the mean height-for-age recorded in the surveillance round and the share of children enrolled in the first year of school computed: 81 per cent of areas in the upper band against 62 per cent of those in the lowest. The index reports the nutrition measure and an enrolment item for each area and never publishes the two against each other.\n\nDistance to the nearest secondary school was then read from the same records and equalised, and the distance narrows to 7 points, which leaves distance as a substantial part of what the first reading associates with nutrition.\n\nAreas the index surveyed at fewer than 20 children are excluded rather than pooled with their neighbours, and they are 6 per cent of areas. The bands and the counts behind each share sit with the annex tables.\n\nMention: \"Regional Nutrition Surveillance Index (RNSI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Regional Nutrition Surveillance Index relates height-for-age bands to first-year school enrolment.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:d3392c2f143cfe6236932766", "occurrence_id": "fe7e976ed162d7ee0bfd1eaf", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Transport network figures covering the whole region derive from the District Transport Network Report (DTNR), Release 9, the report compiled for the Amhara region in 2020, and the four values below are transferred here unchanged.\n\nThe report puts the length of all-weather road per thousand residents at 1.4 kilometres, the share of settlements reached by a vehicle in the dry season at 67 per cent, the mean travel time from a rural settlement to the nearest district town at 74 minutes, and the share of its road segments carrying a recorded surface type at 82 per cent. None was recomputed here, and the report's own definition of an all-weather road, which excludes a surface graded only once in the year, appears with the values.\n\nA segment passable only with a four-wheel drive was not counted as all-weather, and a settlement reached by foot track alone was not counted as reached. Where the report gives a figure for one woreda, that woreda is named with the value rather than pooled.\n\nThe report's table and line for each value, the extraction date and the settlements behind each share are recorded in the annex notes.\n\nMention: \"District Transport Network Report (DTNR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The District Transport Network Report supplies road density, dry-season reach, travel time and surfacing.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:8a9ef7c890514f691adfbd0e", "occurrence_id": "99e485434e8a5a7196420fbd", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Skills and employment the numbers held for the country come from the Continuous Skills and Employment Survey (CSES), the survey NSO follows in Costa Rica, and the four values below are given exactly as the survey issues them.\n\nThe survey puts the share of adults reporting a qualification at the second level or above at 49 per cent, the share of workers whose current work bears no relation to that qualification at 31 per cent, the mean hours in training a worker received in the year at 26, and the share of employers reporting a skills gap as their main hiring difficulty at 38 per cent. None was recalculated here, and the survey's own definitions of a qualification and of a relation to work appear with the values.\n\nA course completed without assessment was not counted in the training hours, and a worker retraining for a different occupation was not counted as mismatched. Where the survey reports for one province, that province is named with the value rather than pooled.\n\nThe survey table and line for each value: the extraction date and the respondents behind each share are stored in the annex notes.\n\nMention: \"Continuous Skills and Employment Survey (CSES)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Continuous Skills and Employment Survey supplies qualification, mismatch, training and gap values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:753908473075a1bc2903c449", "occurrence_id": "4f939037da1d3ebfb3c18cd2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Office workload per staff was derived for this chapter from the Linked Regional Operations Index (LROI), the index DiREC compiled for 2023, and the figures below belong to the chapter rather than to the index's.\n\nOffices were grouped by the caseload per clerk the index records and the share of cases closed within the standard the office sets computed: 71 per cent in offices below 10 cases per clerk against 44 per cent in those above 25. The index reports a caseload figure and a closure rate for each office and never relates them.\n\nThat comparison supports one claim, that the second clerk post should go to the overloaded offices before any new case category is opened. The payroll cost the proposal quotes comes from the establishment table, from which the index supplied nothing.\n\nOffices the index holds without a clerk count are excluded rather than given the regional mean, and they are 7 per cent of offices. The bands and the counts behind each share are recorded under the table.\n\nMention: \"Linked Regional Operations Index (LROI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Linked Regional Operations Index supports one staffing claim by relating caseload to closure rates.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:2ce98456d32133c25bbc9874", "occurrence_id": "0a9a37566bd32659ea1698b2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Operations fleet figures are assembled from the Quality-Assured Operations Monitoring Registry (QOMR), the register Uber maintains for 2015, and the four values below are given exactly as the register holds them.\n\nThe register puts the share of units reporting in service on every working day at 83 per cent, the mean hours a unit runs between services at 240, the share of inspections finding a defect at 11 per cent, and the mean days a unit is withdrawn for a defect at 3. None was recomputed here, and the register's own definitions of in service and of a defect appear with the values.\n\nA unit standing idle for want of a driver was not counted as out of service, and a note recorded without a follow-up was not counted as a defect. Where the register reports for one depot, that depot is named with the value rather than pooled.\n\nThe register table and line for each value, the extraction date, and the units behind each share are listed in the annex notes.\n\nMention: \"Quality-Assured Operations Monitoring Registry (QOMR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Quality-Assured Operations Monitoring Registry supplies availability, service interval and defect values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:7933e1b360bccc3e5729e423", "occurrence_id": "3515919fad06183a58d628e9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Teacher qualification and results were compared for this assessment using the Linked Non-State Education Survey, which PRIO/Uppsala conducted across Central Asia in 2022, and the comparison is the assessment's rather than the survey's.\n\nTeachers were grouped by whether they hold a teaching qualification alongside their trade degree and the share of their pupils reaching the end-of-year benchmark computed: 61 per cent of qualified teachers against 47 per cent of unqualified. The survey records the qualification item and pupil results for the same class and never publishes the crossing.\n\nClass size was then read from the same records and equalised, and the distance narrows to 6 points, which leaves class size as a substantial part of what the first reading credits to the qualification.\n\nTeachers with fewer than two terms in the class are excluded rather than counted at their pupils' result, and they are 9 per cent of teachers. The grouping rule and the counts behind each share sit with the annex tables.\n\nMention: \"Linked Non-State Education Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Linked Non-State Education Survey relates teacher qualification to pupil benchmark attainment.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:3511ace66be0bb963c29dd08", "occurrence_id": "03506a83c3f26ef5bee97b7d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two records lines in the provincial annex are quoted from the Provincial Integrated Records Registry (PIRR), Phase 2, the register maintained for Serbia in 2023, and both are reproduced without amendment.\n\nThe registry puts the share of records held electronically at a single location at 57 per cent, the median years a paper record is kept before it is destroyed at 10, and the share of offices able to produce a record within one working day at 49 per cent. Neither of the first two was recalculated here, and the registry's own rule for counting a record as held electronically, which requires it to be queryable rather than scanned, is stated with the figures.\n\nThe annex's other lines are compiled from the office's own retention schedule, so the registry carries two lines while the schedule carries the rest. A figure given for one record class is printed with that class named rather than averaged across the office.\n\nThe registry reference behind each line, the extraction date and the records behind each share are recorded with the annex notes.\n\nMention: \"Provincial Integrated Records Registry (PIRR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Provincial Integrated Records Registry supplies electronic holding, retention and retrieval lines.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:4b8fa383c0415b3bf13f5670", "occurrence_id": "7e59e53d2b5d5a8c6d71acb9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Whether the agency's own altitude figure for its sites holds was tested against the Rural Elevation Model Dataset (REMD), Wave 4, the geospatial dataset Latinobarómetro publishes, and the dataset is the principal source for the test.\n\nThe agency's site forms give a mean altitude of 1 240 metres across its health posts. The dataset gives 1 385 metres for the same posts, because it samples the surface at the coordinate the form records while the agency's figure is the altitude of the nearest settlement as printed in its directory. Neither figure was converted for the comparison.\n\nThe dataset therefore carries the section's altitude figure, and the agency's lower value is printed beside it so the reader can see that the distance is the difference between a settlement name and a recorded coordinate rather than a disagreement about the terrain. Where a post matches a dataset cell cleanly, the dataset's value is used and the two are not averaged.\n\nPosts whose form gives a coordinate outside the dataset's extent are excluded from the comparison rather than assigned the extent mean, 5 per cent of posts. The sampling rule each source applies is recorded with the annex tables.\n\nMention: \"Rural Elevation Model Dataset (REMD)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Rural Elevation Model Dataset is the principal source testing the agency's mean site altitude.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:73e8658b744cbf2d03aabbe3", "occurrence_id": "5ec675f9e6077ca0bd98d660", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Collective bargaining figures spanning the country are taken from the Quality-Assured Collective Bargaining Survey (QCBS), Edition 6, the survey conducted in Malawi in 2018, and the four values below are transferred here unchanged.\n\nThe survey puts the share of workers whose wages are set by an agreement they did not sign personally at 37 per cent, the mean years between renewals of a workplace agreement at 3, the share of agreements containing a grievance procedure at 64 per cent, and the share of workers reporting they know the terms of their agreement at 41 per cent. None was recomputed here, and the survey's own definitions of an agreement and of a renewal appear with the values.\n\nAn agreement signed by a federation but not registered was not counted, and a worker informed of terms verbally was not counted as knowing them. Where the survey reports for one industry, that industry is named with the value rather than pooled.\n\nThe survey table and line for each value, the extraction date as well as the workers behind each share are set down in the annex notes.\n\nMention: \"Quality-Assured Collective Bargaining Survey (QCBS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Quality-Assured Collective Bargaining Survey supplies coverage, renewal, clause and awareness values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:6b4a1cd803c9c978a4213156", "occurrence_id": "40902c41ea1cb5531d977e54", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two network lines in the transport annex are quoted from the High-Resolution Transport Network Report (HTNR), Second Version, compiled by CSA for 2020, and both are given without restatement.\n\nThe report puts the share of the network carrying a surfaced classification at 68 per cent and the mean gap between the design speed a segment carries and the speed observed at its busiest hour at 21 kilometres per hour. Neither was re-derived for this annex, and the report's own counting rule for a segment, which breaks at every change of classification rather than at every junction, appears with the figures.\n\nThe rest of the annex lines come from the authority's own maintenance log, so the report carries two lines while the log carries the rest. A figure given for one corridor is printed with that corridor named rather than averaged across the network.\n\nThe report's table behind each line, the extraction date and the segments behind each share are recorded with the annex notes.\n\nMention: \"High-Resolution Transport Network Report (HTNR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The High-Resolution Transport Network Report supplies surfaced share and design-versus-observed speed lines.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:69da6ae873470968e85d2f6e", "occurrence_id": "5cd6d8f87dcaa31488b1fa75", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Integrated records figures at the region level are taken from the Experimental Integrated Records Index (EIRI), the index compiled for Eastern Europe in 2015, and the four values below are given exactly as the index issues them.\n\nThe index puts the share of registers in its coverage linked to at least one other register at 44 per cent, the mean number of quarters a link has been maintained at 6, the share of links refreshed at least monthly at 29 per cent, and the share of linked registers whose matching rule is published at 35 per cent. None was recomputed here, and the index's own definition of a link, which requires a test record passing rather than a stated intention, appears with the values.\n\nA link operating in one direction only was still counted as a link, and a register awaiting legislation was not counted as linked. Where the index reports for one country, that country is named with the value rather than pooled.\n\nThe index table and line for each value, plus the extraction date and the registers behind each share are written up in the annex notes.\n\nMention: \"Experimental Integrated Records Index (EIRI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Experimental Integrated Records Index supplies linkage, duration, refresh and publication values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:bee0495da7010db01da63c4d", "occurrence_id": "ba564a05f56ff398bac139a4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Movement and health province-level figures come from the Inter-Agency Mixed Movement Registry, the register International Water Association maintains for Punjab as of 2023, and the four values below are given as the register holds them.\n\nThe register puts the number of mobile households listed in the period at 31 400, the share of listed households whose children completed a course of immunisation at 52 per cent, the mean number of service points a household used during the period at 3, and the share of listings made at a point rather than at a residence at 66 per cent. None was recomputed here, and the register's own definitions of a mobile household and of a completed course appear with the values.\n\nA household moving within its own village was not listed as mobile, and a dose given outside the schedule's window was not counted as completing a course. Where the register reports for one district, that district is named with the value rather than pooled.\n\nThe register table and line for each value, along with the extraction date and the households behind each share are carried in the annex notes.\n\nMention: \"Inter-Agency Mixed Movement Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Inter-Agency Mixed Movement Registry supplies mobile listings, immunisation completion and point use.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:f4e18e011bfbeab621ba6194", "occurrence_id": "b1cf676d2f5b00dd60234697", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The inventory count in this chapter is the office's own, taken from its survey register, and the Rapid Administrative Survey Registry (RASR), Round 3, compiled by ZamStats, is cited to test it.\n\nThe office's register lists 218 active surveys. The registry lists 196 for the same period, because it strikes a survey whose last fieldwork fell more than two years before the count while the office keeps it until formally closed. Neither rule was applied to the other's list.\n\nThe 22 distance is smaller than the number of surveys the office opens in a year, so the office's figure is printed as registered and the registry supplies no revision to it and no entries for the agencies it did not canvass.\n\nExcluded are surveys the registry records without a sponsoring agency, 11 of its entries, since neither count can be attributed for them. Which surveys each source counts as active is recorded with the annex tables.\n\nMention: \"Rapid Administrative Survey Registry (RASR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Rapid Administrative Survey Registry checks the office's count of active surveys in the inventory.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:58b98b11936c194b119d21bc", "occurrence_id": "e702192acf76add92dab34e8", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two cadastral lines in the urban annex are quoted from the Monthly Cadastral Frame Census (MCFC), Round 2, the census conducted in Mexico City in 2018, and both are reproduced as the census prints them.\n\nThe census puts the share of parcels whose recorded owner matches the property register at 89 per cent and the mean months since a parcel's last cadastral update at 26. Neither was recalculated here, and the census's own rule for counting an update, which requires a surveyor's visit rather than an administrative edit, is reproduced with the figures.\n\nThe annex's other lines are compiled from the tax office's billing file, so the census carries two lines while the billing file carries the rest. A figure given for one borough is printed with that borough named rather than averaged across the city.\n\nThe census table behind each line, the extraction date and the parcels behind each share are recorded with the annex notes.\n\nMention: \"Monthly Cadastral Frame Census (MCFC)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Monthly Cadastral Frame Census supplies owner-match rates and update intervals for the urban annex.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:862c8a4ed0d3b894fd6a0126", "occurrence_id": "4dad05f31df03ca64e586704", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Irrigation demand was worked out for this chapter from the Harmonised Water and Climate Report (HWCR), Version 8, compiled for Lebanon in 2021, and the figures below are counted by the chapter rather than by the report's.\n\nBasins were grouped by the change in their seasonal rainfall against the report's own thirty-year norm and the drawn volume per irrigated hectare computed: 4 900 cubic metres in basins below the norm against 3 600 in those above it. The report gives rainfall by basin and a drawn volume by basin and never relates the two.\n\nThat comparison supports one claim, that the abstraction permits should be trimmed in the dry basins before the canal schedule is rewritten. The canal lengths the plan uses come from the works department's inventory, from which the report supplied nothing.\n\nBasins the report lists with a rainfall figure but no volume are excluded rather than given the national mean, and they are 8 per cent of basins. The bands and the counts behind each figure are recorded under the table.\n\nMention: \"Harmonised Water and Climate Report (HWCR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Harmonised Water and Climate Report supports one permit claim by relating rainfall to drawn volume.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:d445354455af6cc0c01081c3", "occurrence_id": "8540880bc280f089a0aaf1c8", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The enrolment count in this chapter is the programme's own, taken from its payment roll, and the Small-Area Targeting and Payments Survey conducted by SEDESOL in the Republic of Congo in 2022 is cited to test it.\n\nThe programme's roll carries 62 300 enrolled households. The survey puts 57 800 households it could confirm as eligible for the same areas, because it interviews households directly and strikes those whose composition no longer meets the rule, a step the roll takes only at re-verification. Neither basis was applied to the other's count.\n\nThe 4 500 distance is smaller than the number of households the roll adds in a quarter, so the programme's figure is printed as enrolled and the survey supplies no revision to it and no households for the areas its teams did not reach.\n\nExcluded are households the survey met but could not match to the roll by name or code, 620 of them, since neither count can include them. Which households each source counts is recorded with the annex tables.\n\nMention: \"Small-Area Targeting and Payments Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Small-Area Targeting and Payments Survey checks the programme's enrolled household count against eligibility.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:f80f119b96335c4e1e80dea7", "occurrence_id": "92b26797fcf42a0bdaaf5ef7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Unemployment insurance figures are reported from the Joint Unemployment Insurance Report (JUIR), Sixth Version, the report CBS compiles, and the four values below are given exactly as the report issues them.\n\nThe report puts the share of unemployed persons receiving a benefit at 46 per cent, the mean weeks a benefit lasts at 21, the share of recipients who had been working outside their country of residence before registering at 12 per cent, and the share of recipient households falling below the national adequacy line while out of work at 19 per cent. None was recomputed here, and the report's own definitions of a recipient and of a prior foreign spell appear with the values.\n\nA person offered work and refusing it was not counted as receiving, and a spell of informal work abroad was not counted as prior employment. Where the report gives a figure for one insurance fund, that fund is named with the value rather than pooled.\n\nThe report's table and line for each value, the extraction date and the recipients behind each share are detailed in the annex notes.\n\nMention: \"Joint Unemployment Insurance Report (JUIR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Joint Unemployment Insurance Report supplies benefit receipt, duration, foreign work and poverty values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:bf4ba1f297bafa286d8bad84", "occurrence_id": "539c0f5b46a50ececad443b4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Income instability was measured for this review from the Pilot Living Standards Survey, which OECD followed in the Dominican Rep. in 2023, and the figures below are the review's rather than the survey's.\n\nHouseholds were grouped by the number of months in the year their income varied by more than a quarter and the share falling below the national adequacy line at any point computed: 44 per cent of households with four such months against 15 per cent of those with none. The survey reports each month's income and the adequacy items without publishing a volatility measure.\n\nWhether the household holds any savings was then read from the same records and equalised, and the distance narrows to 13 points, which leaves the absence of savings as a substantial part of what the first reading attributes to the variability itself.\n\nHouseholds surveyed in only one month are excluded rather than treated as stable, and they are 11 per cent of households. The grouping rule and the counts behind each share sit with the annex tables.\n\nMention: \"Pilot Living Standards Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Pilot Living Standards Survey relates monthly income variability to spells below the poverty line.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:fcd227c3ff158f7c7d38e865", "occurrence_id": "2bb7cec9a029b0166cac83a6", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Yield and heat were joined for this assessment from the Recurrent Temperature Records Collection, the collection DANE. maintains for Nigeria in 2019, and the figures below are the assessment's own.\n\nVillages were matched to their station record and grouped by the number of days above 35 degrees in the growing season, and the mean cereal yield per hectare computed: 1.1 tonnes in villages with more than 40 such days against 1.7 tonnes in those with fewer than 15. The collection holds a daily reading for each station and the survey holds a yield for each village, and neither reports the two together.\n\nWhether the village sits on a scheme was then read from the same records and equalised, and the distance narrows to 0.3 tonnes, which leaves heat as the larger part of the loss than the absence of irrigation, though the two are close.\n\nVillages more than 50 kilometres from a station are excluded rather than assigned the nearest reading, and they are 12 per cent of villages. The day bands and the counts behind each mean sit with the annex tables.\n\nMention: \"Recurrent Temperature Records Collection\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Recurrent Temperature Records Collection relates hot days in the season to cereal yield per hectare.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:607aee2d311e093a4c5edb28", "occurrence_id": "ea1a2483c0b205f4b851595f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Operations monitoring figures for each part of the agencies are drawn from the Triennial Operations Monitoring Index (TOMI), Edition 9, the index CNTS compiles for Jordan, and the four values below are given as the index issues them.\n\nThe index puts the share of agencies filing their operations return within the deadline at 64 per cent, the mean number of indicators each agency reports at 18, the share of reported indicators carrying a source note at 71 per cent, and the share of agencies whose figures were audited in the period at 26 per cent. None was recomputed here, and the index's own definitions of a filed return and of an audited agency appear with the values.\n\nA return filed on paper after the deadline was not counted as filed within it, and an internal review by the agency was not counted as an audit. Where the index reports for one agency group, that group is named with the value rather than pooled.\n\nThe index table and line for each value: the extraction date and the agencies behind each share are carried in the annex notes.\n\nMention: \"Triennial Operations Monitoring Index (TOMI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Triennial Operations Monitoring Index supplies filing timeliness, indicator counts, sourcing and audit values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:f0226764852dd46e34210246", "occurrence_id": "2bc56be0dff53c3f409d81a3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two network lines in the logistics annex are quoted from the Subnational Road Segment Dataset (SRSD), the geospatial dataset OCHA compiled for 2015, and both are carried over unchanged.\n\nThe dataset puts the share of its segments passable to a two-wheel-drive vehicle in the wet season at 58 per cent and the mean length of a segment connecting two health facilities at 14 kilometres. Neither was recalculated for this annex, and the dataset's own rule for recording a segment as passable, which requires a driver report rather than a surface class, is reproduced with them.\n\nThe other annex entries are assembled from the convoy office's own journey log, so the dataset carries two lines while the log carries the rest. A figure given for one province is printed with that province named rather than averaged across the country.\n\nThe dataset layer behind each line, the extraction date and the segments behind each share are recorded with the annex notes.\n\nMention: \"Subnational Road Segment Dataset (SRSD)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Subnational Road Segment Dataset supplies wet-season passability and facility-link length lines.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:87f0f5b0a40136c93e1bd0ab", "occurrence_id": "785f92cd5b19180254e63fde", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Sea level figures measured across the coast are read from the Coordinated Sea Level Collection (CSLC), Eighth Round, the collection OSIPTEL maintains for France, and the four values below are given exactly as the collection issues them.\n\nThe collection puts the mean sea level along its gauged coast at 0.42 metres above its own reference surface, the rate of rise recorded over its observation span at 3.1 millimetres a year, the share of gauges whose reading was corrected for land movement at 68 per cent, and the highest still-water level logged in the span at 1.9 metres. None was recomputed here, and the reference surface the collection uses is named with the values rather than converted.\n\nA gauge operating for fewer than ten years was not counted in the rate, and a level recorded during a storm surge was not counted as the annual high. Where the collection reports for one estuary, that estuary is named with the value rather than pooled.\n\nThe collection table and line for each value, the extraction date, and the gauges behind each share are kept in the annex notes.\n\nMention: \"Coordinated Sea Level Collection (CSLC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Coordinated Sea Level Collection supplies mean level, rise rate, correction coverage and peak values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:78011e067ae085cd084b4338", "occurrence_id": "103a7f7a7550a0e9a966aa9f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Whether the department's own count of inspected workshops holds was tested against the Municipal Labour Inspection Survey conducted by the US census in Pakistan in 2019, and the survey is the principal source for the test.\n\nThe department's inspection file lists 8 700 workshops visited in the year. The survey lists 11 200 for the same municipalities, because it asks every establishment whether it was visited while the file records only visits an inspector logged. Establishments that were visited and did not recall it are counted by neither.\n\nThe survey therefore carries the section's count of inspected workshops, and the department's lower figure is printed beside it so the reader can see that the distance is unlogged visits rather than a disagreement about what a visit is. Where a workshop appears in both, the department's findings are used and the two counts are not averaged.\n\nWorkshops with fewer than five workers are excluded from both rather than counted by the survey only, 22 per cent of its listings. The recall rule and the logging rule each source applies are recorded with the annex tables, beside the share of listed workshops running a training class for young workers.\n\nMention: \"Municipal Labour Inspection Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Municipal Labour Inspection Survey is the principal source testing the department's inspected workshop count.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:d51fef18ff5557f6f5b1a06f", "occurrence_id": "7fe034ce6a1536771baa5b6f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Child wellbeing figures for the country are reported from the Ministerial Child Wellbeing Microdata (MCWM), the microdata set compiled for Peru in 2022, and the four values below are given as the set records them.\n\nThe set puts the share of children living in a household below the national adequacy line at 27 per cent, the share of those children in a household with three or more children at 41 per cent, the share of school-age children not enrolled at 6 per cent, and the mean persons per sleeping room in a poor household at 2.4. None was recomputed here, and the set's own equivalence rule for the adequacy line, which scales by age rather than by headcount, appears with the values.\n\nA child counted at a household's dwelling rather than at a school was still counted at the household, and a room used only in the day was not counted as a sleeping room. Where the set reports for one region, that region is named with the value rather than pooled.\n\nThe set's table and variable behind each value, the extraction date and the children behind each share are held in the annex notes.\n\nMention: \"Ministerial Child Wellbeing Microdata (MCWM)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Ministerial Child Wellbeing Microdata supplies child poverty, household size, enrolment and crowding values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:f82f0aedbcb13fa98291c873", "occurrence_id": "2446506478973012b902460e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Retention through the first year was computed for this review from the Comparative Enrolment and Retention Survey (CERS), Collection 7, conducted in Tajikistan in 2024, and the figures below are the review's rather than the collection's.\n\nSchools were grouped by the pupil-teacher ratio they report and the share of first-year enrollees present in the second term computed: 92 per cent at schools below 30 pupils per teacher against 78 per cent at those above 45. The collection records the ratio and the term-by-term enrolment for the same school and publishes no relationship between them.\n\nWhether the school is more than an hour from the district centre was then read from the same records and equalised, and the distance narrows to 5 points, which leaves remoteness as a substantial part of what the first reading attributes to the ratio.\n\nSchools that opened during the year are excluded rather than counted as retaining everyone, and they are 4 per cent of schools. The bands and the counts behind each share sit with the annex tables.\n\nMention: \"Comparative Enrolment and Retention Survey (CERS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Comparative Enrolment and Retention Survey relates pupil-teacher ratio to first-year retention.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:9df00ea0e24856c63160af68", "occurrence_id": "8f46b7eda7484f22d4a36c93", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The household income figures in this chapter are computed from the office's own panel, and the Joint Shock Coping Survey conducted by World Bank-IFC in Malaysia in 2020 is referred to only as background to them.\n\nThe survey puts the share of households losing income in the reference year at 34 per cent, the share drawing on savings to cover the loss at 57 per cent, and the share of workers whose hours were cut rather than whose contract ended at 61 per cent. All three are given as the survey issues them, with its own reference period stated, and no chapter figure rests on them.\n\nThe chapter builds no table from its responses. The figures the chapter argues from come from the panel, which follows the same households before and after the loss rather than describing the year once, and where the two accounts of income loss differ the difference is noted rather than reconciled.\n\nThe annex lists the survey tables consulted and the date of extraction, and the survey is used for nothing beyond that background.\n\nMention: \"Joint Shock Coping Survey\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The Joint Shock Coping Survey is background while the chapter's income figures come from the office panel.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:ede880b3e397068f2d770322", "occurrence_id": "3a32af11343c1968632a0a63", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The poverty estimate in this chapter is the office's own, computed from its modelled household series, and the Rural Household Consumption Survey (RHCS), Phase 8, compiled by Planning Commission, GoI for Colombia, is cited to test it.\n\nThe office's model puts 31 per cent of rural households below the national adequacy line. The survey puts 36 per cent for the same areas, because it measures consumption from a two-week diary while the model imputes from asset ownership and household composition. Neither measure was applied to the other's data.\n\nThe 5-point distance is smaller than the movement the office's series shows between two consecutive years, so the office's figure is printed as modelled and the survey supplies no revision to it and no households for the departments its teams did not enter.\n\nExcluded are households the survey entered but whose diary covers fewer than ten days, 4 per cent of its sample, since a short diary cannot be compared with a full one. How each source measures welfare is recorded with the annex tables.\n\nMention: \"Rural Household Consumption Survey (RHCS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Rural Household Consumption Survey checks the office's modelled rural poverty estimate.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:a14edd9e05ac242a6871ddd6", "occurrence_id": "e6bf0daa8b837c779f04e026", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Earnings stability was measured for this assessment from the Annual Informal Trade Survey (AITS), Wave 5, compiled by Berne Union, and the figures below are the assessment's own.\n\nTraders were grouped by the number of days in the week they trade and the share whose weekly earnings vary by more than half their own mean computed: 22 per cent of traders working six or seven days against 49 per cent of those working three or fewer. The survey records days traded and each week's earnings without publishing any measure of variability.\n\nWhether the trader holds a fixed stall was then read from the same records and equalised, and the distance narrows to 12 points, which leaves the stall as a substantial part of what the first reading attributes to the number of days.\n\nTraders who report a second wage job are excluded rather than counted as stable, and they are 8 per cent of traders. The grouping rule and the counts behind each share sit with the annex tables.\n\nMention: \"Annual Informal Trade Survey (AITS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Annual Informal Trade Survey relates days traded per week to the variability of weekly earnings.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:bf402300f3874223252621d1", "occurrence_id": "e84f6654a21c00a7f19568ce", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Graduation programme figures for the countries covered are reported from the Decennial Graduation Programmes Survey (DGPS), Collection 4, compiled by World Bank/IFC for 2017, and the four values below are transferred here unchanged.\n\nThe survey puts the share of enrolled households holding a productive asset two years after enrolment at 71 per cent, the share of those households with a member in stable paid work at 34 per cent, the mean transfer value per enrolled household per year at 190, and the share of households still above the exit line three years after graduation at 62 per cent. None was recomputed here, and the survey's own definitions of a productive asset and of stable work appear with the values.\n\nAn asset acquired on credit and not yet paid for was still counted as held, and seasonal work of fewer than six months was not counted as stable. Where the survey reports for one programme, that programme is named with the value rather than pooled.\n\nThe survey table and line for each value, the extraction date as well as the households behind each share are laid out in the annex notes.\n\nMention: \"Decennial Graduation Programmes Survey (DGPS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Decennial Graduation Programmes Survey supplies asset, employment, transfer and graduation durability values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:faff2a7ed0cea8d1d8f5ad1b", "occurrence_id": "2569d2af08b114ca66b71d72", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two operations lines in the agency annex are quoted from the Integrated Operations Index Registry (IOIR), Eighth Phase, the register Fraser Institute maintains, and both are given without restatement.\n\nThe registry puts the share of units whose operational data reach a central platform at 63 per cent and the median days between collection and publication at 12. Neither was recalculated for this annex, and the registry's own rule for counting a unit as reaching the platform, which requires an automated feed rather than a filed file, is reproduced with them.\n\nThe annex's other lines are taken from the ministry's own circular returns, so the registry carries two lines while the returns carry the rest. A figure given for one unit class is printed with that class named rather than averaged across the service.\n\nThe registry reference behind each line, the extraction date and the units behind each share are recorded with the annex notes.\n\nMention: \"Integrated Operations Index Registry (IOIR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Integrated Operations Index Registry supplies platform-feed coverage and publication delay lines.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:43894bf707d4c5ef1977429c", "occurrence_id": "330243f2bd4ea60701f1f9db", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two register lines in the legal annex are quoted from the Government Service Register Survey (GSRS), Round 3, compiled by Berkowitz, Pistor and Richard for 2024, and both are carried over unchanged.\n\nThe survey puts the share of registers whose governing instrument is published online at 78 per cent and the mean number of agencies with a legal claim on the same register at 3. Neither was recomputed here, and the survey's own rule for counting a register, which requires it to record a legal status rather than an administrative note, appears with the figures.\n\nThe annex's other lines are compiled from the ministry's own statute list, so the survey carries two lines while the statute list carries the rest. A figure given for one register type is printed with that type named rather than averaged across registers.\n\nThe survey table behind each line, the extraction date and the registers behind each share are recorded with the annex notes.\n\nMention: \"Government Service Register Survey (GSRS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Government Service Register Survey supplies published-instrument and shared-claim lines.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:68f581178a49b8759e9fc4a5", "occurrence_id": "0855504b2c1571144dd4b5b2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Built-up surface figures for the metropolitan areas are reported from the Metropolitan Built-Up Surface Report (MBSR), the report the Ministry of Finance issues for South Africa, and the four values below are given exactly as the report prints them.\n\nThe report puts the built-up surface of the eight metropolitan areas at 4 120 square kilometres, the share of that surface added since its own baseline at 9 per cent, the share of new development falling inside a previously serviced area at 37 per cent, and the mean density of residents per built-up hectare at 24. None was recomputed here, and the report's own rule for counting built-up surface, which requires a permanent roof and a serviced parcel, appears with the values.\n\nA structure on land awaiting proclamation was not counted as built-up, and a caravan standing the year round was not counted as a dwelling. Where the report gives a figure for one metropolitan area, that area is named with the value rather than pooled.\n\nThe report's page and table behind each value, the extraction date and the areas behind each share are filed in the annex notes.\n\nMention: \"Metropolitan Built-Up Surface Report (MBSR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Metropolitan Built-Up Surface Report supplies surface totals, growth, infill share and density values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:5a98d0197f4cd219f5894631", "occurrence_id": "94393ed5ca501333262aa208", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The stock figure in this chapter is the programme's own, taken from its warehouse returns, and the Local Supply Chain System (LSCS), the register TURKSTAT maintains for Mali, is cited to test it.\n\nThe programme's returns put 118 days of cover for a first-line commodity across its depots. The system puts 96 days for the same depots, because it counts stock held at the point of use as well as the depot balance and divides by a mean drawn from actual distribution rather than from the plan. No basis was used on the other's data.\n\nThe 22-day distance is smaller than the movement the programme's own series shows between two quarters, so the programme's figure is printed as returned and the system supplies no revision to it and no depots for the regions its collection does not reach.\n\nExcluded are depots the system lists with a balance but no distribution record, 7 of its entries, since cover cannot be computed for them. Which stocks each source counts is recorded with the annex tables.\n\nMention: \"Local Supply Chain System (LSCS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Local Supply Chain System checks the programme's days-of-cover figure for a first-line commodity.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:6d13071e574e48d9a0cb00db", "occurrence_id": "ce6c224688fc05f2df7803be", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The employment figures in this chapter are computed from the fund's own registration file, and the Nationwide Poverty and Vulnerability Survey (NPVS), compiled by UNHCR for Ukraine, is quoted only to place those figures in context.\n\nThe survey puts the share of working-age adults in paid work at 54 per cent, the share of those whose work pays below the national adequacy line at 18 per cent, and the share reporting an interrupted work history of more than six months at 29 per cent. All three are given as the survey issues them, with its own definition of paid work stated, and no chapter figure was built from them.\n\nTheir use here is to show what the labour market looks like beside the caseload the chapter reports. The chapter's argument rests on the registration file, which records who registered rather than who works, and the two are kept apart rather than merged.\n\nThe annex lists the survey tables consulted and the date of extraction, which is the full extent of the survey's use in this chapter.\n\nMention: \"Nationwide Poverty and Vulnerability Survey (NPVS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Nationwide Poverty and Vulnerability Survey is quoted as context beside the fund's registration figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:0ba48e086e5d06f7a1e223a5", "occurrence_id": "348f3b130ee55df268d85dc9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Expenditure shares and poverty were compared for this assessment using the Coordinated Household Income and Expenditure Microdata (CHIEM), Fourth Phase, a microdata set covering Thailand in 2016, and the comparison is the assessment's rather than the set's.\n\nHouseholds were grouped by the share of their spending taken by food and the share below the national adequacy line computed: 58 per cent of households spending more than half their outlay on food against 9 per cent of those spending less than a third. The set publishes each spending line and the adequacy measure without crossing them.\n\nWhether the household lives in a municipal or a rural area was then read from the same records and equalised, and the distance narrows to 17 points, which leaves where the household lives as a substantial part of what the first reading attributes to the food share.\n\nHouseholds reporting no outlay in the reference period are excluded rather than counted at zero, and they are 2 per cent of households. The bands and the counts behind each share sit with the annex tables.\n\nMention: \"Coordinated Household Income and Expenditure Microdata (CHIEM)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Coordinated Household Income and Expenditure Microdata relates the food share of outlay to poverty.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:6b186d9e717ac61ba88717ad", "occurrence_id": "517d2857961ea5eed2e585a1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Targeting and payment the programme figures are compiled from the Stratified Targeting and Payments Survey (STPS), the survey conducted in Lebanon in 2020, and the four values below are given as the survey issues them.\n\nThe survey puts the share of enrolled households whose score was verified by an interview at 73 per cent, the share of payments reaching the household within the announced window at 66 per cent, the mean months between enrolment and first payment at 5, and the share of households reporting the score as fair to them at 58 per cent. None was recomputed here, and the survey's own definitions of a verified score and of an announced window appear with the values.\n\nA score verified only against a previous register was not counted as verified by interview, and a payment held at the bank for collection was not counted as reaching the household. Where the survey reports for one stratum, that stratum is named with the value rather than pooled.\n\nThe survey table and line for each value, plus the extraction date and the households behind each share are recorded in the annex notes.\n\nMention: \"Stratified Targeting and Payments Survey (STPS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Stratified Targeting and Payments Survey supplies verification, payment timeliness, delay and fairness values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:acd917805c2f6a842a27625b", "occurrence_id": "0a29543bbeb7d1e22b764b84", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two water lines in the climate annex are quoted from the Household Water and Climate Index (HWCI), the index Bureau van Dijk compiled for 2022, and both are reproduced without amendment.\n\nThe index puts the share of households whose supply fails for more than a day in a dry month at 43 per cent, the share drawing on a private tank or vendor when it fails at 31 per cent, and the mean additional litres a household reports buying in a failed month at 120. Neither of the first two was recalculated here, and the index's own rule for counting a failure, which requires the household to report no supply rather than low pressure, is stated with the figures.\n\nThe annex's other lines are compiled from the utility's own outage log, so the index carries two lines while the log carries the rest. A figure given for one water region is printed with that region named rather than averaged across the sample.\n\nThe index entry behind each line, the extraction date and the households behind each share are recorded with the annex notes.\n\nMention: \"Household Water and Climate Index (HWCI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Household Water and Climate Index supplies supply-failure, coping source and coping cost lines.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:a5c92caffe3b839a421e3149", "occurrence_id": "ebdb59a384ccb599f42a56b7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Boundary change was measured for this review from the Rolling Administrative Boundary Index (RABI), Version 8, the index the Federal Reserve maintains, and the figures below are the review's rather than the index's.\n\nUnits were grouped by the number of times their boundary was redrawn since the index's own baseline and the share whose population figures were republished after the change computed: 22 per cent of units redrawn once against 61 per cent of those redrawn three times or more. The index records each boundary change and each republication for the same unit and never reports the two together.\n\nWhether the unit sits above the median population for its level was then read from the same records and equalised, and the distance narrows to 13 points, which leaves size as a substantial part of what the first reading attributes to the number of changes.\n\nUnits whose boundary changed without a published instrument are excluded rather than counted as unchanged, and they are 5 per cent of units. The grouping rule and the counts behind each share sit with the annex tables.\n\nMention: \"Rolling Administrative Boundary Index (RABI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Rolling Administrative Boundary Index relates boundary redraws to republished population figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:f71b30e676eb47b8e130843b", "occurrence_id": "5265f64b7477d061605f4907", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Bulletin revision practice was built for this assessment from the Pilot Statistical Bulletin Registry (PSBR), Fifth Version, the register CEPAL compiles for 2023, and the readings below are the assessment's own.\n\nBulletins were grouped by the number of source registers behind them and the share revising a previously published figure within a year computed: 12 per cent of bulletins drawing on one register against 34 per cent of those drawing on four or more. The registry records the sources for each bulletin and a revision flag for each bulletin without publishing the crossing.\n\nWhether the bulletin carries a revision policy was then read from the same records and equalised, and the distance narrows to 9 points, which leaves the presence of a policy as a substantial part of what the first reading attributes to the number of sources.\n\nBulletins suspended rather than cancelled during the period are excluded rather than counted as revised, and they are 3 per cent of bulletins. The grouping rule and the counts behind each share are recorded under the table.\n\nMention: \"Pilot Statistical Bulletin Registry (PSBR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Pilot Statistical Bulletin Registry relates the number of source registers to later revisions.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:0c34c94b1f6945e8240b4923", "occurrence_id": "4696cffa150b82f5f38ae087", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Records completeness was assembled for this review out of the Biennial Integrated Records Index (BIRI), Fourth Version, the index compiled for Syria in 2018, and what follows is the review's arithmetic, not the index's.\n\nOffices were sorted by how many of their registers are held electronically and the share of client files that could be produced on request computed: 74 per cent among offices with most registers electronic against 39 per cent among those with hardly any. The index publishes a completeness score and a production rate for each office and never sets one against the other.\n\nMatching the two groups on staff per office brings the gap down to 14 points, which leaves staffing as part of the gap the opening contrast assigned to digitisation.\n\nOffices the index scores on fewer than five registers are dropped rather than pooled with their neighbours, 6 per cent of offices. Working notes list the sorting rule, the extraction day and the denominators each share rests on.\n\nMention: \"Biennial Integrated Records Index (BIRI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Biennial Integrated Records Index relates electronic registers to files a office can produce on request.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:3c3627a241b46b2293a0431b", "occurrence_id": "1b620bc64cf96cc6e29ae5f7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Livelihood figures for the country are drawn from the Experimental Rural Livelihoods Survey, which Supreme Court followed in the Republic of Zambia in 2017, and the four values are quoted here as printed.\n\nThe survey puts the share of rural households with any wage employment at 23 per cent, the mean months a household's head works away from home in the year at 4, the share of households whose main income is own-farm production at 55 per cent, and the share reporting wages in arrears at 17 per cent. No arithmetic was applied to them here, and the survey's own reading of wage employment and of arrears goes with the figures.\n\nA season of casual farm labour was not counted as wage employment, and produce eaten rather than sold was counted as own-farm income. Each figure keeps the province it was collected in attached, instead of being blended into a national average.\n\nUnder the table you will find the survey line for each value, along with the day of the extract and the households each share divides.\n\nMention: \"Experimental Rural Livelihoods Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Experimental Rural Livelihoods Survey supplies wage work, absence, farm dependence and arrears values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:cdd4255d55f54d41f3c39d96", "occurrence_id": "1f22c51c7bea874ff5660cf5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Whether the planning office's own figure for built-up land survives was tested against the Analytical Land Cover System (ALCS), Second Round, the register WITS maintains for Accra, and the register is the principal source for that test.\n\nThe office's development control file counts 118 square kilometres of built-up land. The register counts 137 for the same year, because it classes a yard with a roofed structure and a walled boundary as built while the office requires an approved plan before a parcel is entered. Neither reading was converted for the comparison.\n\nThe register therefore carries the section's figure for built-up land, and the office's smaller total is printed next to it so the reader can see the distance is parcels awaiting approval rather than a disagreement about what a building is. Where a parcel appears in both, the office's approval date is kept and the two totals are not averaged.\n\nParcels the register classes as bare rock are dropped from the comparison rather than counted as unbuilt by both, 3 per cent of parcels. The classing rule each source works to is laid out in the working papers behind the annex.\n\nMention: \"Analytical Land Cover System (ALCS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Analytical Land Cover System is the principal source testing the planning office's built-up area.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:567ae470eab9e989326fd8df", "occurrence_id": "2facbabe003038abef859d02", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Unpaid work and income adequacy were estimated for this assessment using the High-Frequency Time Use Survey (HTUS), Third Wave, the survey the US census follows in Germany, and the figures are the assessment's own.\n\nHouseholds were sorted by the hours per week the second adult spends on unpaid care and the share whose disposable income falls below the national adequacy line computed: 21 per cent where the second adult gives more than 20 hours a week against 8 per cent where they give fewer than 5. The survey publishes a care-hours item and an income item for the same household and never crosses them.\n\nHolding employment status even across the two groups leaves a gap of only 6 points, so the hours themselves, rather than whether the adult works at all, carry much of the difference.\n\nHouseholds with no adult of working age are dropped rather than counted as giving no care, and they are 18 per cent of households. A note in the annex gives the sorting rule, the pull date and, alongside it, the base for each share.\n\nMention: \"High-Frequency Time Use Survey (HTUS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The High-Frequency Time Use Survey relates unpaid care hours to households' income adequacy.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:640c2545d43ce1ebc0dea0fd", "occurrence_id": "caac5486eb5954e8f2400646", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Whether the agency's own count of settlements inside the health catchment holds was tested against the Representative Settlement Location Registry, the register MICT/DTO compiled for Brazil in 2015, and the register is the principal source for the test.\n\nThe agency's outreach list shows 412 settlements within the campaign's catchment. The register shows 356 for the same districts, because it places a settlement by its recorded coordinates and excludes those whose position falls outside the buffer, while the agency assigns a settlement to the nearest post whatever the distance. No buffer was redrawn for the comparison.\n\nThe register therefore carries the section's catchment count, and the agency's larger list is printed beside it so the reader can see that the distance is far-flung settlements rather than a disagreement about which settlements exist. Where a settlement falls inside the buffer in both, the agency's health figures are kept and the two counts are not averaged.\n\nSettlements the register lists without a name are dropped from the comparison rather than counted by the agency alone, 9 of them. The placement rule each source applies is set out in the annex working papers.\n\nMention: \"Representative Settlement Location Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Representative Settlement Location Registry tests the agency's count of settlements in the catchment.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:4e33dcb7f2b480f363fdfaa8", "occurrence_id": "d9297aa4ad652a0c51a4806b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two small-area lines in the statistics annex come from the Departmental Output Area Census (DOAC), Seventh Edition, the census compiled for South Asia in 2020, and both are transcribed without alteration.\n\nThe census puts the mean population of an output area at 1 850 residents and the share of areas whose boundary was last revised more than ten years before the count at 42 per cent. Nothing here was computed afresh for this report, and the census's own rule for closing an area, which requires it to fall below 500 residents, travels with the figures.\n\nThe annex's other entries are drawn from the office's own listing of localities, so the census carries two lines while the listing accounts for the remaining lines. A figure given for a single division stays with that division instead of being averaged over the region.\n\nThe annex log holds the census table, the date of the pull as well as the counts each share rests on.\n\nMention: \"Departmental Output Area Census (DOAC)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Departmental Output Area Census supplies mean area population and boundary revision age for the annex.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:82114ada7f785097b6d15bbb", "occurrence_id": "cf644650db4b0d0c8f90ea35", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Pupil sanitation and attendance were reconstructed for this review from the Rural Facilities and WASH Survey (RFWS), Second Phase, conducted across the CHT and Sylhet divisions in 2015, and the readings below are the review's, not the survey's.\n\nSchools were sorted by whether they operate a separate toilet for girls and the mean attendance on the last day of the term computed: 86 per cent at schools with one against 74 per cent at schools without. The survey records the facility item and the attendance count for the same school and never sets one against the other.\n\nMaking the two groups alike on distance to the nearest all-weather road brings the gap down to 5 points, which leaves distance as part of the distance the first comparison ascribed to the toilet.\n\nSchools that sat no examination that term are dropped rather than counted at zero attendance, and they are 6 per cent of schools. Those references, the extraction day and the denominators are listed in the method note in the annex.\n\nMention: \"Rural Facilities and WASH Survey (RFWS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Rural Facilities and WASH Survey relates girls' toilets to school attendance on the last term day.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:51af1ee40bc152328307cb25", "occurrence_id": "712461aaee5240810defaa6d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The coverage figure in this chapter is the office's own, taken from its enumeration return, and the Midline Administrative Survey System (MASS), Wave 3, the register kept for Viet Nam in 2022, is cited to test it.\n\nThe office's return puts 96 per cent of households listed in the last enumeration. The system puts 91 per cent for the same districts, because it counts a household as listed only where an interviewer's signature and a date appear together, and the office accepts a listing with either. Neither rule was applied to the other's returns.\n\nThe 5-point distance is smaller than the movement the office's own return shows between two waves, so the office's figure is printed as returned and the system supplies no revision to it and no districts for the provinces its wave did not reach.\n\nExcluded are returns the system holds without an interviewer identifier, 640 of them, since neither count can be checked for them. Which listings each source accepts is stored in the annex's method note.\n\nMention: \"Midline Administrative Survey System (MASS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Midline Administrative Survey System checks the office's share of households listed in the enumeration.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:2bffd9cd965ef3bcb1dff976", "occurrence_id": "7fe09a6d7ba8bb3b75aef34b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The request statistics in this chapter are compiled from the agency's own correspondence file, and the Clustered Public Records Index (CPRI), Fifth Collection, the index OPT Protection Cluster maintains, is quoted only to set those statistics in context.\n\nThe index puts the share of requests answered within the statutory period at 52 per cent, the mean days a refused request takes to reach a decision at 44, and the share of decisions carrying a reason at 63 per cent. All three are given as the index issues them, with its own scope of agencies stated, and nothing in the chapter is computed from them.\n\nTheir use here is to show how the wider service behaves beside the one agency whose file the chapter works from. The chapter's argument rests on that file, which records individual letters rather than a sample of agencies, and the two accounts are kept apart rather than merged.\n\nThe annex lists the index tables consulted and the date on the extract, and the index is used for nothing beyond that context.\n\nMention: \"Clustered Public Records Index (CPRI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Clustered Public Records Index is quoted as context beside the agency's own request statistics.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:1fdcdc39e31136e0dee607cf", "occurrence_id": "947863da1ff81d827cf6ca03", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Shift length and injury were joined for this assessment from the Sentinel Working Conditions Index (SWCI), Collection 4, the index FFMS compiles, and the figures below are the assessment's own rather than the index's.\n\nEstablishments were sorted by the mean weekly hours a shift worker works and the share recording an injury with lost time in the year computed: 9 per cent among establishments above 48 hours against 3 per cent among those below 40. The index holds an hours figure and an injury count for each establishment and never sets one against the other.\n\nHolding the size of the establishment even narrows the gap to 3 points, so hours carry the larger part of what the first comparison shows.\n\nEstablishments reporting no shift worker at all are dropped rather than counted as injury-free, and they are 12 per cent of establishments. A note in the annex gives the sorting rule, the pull date together with the base for each share.\n\nMention: \"Sentinel Working Conditions Index (SWCI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Sentinel Working Conditions Index relates shift hours to injuries causing lost time.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:c4cdd57679e1265811b468b8", "occurrence_id": "b39a7e43c2cdfcd942a1c84b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The trader licensing figure in this chapter is the council's own, taken from its permit ledger, and the Regional Informal Trade Survey (RITS), Round 5, conducted by ICAO for 2022, is cited to test it.\n\nThe council's ledger shows 9 200 traders holding a current permit. The survey shows 7 400 traders whose permit the trader could produce at interview, because a permit held at home or expired without renewal still stands in the ledger until the next sweep. The bases stay separate: neither is applied to the other's count.\n\nThe 1 800 distance is smaller than the number of permits the council issues in a quarter, so the council's figure is printed as issued and the survey supplies no revision to it and no traders for the markets its teams did not enter.\n\nExcluded are traders the survey met at two markets in the same week, 210 of them, since the ledger cannot show which entry is theirs. Which permits each source counts is detailed in the annex's method sheet.\n\nMention: \"Regional Informal Trade Survey (RITS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Regional Informal Trade Survey checks the council's ledger of traders holding a current permit.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:16520879e1c3d11ace985b91", "occurrence_id": "94e47290f9fa9a26c69a14f2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two statelessness lines in the protection annex come from the Civil Statelessness Index (CSI), Sixth Collection, the index the Central Bank of Peru compiles, and both are transcribed without alteration.\n\nThe index puts the share of recorded persons in its coverage holding no recognised identity document at 14 per cent and the mean years such a person has lived without one at 9. Nothing here was recomputed for this report, and the index's own rule for counting a person as recorded, which requires an interview rather than a site count, travels with the figures.\n\nThe remaining annex lines come from the agency's own registration sheets, so the index carries two lines while the sheets carry the balance. A figure reported for one host region keeps that region attached rather than being averaged over the programme.\n\nThe annex log holds the index entry, the day of the pull plus the counts each share rests on.\n\nMention: \"Civil Statelessness Index (CSI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Civil Statelessness Index supplies undocumented share and years-undocumented lines for the annex.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:a9813e1a745ebafc5709a1c7", "occurrence_id": "5c0b271fda60aa80baabc9a2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Coping after a storm was estimated for this review using the Rapid Shock Coping Collection, the collection WB/Makerere/FASID ran in Papua New Guinea in 2017, and the figures are the review's own.\n\nHouseholds were sorted by the value of the loss they reported as a multiple of monthly income and the share reducing meals for adults so children ate first computed: 52 per cent among households losing more than three months' income against 17 per cent among those losing less than half a month's. The collection holds a loss estimate and each coping item for the same household and never crosses them.\n\nMaking the groups alike on whether they received any external help brings the gap down to 19 points, which leaves outside help as part of what the raw comparison laid at the door of the size of the loss.\n\nHouseholds that reported a loss but could not date it are dropped rather than placed in the nearest band, and they are 7 per cent of households. Working notes list the sorting rule, the extraction day, and the denominators each share rests on.\n\nMention: \"Rapid Shock Coping Collection\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Rapid Shock Coping Collection relates loss size to adults cutting meals for children.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:dbea3ad190b0c9409d632afd", "occurrence_id": "6d63e4f29c98e138aeb97bf0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Cadastral figures for the whole municipality are sourced from the Municipal Cadastral Frame Dataset (MCFD), the geospatial dataset compiled for the Philippines in 2018, and the four values are carried here as tabulated.\n\nThe dataset puts the share of parcels carrying a surveyed boundary at 58 per cent, the share whose recorded owner agrees with the tax declaration at 72 per cent, the mean parcel area in the built-up frame at 410 square metres, and the share of parcels with no structure recorded at 26 per cent. No reweighting was applied to them, and the dataset's own reading of a surveyed boundary, which requires a coordinate traverse rather than a sketch, travels with the values.\n\nA boundary drawn from an older plan was not counted as surveyed, and a structure under framing was still counted as a structure. Each figure keeps the barangay it was drawn from attached, instead of being blended into a municipal average.\n\nThe annex keeps the dataset layer for each value, the extraction day, with the parcels each share divides alongside.\n\nMention: \"Municipal Cadastral Frame Dataset (MCFD)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Municipal Cadastral Frame Dataset supplies surveyed boundaries, owner agreement, parcel size and vacancy.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:18974d50d2fa3be3dae023c7", "occurrence_id": "28ab53d4568f539e86943871", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The staffing figure in this chapter is the ministry's own, taken from its establishment return, and the Household Public Sector Monitoring Registry (HPSMR), Third Edition, the registry Cingranelli and Richards compile, is cited to test it.\n\nThe ministry's return puts 12 400 filled posts across its offices. The registry puts 11 600 for the same offices, because it counts a post as filled only where the salary was drawn for the whole of the reference quarter, and the return counts a post filled from the first day of pay. Neither rule was applied to the other's return.\n\nThe 800 distance is smaller than the turnover the ministry's return shows between two quarters, so the ministry's figure is printed as returned and the registry supplies no revision to it and no posts for the departments it did not canvass.\n\nExcluded are posts the registry lists without an office code, 240 of them, since neither count can be placed for them. Which posts each source counts is listed in the annex notes.\n\nMention: \"Household Public Sector Monitoring Registry (HPSMR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Household Public Sector Monitoring Registry checks the ministry's count of filled posts.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:c3d695675b71cf43400103ff", "occurrence_id": "e207f24f9f46703ad3b8bab9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Record linkage was assembled for this review out of the Endline Integrated Records Index (EIRI), the index the Ministry of Investment compiled for 2024, and what follows is the review's arithmetic, not the index's.\n\nRegisters were sorted by how often they are refreshed and the share of records that match a second register on the index's own test computed: 81 per cent among registers refreshed monthly against 46 per cent among those refreshed annually or less often. The index publishes a refresh interval and a match rate for each register and never sets one against the other.\n\nThe difference falls to 12 points when the groups are matched on the size of the register, which leaves register size as part of the gap the opening comparison put down to refresh frequency.\n\nRegisters the index holds without a refresh interval are dropped rather than placed in the slowest band, 8 per cent of registers. Those references, the extraction day and the denominators are set down in the annex file.\n\nMention: \"Endline Integrated Records Index (EIRI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Endline Integrated Records Index relates refresh frequency to the share of records matching elsewhere.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:d483c110b5768ada7386beb0", "occurrence_id": "f95d62f9a9175dfcadd9410d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Request handling was reconstructed for this assessment from the Rapid Public Records Survey (RPRS), Collection 9, the survey conducted in Peru in 2015, and the readings below are the assessment's own.\n\nAgencies were sorted by the number of request channels they offer and the median days before a request is answered computed: 19 days among agencies offering a single channel against 31 days among those offering three or more. The survey records the channels for each agency and a handling time for each agency and never sets one against the other.\n\nHolding the size of the agency's caseload even narrows the gap to 4 days, so the number of channels carries less of the delay than the volume of requests an agency receives.\n\nAgencies that answered no request in the reference period are dropped rather than counted at zero days, and they are 5 per cent of agencies. A note in the annex gives the sorting rule, the pull date and the base for each share.\n\nMention: \"Rapid Public Records Survey (RPRS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Rapid Public Records Survey relates the number of request channels to how long answers take.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:51e0265972c0f9aafab05a69", "occurrence_id": "643cbd692b4df9885b235cbb", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Hiring and vacancy figures for the programme areas are reported from the Integrated Hiring and Vacancies System (IHVS), Second Wave, the register IEA maintains, and the four values pass into this report as printed.\n\nThe register puts the share of establishments with at least one vacancy open at the time of the wave at 21 per cent, the mean days a vacancy stands open at 47, the share of vacancies advertised only by word of mouth at 34 per cent, and the share of filled posts where the employer hired from outside at 58 per cent. Nothing was re-estimated here, and the register's own reading of an open vacancy and of an external hire follows the figures.\n\nA post frozen by the employer was not counted as open, and a promotion into the post was not counted as an external hire. Each figure keeps the area it was collected in attached, instead of being blended into a national average.\n\nBeneath the table sit the register line for each value: the extraction date and the establishments each share divides.\n\nMention: \"Integrated Hiring and Vacancies System (IHVS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Integrated Hiring and Vacancies System supplies vacancy incidence, duration, advertising and hiring source.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:eb4de08e7d7f151e74257f37", "occurrence_id": "e34760da464a8e242882321a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Return and relocation the province figures are tabulated by the Monthly Return and Relocation Registry (MRRR), the register kept for Vietnam as of 2018, and the four values come into this report unchanged.\n\nThe register puts the number of households listed as displaced in the month at 6 800, the share expressing a firm intention to return within six months at 44 per cent, the mean kilometres between a household's former dwelling and its present one at 31, and the share whose former dwelling sits in a zone the register marks as unsafe at 27 per cent. No figure was recalculated in this report, and the register's own reading of a firm intention and of an unsafe zone is reported with them.\n\nA household that had already returned once was not counted as intending to return, and a zone under a standing advisory was counted as unsafe whatever its damage grade. Each figure keeps the district it was collected in attached, instead of being blended into a provincial average.\n\nWorking notes record the register line for each value, the day the table was pulled, and the households each share divides.\n\nMention: \"Monthly Return and Relocation Registry (MRRR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Monthly Return and Relocation Registry supplies displaced counts, return intent, distance and risk zone shares.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:5599b79f0cfccc90e7b5cca1", "occurrence_id": "b8b9f2c20efcdc43ece6e1c0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Administrative records the figures for the province are reported by the Standardised Administrative Records System, the register Booking.com maintains for Xinjiang province as of 2017, and the four values are transcribed without alteration.\n\nThe register puts the share of registers held in a single system rather than on paper at 69 per cent, the mean years since a register's fields were last reviewed at 3, the share of registers with a named custodian at 81 per cent, and the share whose contents were checked against a second source in the year at 22 per cent. Nothing here was recomputed here, and the register's own reading of a review and of a check is attached to them.\n\nA review confined to a change of format was not counted, and a check carried out by the custodian's own office was not counted as against a second source. Each figure keeps the register class it was drawn from with it, rather than pooled across the province.\n\nThe annex log holds the register line for each value, the day of the pull as well as the counts each share rests on.\n\nMention: \"Standardised Administrative Records System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Standardised Administrative Records System supplies digitisation, review age, custody and checking values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:f0b19b5af42f0513edfe9694", "occurrence_id": "78cf9f5575412b4725937ebe", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Wages on public works were estimated for this review using the Government Public Works Survey (GPWS), the survey conducted in Rio de Janeiro in 2024, and the figures are the review's own.\n\nWorks were sorted by the number of days a worker is engaged for and the hourly wage the worker reports computed: 14 for engagements of fewer than 20 days against 19 for engagements longer than three months. The survey records the length of engagement and the wage for the same worker and never sets one against the other.\n\nMaking the groups alike on the trade the worker practises brings the gap down to 2, which leaves the trade as part of the distance the first comparison charged to the length of the engagement.\n\nWorkers engaged through a subcontractor rather than the works office are dropped rather than pooled with direct engagements, and they are 9 per cent of workers. Working notes list the sorting rule, the extraction day and, alongside it, the denominators each share rests on.\n\nMention: \"Government Public Works Survey (GPWS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Government Public Works Survey relates engagement length to the hourly wage on public works.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:3d699427e22f8f555d5ab4d9", "occurrence_id": "24ec23e614f592b2b64fe122", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Referral figures for the health network are reported from the Sentinel Referral Pathways Registry (SRPR), Version 8, the database FFMS maintains for Cyprus, and the four values enter this report as they stand.\n\nThe registry puts the share of primary facilities with a written agreement naming a receiving hospital at 63 per cent, the mean days a referred patient waits for an appointment at 12, the share of referrals returned to the sender for want of a document at 14 per cent, and the share of receiving hospitals reporting the referral note as complete at 71 per cent. They are carried as published, and the registry's own reading of a written agreement and of a complete note is stated with them.\n\nAn arrangement resting on custom between two doctors was not counted as an agreement, and a note missing only the patient's identifier was counted as incomplete. Each figure keeps the health region it was drawn from attached, instead of being blended across the island.\n\nThe documentation under the table gives the registry line for each value, plus the day the rows were taken and the facilities each share divides.\n\nMention: \"Sentinel Referral Pathways Registry (SRPR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Sentinel Referral Pathways Registry supplies agreement coverage, waits, returns and note completeness.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:d7967785922ba23d3ef2a8d4", "occurrence_id": "ed4cbb2273dd65d3aa81023e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Exposure and illness were joined for this assessment from the Enumerative Disaster Exposure Dataset, the geospatial dataset the World Food Programme compiled for Sudan in 2023, and the readings below are the assessment's own.\n\nLocalities were matched to their flood class and grouped by the depth the dataset records for the last event, and the share of households reporting acute watery diarrhoea in the following month computed: 21 per cent of localities in the deepest class against 6 per cent of those in the shallowest. The dataset grades depth per locality and the health survey records the illness per household, and neither reports the two together.\n\nHolding the density of households per compound even narrows the gap to 9 points, so crowding carries part of what the unadjusted contrast attributed to the depth of the water.\n\nLocalities the dataset classes as exposed but which the health team did not reach are dropped rather than counted as clear, and they are 7 per cent of localities. A note in the annex gives the classing rule, the pull date as well as the base for each share.\n\nMention: \"Enumerative Disaster Exposure Dataset\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Enumerative Disaster Exposure Dataset relates flood depth to reported acute watery diarrhoea.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:5d23efddb6900063a7cfcb14", "occurrence_id": "c760d7e2335d0186bf64ae24", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The register statistics in this chapter are compiled from the office's own inventory, and the Harmonised Integrated Records Registry (HIRR), Sixth Collection, the register GTAP maintains, is referred to only as background to them.\n\nThe registry puts the share of registers in its coverage with a published data dictionary at 47 per cent, the mean number of other registers a register draws on at 2, and the share whose updating is governed by a written rule at 39 per cent. All three are given as the registry issues them, with its own scope of registers stated, and no chapter figure rests on them.\n\nNo chapter table is built from its entries. The figures the chapter argues from come from the inventory, which lists one administration's registers rather than a comparable panel, and where the two accounts of documentation differ the difference is noted rather than reconciled.\n\nThe annex lists the registry tables consulted and the day the extract ran, and the registry is used for nothing beyond that background.\n\nMention: \"Harmonised Integrated Records Registry (HIRR)\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The Harmonised Integrated Records Registry is background to the office's own register inventory figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:1505f063fe872c2b72c11de3", "occurrence_id": "6489455c58529ebfb5c4a8d7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Hiring and vacancy figures covering the state are drawn from the Coordinated Hiring and Vacancies Survey (CHVS), the survey conducted in Bihar in 2016, and the four values reach this report as published.\n\nThe survey puts the share of establishments seeking a worker in the last quarter at 18 per cent, the mean weeks a vacancy stands unfilled at 8, the share of employers willing to train an inexperienced hire at 52 per cent, and the share of hires found through a relative or neighbour at 61 per cent. Nothing here was recomputed, and the survey's own reading of a vacancy and of a hire comes with them.\n\nA post frozen after the search began was not counted as a vacancy, and an apprentice taken on trial was counted as a hire. Each figure keeps the district it was collected in shown with it, instead of combined into one state figure.\n\nThe survey line for each value, along with the pull date and the establishments each share divides sit with the annex tables.\n\nMention: \"Coordinated Hiring and Vacancies Survey (CHVS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Coordinated Hiring and Vacancies Survey supplies search incidence, vacancy length, training and sourcing.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:a018d3be0800c8b9d1586dbf", "occurrence_id": "61fd853b5279ad7de938d7ef", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Flood risk and response were assembled for this review out of the Regional Climate and Flood Risk Registry (RCFRR), the register compiled for the Republic of Congo in 2020, and what follows is the review's arithmetic, not the register's.\n\nSectors were sorted by the return period the register assigns to their flood hazard and the share with a drainage plan adopted by the authority computed: 31 per cent of sectors inside the 25-year hazard against 9 per cent of those outside it. The register publishes a hazard class and the authority publishes a plan flag for each sector, and neither reports the two together.\n\nMaking the groups alike on population brings the gap down to 11 points, which leaves population as part of what the unadjusted difference owed to the hazard.\n\nSectors the register classes without a completed survey are dropped rather than placed in the nearest class, 6 per cent of sectors. Those references, the extraction day and the denominators are held in the papers behind the annex.\n\nMention: \"Regional Climate and Flood Risk Registry (RCFRR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Regional Climate and Flood Risk Registry relates flood hazard class to adopted drainage plans.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:7f7a2aeae7309efbc807fa28", "occurrence_id": "ba632ce69cb29c9018ea486d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Graduation from the programme was reconstructed for this review from the Stratified Graduation Programmes Survey, which the Chinese Wind Energy Association followed in the Tigray Region in 2020, and the readings below are the review's, not the survey's.\n\nHouseholds were sorted by the number of programme components they received and the share clearing the exit income test at the follow-up computed: 58 per cent of households receiving four components or more against 26 per cent of those receiving two or fewer. The survey records each component and the follow-up result for the same household and never sets one against the other.\n\nHolding whether the household holds livestock even narrows the gap to 14 points, so the asset carries part of what the first comparison credits to the number of components.\n\nHouseholds displaced during the programme period are dropped rather than counted as failing the test, and they are 12 per cent of households. A note in the annex gives the sorting rule, the pull date together with the base for each share.\n\nMention: \"Stratified Graduation Programmes Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Stratified Graduation Programmes Survey relates programme components received to clearing the exit test.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:c7a96167b7f701609dca5020", "occurrence_id": "4aa5b542461e7cd8a9a7bc0d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two monitoring lines in the delivery annex come from the Pilot Operations Monitoring Registry, the register the National Statistics Office maintains for Colombia as of 2020, and both are transcribed without alteration.\n\nThe register puts the share of projects reporting a physical progress figure on time at 61 per cent and the mean months between a project's planned completion and its reported completion at 7. Nothing here was recalculated for this report, and the register's own reading of a reported completion, which requires an acceptance certificate rather than a site note, travels with the figures.\n\nThe annex's remaining lines are compiled from the fund's own disbursement ledger, so the register carries two lines while the ledger carries the rest. A figure reported for one department keeps that department attached rather than being averaged over the country.\n\nThe annex log holds the register entry, the day of extraction plus the counts each share rests on.\n\nMention: \"Pilot Operations Monitoring Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Pilot Operations Monitoring Registry supplies progress-reporting timeliness and completion delay lines.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:a05fac447b25492d9f37d869", "occurrence_id": "9376928e266fae6c1f101423", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Enterprise growth was estimated for this review using the Experimental Enterprise and Jobs Registry, the register the Bank of Italy compiled for India in 2023, and the figures are the review's own.\n\nEnterprises were sorted by the year they registered and the share adding a worker in the following two years computed: 44 per cent of enterprises registering in the first year of the window against 29 per cent of those registering in the last. The register holds a registration date and a headcount for each enterprise and never sets one against the other.\n\nMaking the groups alike on the sector the enterprise works in brings the gap down to 9 points, which leaves the sector as part of the gap the opening contrast assigned to the timing of registration.\n\nEnterprises that changed proprietor within the window are dropped rather than followed under the new proprietor, and they are 5 per cent of enterprises. Working notes list the sorting rule, the extraction day, and the denominators each share rests on.\n\nMention: \"Experimental Enterprise and Jobs Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Experimental Enterprise and Jobs Registry relates registration year to whether an enterprise adds staff.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:70ae0600829b6a58d5522b3d", "occurrence_id": "9e31044c13789ae2e081db7a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Unmet need and waiting were reconstructed for this assessment from the Monthly Health and Household Survey (MHHS), Sixth Round, conducted in Croatia in 2022, and the readings below are the assessment's own.\n\nHouseholds were sorted by the months a patient waited for a specialist appointment and the share reporting they abandoned the appointment altogether computed: 24 per cent among those waiting more than four months against 6 per cent among those waiting under one. The survey records the wait and the abandonment item for the same patient and never sets one against the other.\n\nHolding the presence of private insurance even narrows the gap to 9 points, so insurance carries part of the distance the first comparison ascribed to the length of the wait.\n\nPatients whose appointment was cancelled by the clinic are dropped rather than counted as abandoned, and they are 4 per cent of patients. A note in the annex gives the sorting rule, the pull date, with the base for each share alongside.\n\nMention: \"Monthly Health and Household Survey (MHHS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Monthly Health and Household Survey relates specialist waiting time to abandoned appointments.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:27b442d6efbba78b839dd8e3", "occurrence_id": "903867d37476eb250bd993f0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The consumption figures in this chapter are computed from the office's own panel, and the Household Shock Coping Survey (HSCS), Edition 9, the survey ADCS follows in the Philippines, is quoted only to place those figures in context.\n\nThe survey puts the share of households reporting a shock in the previous twelve months at 41 per cent, the share selling an asset to absorb it at 19 per cent, and the share whose consumption fell after the shock at 27 per cent. All three are given as the survey issues them, with its own list of shocks stated, and nothing in the chapter is computed from them.\n\nTheir use here is to show what a shock does to a household beside the path the chapter's panel follows. The chapter's argument rests on the panel, which measures the same households month after month rather than once after the event, and the two accounts are kept apart rather than merged.\n\nThe annex lists the survey tables consulted and the date it was drawn, which is the full extent of the survey's use in this chapter.\n\nMention: \"Household Shock Coping Survey (HSCS)\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The Household Shock Coping Survey is quoted as context beside the office's consumption panel.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:333f9ff31d07da496ecc5058", "occurrence_id": "4403d2edadc2b3508c34c457", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Whether the ministry's own count of inspected establishments holds was tested against the Rolling Labour Inspection Survey, which Growth Lab at Harvard conducted in Colombia in 2022, and the survey is the principal source for the test.\n\nThe ministry's case file lists 22 600 establishments inspected in the year. The survey lists 27 900 for the same regions, because it asks each establishment whether an inspector attended while the case file counts only inspections that produced a written notice. Establishments visited and warned verbally appear in neither.\n\nThe survey therefore carries the section's count of inspected establishments, and the ministry's smaller figure is printed beside it so the reader can see that the distance is notices not written rather than a disagreement about what an inspection is. Where an establishment appears in both, the ministry's finding is kept and the two counts are not averaged.\n\nEstablishments with fewer than ten workers are dropped from both rather than counted by the survey alone, 18 per cent of its listings. The reporting rule each source works to is documented in the working papers behind the annex.\n\nMention: \"Rolling Labour Inspection Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Rolling Labour Inspection Survey tests the ministry's count of inspected establishments.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:b32e8fea9a6c42970b0cba42", "occurrence_id": "546b4ac6360600e8e0935d82", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Whether the ministry's own figure for schools with digital materials was tested against the Statistical Digital Learning System (SDLS), Release 6, the register UBOS maintains for Albania, and the register is the principal source for that test.\n\nThe ministry's returns give 76 per cent of schools as equipped with digital teaching materials. The register gives 62 per cent for the same schools, because it requires the equipment to have been used in a lesson within the term while the ministry counts delivery to the school. No usage test was applied to the ministry's returns for the comparison.\n\nThe register therefore carries the section's figure for equipment actually in use, and the ministry's larger share is printed beside it so the reader can see that the distance is delivered but unused equipment rather than a disagreement about what a school owns. Where a school appears in both, the ministry's inventory list is kept and the two shares are not averaged.\n\nSchools closed for renovation through the whole term are dropped from both rather than counted as unequipped by one only, 3 per cent of schools. The test each source applies is written up in the annex working papers.\n\nMention: \"Statistical Digital Learning System (SDLS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Statistical Digital Learning System tests the ministry's share of schools using digital materials.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:894c5e6c4496019fe80b5587", "occurrence_id": "1baf9853a6a4565e4eac5f4b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Skills and employment figures covering the whole province derive from the Household Skills and Employment Survey (HSES), the survey INEP follows in Formosa, and the four values appear here exactly as printed.\n\nThe survey puts the share of adults reporting a completed secondary education at 57 per cent, the share of workers in an activity with no written contract at 38 per cent, the mean monthly household income at 1 950, and the share of households with at least one member working outside the province at 16 per cent. They are reproduced as tabulated, and the survey's own reading of an activity and of a contract is kept with them.\n\nA person working in the family shop for food rather than pay was not counted as under contract, and income earned by a member living elsewhere was counted for the household. Each figure keeps the department it was collected in reported with it, rather than blended into a single province share.\n\nUnder the table are the survey line for each value, the date the file was taken and the households each share divides.\n\nMention: \"Household Skills and Employment Survey (HSES)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Household Skills and Employment Survey supplies schooling, contract, income and migration values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:1714da18903c92871913342b", "occurrence_id": "e2a7b5d4cc85218acf93d711", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Clinic staffing was assembled for this review out of the Quarterly Health Workforce Survey (QHWS), Collection 3, conducted in Nairobi County in 2016, and what follows is the review's arithmetic, not the survey's.\n\nFacilities were sorted by the number of clinical staff rostered and the share of sessions actually held in the quarter computed: 91 per cent at facilities rostering four staff or more against 64 per cent at those rostering one or two. The survey records a roster and a session count for the same facility and never sets one against the other.\n\nMaking the groups alike on the distance from the county road brings the gap down to 12 points, which leaves remoteness as part of what the raw comparison laid at the door of the roster.\n\nFacilities that closed for part of the quarter are dropped rather than counted as holding no sessions, and they are 6 per cent of facilities. Those references, the extraction day and the denominators are kept in the method note in the annex.\n\nMention: \"Quarterly Health Workforce Survey (QHWS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Quarterly Health Workforce Survey relates rostered clinical staff to sessions actually held.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:7e56fbd9fd626a78c543293f", "occurrence_id": "ed50d91b50185077132c600f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The unemployment figure in this chapter is the office's own, computed from its labour force return, and the Stratified Skills and Employment Index (SSEI), Eighth Edition, the index WB/Makerere/FASID compile, is cited to test it.\n\nThe office's return puts 7.4 per cent of the labour force as unemployed. The index puts 9.1 per cent for the same period, because it counts a person seeking work who is not available to start within a week, whom the office's return classes as economically inactive. Neither definition was applied to the other's data.\n\nThe 1.7-point distance is smaller than the movement the office's return shows between two quarters, so the office's figure is printed as returned and the index supplies no revision to it and no persons for the strata it did not sample.\n\nExcluded are the strata the index reports on a different reference year, 4 of its 21, since the two cannot be aligned for this chapter. Which persons each source counts as unemployed is noted in the annex's method note.\n\nMention: \"Stratified Skills and Employment Index (SSEI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Stratified Skills and Employment Index checks the office's unemployment rate for the same period.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:082aa5ea665535db812cd79e", "occurrence_id": "070354a375254657c2f5bd98", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Service workloads were estimated for this review using the Urban Regional Operations Survey, the survey Reinhart & Rogoff ran in Buenos Aires in 2022, and the figures are the review's own.\n\nTeams were sorted by the number of cases carried per worker and the share of cases closed inside the standard the region sets computed: 68 per cent at teams under 15 cases per worker against 41 per cent at teams above 30. The survey records a caseload and a closure flag for the same team and never sets one against the other.\n\nMaking the groups alike on the years of service a team averages brings the gap down to 9 points, which leaves experience as part of the gap the opening comparison put down to the caseload.\n\nTeams formed after the quarter began are dropped rather than counted at the regional average, and they are 5 per cent of teams. Working notes list the sorting rule, the extraction day and the denominators each share rests on.\n\nMention: \"Urban Regional Operations Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Urban Regional Operations Survey relates cases per worker to closures inside the regional standard.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:9b0e27624ad2154d588679b3", "occurrence_id": "790b63be30b9aa0a0fca8a6c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The staffing figures in this chapter come from the authority's own roster, and the Longitudinal Health Workforce Index (LHWI), Sixth Phase, the index OCHA compiled for 2023, is quoted only to set those figures in context.\n\nThe index puts the share of health posts in its coverage with a resident clinician at 58 per cent, the mean months a vacant post stands empty at 9, and the share of staff reporting they worked through a heat event without revised hours at 44 per cent. All three are given as the index issues them, with its own coverage of posts stated, and no chapter figure rests on them.\n\nTheir use here is to show what the wider workforce looks like beside the establishments the chapter tabulates. The chapter's argument rests on the roster, which names each post rather than sampling posts, and the two accounts are kept apart rather than merged.\n\nThe annex lists the index tables consulted and the day of the extract, and the index is used for nothing beyond that context.\n\nMention: \"Longitudinal Health Workforce Index (LHWI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Longitudinal Health Workforce Index is quoted as context beside the authority's staffing roster.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:c5f338661a6f03d552d574ea", "occurrence_id": "f019fb893dc18f26b699c019", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Arrival and registration figures are reported from the Panel Arrival and Registration Survey (PARS), Fourth Wave, the survey the Brazilian Ministry of Labor follows, and the four values are reproduced here as issued.\n\nThe survey puts the number of arrivals registered in the wave at 41 300, the share registering within thirty days of arrival at 62 per cent, the share of registered children enrolled in a school by the following term at 48 per cent, and the mean months a household waits before its registration is complete at 3. They were not re-derived here, and the survey's own reading of an arrival and of a complete registration accompanies the figures.\n\nAn arrival registered by a relative on the household's behalf was still counted as registered, and a child enrolled but never attending was not counted as enrolled. Each figure keeps the reception office it was taken at kept with it, instead of being rolled up to the wave.\n\nThe survey line for each value: the date of the pull and the households each share divides are recorded under the table.\n\nMention: \"Panel Arrival and Registration Survey (PARS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Panel Arrival and Registration Survey supplies registered arrivals, timeliness, enrolment and wait values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:1a8d979bde41c78ab84b620c", "occurrence_id": "78c1b62d51ef3aaac18ecd0f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Graduation and work were reconstructed for this review from the Record-Based Graduation Programmes Registry, which Barro and Lee maintain for the Republic of Zambia as of 2018, and the readings below are the review's, not the registry's.\n\nHouseholds were sorted by whether any member holds a written employment contract and the share still above the programme's exit line two years later computed: 74 per cent of households with a contract against 39 per cent of those without. The registry holds a contract flag and an exit result for the same household and never sets one against the other.\n\nHolding the number of productive assets the household owns even narrows the gap to 16 points, so the assets carry part of the distance the first comparison charged to the contract.\n\nHouseholds that left the programme before the follow-up are dropped rather than counted as failing the line, and they are 8 per cent of households. A note in the annex gives the sorting rule, the pull date and, alongside it, the base for each share.\n\nMention: \"Record-Based Graduation Programmes Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Record-Based Graduation Programmes Registry relates written contracts to staying above the exit line.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:2e0b9f494d5e2db9446f3ea2", "occurrence_id": "35c495da4d77615a519d999e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Settlement the numbers held for the region come from the Standardised Settlement Location Registry (SSLR), Version 2, the register kept for Ile de France in 2020, and the four values are quoted here as printed.\n\nThe register puts the number of informal settlements recorded in the region at 168, the share whose residents hold a written occupancy agreement at 22 per cent, the mean number of households in a settlement at 240, and the share of settlements lying within a declared risk zone at 31 per cent. No arithmetic was applied to them here, and the register's own reading of a settlement, which requires eight or more households on one parcel, comes down with them.\n\nA building occupied by fewer than eight households was not counted as a settlement, and an agreement under appeal was still counted as written. Each figure keeps the arrondissement it was drawn from named with it, instead of averaged over the region.\n\nThe annex note carries the register line for each value, the date on the extract, and the settlements each share divides.\n\nMention: \"Standardised Settlement Location Registry (SSLR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Standardised Settlement Location Registry supplies settlement counts, agreements, size and risk-zone shares.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:26080c630d771dcd4942e917", "occurrence_id": "f405d7a360cd1ebe89527271", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The caseload figures in this chapter are computed from the service's own intake log, and the Government Mental Health Survey (GMHS), the survey Booking.com conducted for 2024, is referred to only as background to them.\n\nThe survey puts the share of adults reporting a common mental disorder in the previous month at 12 per cent, the share of those who sought any professional help at 31 per cent, and the mean weeks between first seeking help and first contact at 7. All three are given as the survey issues them, with its own screening instrument stated, and no chapter figure rests on them.\n\nIts responses do not appear in any chapter table. The figures the chapter argues from come from the intake log, which records contacts actually made rather than need reported in a household, and where the two accounts of demand differ the difference is noted rather than reconciled.\n\nThe annex lists the survey tables consulted and the day of the pull, and the survey is used for nothing beyond that background.\n\nMention: \"Government Mental Health Survey (GMHS)\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The Government Mental Health Survey is background to the service's own intake caseload figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:0499b5183a11227b96639aa2", "occurrence_id": "b796eba53ae75ab4315b4049", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Household records were catalogued into the inequality file from the Enumerative Inequality Survey, the survey Standard and Poors conducted in Bangladesh in 2016, and the work done on them was cataloguing rather than measurement.\n\nEach household was given the file's own stratum code and its consumption aggregate checked against the file's bands; where a household reports an income the file cannot classify, it was parked for the analyst rather than placed in the nearest band. Households with the consumption module left unanswered were left unclassified rather than assumed to are written up in the lowest band, and 940 of them entered the file in that condition. The survey's own interview date was kept on every household so a later round can be matched without re-contacting it.\n\nThe inequality file's population weights come from the census tabulations, from which the survey supplied nothing; it is the household coding that lets those weights be read against expenditure for the distribution the report draws up.\n\nThe coding sheet, the count of parked households and the day of deposit are detailed in the annex log.\n\nMention: \"Enumerative Inequality Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The Enumerative Inequality Survey supplies the coded household records behind the inequality file.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:bf447bde2a40dbce752ceb86", "occurrence_id": "06a6d5a50712d292fb8575dd", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Operations figures for the line ministries are reported from the Periodic Operations Index System (POIS), the register kept for Lao PDR as of 2019, and the four values are carried here as tabulated.\n\nThe register puts the share of ministries filing a monthly operations return at 72 per cent, the mean days between a period closing and its return arriving at 18, the share of returns carrying a figure for each planned activity at 54 per cent, and the share of ministries whose return was queried once filed at 29 per cent. No reweighting was applied to them, and the register's own reading of a filed return and of a query is carried alongside them.\n\nA return sent by telephone and never confirmed in writing was not counted as filed, and a query raised by the ministry's own internal audit was not counted. Each figure keeps the ministry it was drawn from attached, instead of being blended across government.\n\nThe table's footnotes hold the register line for each value, the extraction day as well as the ministries each share divides.\n\nMention: \"Periodic Operations Index System (POIS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Periodic Operations Index System supplies filing, delay, completeness and query values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:dbbb1e858c2dfbfade5fe1da", "occurrence_id": "ff904971bd21db376e5fbea7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Seasonal shortfall was assembled for this review out of the Pilot Poverty and Vulnerability Survey (PPVS), Fifth Release, the survey conducted in Papua New Guinea in 2019, and what follows is the review's arithmetic, not the survey's.\n\nHouseholds were sorted by the number of lean months they reported after the last cyclone and the share borrowing for food in the same period computed: 57 per cent of households reporting three lean months or more against 21 per cent of those reporting none. The survey holds a lean-month count and each coping item for the same household and never sets one against the other.\n\nMaking the groups alike on whether the household received any relief brings the gap down to 18 points, which leaves relief as part of what the unadjusted contrast attributed to the length of the lean season.\n\nHouseholds whose garden was destroyed but who moved before the interview are dropped rather than counted as unaffected, and they are 6 per cent of households. Those references, the extraction day and the denominators are carried in the annex's method sheet.\n\nMention: \"Pilot Poverty and Vulnerability Survey (PPVS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Pilot Poverty and Vulnerability Survey relates lean months after a cyclone to borrowing for food.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:385c1ea80118a917c02b080b", "occurrence_id": "644dc0007e4b1213d322fdd3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Administrative record quality country figures are assembled from the Verified Administrative Records Index (VARI), Sixth Phase, the index compiled for Iraq in 2015, and the four values pass into this report as printed.\n\nThe index puts the share of registers subject to a documented completeness check at 49 per cent, the mean years since a register's coverage was last estimated at 5, the share of registers whose errors are logged when found at 33 per cent, and the share of registers reconciled against a survey in the year at 17 per cent. Nothing was re-estimated here, and the index's own reading of a documented check and of a logged error is printed beside them.\n\nA check carried out for a single province was not counted as covering the register, and an error corrected without being recorded was not counted as logged. Each figure keeps the register family it was drawn from listed with it, rather than smoothed across the country.\n\nUnder the table you will find the index line for each value, plus the extraction date and the registers each share divides.\n\nMention: \"Verified Administrative Records Index (VARI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Verified Administrative Records Index supplies checking, coverage estimation, logging and reconciliation values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:074bef73af6d5bd16ae63d65", "occurrence_id": "0a3f19a7cd700ab84b81ef23", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Unpaid and paid hours were estimated for this review using the Metropolitan Time Use Survey, the survey WIDER followed in Comoros in 2015, and the figures are the review's own.\n\nAdults were sorted by the hours they record for paid work in the reference week and the share giving more than 20 hours to fetching water or fuel computed: 34 per cent of adults working fewer than 10 paid hours against 9 per cent of those working more than 40. The survey holds a paid-hours item and each unpaid activity for the same adult and never sets one against the other.\n\nMaking the groups alike on the distance from the household's dwelling to the nearest stand brings the gap down to 11 points, which leaves distance as part of what the unadjusted difference owed to the paid hours.\n\nAdults who reported no activity of any kind in the week are dropped rather than counted as giving no unpaid hours, and they are 7 per cent of adults. Working notes list the sorting rule, the extraction day as well as the denominators each share rests on.\n\nMention: \"Metropolitan Time Use Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Metropolitan Time Use Survey relates paid hours to hours spent fetching water and fuel.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:c22332647d7df9655f09026d", "occurrence_id": "1ff274cc6032e8bea90b6d74", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Survey delay was derived for this chapter from the Annual Administrative Survey System (AASS), Seventh Round, the system compiled for South Asia in 2024, and the figures below are the chapter's own figures and not the system's.\n\nSurveys were sorted by the number of reporting units they cover and the median months between fieldwork closing and publication computed: 7 months among surveys below 5 000 units against 15 months among those above 25 000. The system holds a unit count and a publication date for each survey and never sets one against the other.\n\nThat comparison supports one claim, that the smaller health and education surveys should be tabulated centrally before their larger counterparts are re-planned. The budget lines the proposal quotes come from the office's own vote, from which the system supplied nothing.\n\nSurveys still in the field at the round's close are dropped rather than counted as unpublished, and they are 8 per cent of surveys. The bands and the counts behind each median are recorded under the table.\n\nMention: \"Annual Administrative Survey System (AASS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Annual Administrative Survey System supports one tabulation claim built from survey size and delay.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:100ab1607022cbd367684792", "occurrence_id": "7572909bdb716e5c832a4415", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Service register figures spanning the programme are taken from the Standardised Service Register Registry (SSRR), the database VIIRS maintains for 2022, and the four values come into this report unchanged.\n\nThe register puts the share of service points listed in the register with a named operator at 84 per cent, the mean months since a listing was last verified on site at 14, the share of listings whose capacity figure was revised in the year at 27 per cent, and the share of points reachable by a public vehicle at 58 per cent. No figure was recalculated in this report, and the register's own reading of a verified listing is given next to them.\n\nA listing verified by telephone was not counted as verified on site, and a capacity revised upward without a visit was still counted as revised. Each figure keeps the service type it was drawn from carried with it, rather than mixed across the register.\n\nThe annex keeps the register line for each value, along with the day the table was pulled and the points each share divides.\n\nMention: \"Standardised Service Register Registry (SSRR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Standardised Service Register Registry supplies operator, verification, revision and reachability values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:ca3c626b2e76ed8969a7b380", "occurrence_id": "ad118fea80f52b6d5bc68e9b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Operations figures for the field offices are reported from the Diagnostic Operations Index Registry, the register MapX maintains for Mali as of 2016, and the four values are transcribed without alteration.\n\nThe register puts the share of offices whose monthly report reached the capital by the tenth at 63 per cent, the mean number of staff an office reports as available against its sanctioned strength at 0.8, the share of offices holding a functioning vehicle at 47 per cent, and the share of reports carrying a figure for each active project at 71 per cent. Nothing here was recomputed in this report, and the register's own reading of an available worker and of an active project travels with the figures.\n\nAn office whose report arrived late for one month in the year was still counted as timely, and a vehicle under repair for more than a quarter was not counted as functioning. Each figure keeps the region it was drawn from with it, rather than pooled across the country.\n\nThe annex log holds the register line for each value, the day of the pull together with the offices each share rests on.\n\nMention: \"Diagnostic Operations Index Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Diagnostic Operations Index Registry supplies reporting timeliness, staffing, vehicles and completeness values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:4e31b477063799bf8a8f8f66", "occurrence_id": "7ce4b8f008ba9f3cf183a8a7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The vacancy figures in this chapter are compiled from the ministry's own establishment return, and the Ministerial Public Sector Monitoring Survey (MPSMS), conducted in Germany in 2021, is referred to only as background to them.\n\nThe survey puts the share of posts vacant at the time of reporting at 9 per cent, the mean months a vacancy takes to fill at 6, and the share of vacancies filled without external advertisement at 41 per cent. All three are given as the survey issues them, with its own coverage of ministries stated, and no chapter figure rests on them.\n\nIts responses are kept out of the chapter's tables. The figures the chapter argues from come from the establishment return, which names each post and its holder rather than sampling a service, and where the two accounts of vacancy differ the difference is noted rather than reconciled.\n\nThe annex lists the survey tables consulted and the day the rows were taken, and the survey is used for nothing beyond that background.\n\nMention: \"Ministerial Public Sector Monitoring Survey (MPSMS)\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The Ministerial Public Sector Monitoring Survey is background to the ministry's establishment vacancy figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:2c1a5a18fd47c31ca0413192", "occurrence_id": "08ade44c89db6e234e1bbb74", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Facility figures at the state level are taken from the Recurrent Facility Assessment Survey (RFAS), Version 9, the survey conducted in Johor in 2015, and the four values enter this report as they stand.\n\nThe survey puts the share of facilities offering antenatal care on the day of visit at 79 per cent, the share with electricity through the whole previous month at 88 per cent, the mean number of deliveries a facility attends in a month at 34, and the share with a functioning labour room at 72 per cent. They are carried as published, and the survey's own reading of a functioning room is bound to the figures.\n\nA room usable only after arrangement with another ward was not counted as functioning, and a facility attended by a midwife only was still counted as offering antenatal care. Each figure keeps the district it was collected in attached, instead of being blended across the state.\n\nBeneath the table sit the survey line for each value: the day the extract ran and the facilities each share divides.\n\nMention: \"Recurrent Facility Assessment Survey (RFAS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Recurrent Facility Assessment Survey supplies antenatal cover, power, deliveries and labour-room values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:b5d0ecd1b6dc179163af3599", "occurrence_id": "e89d453799ad367cd613b4a1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The performance figures in this chapter are calculated from the authority's own returns, and the Experimental Operations Index System (EOIS), the register maintained for Zambia in 2021, is quoted only to place those figures in context.\n\nThe register puts the share of offices reporting against a target in the period at 66 per cent, the mean distance between a reported figure and its target at 12 per cent, and the share of offices whose targets were revised mid-period at 9 per cent. All three are given as the register issues them, with its own scope of offices stated, and no chapter figure was built from them.\n\nTheir use here is to show how other offices behave beside the one authority whose returns the chapter works from. The chapter's argument rests on those returns, which record what the authority planned and did rather than a sample of comparable offices, and the two are kept apart rather than merged.\n\nThe annex lists the register tables consulted and the pull date, which is the full extent of the register's use in this chapter.\n\nMention: \"Experimental Operations Index System (EOIS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Experimental Operations Index System is quoted as context beside the authority's performance returns.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:67c56a46a4dfc072b5557cf2", "occurrence_id": "3e1722cd238cc3dc101f3f76", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The stock figure in this chapter is the programme's own, taken from its depot cards, and the Triennial Supply Chain Survey conducted by DLLS in Guatemala in 2016 is cited to test it.\n\nThe programme's cards put 214 days of cover for a first-line commodity at the central depot. The survey puts 189 days for the same depot, because it counts stock in transit to the regions as committed rather than available, while the cards count it as available until it leaves the depot. Each basis keeps its own figures untouched.\n\nThe 25-day distance is smaller than the movement the programme's own cards show between two quarters, so the programme's figure is printed as carded and the survey supplies no revision to it and no depots for the departments its teams did not reach.\n\nExcluded are the depots the survey visited once in the year, 5 of its 31, since a single visit cannot give a mean. Which stocks each source counts is laid out in the annex notes.\n\nMention: \"Triennial Supply Chain Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Triennial Supply Chain Survey checks the programme's days-of-cover figure at the central depot.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:29c43d7eb098a8c26f47dd82", "occurrence_id": "641f472d74a7b2a753c7034d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Parcel footprints for the country are reported from the Rolling Parcel Footprint Collection, the collection ALCS maintains for the Dominican Rep. as of 2022, and the four values reach this report as published.\n\nThe collection puts the share of parcels in the surveyed frame carrying a footprint outline at 76 per cent, the mean built area of a parcel inside the urban frame at 260 square metres, the share of parcels whose outline was redrawn between collections at 8 per cent, and the share of parcels with no outline because of canopy cover at 11 per cent. Nothing here was recomputed, and the collection's own reading of an outline, which requires a wall line rather than a roof shadow, goes with the figures.\n\nA structure still framing was outlined but not counted as built area, and a parcel hidden by tree cover was counted as unoutlined rather than empty. Each figure keeps the municipality it was drawn from shown with it, instead of combined into one country figure.\n\nWorking notes record the collection layer for each value, the day of extraction, and the parcels each share divides.\n\nMention: \"Rolling Parcel Footprint Collection\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Rolling Parcel Footprint Collection supplies outlined parcels, built area, redraws and canopy gaps.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:8f8bd7781448e2dfeed2d323", "occurrence_id": "8c3ee02055831da304242379", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Shift work and absence were estimated for this review using the Coordinated Working Conditions Survey, the survey NSSO followed in Brazil in 2017, and the figures are the review's own.\n\nWorkers were sorted by whether their schedule changes week to week and the number of days they report missing through illness in the year computed: 5 days among workers on a changing schedule against 3 days among those on a fixed one. The survey holds a schedule item and an absence item for the same worker and never sets one against the other.\n\nMaking the groups alike on the hours a worker stands per shift brings the gap down to 1 day, which leaves standing hours as the larger part of what the first comparison credits to the changing schedule.\n\nWorkers holding two jobs at once are dropped rather than counted at their larger schedule, and they are 6 per cent of workers. Working notes list the sorting rule, the extraction day plus the denominators each share rests on.\n\nMention: \"Coordinated Working Conditions Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Coordinated Working Conditions Survey relates a changing schedule to days lost through illness.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:892b2038a4b1c982375e709f", "occurrence_id": "cf49c8a0ef30526c55ad6ab8", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Deprivation and hardship country-level figures come from the Annual Deprivation and Hardship Survey, which Germanwatch e.V. conducted in Turkey in 2021, and the four values are transcribed without alteration.\n\nThe survey puts the share of households unable to keep the home adequately warm in winter at 22 per cent, the share that skipped a meal in the previous month for want of money at 9 per cent, the mean months a household went without heating fuel in the year at 2, and the share reporting debt payments take more than a tenth of income at 27 per cent. Nothing here was recalculated in this report, and the survey's own reading of adequate warmth travels with the figures.\n\nA household that burned furniture rather than go cold was counted as unable to keep the home warm, and a meal skipped for health reasons was not counted as hardship. Each figure keeps the region it was collected in reported with it, rather than blended into a single country share.\n\nThe annex log holds the survey line for each value, the date it was drawn as well as the households each share rests on.\n\nMention: \"Annual Deprivation and Hardship Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Annual Deprivation and Hardship Survey supplies warmth, skipped meals, fuel gap and debt burden values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:42d54c1b6070ca3cf7e623fd", "occurrence_id": "b800322501e46b8e2ae745c7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The stock figures in this chapter are computed from the depot's own ledger, and the Continuous Supply Chain Survey (CSCS), Sixth Release, the survey conducted in Romania in 2023, is referred to only as background to them.\n\nThe survey puts the share of depots reporting a stock-out of a first-line commodity in the period at 17 per cent, the mean weeks a stock-out lasts at 4, and the share of depots whose balance was checked physically at the visit at 63 per cent. All three are given as the survey issues them, with its own list of commodities stated, and no chapter figure rests on them.\n\nNo chapter table rests on its responses. The figures the chapter argues from come from the ledger, which records each movement rather than a snapshot taken once a year, and where the two accounts of cover differ the difference is noted rather than reconciled.\n\nThe annex lists the survey tables consulted and the date the file was taken, and the survey is used for nothing beyond that background.\n\nMention: \"Continuous Supply Chain Survey (CSCS)\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The Continuous Supply Chain Survey is background to the depot ledger's stock cover figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:4da45514ba38b056b4f006dc", "occurrence_id": "411b8f3c0688f118ab77dda6", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Crowding and adequacy were reconstructed for this review from the Clustered Housing Conditions Survey, which ILO followed in Uzbekistan in 2015, and the readings below are the review's, not the survey's.\n\nHouseholds were sorted by persons per room and the share reporting the dwelling lacks a working bath or toilet computed: 31 per cent of households above two persons per room against 7 per cent of those below one. The survey holds a room count and each amenity item for the same household and never sets one against the other.\n\nHolding the age of the building even narrows the gap to 9 points, so the building's age carries part of the gap the opening contrast assigned to the crowding.\n\nHouseholds sharing a dwelling with another household but reporting separate rooms are dropped rather than pooled with the uncrowded, and they are 5 per cent of households. A note in the annex gives the sorting rule, the pull date, and the base for each share.\n\nMention: \"Clustered Housing Conditions Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Clustered Housing Conditions Survey relates persons per room to missing bath or toilet facilities.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:be7251363c0bca27089a50c9", "occurrence_id": "be26b2dec922d8c2101776f2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two monitoring lines in the response annex come from the Cross-Sectoral Programme Monitoring Registry (CPMR), Collection 7, the register WFP-UNHCR maintains, and both are transcribed without alteration.\n\nThe register puts the share of partner reports reaching the coordination platform on time at 74 per cent and the mean weeks between an activity ending and its report arriving at 5. Nothing here was re-derived for this report, and the register's own reading of an on-time report, which runs from the period close rather than the date of signature, travels with the figures.\n\nThe other lines in the annex are copied from the cluster's own financial tracking, so the register carries two lines while the tracking holds the other lines. The report keeps each sector apart rather than averaging across the response.\n\nThe annex log holds the register entry, the day of the extract, with the counts each share rests on alongside.\n\nMention: \"Cross-Sectoral Programme Monitoring Registry (CPMR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Cross-Sectoral Programme Monitoring Registry supplies report timeliness and closing-to-report delay lines.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:732a455b911c651c08e907f3", "occurrence_id": "9ab88e71395777c07582504d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Displacement and work figures for each part of the region are drawn from the Subnational Displacement Tracking Survey (SDTS), the survey conducted in Eritrea in 2024, and the four values appear here exactly as printed.\n\nThe survey puts the number of displaced households tracked in the period at 18 900, the share of those households with a member engaged in seasonal labour at 46 per cent, the mean months a household has been displaced at 19, and the share receiving any regular transfer at 23 per cent. They are reproduced as tabulated, and the survey's own reading of a displaced household and of seasonal labour travels with the values.\n\nA household that moved within its own town was not tracked as displaced, and work paid in kind was counted as seasonal labour. Each figure keeps the sub-zone it was collected in kept with it, instead of being rolled up to the region.\n\nThe documentation under the table gives the survey line for each value, plus the date of the pull and the households each share divides.\n\nMention: \"Subnational Displacement Tracking Survey (SDTS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Subnational Displacement Tracking Survey supplies tracked households, seasonal work, duration and transfers.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:6a90d1e3d74e85e4765a6663", "occurrence_id": "f2ef59d5f72c4a06335b0009", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Whether the registry's own count of receiving households holds was tested against the Municipal Migration and Remittance Survey (MMRS), Sixth Wave, the survey the World Bank Group follows, and the survey is the principal source for the test.\n\nThe registry's client list carries 26 400 households it records as receiving a remittance in the year. The survey lists 31 800 households for the same municipalities, because it asks every household whether money arrived from a migrant, including transfers the registry sees only when they pass through a named agent. Neither basis was carried across to the other's list.\n\nThe survey therefore carries the section's count of receiving households, and the registry's smaller figure is printed beside it so the reader can see that the distance is transfers carried by hand rather than a disagreement about who a migrant is. Where a household appears in both, the registry's payment record is kept and the two counts are not averaged.\n\nHouseholds whose member migrated more than five years before the wave are dropped from both rather than counted by the survey only, 9 per cent of its listings. Which transfers each source can see is set out in the annex file.\n\nMention: \"Municipal Migration and Remittance Survey (MMRS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Municipal Migration and Remittance Survey tests the registry's count of remittance-receiving households.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:9df860aceae175e9598b8127", "occurrence_id": "f77b0d7568130c36a69850c9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Social protection figures measured across the country are read from the Standard Social Protection Report (SSPR), Release 8, the report issued for Nepal in 2017, and the four values are transcribed without alteration.\n\nThe report puts the share of households receiving any social transfer at 38 per cent, the mean annual value of a transfer as a share of the national adequacy line at 21 per cent, the share of transfer recipients who also farm for a living at 64 per cent, and the share of eligible households reporting they were never told of the programme at 17 per cent. Nothing here was worked out again here, and the report's own reading of an eligible household travels with the figures.\n\nA household receiving only a fee waiver was not counted as receiving a transfer, and farming for a season was counted as farming for a living. Each figure keeps the district it was drawn from named with it, instead of averaged over the country.\n\nThe annex log holds the report page for each value, along with the date on the extract and the households each share rests on.\n\nMention: \"Standard Social Protection Report (SSPR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Standard Social Protection Report supplies transfer coverage, value, farming overlap and awareness.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:9c27ed7a9441eec126eec851", "occurrence_id": "af593907ca6ce479c824ea90", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Insurance coverage was assembled for this review out of the Ministerial Unemployment Insurance Census (MUIC), the census conducted in Croatia in 2016, and what follows is the review's arithmetic, not the census's.\n\nAreas were sorted by the share of residents working across the border and the share of the local labour force insured at the reference date computed: 48 per cent in areas where cross-border work exceeds a quarter against 71 per cent where it falls below a tenth. The census holds a cross-border share and an insured share for each area and never sets one against the other.\n\nMaking the groups alike on the mean size of the establishments in the area brings the gap down to 9 points, which leaves establishment size as part of the distance the first comparison ascribed to cross-border work.\n\nAreas with fewer than 200 residents of working age are dropped rather than pooled with their neighbours, and they are 12 per cent of areas. Those references, the extraction day and the denominators are recorded in the papers behind the annex.\n\nMention: \"Ministerial Unemployment Insurance Census (MUIC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Ministerial Unemployment Insurance Census relates cross-border work to local insurance coverage.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:9efc0812d10aca2ed7f1566e", "occurrence_id": "a1cacec8cfecada7457ed2a3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two temperature lines in the protection annex come from the Endline Heat and Cold Extremes Report (EHCER), Third Phase, the report FFMS issued, and both are transcribed without alteration.\n\nThe report puts the mean days above 35 degrees at the stations it reads in the season at 41 and the share of displacement movements it records that follow such a spell by less than a month at 28 per cent. Nothing here was computed afresh for this report, and the report's own reading of a movement, which requires a change of dwelling rather than of work, travels with the figures.\n\nThe annex's remaining entries follow the camp's own registration sheets, so the report carries two lines while the sheets carry the balance. A site figure carries its site with it instead of being smoothed across the operation.\n\nThe annex log holds the report page for each value, the day of the pull and the counts each share rests on.\n\nMention: \"Endline Heat and Cold Extremes Report (EHCER)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Endline Heat and Cold Extremes Report supplies hot-spell days and the share of movements following them.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:d69bbf5619d07500bc27568a", "occurrence_id": "3206b13275fc45315795ace1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Record timeliness was derived for this chapter from the Metropolitan Integrated Records Survey (MIRS), First Edition, the survey ICP follows, and the figures below come from the chapter and not from the survey's.\n\nOffices were sorted by the number of records they hold and the median days before a record reaches the central index computed: 11 days among offices below 10 000 records against 34 days among those above 100 000. The survey holds a record count and a posting date for each office and never sets one against the other.\n\nThat comparison supports one claim, that indexing should be delegated to the regional bureaux before the central queue is enlarged. The staffing lines the proposal quotes come from the ministry's own establishment return, from which the survey supplied nothing.\n\nOffices that closed their index for part of the period are dropped rather than counted as slow, and they are 4 per cent of offices. The bands and the counts behind each median are recorded under the table.\n\nMention: \"Metropolitan Integrated Records Survey (MIRS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Metropolitan Integrated Records Survey supports one delegation claim built from record load and delay.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:e69cda7217a41e63fd22375a", "occurrence_id": "4a77f25e3007abe1d5ef7a5d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Internal movement figures for the city are reported from the Harmonised Internal Movement Registry (HIMR), Round 7, the register CGAP maintains for Chengdu, and the four values are reproduced here as issued.\n\nThe register puts the number of moves recorded in the round at 128 600, the share of movers whose previous dwelling lies outside the municipality at 47 per cent, the mean months between a move and its registration at 2, and the share of movers holding a rental contract at the destination at 61 per cent. They were not re-derived here, and the register's own reading of a move, which requires a change of residence rather than of workplace, follows the figures.\n\nA student boarding only during term was not counted as a mover, and an oral arrangement at the destination was not counted as a contract. Each figure keeps the district it was recorded in listed with it, rather than smoothed across the city.\n\nThe register line for each value: the extraction day and the movers each share divides sit with the annex tables.\n\nMention: \"Harmonised Internal Movement Registry (HIMR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Harmonised Internal Movement Registry supplies moves, origin, registration delay and contract shares.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:ae1412064d62393b43edbd6a", "occurrence_id": "2c5cb7237c2612e511ddbe8e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two service-coverage lines in the delivery annex come from the Periodic Service Register Index (PSRI), the index the Department of National Planning compiled for 2020, and both are transcribed without alteration.\n\nThe index puts the share of administrative units with a service point of the kind in question at 83 per cent and the mean headroom population behind each such point at 6 400. Nothing here was recomputed for this report, and the index's own reading of a service point, which requires staffing on the day of the round rather than a building alone, travels with the figures.\n\nThe rest of the annex lines come from the ministry's own facility listing, so the index carries two lines while the listing carries the lines the table does not. No figure is folded over the country; each one names the province it came from.\n\nThe annex log holds the index entry, the extraction date and, alongside it, the counts each share rests on.\n\nMention: \"Periodic Service Register Index (PSRI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Periodic Service Register Index supplies service-point coverage and headroom population lines.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:bb6d5bdd9bcca35f236cb7cd", "occurrence_id": "5cf0b7198aee2be4ff899355", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Temperature exposure for the city is reported from the Metropolitan Heat and Cold Extremes Dataset (MHCED), the geospatial dataset compiled for Riga in 2017, and the four values are quoted here as printed.\n\nThe dataset puts the share of the built-up area classed as a heat surface at 29 per cent, the mean difference in night temperature between that area and the surrounding green belt at 3 degrees, the share of residents living within 300 metres of cooling cover at 34 per cent, and the share of dwellings in the hottest class lacking any shading at 46 per cent. No arithmetic was applied to them here, and the dataset's own reading of a heat surface, which requires both a daytime threshold and a night one, is reported with them.\n\nA courtyard shaded only after sunset was not counted as cooling cover, and a dwelling shaded on one face only was still counted as shaded. Each figure keeps the district it was drawn from carried with it, rather than mixed across the city.\n\nUnder the table are the dataset layer for each value, the day the table was pulled, and the residents each share divides.\n\nMention: \"Metropolitan Heat and Cold Extremes Dataset (MHCED)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Metropolitan Heat and Cold Extremes Dataset supplies heat-surface share, night contrast, cooling access and shading.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:5cb60f990a6e86cb90d587da", "occurrence_id": "a8962da7a62bc491cd817cef", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Coping after a flood was reconstructed for this review from the Departmental Shock Coping Survey (DSCS), Edition 2, conducted in Lagos State in 2022, and the readings below are the review's, not the survey's.\n\nHouseholds were sorted by the days their dwelling stayed under water and the share whose children missed a week of school afterwards computed: 43 per cent of households with three days or more against 12 per cent of those with less than a day. The survey holds a flooding duration and a schooling item for the same household and never sets one against the other.\n\nHolding whether the household rebuilt on its own even narrows the gap to 17 points, so rebuilding carries part of what the raw comparison laid at the door of the duration of the flooding.\n\nHouseholds that had moved out of the state before the interview are dropped rather than counted as unaffected, and they are 9 per cent of households. A note in the annex gives the sorting rule, the pull date as well as the base for each share.\n\nMention: \"Departmental Shock Coping Survey (DSCS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Departmental Shock Coping Survey relates flooding duration to children missing a week of school.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:036c800cdb187f4cfa5bfcbe", "occurrence_id": "c76ce3e573e3f1391a0d2235", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Learning and equity were estimated for this review using the Subnational Learning and Equity Survey, which WB/Makerere/FASID followed in Punjab in 2022, and the figures are the review's own.\n\nChildren were sorted by the household's distance from the nearest school and the share unable to read a short sentence at grade level computed: 38 per cent of children living more than 3 kilometres away against 21 per cent of those living within half a kilometre. The survey holds a distance for each child and a reading result for each child and never sets one against the other.\n\nMaking the groups alike on the mother's schooling brings the gap down to 9 points, which leaves the household's schooling as part of the gap the opening comparison put down to distance.\n\nChildren enrolled in a school that refused the assessment are dropped rather than counted as unable to read, and they are 4 per cent of children. Working notes list the sorting rule, the extraction day together with the denominators each share rests on.\n\nMention: \"Subnational Learning and Equity Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Subnational Learning and Equity Survey relates distance to school and reading at grade level.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:7f2c1b7af99f05abc3e1e794", "occurrence_id": "3f12b47d1b41338a5e9dad39", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Operations the regions figures are compiled from the Quarterly Regional Operations Index (QROI), Fourth Wave, the index ILO compiles, and the four values are carried here as tabulated.\n\nThe index puts the share of regions filing a complete quarterly return at 69 per cent, the mean days between the quarter closing and a return arriving at 21, the share of returns whose figures were revised after filing at 18 per cent, and the share of regions whose return carries a figure for each service line at 57 per cent. No reweighting was applied to them, and the index's own reading of a complete return is attached to them.\n\nA return missing one service line was counted as incomplete, and a revision made by the index itself for arithmetic error was not counted as a revision. Each figure keeps the wave it was collected in attached, instead of being blended across regions.\n\nThe index line for each value, the day of the pull as well as the regions each share divides are recorded under the table.\n\nMention: \"Quarterly Regional Operations Index (QROI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Quarterly Regional Operations Index supplies filing completeness, delay, revisions and line coverage.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:a440394e7e57ab92ec9c5238", "occurrence_id": "6e1a8214bff8d8c96ff31218", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Whether the agency's own count of displaced households holds was tested against the Decennial Forced Population Report (DFPR), Version 3, the report issued for Haiti in 2016, and the report is the principal source for the test.\n\nThe agency's cards count 61 200 households in its sites. The report counts 74 500 for the same communes, because it counts a household displaced by the event whether or not it ever presented at a site, while the cards record only households that registered. No registration requirement was applied to the report's count for the comparison.\n\nThe report therefore carries the section's count of households displaced, and the agency's smaller figure is printed beside it so the reader can see that the distance is households who never came forward rather than a disagreement about what displacement is. Where a household appears in both, the agency's card date is kept and the two counts are not averaged.\n\nHouseholds displaced by later storms are dropped from both rather than counted by the report only, 11 per cent of its listings. The counting rule each source works to is written up in the working papers behind the annex.\n\nMention: \"Decennial Forced Population Report (DFPR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Decennial Forced Population Report tests the agency's count of displaced households.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:bcf12152691b8724f816b9a8", "occurrence_id": "5e33082703c297ed187db391", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Warning coverage and loss were assembled for this review out of the Independent Early Warning Dataset (IEWD), Seventh Collection, the geospatial dataset compiled for Punjab in 2024, and what follows is the review's arithmetic, not the dataset's.\n\nVillages were matched to their alert class and grouped by how many hours before the event the alert reached them, and the share of households reporting a loss of stored grain computed: 12 per cent of villages alerted more than 24 hours ahead against 39 per cent of those alerted under 6 hours. The dataset times an alert per village and the assessment records the loss per household, and neither reports the two together.\n\nHolding the height of the water at the village even narrows the gap to 14 points, so the height carries part of the distance the first comparison charged to the warning time.\n\nVillages the dataset lists an alert for but where no household was reached in the assessment are dropped rather than counted as loss-free, 5 per cent of villages. Those references, the extraction day and the denominators sit in the annex working papers.\n\nMention: \"Independent Early Warning Dataset (IEWD)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Independent Early Warning Dataset relates alert lead time to households losing stored grain.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:8102cc5bfcd4259984998035", "occurrence_id": "53722ca08d6149cfcd5c2380", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Poverty and access were estimated for this review using the Quality-Assured Socioeconomic Survey (QSS), Wave 2, the survey AICD follows, and the figures are the review's own.\n\nHouseholds were sorted by the nearest distance to a market and the share falling below the national poverty line computed: 47 per cent of households more than 10 kilometres from a market against 24 per cent of those within 2. The survey holds a distance for each household and a consumption aggregate for each household and never sets one against the other.\n\nMaking the groups alike on the land a household farms brings the gap down to 11 points, which leaves land as part of what the unadjusted contrast attributed to distance.\n\nHouseholds whose consumption module failed the survey's own quality check are dropped rather than carried at their reported value, and they are 3 per cent of households. Working notes list the sorting rule, the extraction day plus the denominators each share rests on.\n\nMention: \"Quality-Assured Socioeconomic Survey (QSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Quality-Assured Socioeconomic Survey relates distance to market and poverty-line incidence.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:028f98688636712440f7cb78", "occurrence_id": "a5d0f1e8871de9edaca5a64a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Household records were catalogued into the hazard file from the Annual Landslide Hazard Registry, the register the Bureau of Statistics maintains for Mongolia in 2024, and the work done on them was cataloguing rather than measurement.\n\nEach record was given the file's own slope band and its date of observation checked against the file's window; where a register carries a movement date the file cannot place, it was parked for the analyst rather than pushed to the nearest season. Records whose area had no population frame at all were left unweighted rather than assumed to are listed in the provincial average, and 1 260 of them entered the file in that condition. The register's own observation method was kept on every record so a later edition can be matched without re-deriving it.\n\nThe hazard file's exposure weights come from the population census, from which the register supplied nothing; it is the record coding that lets those weights be read against movement for the risk table the report draws up.\n\nThe coding sheet, the count of parked records and the day of deposit are kept in the annex log.\n\nMention: \"Annual Landslide Hazard Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The Annual Landslide Hazard Registry supplies the coded hazard records behind the risk file.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:95a8aad3bbe0b19b72f6d5fe", "occurrence_id": "eef701abeff141c8fb127a99", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two service lines in the delivery annex come from the Regional Service Register Report (RSRR), Release 9, the report issued for Samoa in 2024, and both are transcribed without alteration.\n\nThe report puts the share of outer-island centres with a resident service worker at 36 per cent and the mean days in a month on which a centre is attended at 14. Nothing here was recomputed here, and the report's own reading of an attended day, which requires a scheduled service rather than a visit, travels with the figures.\n\nThe annex's remaining entries follow the ministry's own facility listing, so the report carries two lines while the listing carries the lines the table does not. A figure reported for one island group keeps that group attached rather than being averaged over the country.\n\nThe annex log holds the report page for each value, the day the table was pulled as well as the counts each share rests on.\n\nMention: \"Regional Service Register Report (RSRR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Regional Service Register Report supplies resident-staff coverage and attended-day lines.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:8f8265632fe80a94569ad100", "occurrence_id": "238a8cbbfa58f78682f90b06", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Insurance receipt and job-finding were reconstructed for this review from the Standard Unemployment Insurance Survey, which SECEX followed in India in 2020, and the readings below are the review's, not the survey's.\n\nClaimants were sorted by the weeks they drew benefit and the number of applications they sent in the same period computed: 21 applications among claimants drawing more than 12 weeks against 9 among those drawing under 4. The survey holds a duration of receipt and an application count for the same claimant and never sets one against the other.\n\nHolding the claimant's years of previous service even narrows the gap to 5 applications, so previous service carries part of what the unadjusted difference owed to the duration of receipt.\n\nClaimants who took work but did not report it are dropped rather than kept in the pool, and they are 6 per cent of claimants. A note in the annex gives the sorting rule, the pull date, and the base for each figure.\n\nMention: \"Standard Unemployment Insurance Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Standard Unemployment Insurance Survey relates weeks of benefit drawn to applications sent.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:1abc4644cf537433eb1f0abc", "occurrence_id": "80426d69875100615536d652", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The coverage figure in this chapter is the programme's own, taken from its session tally sheets, and the Ministerial Vaccination Campaign Survey (MVCS) conducted in Nicaragua in 2024 is cited to test it.\n\nThe programme's sheets put 92 per cent of the target children reached in the round. The survey puts 84 per cent for the same districts, because it counts a child as reached only where the carer could produce the card or the tally showed the child's name, while the sheets count a child entered at the session whether or not a dose was given. Neither rule was applied to the other's count.\n\nThe 8-point distance is larger than the movement the programme's sheets show between two rounds, so the chapter prints the survey's lower figure for children actually reached and the programme's higher one for sessions attended, and the survey supplies no districts for the posts its teams did not visit.\n\nExcluded are children the survey found at home but too ill to dose, 380 of them, since neither count can settle them. Which children each source counts is detailed in the method note in the annex.\n\nMention: \"Ministerial Vaccination Campaign Survey (MVCS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Ministerial Vaccination Campaign Survey checks the programme's share of target children reached.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:ed8daa90d179a05cd2727c36", "occurrence_id": "5c74741d538fc26076b741d2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The distribution figures in this chapter are drawn from the operation's own waybills, and the Independent Programme Monitoring Registry (IPMR), the register maintained for Myanmar in 2018, is quoted only to place those figures in context.\n\nThe registry puts the share of planned distributions completed on schedule at 61 per cent, the mean weeks a distribution runs behind its plan at 4, and the share of sites where a count was verified by a second observer at 48 per cent. All three are given as the registry issues them, with its own scope of sites stated, and no chapter figure was built from them.\n\nTheir use here is to show how the wider operation behaved beside the corridors whose waybills the chapter works from. The chapter's argument rests on those waybills, which record each consignment rather than a sample of sites, and the two are kept apart rather than merged.\n\nThe annex lists the registry tables consulted and the day the extract ran, which is the full extent of the registry's use in this chapter.\n\nMention: \"Independent Programme Monitoring Registry (IPMR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Independent Programme Monitoring Registry is quoted as context beside the operation's waybill figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:ce3218078dfcc5fff15fcfc7", "occurrence_id": "2a98ea70dd2aa650614ea0ae", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The wage figures in this chapter are computed from the office's own payroll returns, and the Clustered Enterprise and Jobs Survey (CEJS), Collection 2, the survey Enerdata conducts, is referred to only as background to them.\n\nThe survey puts the share of enterprises paying the statutory minimum at 72 per cent, the mean gap between a worker's pay and that minimum where it falls below it at 8 per cent, and the share of enterprises adjusting pay within six months of a change in the law at 39 per cent. All three are given as the survey issues them, with its own enterprise frame stated, and no chapter figure rests on them.\n\nNone of the chapter's tables rest on its responses. The figures the chapter argues from come from the payroll returns, which cover each establishment on the register rather than a sample of enterprises, and where the two accounts of compliance differ the difference is noted rather than reconciled.\n\nThe annex lists the survey tables consulted and the pull date, and the survey is used for nothing beyond that background.\n\nMention: \"Clustered Enterprise and Jobs Survey (CEJS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Clustered Enterprise and Jobs Survey is background to the office's payroll minimum-wage figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:cd38a7bdf9986e9acbdcb4d8", "occurrence_id": "6a0418c657b4b3fdf365e945", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Soil condition and labour were assembled for this review out of the Annual Soil Fertility Survey (ASFS), Round 4, the survey the National Statistics Office ran for 2024, and what follows is the review's arithmetic, not the survey's.\n\nHoldings were sorted by the nitrogen level their sample returned and the number of paid days the holding bought in the season computed: 46 days at holdings in the lowest third against 22 days at those in the highest. The survey holds a sample result and a hired-labour item for the same holding and never sets one against the other.\n\nMaking the groups alike on the area a holding works brings the gap down to 9 days, which leaves area as part of what the first comparison credits to the soil result.\n\nHoldings whose sample failed the laboratory's own test are dropped rather than carried at the plot average, and they are 5 per cent of holdings. Those references, the extraction day and the denominators are held in the annex's method note.\n\nMention: \"Annual Soil Fertility Survey (ASFS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Annual Soil Fertility Survey relates soil nitrogen to the paid labour days a holding buys.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:b084518c4639499b0d783e46", "occurrence_id": "1c75131af9f0e390a0ac1d62", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two operations lines in the annex come from the Decennial Operations Monitoring Registry (DOMR), Collection 4, the register kept for Lao PDR in 2019, and both are transcribed without alteration.\n\nThe register puts the share of units whose ledger was inspected in the collection at 54 per cent and the mean number of entries per ledger found irregular at inspection at 7. Nothing here was recalculated for this report, and the register's own reading of an irregular entry, which requires the auditor's written finding rather than a query, travels with the figures.\n\nThe annex's other lines are taken from the treasury's own payment file, so the register carries two lines while the file supplies the balance. Nothing is averaged over the country: a figure for one sector keeps that sector named.\n\nThe annex log holds the register entry, the day of extraction, with the counts each share rests on alongside.\n\nMention: \"Decennial Operations Monitoring Registry (DOMR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Decennial Operations Monitoring Registry supplies ledger inspection coverage and irregular entries.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:44d2ea764b0539098d608198", "occurrence_id": "e327d1b068782587d184c3ac", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Whether the ministry's own figure for continued schooling holds was tested against the Nationwide Youth Transition Survey (NYTS), the survey conducted in the Republic of Zambia in 2017, and the survey is the principal source for the test.\n\nThe ministry's register shows 71 per cent of form-four leavers enrolled in further study or training a year later. The survey shows 58 per cent for the same leavers, because it counts a young person in study only where they attended through the term, while the register counts an enrolment that lapsed in the first weeks. Neither rule was applied to the other's list.\n\nThe survey therefore carries the section's figure for continued participation, and the ministry's larger share is printed beside it so the reader can see that the distance is enrolments that lapsed rather than a disagreement about who left school. Where a leaver appears in both, the ministry's examination result is kept and the two shares are not averaged.\n\nLeavers who married or migrated within the year are dropped from both rather than counted by the survey only, 7 per cent of its listings. The attendance test each source works to is laid out in the annex's method sheet.\n\nMention: \"Nationwide Youth Transition Survey (NYTS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Nationwide Youth Transition Survey tests the ministry's share of leavers continuing in study.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:144a0b92c1723bb87c8fe2a2", "occurrence_id": "8f0348aff6b769144bb5a6f2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two delivery lines in the service annex come from the Household Service Delivery Registry (HSDR), Second Wave, the register kept for Nepal in 2023, and both are transcribed without alteration.\n\nThe register puts the share of households reporting they obtained the document they applied for at 67 per cent and the mean visits a household makes before a document is issued at 2. Nothing here was recomputed here, and the register's own reading of an issued document, which requires the household to hold it at interview, travels with the figures.\n\nThe annex's remaining entries follow the ward's own application book, so the register carries two lines while the book carries the lines the table does not. A figure reported for one district keeps that district attached rather than being averaged over the country.\n\nThe annex log holds the register entry, the day the table was pulled and the counts each share rests on.\n\nMention: \"Household Service Delivery Registry (HSDR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Household Service Delivery Registry supplies documents obtained and visits needed before issue.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:a738c17a0b6a0f8a4c4ae499", "occurrence_id": "9fb8b370fc1c724d09d6d286", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two insurance lines in the labour annex come from the Multi-Topic Unemployment Insurance Survey (MUIS), Round 5, the survey the Chinese Wind Energy Association follows, and both are transcribed without alteration.\n\nThe survey puts the share of unemployed persons eligible for benefit who claimed it at 55 per cent and the mean weeks between a person's last day worked and their first payment at 7. Nothing here was recalculated in this report, and the survey's own reading of eligibility, which follows the contribution record rather than the claim form, travels with the figures.\n\nThe remaining annex lines come from the office's own claims file, so the survey carries two lines while the file carries what is left over. Each cohort figure is reported for its own cohort, not pooled across the scheme.\n\nThe annex log holds the survey line for each value, along with the date the file was taken and the counts each share rests on.\n\nMention: \"Multi-Topic Unemployment Insurance Survey (MUIS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Multi-Topic Unemployment Insurance Survey supplies take-up among the eligible and weeks to first payment.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:a602a965a5489798e624017b", "occurrence_id": "db05155461eb2b4af6e36d6e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two examination lines in the education annex come from the Quality-Assured Exam Performance Index, the index the Berne Union compiled for France in 2024, and both are transcribed without alteration.\n\nThe index puts the share of candidates passing the terminal examination at 88 per cent and the mean points by which candidates in the lowest decile fall short of a pass at 12. Nothing here was re-derived for this report, and the index's own reading of a candidate, which excludes those entering mid-year and those sitting a single paper only, travels with the figures.\n\nThe annex's remaining lines are compiled from the board's own results file, so the index carries two lines while the file accounts for the remaining lines. A figure reported for one academy keeps that academy attached rather than being averaged over the country, and where the index and the board disagree on a pass rate the index's figure is kept for these two lines and the board's for the rest, with the difference noted under the table rather than smoothed away.\n\nCandidates entered by a centre that closed before the papers were set are excluded from both lines, since neither source can say how many of them sat. The annex log holds the index entry behind each value, the day of the extract and the base each share is taken on.\n\nMention: \"Quality-Assured Exam Performance Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Quality-Assured Exam Performance Index supplies pass rates and the shortfall of the lowest decile.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:200800eb17c2ba6c7b916cd6", "occurrence_id": "ac28ee4abb156fd12135e4fd", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Wage dispersion was estimated for this review using the High-Frequency Wage Structure Microdata (HWSM), the extract compiled for Kinshasa in 2024, and the figures are the review's own.\n\nWorkers were sorted by the number of establishments they had worked for in two years and the pay the current job returns computed: 1 340 for workers on a third or later job against 1 610 for workers on their first. The extract holds a job history and a wage for the same worker and never sets one against the other.\n\nMaking the groups alike on the trade each worker practises brings the gap down to 90, which leaves the trade as part of the gap the opening contrast assigned to the number of moves.\n\nWorkers whose current job began less than a month before the interview are dropped rather than counted at their starting rate, and they are 8 per cent of workers. Working notes list the sorting rule, the extraction day and the denominators each share rests on.\n\nMention: \"High-Frequency Wage Structure Microdata (HWSM)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The High-Frequency Wage Structure Microdata relates job moves to the wage of the current job.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:481818503eab1fb119a985f6", "occurrence_id": "5623968e2c01e76da6332b6d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Wage structure figures for the whole city are sourced from the Joint Wage Structure Survey, which CGAP conducted in Lusaka in 2024, and the four values pass into this report as printed.\n\nThe survey puts the median monthly wage for a full-time worker at 3 400, the share of workers paid below the published minimum at 14 per cent, the mean ratio between the highest and lowest wage in an establishment at 3.6, and the share of establishments paying an end-of-year bonus at 41 per cent. Nothing was re-estimated here, and the survey's own reading of a full-time worker and of a published minimum is stated with them.\n\nA worker paid by the day but working a full week was counted as full-time, and a bonus offered only to managers was still counted as paid. Each figure keeps the ward it was collected in with it, rather than pooled across the city.\n\nThe annex note carries the survey line for each value, the date of the pull and, alongside it, the establishments each share divides.\n\nMention: \"Joint Wage Structure Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Joint Wage Structure Survey supplies median wage, minimum breaches, wage ratio and bonus coverage.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:4b6d9c1a9f6c746836af1d1d", "occurrence_id": "6d99596233aab4a91c9a0878", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Household records were catalogued into the health file from the Midline Health and Household Survey (MHHS), the survey Gallup ran for 2020, and the work done on them was cataloguing rather than measurement.\n\nEach household was given the file's own facility code and its illness dates checked against the file's episode window; where a household reports a treatment the file has no code for, it was parked for the analyst rather than entered under the nearest one. Households whose health module was answered by a proxy were left marked as proxy rather than recorded as self-reported, and 1 780 of them entered the file in that condition. The survey's own interview date was kept on every household so a later round can be matched without re-contacting it.\n\nThe health file's enumeration weights come from the population frame, from which the survey supplied nothing; it is the household coding that lets those weights be read against treatment for the service table the report draws up.\n\nThe coding sheet, the count of parked households and the day of deposit are filed in the annex log.\n\nMention: \"Midline Health and Household Survey (MHHS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The Midline Health and Household Survey supplies the coded household records behind the health file.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:973e98c6e2aaf912a6ab6381", "occurrence_id": "2c7687f3f9c2f4fa567b46b3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Household records were catalogued into the payments file from the High-Frequency Targeting and Payments Survey (HTPS), Version 5, the survey Frontex follows, and the work done on them was cataloguing rather than measurement.\n\nEach household was given the file's own eligibility band and its payment dates checked against the file's cycle; where a household reports a payment the file cannot place in a cycle, it was parked for the analyst rather than pushed into the nearest one. Households whose targeting score the survey did not carry were left unscored rather than assumed to sit at the median, and 640 of them entered the file in that condition. The survey's own round number was kept on every household so a later wave can be matched without re-deriving it.\n\nThe payments file's coverage weights come from the registration register, from which the survey supplied nothing; it is the household coding that lets those weights be read against receipt for the transfer table the report draws up.\n\nThe coding sheet, the count of parked households and the day of deposit are recorded in the annex log.\n\nMention: \"High-Frequency Targeting and Payments Survey (HTPS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The High-Frequency Targeting and Payments Survey supplies the coded household records behind the payments file.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:e1c94b3955ea6aab3ce748a1", "occurrence_id": "5199991d7ba006482c75a9e6", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Whether the camp's own count of poor households holds was tested against the Endline Displacement Tracking Survey (EDTS), the survey conducted in Morocco in 2019, and the survey is the principal source for the test.\n\nThe camp's ration cards show 44 per cent of its households below the agency's own adequacy line. The survey shows 57 per cent for the same sites, because it measures consumption over a month of interviews while the cards are drawn from what the household reported at arrival, before it had lost its savings. No month of measurement was applied to the cards for the comparison.\n\nThe survey therefore carries the section's figure for poverty among the displaced, and the camp's lower share is printed beside it so the reader can see that the distance is a report taken too early rather than a disagreement about the adequacy line. Where a household appears in both, the camp's family composition is kept and the two shares are not averaged.\n\nHouseholds that left the site before the survey's last round are dropped from both rather than counted by the survey only, 10 per cent of its listings. The measurement each source works to is stored in the annex notes.\n\nMention: \"Endline Displacement Tracking Survey (EDTS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Endline Displacement Tracking Survey tests the camp's share of households below the adequacy line.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:c8a337207af7a95c451ef292", "occurrence_id": "0115c88618044d452e0b9697", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The alert figure in this chapter is the service's own, taken from its dispatch log, and the Sentinel Early Warning Registry (SEWR), the register kept for Burkina-Faso in 2023, is cited to test it.\n\nThe service's log records 96 alerts issued in the season. The registry records 118 for the same departments, because it counts an alert passed on by a radio station or a chief's office, which never returns to the dispatch log. One basis was never used to adjust the other's count.\n\nThe 22 distance is smaller than the number of alerts the service issues in a week at the season's peak, so the service's figure is printed as logged and the registry supplies no revision to it and no alerts for the departments its monitors did not cover.\n\nExcluded are alerts the registry dates only to a month, 31 of them, since the season's two spells cannot be separated for them. Which alerts each source can see is listed in the annex file.\n\nMention: \"Sentinel Early Warning Registry (SEWR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Sentinel Early Warning Registry checks the service's count of alerts issued in the season.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:26988247c7b93b7b31a708ae", "occurrence_id": "2bb126142076c790f31fff2d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The support figures in this chapter are compiled from the office's own benefit file, and the Statistical Disability and Support Survey (SDSS), Third Phase, the survey Doing Business conducts, is referred to only as background to them.\n\nThe survey puts the share of adults reporting difficulty in one or more everyday activities at 19 per cent, the share of those adults receiving any state support at 27 per cent, and the mean monthly value of that support as a share of the national minimum wage at 46 per cent. All three are given as the survey issues them, with its own question set stated, and no chapter figure rests on them.\n\nNo table in the chapter uses its responses. The figures the chapter argues from come from the benefit file, which lists each recipient rather than a sample of households, and where the two accounts of coverage differ the difference is noted rather than reconciled.\n\nThe annex lists the survey tables consulted and the date on the extract, and the survey is used for nothing beyond that background.\n\nMention: \"Statistical Disability and Support Survey (SDSS)\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The Statistical Disability and Support Survey is background to the office's benefit coverage figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:dd6d2e45bf05567564b31c65", "occurrence_id": "c043b3a5bdf80c3739561499", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Literacy and numeracy the region figures are tabulated by the Rolling Literacy and Numeracy Index (RLNI), Wave 3, the index Mercer compiles for the Tigray Region, and the four values come into this report unchanged.\n\nThe index puts the share of children in grade four able to read a passage of text at 41 per cent, the share able to solve a two-step arithmetic problem at 36 per cent, the mean years of schooling a child's teacher has at 11, and the share of classes with a pupil-textbook ratio below forty at 22 per cent. No figure was recalculated in this report, and the index's own reading of a solved problem, which allows no working error, comes with them.\n\nA child who read aloud but could not answer a question about the passage was not counted as able to read, and a teacher studying for a diploma was counted at years already completed. Each figure keeps the woreda it was collected in attached, instead of being blended across the region.\n\nThe table's footnotes hold the index line for each value: the day of the pull and the children each share divides.\n\nMention: \"Rolling Literacy and Numeracy Index (RLNI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Rolling Literacy and Numeracy Index supplies reading, arithmetic, teacher schooling and textbook ratios.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:e739dcf713f1391d2acb178b", "occurrence_id": "2e25fcdcb3159530d1b17acc", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Vacancy duration and size were assembled for this review out of the Standardised Hiring and Vacancies Survey, which ATO followed in Botswana in 2016, and what follows is the review's arithmetic, not the survey's.\n\nEstablishments were sorted by the number of workers they employ and the median weeks a vacancy stood open computed: 6 weeks at establishments below 20 workers against 15 weeks at those above 200. The survey holds a size for each establishment and an open period for each vacancy and never sets one against the other.\n\nMaking the groups alike on the skill level the vacancy asks for brings the gap down to 4 weeks, which leaves the skill asked for as part of the distance the first comparison ascribed to the size of the establishment.\n\nVacancies the employer withdrew rather than filled are dropped rather than counted as filled at their closing date, and they are 7 per cent of vacancies. Those references, the extraction day and the denominators are held in the papers behind the annex.\n\nMention: \"Standardised Hiring and Vacancies Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Standardised Hiring and Vacancies Survey relates establishment size to weeks a vacancy stands open.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:3f03e0787da2f352ab767e5c", "occurrence_id": "6779103452c7084e221f1723", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Clinic quality the figures for the island are reported by the Small-Area Clinic Quality System (SCQS), Collection 9, the register maintained for Sri Lanka in 2020, and the four values enter this report as they stand.\n\nThe register puts the share of clinic areas meeting the waiting-time standard at 63 per cent, the mean minutes a patient waits before being seen at 48, the share of areas with the full panel of essential medicines in stock at 78 per cent, and the share of areas whose cleanliness score reached the second grade at 71 per cent. They are carried as published, and the register's own reading of a waiting time, which runs from registration rather than arrival, is kept with them.\n\nA patient who left without being seen was not counted in the waiting time, and a medicine stocked at half the panel quantity was counted as out. Each figure keeps the medical office it belongs to shown with it, instead of combined into one island figure.\n\nUnder the table you will find the register line for each value, the extraction day, and the patients each share divides.\n\nMention: \"Small-Area Clinic Quality System (SCQS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Small-Area Clinic Quality System supplies waiting standards, waits, medicine stock and cleanliness grades.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:deb61878956925c69075bd47", "occurrence_id": "43db61a523cd000161f47982", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Training and re-employment were reconstructed for this review from the Biennial Unemployment Insurance Survey (BUIS), Wave 6, the survey NBS follows, and the readings below are the review's, not the survey's.\n\nClaimants were sorted by whether they took a short vocational course while claiming and the weeks before they returned to work computed: 14 weeks among claimants who took a course against 23 weeks among those who did not. The survey holds a course item and a return date for the same claimant and never sets one against the other.\n\nHolding the years of previous service even narrows the gap to 5 weeks, so previous service carries part of what the raw comparison laid at the door of the course.\n\nClaimants whose benefit ended because they reached its limit are dropped rather than counted as still seeking work, and they are 6 per cent of claimants. A note in the annex gives the sorting rule, the pull date as well as the base for each figure.\n\nMention: \"Biennial Unemployment Insurance Survey (BUIS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Biennial Unemployment Insurance Survey relates taking a course while claiming to weeks to re-employment.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:853ccbb04633fc5305d5d41f", "occurrence_id": "f9f137142f4ce86c17aecd93", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Safety practice and incidents were estimated for this review using the Pilot Occupational Safety Microdata, the extract RVRN compiled for Slovakia in 2016, and the figures are the review's own.\n\nEstablishments were sorted by the share of workers given a safety briefing on joining and the number of reportable incidents in two years computed: 11 incidents per hundred workers at establishments briefing more than nine in ten against 3 per hundred at those briefing fewer than half. The extract holds a briefing share and an incident count for the same establishment and never sets one against the other.\n\nMaking the groups alike on the number of hours a worker stands per week brings the gap down to 3 incidents, so standing hours carry part of the gap the opening comparison put down to the briefing.\n\nEstablishments with no worker present for most of the period are dropped rather than counted as incident-free, and they are 4 per cent of establishments. Working notes list the sorting rule, the extraction day together with the denominators each share rests on.\n\nMention: \"Pilot Occupational Safety Microdata\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Pilot Occupational Safety Microdata relates on-joining briefings to reportable incidents.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:9d10a662f0bf126172606ed2", "occurrence_id": "a47f548b53a8f7d85a8a6939", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Plot records were catalogued into the inputs file from the Civil Irrigation and Inputs Registry (CIIR), Third Version, the register kept for Gabon in 2015, and the work done on them was cataloguing rather than measurement.\n\nEach plot was given the file's own water-source class and its input purchases checked against the file's season window; where a record carries an input the file has no class for, it was parked for the analyst rather than entered under the nearest one. Plots whose irrigated area the registry left blank were left unclassified rather than assumed to sit at the group median, and 910 of them entered the file in that condition. The registry's own verification status was kept on every plot so a later version can be matched without re-visiting it.\n\nThe inputs file's area weights come from the agricultural census, from which the registry supplied nothing; it is the plot coding that lets those weights be read against input use for the seasonal table the report draws up.\n\nThe coding sheet, the count of parked plots and the day of deposit are set down in the annex log.\n\nMention: \"Civil Irrigation and Inputs Registry (CIIR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The Civil Irrigation and Inputs Registry supplies the coded plot records behind the inputs file.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:e797d70a03ca30c10bab3f52", "occurrence_id": "364ba9a6999d3c84a0a738d7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Nutrition and child deaths were assembled for this review out of the Monthly Nutrition and Mortality Survey (MNMS), Sixth Collection, conducted in Sud and Nippes in 2023, and what follows is the review's arithmetic, not the survey's.\n\nHouseholds were sorted by the months the child went without the supplementary ration and the share showing moderate acute malnutrition at measurement computed: 29 per cent of children missing four months or more against 8 per cent of those missing none. The survey holds a ration history and a measurement for the same child and never sets one against the other.\n\nMaking the groups alike on the household's distance from the distribution point brings the gap down to 11 points, which leaves distance as part of the distance the first comparison charged to the missed months.\n\nChildren measured more than once in the collection are counted at their first measurement rather than their worst, and they are 9 per cent of children. Those references, the extraction day and the denominators are kept in the working papers behind the annex.\n\nMention: \"Monthly Nutrition and Mortality Survey (MNMS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Monthly Nutrition and Mortality Survey relates missed ration months to moderate acute malnutrition.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:4aae4de1a13a0dcb5a54476e", "occurrence_id": "b596540a92054da42cce3643", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two public-works lines in the labour annex come from the Comparative Public Works Index (CPWI), Second Phase, the index compiled for the CHT and Sylhet divisions in 2020, and both are transcribed without alteration.\n\nThe index puts the share of works paying the published wage rate in full at 68 per cent and the mean days in the year a worker is engaged on a work at 96. Nothing here was worked out again here, and the index's own reading of a day engaged, which requires the worker to have attended rather than to have been enrolled, travels with the figures.\n\nThe other lines in the annex are copied from the programme's own muster rolls, so the index carries two lines while the rolls carry the balance. A figure reported for one division keeps that division attached rather than being averaged across both, and where the index and the rolls disagree on days engaged the index's figure is kept for these two lines and the rolls' for the rest, with the difference noted under the table rather than smoothed away.\n\nWorks suspended before their first payday are excluded from both lines, since neither source can give a wage for them. The annex log holds the index entry behind each value, the extraction date plus the counts the shares rest on.\n\nMention: \"Comparative Public Works Index (CPWI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Comparative Public Works Index supplies full-rate payment and days engaged on public works.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:b0e9a570881e5f9b056bfabf", "occurrence_id": "c6ab42ee33ddbd13b0eacf36", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Reading and teaching were reconstructed for this review from the Nationwide Early Grade Reading Census (NEGRC), Wave 4, the census Coriolis conducts, and the readings below are the review's, not the census's.\n\nClasses were sorted by the hours of literacy instruction a week the teacher reports and the share of pupils reading fluently at the end of the year computed: 44 per cent in classes with five hours or more against 19 per cent in classes with two hours or less. The census holds an instruction figure for each class and a fluency result for each pupil and never sets one against the other.\n\nHolding the number of pupils in the class even narrows the gap to 13 points, so class size carries part of what the unadjusted contrast attributed to the hours taught.\n\nClasses whose teacher was absent for more than a month of the year are dropped rather than counted at the school's average, and they are 5 per cent of classes. A note in the annex gives the sorting rule, the pull date, and the base for each share.\n\nMention: \"Nationwide Early Grade Reading Census (NEGRC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Nationwide Early Grade Reading Census relates weekly literacy instruction to fluent reading.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:cc31795bbd752339824b2bf7", "occurrence_id": "885e05d68a7a762520c82eb5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Hardship and work were estimated for this review using the Continuous Deprivation and Hardship Survey (CDHS), Phase 2, the survey Jacobs et al follow in South Asia, and the figures are the review's own.\n\nHouseholds were sorted by the number of adults with no earnings in the month and the share unable to meet an unexpected medical cost computed: 71 per cent of households with no earner against 34 per cent of those with two or more. The survey holds an earnings item for each adult and the cost question for the same household and never sets one against the other.\n\nMaking the groups alike on whether the household holds a savings account brings the gap down to 18 points, which leaves the account as part of what the unadjusted difference owed to the absence of an earner.\n\nHouseholds whose head reported a job but gave no hours for it are dropped rather than counted as working, and they are 7 per cent of households. Working notes list the sorting rule, the extraction day, with the denominators each share rests on alongside.\n\nMention: \"Continuous Deprivation and Hardship Survey (CDHS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Continuous Deprivation and Hardship Survey relates household earners to meeting a medical cost.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:a79575f487142ba781eda96e", "occurrence_id": "e70feef6bb177538c71fadae", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: School continuation was reconstructed for this review from the Civil Youth Transition System (CYTS), Round 5, the register kept for Fortaleza in 2017, and the readings below are the review's, not the register's.\n\nSchools were sorted by the share of pupils leaving at the end of the lower cycle and the share of those leavers found in a course or a job by the following round computed: 62 per cent where leavers are under a tenth against 34 per cent where they are above a quarter. The register holds a leaving flag for each pupil and a destination flag for the same pupil and never sets one against the other.\n\nHolding the mean household size of the school's catchment even narrows the gap to 12 points, so household size carries part of what the first comparison credits to the leaving rate.\n\nLeavers with no trace in the following round are dropped rather than counted as unemployed, and they are 8 per cent of leavers. A note in the annex gives the sorting rule, the pull date and the base for each share.\n\nMention: \"Civil Youth Transition System (CYTS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Civil Youth Transition System relates early leaving to leavers found in a course or job.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:ae3db96fdd79149ff0d6a601", "occurrence_id": "f33e6d1030e6e189e16e3167", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Inspection exposure and compliance were estimated for this review using the Household Labour Inspection Survey (HLIS), Wave 6, the survey IRI Worldwide follows for 2019, and the figures are the review's own.\n\nHouseholds were sorted by whether any member works in an establishment ever inspected and the share of those members holding a written contract computed: 77 per cent where an inspection had occurred against 49 per cent where none had. The survey holds an inspection item for each establishment and a contract item for the worker and never sets one against the other.\n\nMaking the groups alike on the size of the establishment brings the gap down to 10 points, which leaves size as part of the gap the opening contrast assigned to the inspection.\n\nWorkers employed in a private household are dropped rather than pooled with establishment workers, and they are 11 per cent of workers. Working notes list the sorting rule, the extraction day and, alongside it, the denominators each share rests on.\n\nMention: \"Household Labour Inspection Survey (HLIS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Household Labour Inspection Survey relates ever-inspected workplaces to written contracts.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:66a8a71507acaeefcf193410", "occurrence_id": "8667640e652b322133262488", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Hiring and vacancy figures covering the country are drawn from the High-Resolution Hiring and Vacancies Survey, which NRC-IDMC conducted in Lebanon in 2018, and the four values reach this report as published.\n\nThe survey puts the share of establishments with a vacancy open at the visit at 24 per cent, the mean weeks before a vacancy is filled at 11, the share of vacancies open to workers without a residence permit at 37 per cent, and the share of filled posts filled from outside the branch at 52 per cent. Nothing here was recomputed, and the survey's own reading of an open vacancy and of an outside hire accompanies the figures.\n\nA post held open pending a permit decision was counted as open, and a transfer from another branch was counted as an outside hire. Each figure keeps the governorate it was collected in reported with it, rather than blended into a single country share.\n\nThe annex keeps the survey line for each value, the day the table was pulled as well as the establishments each share divides.\n\nMention: \"High-Resolution Hiring and Vacancies Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The High-Resolution Hiring and Vacancies Survey supplies vacancy incidence, duration, permit openness and hiring source.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:cc1948a3540a9719e7cfa27a", "occurrence_id": "655ac07ac364e5b577145ba6", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Reporting delay and workload were assembled for this review out of the Joint Operations Monitoring Registry, the register US-EPA maintains for Viet Nam as of 2016, and what follows is the review's arithmetic, not the register's.\n\nUnits were sorted by the number of cases they carry and the median weeks before their return reaches the centre computed: 3 weeks at units below 300 cases against 12 weeks at those above 1 500. The register holds a case count and a receipt date for each unit and never sets one against the other.\n\nMaking the groups alike on the staff a unit has for the work brings the gap down to 4 weeks, which leaves staffing as part of the distance the first comparison ascribed to the case load.\n\nUnits that filed no return in the period are dropped rather than counted as maximally late, and they are 6 per cent of units. Those references, the extraction day and the denominators are written up in the annex working papers.\n\nMention: \"Joint Operations Monitoring Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Joint Operations Monitoring Registry relates case load to weeks before a return reaches the centre.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:6f5d2bb819bfcb1b5b86f9aa", "occurrence_id": "327028dca5d5e0601336c3cb", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Transfer receipt and hardship were reconstructed for this review from the Standardised Cash Transfer Registry Survey, which O*NET followed in Rwanda in 2024, and the readings below are the review's, not the survey's.\n\nHouseholds were sorted by the months they had been on the transfer roll and the share reporting they went a whole day without food in the previous month computed: 9 per cent of households on the roll two years or more against 26 per cent of those enrolled under six months. The survey holds an enrolment date and a food item for the same household and never sets one against the other.\n\nHolding the household's livestock count even narrows the gap to 8 points, so the livestock carries part of what the raw comparison laid at the door of the time on the roll.\n\nHouseholds whose payment was recorded but never collected are dropped rather than counted as receiving, and they are 5 per cent of households. A note in the annex gives the sorting rule, the pull date as well as the base for each share.\n\nMention: \"Standardised Cash Transfer Registry Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Standardised Cash Transfer Registry Survey relates time on the roll to going a day without food.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:351a30eb67355666caff7fb5", "occurrence_id": "a62404ec17e3af386bbbe10b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The register figures in this chapter are drawn from the office's own inventory, and the Midline Administrative Records System, the system CIA maintains for Kosovo as of 2019, is quoted only to place those figures in context.\n\nThe system puts the share of registers with a defined retention period at 58 per cent, the mean years since a register's layout was last revised at 6, and the share of registers whose contents are open to another office on request at 31 per cent. All three are given as the system issues them, with its own list of registers stated, and no chapter figure was built from them.\n\nTheir use here is to show how registers elsewhere are kept beside the ones the chapter's inventory covers. The chapter's argument rests on that inventory, which lists one ministry's registers rather than a comparable panel, and the two are kept apart rather than merged.\n\nThe annex lists the system tables consulted and the day of the pull, which is the full extent of the system's use in this chapter.\n\nMention: \"Midline Administrative Records System\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Midline Administrative Records System is quoted as context beside the office's register inventory.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:8d1e5835addc1646b507884f", "occurrence_id": "3fbae1896c6c75a9c06580dc", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two monitoring lines in the regional annex come from the Endline Programme Monitoring Registry, the register the International Labor Organization keeps for Ile de France as of 2021, and both are transcribed without alteration.\n\nThe registry puts the share of measures in the programme with a reported output figure at 74 per cent and the mean months between a measure starting and its first figure arriving at 8. Nothing here was computed afresh for this report, and the registry's own reading of a first figure, which must be audited rather than merely filed, travels with the figures.\n\nThe annex's remaining entries follow the council's own project file, so the registry carries two lines while the file carries the rest. A figure given for a single department stays with that department instead of being averaged across the region.\n\nThe annex log holds the registry entry, the day the rows were taken and the counts each share rests on.\n\nMention: \"Endline Programme Monitoring Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Endline Programme Monitoring Registry supplies output-reporting coverage and time to first figure.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:d74fb63310b551141c8db9ca", "occurrence_id": "33e9ee4abc2482e459c46db1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The register count in this chapter is the bureau's own, taken from its own census of offices, and the Metropolitan Integrated Records Index, the index MODIS compiled for Hungary in 2019, is cited to test it.\n\nThe bureau's census lists 1 240 registers held across the offices it can reach. The index lists 1 385 for the same offices, because it counts a register held on a spreadsheet in a single office, which the bureau's census asks only about systems entered in its catalogue. The two bases were not applied to each other's list.\n\nThe 145 distance is smaller than the number of registers the bureau's census finds added in a year, so the bureau's figure is printed as enumerated and the index supplies no revision to it and no registers for the offices it did not canvass.\n\nExcluded are registers the index dates only to a decade, 62 of them, since no change can be shown for them. Which registers each source counts is noted in the method note in the annex.\n\nMention: \"Metropolitan Integrated Records Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Metropolitan Integrated Records Index checks the bureau's census count of registers held.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:740f22c85abbe7d44adfa64e", "occurrence_id": "78faa1d75015df6aaadc762f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Supply figures covering the whole regions derive from the Provincial Supply Chain System, the register UNODC maintains for the Russian Federation as of 2023, and the four values appear here exactly as printed.\n\nThe system puts the share of depots holding a month of the essential medicine list at the reference date at 69 per cent, the mean days a consignment waits at a provincial boundary at 4, the share of consignments to transit sites arriving intact at 88 per cent, and the share of depots reporting a cold-chain failure in the year at 12 per cent. They are reproduced as tabulated, and the system's own reading of an intact consignment comes down with them.\n\nA consignment short by one item was counted as not intact, and a failure reported without a temperature record was still counted. Each figure keeps the region it was collected in kept with it, instead of being rolled up to the federation.\n\nBeneath the table sit the system line for each value, plus the day the extract ran and the consignments each share divides.\n\nMention: \"Provincial Supply Chain System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Provincial Supply Chain System supplies medicine cover, border waits, intact arrivals and cold-chain failures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:0eaf1f539663a6558cdf883c", "occurrence_id": "5c536abdfd47de3a27c20790", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Household records were catalogued into the service file from the Follow-Up Service Delivery Index (FSDI), Edition 7, the index INCAJU compiles for Moldova, and the work done on them was cataloguing rather than measurement.\n\nEach household was given the file's own service class and its follow-up dates checked against the file's interval; where a household reports a service the file has no class for, it was parked for the analyst rather than entered under the nearest one. Households whose follow-up the index records as refused were left marked as refused rather than counted as missed, and 480 of them entered the file in that condition. The index's own edition number was kept on every household so a later edition can be matched without re-deriving it.\n\nThe service file's coverage weights come from the population register, from which the index supplied nothing; it is the household coding that lets those weights be read against contact for the delivery table the report draws up.\n\nThe coding sheet, the count of parked households and the day of deposit are held in the annex log.\n\nMention: \"Follow-Up Service Delivery Index (FSDI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The Follow-Up Service Delivery Index supplies the coded household records behind the service file.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:c8dc1f004ace80a91858d629", "occurrence_id": "aec8aadac0f094935a5c7398", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Household records were catalogued into the rural livelihoods file from the Municipal Rural Livelihoods Registry (MRLR), the register BVC maintains for Fortaleza, and the work done on them was cataloguing rather than measurement.\n\nEach household was given the file's own agro-ecological zone and its employment dates checked against the file's season window; where a household reports work the file cannot place in a season, it was parked for the analyst rather than pushed into the nearest one. Households whose plot the registry lists without coordinates were left unlocated rather than placed at the ward centroid, and 1 040 of them entered the file in that condition. The registry's own listing date was kept on every household so a later round can be matched without re-deriving it.\n\nThe livelihoods file's area weights come from the agricultural census, from which the registry supplied nothing; it is the household coding that lets those weights be read against work for the seasonal table the report draws up.\n\nThe coding sheet, the count of parked households and the day of deposit are detailed in the annex log.\n\nMention: \"Municipal Rural Livelihoods Registry (MRLR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The Municipal Rural Livelihoods Registry supplies the coded household records behind the livelihoods file.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:cbb36fa80311aa0dea41655d", "occurrence_id": "a482f1ab6ba83193fa1f59d3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Service readiness the numbers held for the divisions come from the Clustered Service Delivery and Readiness Survey (CSDRS), the survey conducted in the CHT and Sylhet divisions in 2020, and the four values are reproduced here as issued.\n\nThe survey puts the share of facilities able to treat a diarrhoeal case on the day of visit at 71 per cent, the share of the poor quintile reaching a facility within an hour at 38 per cent, the mean hours in a day a facility is open at 7, and the share of facilities with a birth attendant present at night at 46 per cent. They were not re-derived here, and the survey's own reading of ability to treat, which requires both a supply and a staff member, is carried alongside them.\n\nA facility stocked but unstaffed was counted as unable to treat, and an attendant sleeping on site was counted as present at night. Each figure keeps the upazila it was collected in named with it, instead of averaged over the divisions.\n\nWorking notes record the survey line for each value, along with the pull date and the facilities each share divides.\n\nMention: \"Clustered Service Delivery and Readiness Survey (CSDRS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Clustered Service Delivery and Readiness Survey supplies treatment ability, poor-quintile access, hours and night attendance.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:273bcd4e52036c40f4d496b5", "occurrence_id": "586474c607e2f9a634b680a4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two operations lines in the annex come from the Quarterly Operations Monitoring Registry, the register the Commerce Department keeps for Zambia as of 2020, and both are transcribed without alteration.\n\nThe registry puts the share of offices whose quarterly figures pass its internal check at 82 per cent and the mean number of revisions a quarter's figures pass through before publication at 2. Nothing here was recomputed for this report, and the registry's own reading of a revision, which counts only changes made after a check has failed, travels with the figures.\n\nThe rest of the annex lines come from the department's own returns, so the registry carries two lines while the returns carry the balance. A figure reported for one province keeps that province attached rather than being averaged over the country.\n\nThe annex log holds the registry entry, the day of extraction together with the counts each share rests on.\n\nMention: \"Quarterly Operations Monitoring Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Quarterly Operations Monitoring Registry supplies check-passing rates and revisions before publication.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:896ec2829575bf1205878aa0", "occurrence_id": "8ed2e24daa0b1189da4887a7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Social protection figures for the resident population are reported from the Civil Social Protection Survey, which FEPB conducted in Singapore in 2022, and the four values are quoted here as printed.\n\nThe survey puts the share of households receiving any means-tested support at 17 per cent, the share of recent arrivals reporting no entitlement to support at 29 per cent, the mean monthly value of support received as a share of household income at 12 per cent, and the share of recipient households reporting the application took more than a month at 21 per cent. No arithmetic was applied to them here, and the survey's own reading of a recent arrival, which follows the date of first residence, is printed beside them.\n\nA household receiving a concession rather than a payment was counted as receiving support, and an application still open at interview was counted by the time already run. Each figure keeps the planning area it was drawn from listed with it, rather than smoothed across the island.\n\nThe documentation under the table gives the survey line for each value, the date it was drawn and the households each share divides.\n\nMention: \"Civil Social Protection Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Civil Social Protection Survey supplies support receipt, entitlement gaps, value and application delay.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:42f5d723b93f21889629a8bc", "occurrence_id": "45999e5ccfe2db928ca4120b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Delivery and distance were assembled for this review out of the Linked Service Delivery Index, the index UNIDO compiled for Sierra Leone in 2017, and what follows is the review's arithmetic, not the index's.\n\nWards were sorted by the kilometres to the nearest functioning point and the share of births attended by a skilled worker computed: 74 per cent in wards under 5 kilometres against 41 per cent in wards beyond 20. The index holds a distance for each ward and an attendance share for each ward and never sets one against the other.\n\nMaking the groups alike on the number of attendants the ward is assigned brings the gap down to 12 points, which leaves the staffing carried as part of the gap the opening comparison put down to distance.\n\nWards the index lists a point for but scores as closed are dropped rather than counted as unattended, and they are 4 per cent of wards. Those references, the extraction day and the denominators are carried in the annex's method note.\n\nMention: \"Linked Service Delivery Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Linked Service Delivery Index relates distance to a service point and skilled attendance at birth.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:d0f33ee5a52f546f4232bb7a", "occurrence_id": "3e6f6265653e9ce199cc3f8f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Bargaining coverage and pay were reconstructed for this review from the Multi-Topic Collective Bargaining Survey (MCBS), Third Collection, conducted in Slovensku in 2024, and the readings below are the review's, not the survey's.\n\nEstablishments were sorted by whether a collective agreement covers their workers and the hourly wage a typical worker earns computed: 9.40 among covered establishments against 7.10 among those with no agreement. The survey holds a coverage item for each establishment and a wage for each worker and never sets one against the other.\n\nHolding the industry the establishment works in even narrows the gap to 1.20, so the industry carries part of the distance the first comparison charged to the agreement.\n\nEstablishments whose agreement had lapsed but was still being applied are dropped rather than counted as covered, and they are 6 per cent of establishments. A note in the annex gives the sorting rule, the pull date plus the base for each figure.\n\nMention: \"Multi-Topic Collective Bargaining Survey (MCBS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Multi-Topic Collective Bargaining Survey relates agreement coverage to the hourly wage.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:68715d97345552f57a1bd28b", "occurrence_id": "87580fe1c8435618ddeb0d70", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Farm income and poverty were estimated for this review using the Integrated Living Standards Survey (ILSS), the survey CDR follows in Cairo, and the figures are the review's own.\n\nHouseholds were sorted by the area they cultivate and the share falling below the national poverty line computed: 44 per cent of households cultivating under half a hectare against 12 per cent of those cultivating more than three. The survey holds a cultivated area for each household and a consumption aggregate for each household and never sets one against the other.\n\nMaking the groups alike on whether the household irrigates brings the gap down to 13 points, which leaves irrigation as part of what the unadjusted contrast attributed to the area cultivated.\n\nHouseholds that cultivate land they do not report as their own are dropped rather than counted at zero area, and they are 8 per cent of households. Working notes list the sorting rule, the extraction day, and the denominators each share rests on.\n\nMention: \"Integrated Living Standards Survey (ILSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Integrated Living Standards Survey relates cultivated area to poverty-line incidence.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:7d38650f9eb1aa6dd46cdc71", "occurrence_id": "5b43a7e4a81c05f38e16282c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The enrolment figures in this chapter are computed from the ministry's own school registers, and the Independent Learning and Equity Survey (ILES), Sixth Edition, the survey conducted in Lao PDR in 2015, is referred to only as background to them.\n\nThe survey puts the share of children of lower-secondary age enrolled at any level at 78 per cent, the share of those enrolled who are girls at 49 per cent, and the mean years of schooling a household head has completed at 5. All three are given as the survey issues them, with its own age window stated, and no chapter figure rests on them.\n\nThe chapter's tables draw on none of its responses. The figures the chapter argues from come from the school registers, which count each enrolled pupil rather than a sample of households, and where the two accounts of enrolment differ the difference is noted rather than reconciled.\n\nThe annex lists the survey tables consulted and the date the file was taken, and the survey is used for nothing beyond that background.\n\nMention: \"Independent Learning and Equity Survey (ILES)\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The Independent Learning and Equity Survey is background to the ministry's enrolment figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:7182bee20a5711a95ef2b8a9", "occurrence_id": "3b37d392c1a88b4142181af2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The transfer coverage figure in this chapter is the programme's own, taken from its payment file, and the Midline Targeting and Payments Microdata (MTPM), the extract compiled for Serbia in 2019, is cited to test it.\n\nThe programme's file shows 81 per cent of eligible households paid in the quarter. The extract shows 73 per cent for the same municipalities, because it takes eligibility from the household's own reported income at interview while the file uses the income recorded when the household enrolled, which is often older. Neither basis reworks the other's list.\n\nThe 8-point distance is smaller than the movement the programme's file shows between two quarters, so the programme's figure is printed as paid and the extract supplies no revision to it and no households for the areas its interviewers did not reach.\n\nExcluded are households the extract lists with no income item at all, 1 930 of them, since eligibility cannot be settled for them. Which households each source counts as eligible is laid out in the annex's method sheet.\n\nMention: \"Midline Targeting and Payments Microdata (MTPM)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Midline Targeting and Payments Microdata checks the programme's share of eligible households paid.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:e0ca6d6577348a8bfc593348", "occurrence_id": "2f08b1b8f9499d5759f380f6", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The background table in this section is transcribed from the Standard Education Finance Survey (SEFS), the survey Baier and Bergstrand follow in Saint Lucia, and its figures enter the report as issued.\n\nThe survey puts the share of school spending reaching the school gate at 74 per cent, the mean annual grant per pupil at 1 120, and the share of schools whose audited accounts agree with the grant received at 66 per cent. Each is given with the survey's own definition of school spending attached, and no figure elsewhere in the report is computed from them.\n\nThe section's own argument rests on the treasury's vote papers, which show what was allocated rather than what arrived, and the survey's lines stand beside them as the background against which the shortfall is read. Where the two accounts of a grant differ, both are printed and neither is adjusted to the other.\n\nThe annex lists the survey tables copied, the day of the extract, with the schools the survey's shares divide alongside.\n\nMention: \"Standard Education Finance Survey (SEFS)\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"source\", \"summary\": \"The Standard Education Finance Survey supplies the section's background table on school spending.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:501021eeb8f58972b911f098", "occurrence_id": "77cf44df77c89566186cf3e1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Movement intention municipalities figures are assembled from the Municipal Movement Intentions Registry (MMIR), Third Round, the register Fed maintains for Malaysia, and the four values are carried here as tabulated.\n\nThe register puts the share of households naming an intention to move within a year at 14 per cent, the share of those households naming a destination outside their own state at 39 per cent, the mean months a household has held its intention at 8, and the share citing the cost of housing as the reason at 31 per cent. No reweighting was applied to them, and the register's own reading of an intention, which requires the household to name a destination, is given next to them.\n\nA household wishing to move but naming no place was not counted as holding an intention, and a reason offered second was still counted. Each figure keeps the municipality it was collected in carried with it, rather than mixed across the country.\n\nThe register line for each value: the date of the pull and the households each share divides sit with the annex tables.\n\nMention: \"Municipal Movement Intentions Registry (MMIR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Municipal Movement Intentions Registry supplies intention incidence, destination, duration and stated reasons.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:668f945d689dc25013b3294d", "occurrence_id": "772e1e4827ed907f47eac516", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Hiring and vacancy figures spanning the regions are taken from the Regional Hiring and Vacancies System (RHVS), Round 2, the register kept for Moldova in 2019, and the four values pass into this report as printed.\n\nThe register puts the number of vacancies notified to the employment office in the round at 21 400, the share of notifications withdrawn by the employer at 12 per cent, the mean days a notification stands before it is filled at 42, and the share of notifications asking for a qualification at level four or above at 27 per cent. Nothing was re-estimated here, and the register's own reading of a filled notification is bound to the figures.\n\nA vacancy filled by the employer on its own and never reported back was not counted as filled, and a withdrawn notification was counted as withdrawn rather than unfilled. Each figure keeps the region it was notified in with it, rather than pooled across the country.\n\nUnder the table are the register line for each value, the date on the extract, and the notifications each share divides.\n\nMention: \"Regional Hiring and Vacancies System (RHVS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Regional Hiring and Vacancies System supplies notified vacancies, withdrawals, time to fill and skill demand.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:fcf31e86e94ba4d629ab7bf7", "occurrence_id": "340f85fc0fb4d8578ade7c4f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Whether the programme's own count of households in shock holds was tested against the Experimental Shock Coping Survey (ESCS), Phase 6, the survey NBRM follows, and the survey is the principal source for the test.\n\nThe programme's intake book records 34 500 households reporting a shock large enough to qualify for help. The survey records 41 200 for the same areas, because it asks each household about every shock in twelve months while the book records only shocks the household chose to present at a desk. Neither question was put to the other's data.\n\nThe survey therefore carries the section's count of households in shock, and the programme's smaller figure is printed beside it so the reader can see that the distance is shocks never brought forward rather than a disagreement about what a shock is. Where a household appears in both, the programme's grant record is kept and the two counts are not averaged.\n\nHouseholds reporting a shock that ended before the reference year are dropped from both rather than counted by the survey only, 8 per cent of its listings. The question each source works to is set out in the annex notes.\n\nMention: \"Experimental Shock Coping Survey (ESCS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Experimental Shock Coping Survey tests the programme's count of households reporting a shock.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:759ad9a3d8e73efa12b83a81", "occurrence_id": "6df22e664ff42e18cac629fc", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Carbon figures for the surveyed land are reported from the Enumerative Ecosystem Carbon Dataset (EECD), Wave 8, the geospatial dataset the People’s Bank of China compiled for 2017, and the four values come into this report unchanged.\n\nThe dataset puts the share of the surveyed frame under tree cover at 34 per cent, the mean tonnes of carbon held per hectare in that cover at 96, the share of the frame whose stock fell between the two waves at 21 per cent, and the share of the loss falling on peat soils at 17 per cent. No figure was recalculated in this report, and the dataset's own reading of tree cover, which requires a canopy above a fixed height, goes with the figures.\n\nA plantation below that height was not counted as cover, and a stock change inside the dataset's margin of error was counted as no change. Each figure keeps the basin it was drawn from attached, instead of being blended across the frame.\n\nThe dataset layer for each value, the day of the pull as well as the hectares each share divides are recorded under the table.\n\nMention: \"Enumerative Ecosystem Carbon Dataset (EECD)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Enumerative Ecosystem Carbon Dataset supplies canopy share, stock density, stock decline and peat loss.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:399260f136a3643e34a62dd2", "occurrence_id": "e49fec086ca7c99ff9f35ce8", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Plot records were catalogued into the coastal file from the Inter-Agency Coastal Erosion Registry (ICER), First Wave, the register maintained for Peru in 2022, and the work done on them was cataloguing rather than measurement.\n\nEach plot was given the file's own shoreline band and its survey dates checked against the file's tide window; where the registry carries a date the file cannot place against a tide table, it was parked for the analyst rather than pushed to the nearest season. Plots whose frontage the registry left blank were left unbanded rather than assumed to sit at the group median, and 730 of them entered the file in that condition. The registry's own source method was kept on every plot so a later wave can be matched without re-deriving it.\n\nThe coastal file's exposure weights come from the property census, from which the registry supplied nothing; it is the plot coding that lets those weights be read against retreat for the hazard table the report draws up.\n\nThe coding sheet, the count of parked plots and the day of deposit are kept in the annex log.\n\nMention: \"Inter-Agency Coastal Erosion Registry (ICER)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The Inter-Agency Coastal Erosion Registry supplies the coded plot records behind the coastal file.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:3f4fd42da70467deeec11f69", "occurrence_id": "799675fc2c87936da6f80325", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Coping after a flood was derived for this chapter from the Follow-Up Shock Coping Survey (FSCS), Round 9, the survey NIRA follows in Johor, and the figures below come from the chapter, not from the survey's.\n\nHouseholds were sorted by the value of property they lost and the share whose income had not returned to its pre-flood level a year later computed: 62 per cent of households losing more than half a year's income against 24 per cent of those losing less than a month's. The survey holds a loss estimate and an income item for the same household and never sets one against the other.\n\nThat comparison supports one claim, that the small-loss households should be settled first so the larger cases are not crowded out of the queue. The budget lines the proposal quotes come from the state's own relief vote, from which the survey supplied nothing.\n\nHouseholds that had not rebuilt at follow-up are dropped rather than counted as unrecovered, and they are 9 per cent of households. The bands and the counts behind each share are recorded under the table.\n\nMention: \"Follow-Up Shock Coping Survey (FSCS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Follow-Up Shock Coping Survey supports a queue-order claim built from loss size and recovery.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:29613181fa046391cd87ed5a", "occurrence_id": "264b413fa7db2bd3d20c8571", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two records lines in the county annex come from the Municipal Public Records Registry (MPRR), the register kept for Nairobi County in 2023, and both are transcribed without alteration.\n\nThe registry puts the share of requests for a public record answered within the period the law sets at 57 per cent and the mean days a refusal takes to reach the applicant at 26. Nothing here was recomputed here, and the registry's own reading of a refusal, which requires a written reply rather than silence, travels with the figures.\n\nThe annex's remaining entries follow the county's own correspondence file, so the registry carries two lines while the file carries the lines the table does not. A figure reported for one sub-county keeps that sub-county attached rather than being averaged over the county.\n\nThe annex log holds the registry entry, the day the table was pulled and the counts each share rests on.\n\nMention: \"Municipal Public Records Registry (MPRR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Municipal Public Records Registry supplies timely-answer rates and days to a written refusal.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:d8ba554f8dc175583eb9af17", "occurrence_id": "6711bfe3fab48845e6ebca13", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two absence lines in the school annex come from the Baseline Teacher Absence Survey (BTAS), Edition 3, the survey conducted in Brazil in 2024, and both are transcribed without alteration.\n\nThe survey puts the share of school days on which a class had no assigned teacher at 11 per cent and the mean days a vacancy in a teaching post stood unfilled in the year at 38. Nothing here was recalculated for this report, and the survey's own reading of an uncovered class, which requires the class to have run without a teacher rather than with a substitute, travels with the figures.\n\nThe annex's other lines are taken from the secretariat's own payroll, so the survey carries two lines while the payroll carries the lines the table does not. A municipality figure carries its municipality with it instead of being smoothed over the state.\n\nClasses covered by a teacher borrowed from another school were counted as covered, and a post whose holder was on long sick leave was still counted as unfilled. The annex log holds the survey line behind each value, the extraction date and, alongside it, the base of each share.\n\nMention: \"Baseline Teacher Absence Survey (BTAS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Baseline Teacher Absence Survey supplies uncovered class days and days a teaching post stands unfilled.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:606f3497fad766f8c17e1336", "occurrence_id": "653caec4553e0bdf2b6381d9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Informal activity and protection were assembled for this review out of the Subnational Informal Sector System, the register NBRM maintains for Costa Rica as of 2019, and what follows is the review's arithmetic, not the register's.\n\nDistricts were sorted by the share of workers the register classes as informal and the share of those workers holding any health coverage computed: 46 per cent in districts where informality exceeds a third against 72 per cent where it falls below a tenth. The register holds a classification for each worker and a coverage flag for each worker and never sets one against the other.\n\nMaking the groups alike on the mean age of workers in the district brings the gap down to 9 points, which leaves age as part of what the unadjusted difference owed to informality.\n\nWorkers the register lists in both an informal and a registered activity are dropped rather than counted as informal, and they are 5 per cent of workers. Those references, the extraction day and the denominators are held in the annex file.\n\nMention: \"Subnational Informal Sector System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Subnational Informal Sector System relates informal employment to health coverage among those workers.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:22a172e98986a8ce2f26c7f9", "occurrence_id": "b6ee9b8bcb8c6f3483f4e40f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Administrative delay was reconstructed for this review from the Midline Regional Operations Index (MROI), Collection 7, the index the Berne Union compiled for 2021, and the readings below are the review's, not the index's.\n\nOffices were sorted by the share of their staff on temporary contract and the median days a file takes to move from receipt to decision computed: 44 days at offices where temporary staff exceed a third against 27 days at offices where they fall below a tenth. The index holds a contract share for each office and a decision date for each file and never sets one against the other.\n\nHolding the number of files an office receives in the month even narrows the gap to 11 days, so the volume of work carries part of what the first comparison credits to the contract mix.\n\nOffices that moved premises during the collection are dropped rather than counted at the national median, and they are 3 per cent of offices. A note in the annex gives the sorting rule, the pull date as well as the base for each share.\n\nMention: \"Midline Regional Operations Index (MROI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Midline Regional Operations Index relates temporary staffing to days before a file is decided.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:dd4182e8c4122486c7c960fb", "occurrence_id": "23413c5185c19dc8c470e37a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two hardship lines in the welfare annex come from the Civil Minimum Needs Survey (CMNS), Collection 9, the survey MEF follows, and both are transcribed without alteration.\n\nThe survey puts the share of households that could not replace worn clothing for a child in the year at 23 per cent and the share that went without cooling equipment during the last hot spell at 41 per cent. Nothing here was recomputed in this report, and the survey's own reading of an inability, which requires the household to have wanted the item and gone without it, travels with the figures.\n\nThe annex's other entries are drawn from the ministry's own benefit register, so the survey carries two lines while the register carries the balance. No figure is folded across the sample; each one names the stratum it came from.\n\nThe annex log holds the survey line behind each value, the day the table was pulled and the counts each share divides.\n\nMention: \"Civil Minimum Needs Survey (CMNS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Civil Minimum Needs Survey supplies clothing replacement failure and lack of cooling in a hot spell.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:be6859898cb286981f945585", "occurrence_id": "0fae41269ca2a3f995ab62ef", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Protective equipment and injury were estimated for this review using the Continuous Occupational Safety Survey (COSS), Sixth Collection, the survey Treasury follows, and the figures are the review's own.\n\nEstablishments were sorted by the share of workers reported as wearing the required protective equipment and the rate of skin and eye injuries in two years computed: 2 per hundred workers at establishments above nine in ten against 7 per hundred at those below six in ten. The survey holds an equipment share for each establishment and an injury count for each establishment and never sets one against the other.\n\nMaking the groups alike on the substance a worker handles brings the gap down to 2 injuries, so the substance handled carries part of the gap the opening contrast assigned to the equipment.\n\nEstablishments that reported no exposure to any hazard were dropped rather than counted as injury-free, and they are 6 per cent of establishments. Working notes list the sorting rule, the extraction day together with the denominators each share rests on.\n\nMention: \"Continuous Occupational Safety Survey (COSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Continuous Occupational Safety Survey relates protective equipment use to skin and eye injuries.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:9945e6c1596cbdc85e95ac3d", "occurrence_id": "ed3af8ff17b408cbf1d4cf75", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Service timeliness was assembled for this review out of the High-Frequency Programme Monitoring Survey (HPMS), Third Version, the survey GBVIMS follows, and what follows is the review's arithmetic, not the survey's.\n\nSites were sorted by how often the programme visits them and the share of scheduled sessions actually held in the period computed: 88 per cent at sites visited monthly or more often against 59 per cent at sites visited twice a year or less. The survey holds a visit count for each site and a session flag for each session and never sets one against the other.\n\nMaking the groups alike on the distance from the programme's office brings the gap down to 12 points, which leaves remoteness as part of the distance the first comparison ascribed to the visit frequency.\n\nSites opened within the last two months of the period are dropped rather than counted at the average, and they are 7 per cent of sites. Those references, the extraction day and the denominators sit in the papers behind the annex.\n\nMention: \"High-Frequency Programme Monitoring Survey (HPMS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The High-Frequency Programme Monitoring Survey relates visit frequency to sessions actually held.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:2f6899ff29fd734e2609a3cf", "occurrence_id": "fa12f9b846c3583d13632e62", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Parcel records were catalogued into the land file from the Cross-Sectoral Remote Sensing Dataset (CRSD), Wave 7, the geospatial dataset compiled for Moldova in 2015, and the work done on them was cataloguing rather than measurement.\n\nEach parcel was given the file's own cover class and its acquisition dates checked against the file's season window; where the dataset carries a class the file has no code for, it was parked for the analyst rather than entered under the nearest one. Parcels the dataset flags as cloud-obscured were left unclassified rather than assumed to carry the previous wave's class, and 1 420 of them entered the file in that condition. The dataset's own sensor and pass were kept on every parcel so a later wave can be matched without re-deriving it.\n\nThe land file's area weights come from the cadastral census, from which the dataset supplied nothing; it is the parcel coding that lets those weights be read against cover change for the land-use table the report draws up.\n\nThe coding sheet, the count of parked parcels and the day of deposit are laid out in the annex log.\n\nMention: \"Cross-Sectoral Remote Sensing Dataset (CRSD)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The Cross-Sectoral Remote Sensing Dataset supplies the coded parcel records behind the land file.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:3d99a73f1173b2b4c7ca528e", "occurrence_id": "2fc45e371618831d7d680e3d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Insurance figures at the region level are taken from the Analytical Unemployment Insurance System, the register EGSII maintains for Ile de France in 2021, and the four values enter this report as they stand.\n\nThe register puts the share of registered unemployed persons drawing a benefit at 46 per cent, the mean monthly benefit as a share of the regional median wage at 38 per cent, the share of claimants whose entitlement ended without work at 22 per cent, and the mean months a claim runs before it closes at 7. They are carried as published, and the register's own reading of a closed claim, which requires the office to have recorded a reason, travels with the values.\n\nA claim closed for a missing document was counted as closed without work, and a person working a short contract while claiming was still counted as drawing. Each figure keeps the department it was collected in shown with it, instead of combined into one region figure.\n\nThe annex note carries the register line for each value, plus the day of the pull and the claimants each share divides.\n\nMention: \"Analytical Unemployment Insurance System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Analytical Unemployment Insurance System supplies benefit receipt, benefit level, closure reason and claim length.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:189c07fe4f91922e2d462f2d", "occurrence_id": "e7b98c5c1f26d46703940c42", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Programme coverage was derived for this chapter from the Municipal Programme Monitoring Census (MPMC), Edition 7, the census the Ministry of Finance follows in Pakistan, and the figures below are the chapter's own rather than the census's.\n\nMunicipalities were sorted by the number of schemes they implement and the share of planned activities recorded as complete computed: 79 per cent at municipalities running three schemes or fewer against 54 per cent at those running eight or more. The census holds a scheme count and an activity flag for each municipality and never sets one against the other.\n\nThat comparison supports one claim, that the consolidation of small schemes should precede any increase in their grants. The staffing lines the proposal quotes come from the ministry's own establishment return, from which the census supplied nothing.\n\nMunicipalities that did not return the census form are dropped rather than counted as incomplete, and they are 6 per cent of municipalities. The bands and the counts behind each share are recorded under the table.\n\nMention: \"Municipal Programme Monitoring Census (MPMC)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Municipal Programme Monitoring Census supports a scheme-consolidation claim built from scheme count and completion.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:acc28d9dd7b0e02456f9fde6", "occurrence_id": "919deb860a5fb986e22c1fac", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two land-cover lines in the annex come from the Small-Area Remote Sensing Survey (SRSS), the survey the German Socioeconomic Panel follows in Timor-Leste, and both are transcribed without alteration.\n\nThe survey puts the share of the surveyed area under rainfed cropping at the last observation at 37 per cent and the share whose cover changed class between the two observations at 9 per cent. Nothing here was recalculated in this report, and the survey's own reading of a class change, which requires the change to exceed its minimum mapping area, travels with the figures.\n\nThe remaining annex lines come from the ministry's own field cards, so the survey carries two lines while the cards carry the balance. Nothing is averaged over the country: a figure for one municipality keeps that municipality named.\n\nThe annex log holds the survey line behind each value, the day the rows were taken and the counts under each share.\n\nMention: \"Small-Area Remote Sensing Survey (SRSS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Small-Area Remote Sensing Survey supplies rainfed cropping share and cover-change share.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:71d1ef9ee6f51135227e2971", "occurrence_id": "b4dcdd81607eca482faeed15", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The benefit figure in this chapter is the office's own, taken from its payment ledger, and the Endline Unemployment Insurance Registry, the register Chinese Customs maintains for Uruguay as of 2023, is cited to test it.\n\nThe office's ledger shows 58 300 people drawing a benefit in the reference month. The registry shows 52 900 for the same offices, because it drops a person whose payment was recalled within the month, which the ledger keeps until the recall is settled. Neither rule was applied to the other's count.\n\nThe 5 400 distance is smaller than the movement the office's ledger shows between two months, so the office's figure is printed as paid and the registry supplies no revision to it and no persons for the offices it did not canvass.\n\nExcluded are persons the registry dates only to a quarter, 810 of them, since the reference month cannot be fixed for them. Which persons each source counts is written up in the working papers behind the annex.\n\nMention: \"Endline Unemployment Insurance Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Endline Unemployment Insurance Registry checks the office's count of people drawing a benefit.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:b298bb3b7730fc4aceb9f6f1", "occurrence_id": "2e97bac1c462ffd57689657a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Household records were catalogued into the exposure file from the Civil Disaster Exposure System (CDES), Sixth Wave, the register NIRA maintains, and the work done on them was cataloguing rather than measurement.\n\nEach household was given the file's own hazard band and its damage dates checked against the file's event window; where a household reports damage the file cannot attach to an event, it was parked for the analyst rather than pushed to the nearest spell. Households whose exposure the system lists without a structure count were left unbanded rather than assumed to sit at the local median, and 1 150 of them entered the file in that condition. The system's own wave number was kept on every household so a later wave can be matched without re-deriving it.\n\nThe exposure file's loss weights come from the population and housing census, from which the system supplied nothing; it is the household coding that lets those weights be read against damage for the risk table the report draws up.\n\nThe coding sheet, the count of parked households and the day of deposit are filed in the annex log.\n\nMention: \"Civil Disaster Exposure System (CDES)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The Civil Disaster Exposure System supplies the coded household records behind the exposure file.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:9fe255db436365852e7398a2", "occurrence_id": "4bd59f1f73f8d02468b4263d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Statelessness figures for the surveyed areas are reported from the Comparative Statelessness Registry (CSR), Edition 5, the register ITESM maintains, and the four values reach this report as published.\n\nThe register puts the share of recorded persons holding no recognised identity document at 12 per cent, the share whose place of enumeration lies outside any registered settlement at 8 per cent, the mean years a person has lived in the area without documentation at 11, and the share of areas where a documentation office sits within 10 kilometres at 44 per cent. Nothing here was recomputed, and the register's own reading of a recognised document follows the figures.\n\nA document expired but renewable was still counted as recognised, and a settlement registered only on paper was counted as registered. Each figure keeps the enumeration area it was drawn from reported with it, rather than blended into a single register share.\n\nThe table's footnotes hold the register line for each value, along with the day the extract ran and the persons each share divides.\n\nMention: \"Comparative Statelessness Registry (CSR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Comparative Statelessness Registry supplies undocumented shares, settlement status, duration and office access.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:6d955998b4a7f295dd3782e7", "occurrence_id": "3ec59d421e336f4256a367ce", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The laboratory figures in this chapter are drawn from the network's own returns, and the Experimental Diagnostic Capacity Report (EDCR), the report ITESM issued for 2022, is referred to only as background to them.\n\nThe report puts the share of facilities in its coverage able to run a basic panel of tests at 51 per cent, the mean days a sample waits before it is tested at 6, and the share of results returned to the requesting clinician within a fortnight at 63 per cent. All three are given as the report issues them, with its own list of tests stated, and no chapter figure rests on them.\n\nThe chapter builds no table from its entries. The figures the chapter argues from come from the network returns, which cover each member laboratory rather than a sample of facilities, and where the two accounts of capacity differ the difference is noted rather than reconciled.\n\nThe annex lists the report tables consulted and the pull date, and the report is used for nothing beyond that background.\n\nMention: \"Experimental Diagnostic Capacity Report (EDCR)\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The Experimental Diagnostic Capacity Report is background to the network's own laboratory figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:701ddbb1cad1b01d71a559bb", "occurrence_id": "017e51031997e02a14e925f2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The all-weather coverage figure in this chapter is the council's own, taken from its road register, and the Linked Road Segment Dataset (LRSD), the geospatial dataset compiled for Lusaka in 2017, is cited to test it.\n\nThe council's register shows 72 per cent of its segments surfaced to an all-weather standard. The dataset shows 61 per cent for the same segments, because it classes a segment as all-weather only where the surface holds through the wet season, which the register does not test. Neither rule was applied to the other's list.\n\nThe 11-point distance is larger than the change the council's register shows between two years, so the chapter prints the dataset's lower figure for segments that hold in the wet season and the council's higher one for segments surfaced at the year's start, and the dataset supplies no segments for the wards its survey did not enter.\n\nExcluded are segments the dataset records as washed out at one observation, 46 of them, since neither standard can be settled for them. Which segments each source counts is detailed in the annex working papers.\n\nMention: \"Linked Road Segment Dataset (LRSD)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Linked Road Segment Dataset checks the council's share of all-weather road segments.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:620872672beaa5967d4044dd", "occurrence_id": "2515bc3da9878ce20f831f5c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Campaign reach and refusal were assembled for this review out of the Subnational Vaccination Campaign Census, the census MSCI conducted in Turkey in 2024, and what follows is the review's arithmetic, not the census's.\n\nNeighbourhoods were sorted by the number of teams the campaign posted and the share of children the census finds unvaccinated computed: 14 per cent in neighbourhoods with three teams or more against 29 per cent in those with a single team. The census holds a team count for each neighbourhood and a vaccination flag for each child and never sets one against the other.\n\nMaking the groups alike on the share of households new to the neighbourhood brings the gap down to 8 points, which leaves recent arrival as part of what the raw comparison laid at the door of the number of teams.\n\nNeighbourhoods where the census team was refused entry are dropped rather than counted as wholly unvaccinated, and they are 3 per cent of neighbourhoods. Those references, the extraction day and the denominators are listed in the method note in the annex.\n\nMention: \"Subnational Vaccination Campaign Census\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Subnational Vaccination Campaign Census relates teams posted to the share of children unvaccinated.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:7ae6a7926e539a19f50b8430", "occurrence_id": "13e79ff4edc54f56c9c7cfc6", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Whether the ministry's own figure for reported incidents holds was tested against the Midline Statistical Bulletin Report (MSBR), Fifth Release, the report the Uppsala Conflict Data Programme issues, and the report is the principal source for the test.\n\nThe ministry's returns give 4 310 incidents recorded in the period. The report gives 5 720 for the same districts, because it counts an incident reported by any of three channels while the returns count only those reaching the ministry's own desk. No channel test was applied to the report's count for the comparison.\n\nThe report therefore carries the section's count of incidents, and the ministry's smaller figure is printed beside it so the reader can see that the distance is reports arriving elsewhere rather than a disagreement about what an incident is. Where an incident appears in both, the ministry's classification is kept and the two counts are not averaged.\n\nIncidents dated by the report to a month only are dropped from both rather than counted by the report alone, 9 per cent of its listings. The channels each source reads are set out in the annex's method note.\n\nMention: \"Midline Statistical Bulletin Report (MSBR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Midline Statistical Bulletin Report tests the ministry's count of recorded incidents.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:0a2a662ab4a9c4a1229d2023", "occurrence_id": "9f3690156441ec8ceb9aaaf3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two boundary lines in the annex come from the Pilot Ward Outline Report (PWOR), the report issued for Lusaka in 2017, and both are transcribed without alteration.\n\nThe report puts the share of wards whose outline was agreed by both the council and the statistics office at 68 per cent and the mean number of disputes recorded per ward outline at 1. Nothing here was re-derived for this report, and the report's own reading of an agreed outline, which requires a signature from each side rather than a filed objection, travels with the figures.\n\nThe annex's remaining lines are compiled from the census office's own frame listing, so the report carries two lines while the listing supplies the balance. Figures stay at the constituency level rather than being pooled over the city.\n\nAn outline settled at a meeting but never signed was counted as disputed, and a boundary moved after the report's cut-off date was not counted at all. The annex log holds the report section behind each value, the day of extraction plus the wards the shares rest on.\n\nMention: \"Pilot Ward Outline Report (PWOR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Pilot Ward Outline Report supplies agreed-outline rates and disputes recorded per outline.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:84fa374ab461e0776867851e", "occurrence_id": "999b6eb9bb2b6f3ba260069f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Public records country-level figures come from the Recurrent Public Records Registry (RPRR), Wave 9, the register Townsend maintains for France, and the four values appear here exactly as printed.\n\nThe register puts the share of authorities publishing a register of their holdings at 57 per cent, the mean days before a request is acknowledged at 9, the share of requests refused in whole or in part at 24 per cent, and the share of refusals carrying a reason the applicant could act on at 61 per cent. They are reproduced as tabulated, and the register's own reading of an acknowledgement is reported with them.\n\nAn automatic receipt was not counted as an acknowledgement, and a refusal giving only a statutory heading was not counted as actionable. Each figure keeps the authority class it was drawn from kept with it, instead of being rolled up to the country.\n\nUnder the table you will find the register line for each value, the date it was drawn and the requests each share divides.\n\nMention: \"Recurrent Public Records Registry (RPRR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Recurrent Public Records Registry supplies published holdings, acknowledgement delay, refusals and reasons.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:dbcfe5fa06eeaf2ee9d1a96a", "occurrence_id": "9f23698033cadf9298c3c820", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The site population figure in this chapter is the coordination's own, taken from its registration book, and the Decennial Refugee and Migrant Census (DRMC), the census conducted in the Tigray Region in 2023, is cited to test it.\n\nThe coordination's book carries 96 400 individuals registered across the sites. The census carries 108 900 for the same sites, because it counts a person present at the time of enumeration who had not yet registered, which the book cannot show until the family is written up. Neither basis was applied to the other's count.\n\nThe 12 500 distance is smaller than the number the coordination's book adds in a month at the peak of arrivals, so the coordination's figure is printed as registered and the census supplies no revision to it and no persons for the sites its enumerators could not reach.\n\nExcluded are persons the census records without a site of origin, 2 140 of them, since the two counts cannot be matched for them. Which persons each source counts is recorded in the annex's method sheet.\n\nMention: \"Decennial Refugee and Migrant Census (DRMC)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Decennial Refugee and Migrant Census checks the coordination's registered site population.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:14cf460b59615a23d90abd40", "occurrence_id": "c14c851d43898a487ee1a1a4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Labour force figures for each part of the municipality are drawn from the Monthly Labour Force Survey, which the Central Electricity Boards conducted in Beijing in 2022, and the four values are reproduced here as issued.\n\nThe survey puts the share of adults aged 16 and over in paid work at 68 per cent, the share of those not working who are seeking work at 6 per cent, the share of workers holding a tertiary qualification at 43 per cent, and the mean weeks a jobseeker has been seeking at 13. They were not re-derived here, and the survey's own reading of paid work, which requires an hour worked in the reference week, is attached to them.\n\nA person unpaid in a family firm was counted as not in paid work, and a student who took a weekend job was counted as working. Each figure keeps the district it was collected in named with it, instead of averaged over the municipality.\n\nThe annex keeps the survey line for each value: the date the file was taken and the adults each share divides.\n\nMention: \"Monthly Labour Force Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Monthly Labour Force Survey supplies employment, unemployment, qualification and search-duration values.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:bc32c030a7685c0b46b4f2fa", "occurrence_id": "0425e8c95f2551d8f24c9abe", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Statistical production figures measured across the offices are read from the Inter-Agency Statistical Bulletin Registry (ISBR), First Edition, the register the Hungarian NHIF maintains, and the four values are quoted here as printed.\n\nThe registry puts the share of planned bulletins issued on their announced date at 58 per cent, the mean days a delayed bulletin runs past that date at 26, the share of bulletins issued with a method note attached at 71 per cent, and the share whose figures were revised after issue at 19 per cent. No arithmetic was applied to them here, and the registry's own reading of a revision is stated with them.\n\nA correction of a typographical error was not counted as a revision, and a bulletin issued early was counted as issued on its date. Each figure keeps the office that issued it listed with it, rather than smoothed across the agencies.\n\nBeneath the table sit the registry line for each value, the day of the extract, and the bulletins each share divides.\n\nMention: \"Inter-Agency Statistical Bulletin Registry (ISBR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Inter-Agency Statistical Bulletin Registry supplies on-time issue, delay, method notes and post-issue revisions.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:24a6d4e64308e0ef5af5038b", "occurrence_id": "9da60dc7d3ca80776d4611a1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Migration and remittance figures for the city are reported from the Panel Migration and Remittance Index, the index Coriolis maintains for Chengdu as of 2022, and the four values are carried here as tabulated.\n\nThe index puts the share of households with a member working outside the city at 27 per cent, the mean share of household income arriving from that member at 31 per cent, the mean months a migrant returns in a year at 1, and the share of sending households with no member aged under 16 at 49 per cent. No reweighting was applied to them, and the index's own reading of a sending household comes with them.\n\nA student studying in a neighbouring city was counted as a migrant, and a transfer spent in the city by the migrant was not counted as remitted income. Each figure keeps the county it was drawn from carried with it, rather than mixed across the city.\n\nWorking notes record the index line for each value, the date of the pull as well as the households each share divides.\n\nMention: \"Panel Migration and Remittance Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Panel Migration and Remittance Index supplies sending incidence, remittance share, return and household composition.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:69e8411dd170217f238d8150", "occurrence_id": "7ee1e61c6a3a523bc52e2607", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Training and placement were reconstructed for this review from the Rural Vocational Training Survey (RVTS), Collection 2, conducted in Ecuador in 2015, and the readings below are the review's, not the survey's.\n\nGraduates were sorted by the length of the course they completed and the share working in the trade they trained for two years later computed: 52 per cent of those completing a course of a year or more against 27 per cent of those completing under three months. The survey holds a course length and a destination for the same graduate and never sets one against the other.\n\nHolding the distance from the graduate's home to the nearest worksite even narrows the gap to 12 points, so distance carries part of the gap the opening comparison put down to the length of the course.\n\nGraduates who took up further study rather than work are dropped rather than counted as outside the trade, and they are 6 per cent of graduates. A note in the annex gives the sorting rule, the pull date, and the base for each share.\n\nMention: \"Rural Vocational Training Survey (RVTS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Rural Vocational Training Survey relates course length to working in the trained trade.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:3afa2f13a30cbf099023f1db", "occurrence_id": "f9c15bda1004094d6347f5b0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Insurance figures for the metropolitan area are reported from the Endline Unemployment Insurance Survey (EUIS), Collection 3, the survey conducted in Houston in 2021, and the four values pass into this report as printed.\n\nThe survey puts the share of unemployed persons who have ever claimed at 61 per cent, the mean weeks a claim remained open at 14, the share of claimants who found work before the claim closed at 47 per cent, and the share of claimants leaving the area while claiming at 7 per cent. Nothing was re-estimated here, and the survey's own reading of an open claim is kept with them.\n\nA claim reopened within a year of closing was counted as one claim, and a person who moved but kept claiming was counted as leaving the area. Each figure keeps the ward it was collected in with it, rather than pooled across the area.\n\nThe documentation under the table gives the survey line for each value, plus the date on the extract and the claimants each share divides.\n\nMention: \"Endline Unemployment Insurance Survey (EUIS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Endline Unemployment Insurance Survey supplies claim experience, duration, exit into work and out-migration.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:d196b66dbf8885bf4790eb7c", "occurrence_id": "135ccad73e1e42da69c6dede", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two administrative-quality lines in the annex come from the Rapid Administrative Survey Index (RASI), Second Release, the index compiled for Costa Rica in 2018, and both are transcribed without alteration.\n\nThe index puts the share of registers whose completeness was estimated in the last three years at 44 per cent and the mean share of a register's records missing a date field at 6 per cent. Nothing here was worked out again here, and the index's own reading of an estimate, which requires a published figure rather than an internal note, travels with the figures.\n\nThe other lines in the annex are copied from the statistics office's own register inventory, so the index carries two lines while the inventory holds the remainder. A figure reported for one register family keeps that family attached rather than being averaged across the office.\n\nA register whose estimate covered one province only was not counted as estimated, and a date field carrying a default value was counted as filled. The counts behind both lines, the index section they come from and the day of the pull are stored in the annex log.\n\nMention: \"Rapid Administrative Survey Index (RASI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Rapid Administrative Survey Index supplies completeness-estimation coverage and missing date fields.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:bb978a0ed1a893aba8255a5e", "occurrence_id": "e8986ef6c93736cd2858f656", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Supply figures for the reception sites are reported from the Rapid Supply Chain Survey, which the Ministry of Health conducted in Riga in 2022, and the four values come into this report unchanged.\n\nThe survey puts the share of sites holding a week of essential supplies at the visit at 73 per cent, the mean days a requisition takes to reach the site from the depot at 5, the share of sites whose stock was counted by the site's own worker at 66 per cent, and the share reporting a stock-out of a paediatric item in the month at 18 per cent. No figure was recalculated in this report, and the survey's own reading of a counted stock accompanies the figures.\n\nA stock figure taken from the site's ledger rather than the shelf was not counted as counted by the worker, and a shortage lasting a single day was counted as a stock-out. Each figure keeps the site it was collected at attached, instead of being blended across the city.\n\nThe survey line for each value, along with the extraction day and the sites each share divides sit with the annex tables.\n\nMention: \"Rapid Supply Chain Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Rapid Supply Chain Survey supplies supply cover, requisition delay, stock checks and paediatric stock-outs.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:7d26ee258b7ea985d75f0c9b", "occurrence_id": "76cad505213dd9442717f24a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Operations the districts figures are compiled from the District Operations Index Registry, the register APRA maintains for Lebanon as of 2020, and the four values enter this report as they stand.\n\nThe register puts the share of districts filing a monthly situation report at 81 per cent, the mean days between the month closing and the report arriving at 12, the share of reports carrying a figure for each active service at 58 per cent, and the share of districts whose report was returned for correction at 22 per cent. They are carried as published, and the register's own reading of a filed report comes down with them.\n\nA report sent incomplete and completed later was counted as filed on the later date, and a correction raised by the district itself was counted as a return. Each figure keeps the governorate it belongs to shown with it, instead of combined into one country figure.\n\nUnder the table are the register line for each value, the extraction date, with the districts each share divides alongside.\n\nMention: \"District Operations Index Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The District Operations Index Registry supplies reporting rates, delay, service completeness and returns for correction.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:78888c91dedaa4d6ef720b27", "occurrence_id": "3b78fd46e2c323f9b92c14db", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Whether the office's own count of undocumented residents holds was tested against the Municipal Statelessness Registry, the register Banco Central de Chile maintains for Germany as of 2021, and the registry is the principal source for the test.\n\nThe office's case file lists 18 700 residents recorded without a recognised nationality. The registry lists 24 300 for the same municipalities, because it counts a person whose file was opened on a provisional document, which the office moves to a separate ledger before the year closes. Neither rule was applied to the other's file.\n\nThe registry therefore carries the section's count of undocumented residents, and the office's smaller figure is printed beside it so the reader can see that the distance is provisional files rather than a disagreement about who lacks a nationality. Where a person appears in both, the office's decision date is kept and the two counts are not averaged.\n\nResidents whose file the registry dates only to a year are dropped from both rather than counted by the registry alone, 6 per cent of its listings. The document test each source works to is listed in the annex notes.\n\nMention: \"Municipal Statelessness Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Municipal Statelessness Registry tests the office's count of residents without a recognised nationality.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:aef861a42c846b7ad072fc11", "occurrence_id": "3ede48ac01fdb99238394148", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The employment figure in this chapter is the programme's own, taken from its muster rolls, and the National Public Works Census (NPWC), Collection 2, the census conducted in Ghana in 2021, is cited to test it.\n\nThe programme's rolls show 143 000 workers engaged on public works in the year. The census shows 121 600 for the same works, because it counts a worker only where the worker was present for at least twenty days, which the rolls do not test. Neither rule was applied to the other's count.\n\nThe 21 400 distance is smaller than the number the programme's rolls add between two seasons, so the programme's figure is printed as enrolled and the census supplies no revision to it and no workers for the works its enumerators did not visit.\n\nExcluded are workers the census records at two works in the same season, 1 870 of them, since the rolls cannot show which engagement is theirs. Which workers each source counts is set down in the annex file.\n\nMention: \"National Public Works Census (NPWC)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The National Public Works Census checks the programme's count of workers engaged on public works.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:f3078a1f99d8ecd1330d018d", "occurrence_id": "d8111b3c692b755d68104716", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Teacher workforce figures for the programme areas are reported from the High-Frequency Teacher Workforce System (HTWS), Version 6, the register Univer sity of Minnesota maintains, and the four values reach this report as published.\n\nThe register puts the share of schools with a teacher assigned to every class at 79 per cent, the mean months a vacant teaching post stands unfilled at 6, the share of teachers holding a trained qualification at 68 per cent, and the share of schools whose roll includes children recorded as displaced at 23 per cent. Nothing here was recomputed, and the register's own reading of an assigned teacher is carried alongside them.\n\nA teacher covering two classes in the same hour was counted as assigned to neither, and a qualification under apostille was counted as trained. Each figure keeps the area it was collected in reported with it, rather than blended into a single programme share.\n\nThe register line for each value: the day the table was pulled and the schools each share divides are recorded under the table.\n\nMention: \"High-Frequency Teacher Workforce System (HTWS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The High-Frequency Teacher Workforce System supplies class coverage, post vacancy, qualifications and displaced enrolment.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:725fad94454f9153d5a6add1", "occurrence_id": "72c1949aa3d2565cb625349c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Delivery and contact were assembled for this review out of the Verified Service Delivery Registry, the register GGDC maintains for Papua New Guinea as of 2015, and what follows is the review's arithmetic, not the register's.\n\nService points were sorted by the number of contact days they offer in a month and the share of registered clients reached in the period computed: 71 per cent at points offering sixteen contact days or more against 43 per cent at those offering six or fewer. The register holds a contact-day count for each point and a reached flag for each client and never sets one against the other.\n\nMaking the groups alike on the travel time from the point to its farthest village brings the gap down to 11 points, which leaves travel time as part of the distance the first comparison charged to the contact days.\n\nService points closed for part of the period are dropped rather than counted as reaching no one, and they are 5 per cent of points. Those references, the extraction day and the denominators are filed in the papers behind the annex.\n\nMention: \"Verified Service Delivery Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Verified Service Delivery Registry relates monthly contact days to the share of clients reached.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:02a85da7c9e1db3e600951be", "occurrence_id": "6bf12e7786a26f7eec4700b0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Energy spending and shortfall were estimated for this review using the Biennial Fuel and Energy Use Survey, which US Customs conducted in the Arab Republic of Egypt in 2024, and the figures are the review's own.\n\nHouseholds were sorted by the share of their spending that goes to fuel and power and the share reporting they went without heating or cooling for a day in the last month computed: 38 per cent of households above a tenth against 12 per cent of those below a twentieth. The survey holds a spending item for each household and a deprivation item for the same household and never sets one against the other.\n\nMaking the groups alike on the size of the dwelling brings the gap down to 9 points, which leaves dwelling size as part of what the unadjusted contrast attributed to the fuel share.\n\nHouseholds that receive fuel free under a scheme are dropped rather than counted at zero spending, and they are 7 per cent of households. Working notes list the sorting rule, the extraction day and the denominators each share rests on.\n\nMention: \"Biennial Fuel and Energy Use Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Biennial Fuel and Energy Use Survey relates the fuel spending share to going without heating or cooling.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:34384924ca76b411b15c5d77", "occurrence_id": "89c18abc30125029c51a8b22", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Parcel records were catalogued into the urban file from the Quality-Assured Parcel Footprint System, the register CLGF maintains for Africa as of 2015, and the work done on them was cataloguing rather than measurement.\n\nEach parcel was given the file's own tenure class and its survey dates checked against the file's capture window; where the system carries a tenure statement the file has no class for, it was parked for the analyst rather than entered under the nearest one. Parcels the system lists without an area were left unclassed rather than assumed to sit at the block median, and 2 310 of them entered the file in that condition. The system's own assurance flag was kept on every parcel so a later edition can be matched without re-deriving it.\n\nThe urban file's density weights come from the population and housing census, from which the system supplied nothing; it is the parcel coding that lets those weights be read against tenure for the land table the report draws up.\n\nThe coding sheet, the count of parked parcels and the day of deposit are written up in the annex log.\n\nMention: \"Quality-Assured Parcel Footprint System\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The Quality-Assured Parcel Footprint System supplies the coded parcel records behind the urban land file.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:af5cb2e20116fb0d43515cb8", "occurrence_id": "9c50b5d3d60e9f0254573e1b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Vegetation stress and yield were reconstructed for this review from the Government Vegetation Condition Collection (GVCC), Release 5, the collection compiled for Florida in 2016, and the readings below are the review's, not the collection's.\n\nParish farms were matched to their condition class and grouped by the lowest index the collection records in the season, and the share of the crop the farmer reports as lost computed: 31 per cent of farms whose index fell below the stress threshold against 9 per cent of those that stayed above it. The collection grades an index per parcel and the farm returns record the loss per holding, and neither reports the two together.\n\nHolding whether the parcel was irrigated even narrows the gap to 12 points, so irrigation carries part of what the unadjusted difference owed to the index.\n\nParcels the collection classes as non-crop were dropped rather than counted as losses, and they are 8 per cent of parcels. A note in the annex gives the classing rule, the pull date and, alongside it, the base for each share.\n\nMention: \"Government Vegetation Condition Collection (GVCC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Government Vegetation Condition Collection relates in-season index stress to reported crop loss.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:7dcf73b4d7c3515b32e94691", "occurrence_id": "d8dd5926534309e34862968f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Regional operations figures for the whole offices are sourced from the Government Regional Operations Registry (GROR), Eighth Phase, the register WHO maintains, and the four values appear here exactly as printed.\n\nThe register puts the share of offices submitting a complete monthly return at 76 per cent, the mean days between the month closing and the return arriving at 15, the share of returns carrying a figure for each priority programme at 62 per cent, and the share of offices whose return was queried once filed at 24 per cent. They are reproduced as tabulated, and the register's own reading of a complete return is printed beside them.\n\nA return filed with one programme left blank was counted as incomplete, and a query raised by the office on its own figures was counted as a query. Each figure keeps the region it was drawn from kept with it, instead of being rolled up to the offices.\n\nThe annex note carries the register line for each value, the day of the pull, and the offices each share divides.\n\nMention: \"Government Regional Operations Registry (GROR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Government Regional Operations Registry supplies complete-return rates, delay, programme coverage and queries.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:e1524c88494b4e1c0bda2751", "occurrence_id": "df54ea7e9516f67ce2602b98", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Debt and saving were assembled for this review out of the Monthly Debt and Savings Survey (MDSS), the survey УВКБ ООН, REACH follows in Guatemala, and what follows is the review's arithmetic, not the survey's.\n\nHouseholds were sorted by the share of their income pledged to repayments and the share managing to set anything aside in a month computed: 8 per cent of households above a quarter against 34 per cent of those below a tenth. The survey holds a repayment item for each household and a saving item for the same household and never sets one against the other.\n\nMaking the groups alike on whether the household holds a savings account brings the gap down to 11 points, which leaves the account as part of what the first comparison credits to the repayment share.\n\nHouseholds reporting a repayment but no loan were dropped rather than counted at zero pledge, and they are 4 per cent of households. Those references, the extraction day and the denominators are held in the working papers behind the annex.\n\nMention: \"Monthly Debt and Savings Survey (MDSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Monthly Debt and Savings Survey relates pledged income to households able to save.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:0684d0a220db24c133107749", "occurrence_id": "d5c61f10c64ee75b1abb69ab", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two business lines in the labour annex come from the Decennial Business Demography Registry (DBDR), Version 5, the register kept for Lagos State in 2015, and both are transcribed without alteration.\n\nThe registry puts the share of registered enterprises surviving their third year at 46 per cent and the mean workers an enterprise holds at the point of registration at 4. Nothing here was computed afresh for this report, and the registry's own reading of survival, which requires the enterprise to have filed in each of the three years, travels with the figures.\n\nThe annex's remaining entries follow the state's own licensing file, so the registry carries two lines while the file carries what is left over. Each sector figure is reported for its own sector, not pooled across the state.\n\nThe annex log holds the registry entry behind each value, the day the rows were taken and the population each share divides.\n\nMention: \"Decennial Business Demography Registry (DBDR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Decennial Business Demography Registry supplies three-year survival and workers at registration.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:cdc20bd2b3b00f6b88acb201", "occurrence_id": "5f89376c454219c952ef16a8", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Child wellbeing figures for the metropolitan area are reported from the Experimental Child Wellbeing Survey (ECWS), Phase 9, the survey conducted in Houston in 2017, and the four values are reproduced here as issued.\n\nThe survey puts the share of children living in a household below the area's adequacy line at 21 per cent, the share whose dwelling lies within a flood-marked block at 14 per cent, the share of children with a bedroom of their own at 52 per cent, and the share of households reporting the child missed health care for want of transport at 9 per cent. They were not re-derived here, and the survey's own reading of a bedroom of a child's own is given next to them.\n\nA child sharing a room with a cousin was not counted as having a room, and a block marked only in the survey's own map was counted as flood-marked. Each figure keeps the census tract it was collected in named with it, instead of averaged over the area.\n\nThe table's footnotes hold the survey line for each value, the day the extract ran as well as the children each share divides.\n\nMention: \"Experimental Child Wellbeing Survey (ECWS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Experimental Child Wellbeing Survey supplies adequacy, flood-block location, personal space and missed care.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:151fc7e80f5000aa7f52603e", "occurrence_id": "fa08c6fca802e0b7afb165ee", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Exposure the counties figures are tabulated by the Combined Disaster Exposure System, the register COFECE maintains for Kenya as of 2017, and the four values are quoted here as printed.\n\nThe register puts the share of households in the flood-marked frame at 19 per cent, the share of those households whose dwelling sits on unmapped land at 26 per cent, the mean years a household has lived in the exposed frame at 12, and the share of the frame covered by an active warning point at 41 per cent. No arithmetic was applied to them here, and the register's own reading of the exposed frame, which requires two matched observations, is bound to the figures.\n\nA household present at one observation only was not counted as exposed, and a warning point tested but never used was counted as active. Each figure keeps the county it was drawn from listed with it, rather than smoothed across the country.\n\nUnder the table you will find the register line for each value, plus the pull date and the households each share divides.\n\nMention: \"Combined Disaster Exposure System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Combined Disaster Exposure System supplies exposed-household share, unmapped land, duration and warning cover.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:83ce15a03b1f6b3840d70323", "occurrence_id": "9f4f49b357596cbd74dd3266", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Incidence reporting and outcome were reconstructed for this review from the Small-Area Protection Incident Registry, the register IBGE maintains for Mexico City in 2021, and the readings below are the review's, not the register's.\n\nBlocks were sorted by the number of protection incidents the register records in a quarter and the share of households in the block reporting a forced move in the following quarter computed: 7 per cent of blocks above six incidents against 1 per cent of those recording none. The register holds an incident count per block and the survey holds a move flag per household, and neither reports the two together.\n\nHolding the mean rent of the block even narrows the gap to 3 points, so rent carries part of the gap the opening contrast assigned to the incidents.\n\nBlocks where the register records an incident without a date are dropped rather than counted as incident-free, and they are 4 per cent of blocks. A note in the annex gives the sorting rule, the pull date as well as the base for each share.\n\nMention: \"Small-Area Protection Incident Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Small-Area Protection Incident Registry relates block incident counts to households forced to move.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:2cb6a25edf2cd90ec2658260", "occurrence_id": "72c1797421ad04a52605623b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Early grade reading figures for the programme areas are reported from the Decennial Early Grade Reading Survey (DEGRS), Eighth Edition, the survey DOLISA follows, and the four values are carried here as tabulated.\n\nThe survey puts the share of pupils in grade three able to read a grade-level sentence at 46 per cent, the share able to name all the letters of the alphabet at 71 per cent, the mean words a pupil reads correctly a minute at 38, and the share of pupils taught in a language spoken at home at 54 per cent. No reweighting was applied to them, and the survey's own reading of a correct minute, which disallows a self-correction, goes with the figures.\n\nA pupil who read the sentence after a model was not counted as able to read it, and a language spoken only by one parent was counted as spoken at home. Each figure keeps the area it was collected in carried with it, rather than mixed across the programme.\n\nThe annex keeps the survey line for each value, along with the day of extraction and the pupils each share divides.\n\nMention: \"Decennial Early Grade Reading Survey (DEGRS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Decennial Early Grade Reading Survey supplies sentence reading, letter naming, fluency and language of instruction.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:4af76646fae8557118d9e761", "occurrence_id": "4a880cf2355fc0c314eb2717", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Hardship and support were assembled for this review out of the Nationwide Deprivation and Hardship Registry (NDHR), Seventh Wave, the register kept for Kenya in 2019, and what follows is the review's arithmetic, not the register's.\n\nHouseholds were sorted by the number of deprivation items the register marks them with and the share receiving any public transfer in the year computed: 29 per cent of households with four items or more against 47 per cent of those with none. The register holds a deprivation flag for each item and a transfer flag for the same household and never sets one against the other.\n\nMaking the groups alike on the household's distance from a service point brings the gap down to 9 points, which leaves distance as part of the distance the first comparison ascribed to the count of items.\n\nHouseholds the register lists with a missing item are dropped rather than counted as deprived on that item, and they are 6 per cent of households. Those references, the extraction day and the denominators are kept in the annex working papers.\n\nMention: \"Nationwide Deprivation and Hardship Registry (NDHR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Nationwide Deprivation and Hardship Registry relates deprivation item counts to transfer receipt.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:9d1716b5bc6b4cd1a82e764d", "occurrence_id": "982eb4cfa7c733d1edd1b360", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The consumption figures in this chapter are computed from the office's own diary panel, and the Civil Household Consumption Registry, the register DANE maintains for Nairobi County as of 2021, is quoted only to place those figures in context.\n\nThe registry puts the share of households whose weekly food spending fell in the quarter at 34 per cent, the mean share of income spent on food at 42 per cent, and the share of households reporting they borrowed to buy food at 16 per cent. All three are given as the registry issues them, with its own reference week stated, and no chapter figure was built from them.\n\nTheir use here is to show how spending behaved across the county beside the path the chapter's panel follows. The chapter's argument rests on the panel, which measures the same households week after week rather than once at a visit, and the two are kept apart rather than merged.\n\nThe annex lists the registry tables consulted and the date it was drawn, which is the full extent of the registry's use in this chapter.\n\nMention: \"Civil Household Consumption Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Civil Household Consumption Registry is quoted as context beside the office's consumption panel.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:5775b630e97255116604da9b", "occurrence_id": "f98eb0e5693904e5200a3bfe", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Demographic and health figures for the survey areas are reported from the Civil Demographic and Health Survey (CDHS), Second Wave, the survey the Ministry of Education follows, and the four values pass into this report as printed.\n\nThe survey puts the share of births attended in a facility at 68 per cent, the mean number of living children ever born to women aged 40-49 at 4, the share of households whose nearest facility lies outside their own cluster at 27 per cent, and the share of children fully vaccinated by twelve months at 59 per cent. Nothing was re-estimated here, and the survey's own reading of a birth attended in a facility travels with the values.\n\nA birth begun at home and finished at a facility was counted as attended in a facility, and a vaccination card without a date was counted as unverified rather than as given. Each figure keeps the cluster it was drawn from with it, rather than pooled across the survey areas.\n\nBeneath the table sit the survey line for each value, the date the file was taken and the women each share divides.\n\nMention: \"Civil Demographic and Health Survey (CDHS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Civil Demographic and Health Survey supplies facility births, fertility, facility distance and full vaccination.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:447e87d905a5179f7b48113d", "occurrence_id": "47bfa8acc5a151bca572113a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The coverage figure in this chapter is the municipality's own, taken from its service ledger, and the Urban Service Delivery Registry (USDR), Release 3, the register kept for Ecuador in 2018, is cited to test it.\n\nThe municipality's ledger shows 91 per cent of its plotted stands connected to a piped supply. The registry shows 83 per cent for the same stands, because it counts a stand connected only where water ran at the time of the round, which the ledger does not test. Neither rule was applied to the other's list.\n\nThe 8-point distance is smaller than the movement the municipality's ledger shows between two years, so the municipality's figure is printed as connected and the registry supplies no revision to it and no stands for the districts its surveyors did not enter.\n\nExcluded are stands the registry records as dry for one round only, 240 of them, since neither count can settle them. Which stands each source counts is noted in the method note in the annex.\n\nMention: \"Urban Service Delivery Registry (USDR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Urban Service Delivery Registry checks the municipality's share of stands connected to a piped supply.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:97251a9a1a945d4b05fe25e5", "occurrence_id": "a5db573bf9afc22447a7bc20", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Delivery and contact were reconstructed for this review from the Monthly Service Delivery Index (MSDI), Second Collection, the index compiled for Burundi in 2017, and the readings below are the review's, not the index's.\n\nSites were sorted by the number of staff days the index records in a month and the share of scheduled contacts actually made computed: 84 per cent at sites above eighteen staff days against 57 per cent at those below eight. The index holds a staff-day count for each site and a contact flag for each contact and never sets one against the other.\n\nHolding the distance from the site to the district town even narrows the gap to 11 points, so remoteness carries part of what the raw comparison laid at the door of the staff days.\n\nSites the index lists without a single staff day are dropped rather than counted as making no contact, and they are 5 per cent of sites. A note in the annex gives the sorting rule, the pull date together with the base for each share.\n\nMention: \"Monthly Service Delivery Index (MSDI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Monthly Service Delivery Index relates staff days per month to contacts actually made.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:8f998898472a3800f376e974", "occurrence_id": "ce94e8fb1be730387d9a9a69", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Job search and hire were assembled for this review out of the Record-Based Labour Force Survey (RLFS), Collection 2, the survey CBK follows, and what follows is the review's arithmetic, not the survey's.\n\nJobseekers were sorted by the number of channels they used in the month and the weeks before a job was found computed: 9 weeks among those using three channels or more against 17 weeks among those using a single channel. The survey holds a channel count for each jobseeker and a hire date for the same jobseeker and never sets one against the other.\n\nMaking the groups alike on the years of experience the jobseeker reports brings the gap down to 4 weeks, which leaves experience as part of the gap the opening comparison put down to the number of channels.\n\nJobseekers who stopped looking before the collection closed are dropped rather than counted as still seeking, and they are 8 per cent of jobseekers. Those references, the extraction day and the denominators are detailed in the annex's method note.\n\nMention: \"Record-Based Labour Force Survey (RLFS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Record-Based Labour Force Survey relates search channels used to weeks before a job is found.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:b468edd821bd320b5034e9bf", "occurrence_id": "99cabd8dcf4ad177c78f4c63", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The seed figure in this chapter is the extension office's own, taken from its distribution book, and the Coordinated Seed Use Survey (CSUS), Wave 2, the survey conducted in Colombia in 2024, is cited to test it.\n\nThe office's book shows 62 per cent of holdings sowing certified seed in the season. The survey shows 48 per cent for the same municipalities, because it asks what was actually sown while the book records what was collected at the depot, and seed sold on again never returns to the book. No basis was used on the other's count.\n\nThe 14-point distance is larger than the change the office's book shows between two seasons, so the chapter prints the survey's lower figure for holdings that sowed and the office's higher one for holdings that collected, and the survey supplies no holdings for the veredas its teams did not reach.\n\nExcluded are holdings the survey found with seed stored but not yet sown at the time of the round, 510 of them. Which holdings each source counts is laid out in the annex's method sheet.\n\nMention: \"Coordinated Seed Use Survey (CSUS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Coordinated Seed Use Survey checks the office's share of holdings sowing certified seed.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:daf8326515bb3844b63c18a4", "occurrence_id": "f515cafb565f2f5b154b3dfa", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Night brightness and service were reconstructed for this review from the Endline Nighttime Lights Registry (ENLR), Third Edition, the register kept for Morocco in 2022, and the readings below are the review's, not the register's.\n\nSettlements were matched to their mean radiance and grouped by the lowest sixth the register records, and the share of households reporting a supply that fails most evenings computed: 34 per cent of settlements in the lowest sixth against 6 per cent of those in the highest. The register grades radiance per settlement and the household survey records the failure per household, and neither reports the two together.\n\nHolding the distance from the settlement to the nearest grid line even narrows the gap to 12 points, so the grid distance carries part of the distance the first comparison charged to the radiance.\n\nSettlements the register classes under cloud cover for the whole period are dropped rather than counted in the lowest sixth, and they are 7 per cent of settlements. A note in the annex gives the classing rule, the pull date plus the base for each share.\n\nMention: \"Endline Nighttime Lights Registry (ENLR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Endline Nighttime Lights Registry relates night radiance to households reporting failing supply.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:ae409eb59cd4faa64eed925f", "occurrence_id": "e0f26b8617152e0d0307f912", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Collective bargaining the figures for the country are reported by the Standard Collective Bargaining Survey (SCBS), Collection 4, the survey conducted in Ethiopia in 2021, and the four values come into this report unchanged.\n\nThe survey puts the share of workers covered by a collective agreement at 23 per cent, the share of establishments recognising a union committee at 31 per cent, the mean years between the renewal of an agreement at 3, and the share of agreements setting a wage floor at 68 per cent. No figure was recalculated in this report, and the survey's own reading of coverage, which requires the agreement to name the establishment, follows the figures.\n\nAn industry-wide agreement the employer never signed was not counted as covering the worker, and an expired agreement still applied was counted as in force. Each figure keeps the region it was collected in attached, instead of being blended across the country.\n\nWorking notes record the survey line for each value: the day of the extract and the workers each share divides.\n\nMention: \"Standard Collective Bargaining Survey (SCBS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Standard Collective Bargaining Survey supplies coverage, union recognition, renewal interval and wage floors.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:f1c37ddccf0d693fc2a91d0c", "occurrence_id": "7e24024d31f2665d87f2c997", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two records lines in the annex come from the Monthly Public Records Registry (MPRR), Edition 5, the register Reinhart & Rogoff maintains for 2020, and both are transcribed without alteration.\n\nThe registry puts the share of authorities publishing their decision log within the month at 49 per cent and the mean days a request remains open before it is closed at 33. Nothing here was recomputed for this report, and the registry's own reading of a published log, which requires each decision to carry a date, travels with the figures.\n\nThe rest of the annex lines come from the ombudsman's own complaint file, so the registry carries two lines while the file accounts for the remaining lines. A figure reported for one authority class keeps that class attached rather than being averaged across the sample.\n\nThe annex log holds the registry entry behind each value, the date of the pull, and the denominators each share uses.\n\nMention: \"Monthly Public Records Registry (MPRR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Monthly Public Records Registry supplies decision-log publication and days a request stays open.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:a8c82d1bf5839f392c17e62e", "occurrence_id": "dad95d568d3c624aa057568a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Occupational safety figures covering the woredas are drawn from the Subnational Occupational Safety System, the register World Bank/IFC maintains for the Tigray Region as of 2020, and the four values enter this report as they stand.\n\nThe register puts the share of reported workplaces holding a written safety rule at 42 per cent, the share whose inspector visited in the year at 29 per cent, the mean reportable injuries per hundred workers at 4, and the share of workplaces lying within a mapped industrial plot at 57 per cent. They are carried as published, and the register's own reading of a reportable injury is reported with them.\n\nAn injury treated at home was not counted as reportable, and a workplace operating under a rule drafted for another branch was counted as holding a rule. Each figure keeps the woreda it was collected in shown with it, instead of combined into one region figure.\n\nThe documentation under the table gives the register line for each value, the date on the extract, and the workplaces each share divides.\n\nMention: \"Subnational Occupational Safety System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Subnational Occupational Safety System supplies written rules, inspections, injury rates and plot location.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:255c94fa0652416dab0f73df", "occurrence_id": "66f76ecc405ba849e55045e2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Place name figures for the surveyed areas are reported from the Rolling Place Name Registry (RPNR), Release 3, the register ICRG maintains for 2019, and the four values reach this report as published.\n\nThe registry puts the share of named places carrying a coordinate to the sixth decimal at 86 per cent, the share whose name appears in more than one spelling in official use at 18 per cent, the mean number of alternative names held for a place at 2, and the share of places whose name was recorded in a local language as well at 44 per cent. Nothing here was recomputed, and the registry's own reading of a named place, which requires a feature matched at two releases, is attached to them.\n\nA place named in one release only was not counted as named, and a spelling differing only by a diacritic was counted as a second spelling. Each figure keeps the district it belongs to reported with it, rather than blended into a single surveyed share areas.\n\nThe registry line for each value, the day of the pull as well as the places each share divides sit with the annex tables.\n\nMention: \"Rolling Place Name Registry (RPNR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Rolling Place Name Registry supplies coordinate precision, spelling variants, alternative names and local-language names.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:fe026f66b4ad2b8e1138f009", "occurrence_id": "bd9054d87dcd1a6f2953b29c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Household records were catalogued into the drought file from the Decennial Drought and Resilience Survey (DDRS), Round 7, the survey World Bank/IFC follows, and the work done on them was cataloguing rather than measurement.\n\nEach household was given the file's own rainfall band and its sowing dates checked against the file's season window; where a household reports a sowing the file cannot place in a season, it was parked for the analyst rather than pushed into the nearest one. Households whose livestock count the survey left blank were left unbanded rather than assumed to sit at the village median, and 1 610 of them entered the file in that condition. The survey's own round number was kept on every household so a later round can be matched without re-deriving it.\n\nThe drought file's exposure weights come from the agricultural census, from which the survey supplied nothing; it is the household coding that lets those weights be read against loss for the resilience table the report draws up.\n\nThe coding sheet, the count of parked households and the day of deposit are carried in the annex log.\n\nMention: \"Decennial Drought and Resilience Survey (DDRS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The Decennial Drought and Resilience Survey supplies the coded household records behind the drought file.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:731021beb3b7c4aaba44cc15", "occurrence_id": "90a70c9b388d7a2a62868218", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The reporting figure in this chapter is the coordination's own, taken from its returns ledger, and the Combined Operations Index Registry (COIR), Edition 8, the register kept for Afghanistan in 2024, is cited to test it.\n\nThe coordination's ledger shows 78 per cent of partner returns arriving by the deadline. The registry shows 69 per cent for the same partners, because it counts a return late if any schedule within it is late, while the ledger counts a return on time if the return itself arrives. Neither rule was applied to the other's ledger.\n\nThe 9-point distance is smaller than the movement the coordination's ledger shows between two months, so the coordination's figure is printed as filed and the registry supplies no revision to it and no returns for the partners it did not canvass.\n\nExcluded are returns the registry holds without a partner identifier, 260 of them, since neither count can be matched for them. Which returns each source counts is set out in the annex notes.\n\nMention: \"Combined Operations Index Registry (COIR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Combined Operations Index Registry checks the coordination's share of partner returns filed on time.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:157ebca6760000a2be678f6d", "occurrence_id": "8505e8fc0268f8364d5b8696", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Enterprise survival and location were assembled for this review out of the Coordinated Enterprise and Jobs Census (CEJC), Release 5, the census conducted in Fortaleza in 2022, and what follows is the review's arithmetic, not the census's.\n\nEnterprises were sorted by the distance from the nearest market and the share still trading three years after registration computed: 58 per cent of enterprises within 2 kilometres against 39 per cent of those beyond 8. The census holds a location for each enterprise and a trading status for the same enterprise and never sets one against the other.\n\nMaking the groups alike on the number of workers at registration brings the gap down to 9 points, which leaves size as part of what the unadjusted contrast attributed to the distance.\n\nEnterprises the census finds at a address it cannot match to a frame unit are dropped rather than placed at the unit centroid, and they are 5 per cent of enterprises. Those references, the extraction day and the denominators are kept in the annex file.\n\nMention: \"Coordinated Enterprise and Jobs Census (CEJC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Coordinated Enterprise and Jobs Census relates distance to market and three-year enterprise survival.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:2c71bb4242cee1f21faba715", "occurrence_id": "69acfee5f4582c12b6d16413", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Whether the agency's own count of households in need holds was tested against the Recurrent Shock Coping Survey, which UNODC followed in the Tigray Region in 2021, and the survey is the principal source for the test.\n\nThe agency's distribution list carries 52 800 households classed in need. The survey carries 67 400 for the same woredas, because it asks each household what it did after the shock, which brings in households the agency never saw. Neither question was put to the other's list.\n\nThe survey therefore carries the section's count of households in need, and the agency's smaller figure is printed beside it so the reader can see that the distance is households never reached rather than a disagreement about need. Where a household appears in both, the agency's ration date is kept and the two counts are not averaged.\n\nHouseholds the survey places outside a woreda are dropped from both rather than counted by the survey only, 7 per cent of its listings. The question each source works to is written up in the papers behind the annex.\n\nMention: \"Recurrent Shock Coping Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Recurrent Shock Coping Survey tests the agency's count of households classed in need.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:3e2c4fa25bde2b8e19028144", "occurrence_id": "882a96c0b99decb0d25f1438", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Feeding and enrolment were reconstructed for this review from the Provincial School Feeding Survey (PSFS), Seventh Phase, the survey the Fiscal Affairs Department follows, and the readings below are the review's, not the survey's.\n\nSchools were sorted by the days a week the meal is served and the share of pupils enrolled at the start of the following term computed: 94 per cent at schools serving five days against 86 per cent at schools serving two days or fewer. The survey holds a serving count for each school and an enrolment record for the same school and never sets one against the other.\n\nHolding the distance from the school to the nearest road even narrows the gap to 3 points, so remoteness carries part of what the unadjusted difference owed to the days served.\n\nSchools that closed mid-term for reasons unrelated to the programme are dropped rather than counted as losing their pupils, and they are 4 per cent of schools. A note in the annex gives the sorting rule, the pull date, with the base for each share alongside.\n\nMention: \"Provincial School Feeding Survey (PSFS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Provincial School Feeding Survey relates days fed a week to pupils re-enrolling next term.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:e3a74a632d4d7d015772a5ea", "occurrence_id": "220a4b90f76335a53814e0bd", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Child wellbeing figures covering the whole country derive from the Clustered Child Wellbeing Survey (CCWS), the survey conducted in Azerbaijan in 2016, and the four values appear here exactly as printed.\n\nThe survey puts the share of children living in a household below the national adequacy line at 24 per cent, the share whose household cannot afford school materials at 11 per cent, the share of children reporting they go to bed hungry in a week at 6 per cent, and the share of households with a child registered at birth at 98 per cent. They are reproduced as tabulated, and the survey's own reading of adequacy follows the national line without adjustment.\n\nA household receiving in-kind help was still counted below the line if its cash spending fell short, and a materials cost met by the school was not counted as unaffordable. Each figure keeps the region it was drawn from kept with it, instead of being rolled up to the country.\n\nUnder the table are the survey line for each value, plus the extraction day and the children each share divides.\n\nMention: \"Clustered Child Wellbeing Survey (CCWS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Clustered Child Wellbeing Survey supplies child poverty, materials affordability, hunger and birth registration.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:bb38bcb28c8ec9e2c9314b4d", "occurrence_id": "1417838d67ebc310192e5e40", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Irrigation and input use were estimated for this review using the Municipal Irrigation and Inputs Microdata (MIIM), the extract Kenya ICT Action Network compiles for Morocco, and the figures are the review's own.\n\nPlots were sorted by the method of watering the household reports and the value of fertiliser bought in the season computed: 610 among plots drawing from a canal against 240 among those relying on rainfall. The extract holds a watering method for each plot and a purchase value for the same plot and never sets one against the other.\n\nMaking the groups alike on the area the household works brings the gap down to 190, which leaves area as part of what the first comparison credits to the watering method.\n\nPlots the extract lists under two watering methods are dropped rather than counted at the higher one, and they are 6 per cent of plots. Working notes list the sorting rule, the extraction day and the denominators each share rests on.\n\nMention: \"Municipal Irrigation and Inputs Microdata (MIIM)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Municipal Irrigation and Inputs Microdata relates watering method to fertiliser spending in the season.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:c98adf02d177b06995f47c21", "occurrence_id": "e3f3558d484afac7923d3c2c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The request figure in this chapter is the council's own, taken from its correspondence book, and the District Public Records Registry (DPRR), the register ISTAT maintains for Tanzania, is cited to test it.\n\nThe council's book shows 1 340 requests received in the year. The registry shows 1 585 for the same districts, because it counts a request made in person and never written down, which the council's book records only once a clerk opens a file. The bases stay separate: neither is applied to the other's count.\n\nThe 245 distance is smaller than the number of requests the council's book records in a month at the year's peak, so the council's figure is printed as filed and the registry supplies no revision to it and no requests for the districts its round did not visit.\n\nExcluded are requests the registry dates only to a quarter, 96 of them, since the year cannot be fixed for them. Which requests each source counts is detailed in the working papers behind the annex.\n\nMention: \"District Public Records Registry (DPRR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The District Public Records Registry checks the council's count of information requests received.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:ec3612732f418197310e5ab9", "occurrence_id": "65c78f4aee4b3f501fd771f9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Publication delay was derived for this chapter from the Harmonised Statistical Bulletin System (HSBS), Fourth Collection, the system the Chinese Wind Energy Association maintains, and the figures below are the chapter's rather than the system's.\n\nBulletins were sorted by the number of sources they draw on and the median weeks between the reference period closing and publication computed: 8 weeks for bulletins drawing on two sources or fewer against 19 weeks for those drawing on five or more. The system holds a source count for each bulletin and a publication date for each bulletin and never sets one against the other.\n\nThat comparison supports one claim, that bulletins drawing on a single register should be tabulated on a fixed calendar. The staffing lines the proposal quotes come from the office's own establishment return, from which the system supplied nothing.\n\nBulletins withdrawn and reissued in the collection are dropped rather than counted twice, and they are 3 per cent of bulletins. The bands and the counts behind each median are recorded under the table.\n\nMention: \"Harmonised Statistical Bulletin System (HSBS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Harmonised Statistical Bulletin System supports a fixed-calendar claim built from source count and delay.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:8eae698bc07da33fdfb8c75c", "occurrence_id": "dbf0ea2ea6ea785d9fc7a5b6", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Household consumption figures for the sampled areas are reported from the Small-Area Household Consumption Survey (SHCS), Second Round, the survey The World Bank. follows, and the four values are reproduced here as issued.\n\nThe survey puts the share of households below the national poverty line at 27 per cent, the mean monthly consumption per adult at 890, the share of households whose food spending rose after the last drought at 38 per cent, and the share borrowing at food prices at 19 per cent. They were not re-derived here, and the survey's own reference period of four weeks is stated with them.\n\nA household eating from its own garden was counted at the value of what it ate, and a loan taken for a wedding was not counted as borrowing at food prices. Each figure keeps the small area it was drawn from named with it, instead of averaged over the sampled areas.\n\nThe survey line for each value, along with the extraction date and the households each share divides are recorded under the table.\n\nMention: \"Small-Area Household Consumption Survey (SHCS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Small-Area Household Consumption Survey supplies poverty incidence, consumption, food-spending rises and food borrowing.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:4f719fcf73565336f00d5791", "occurrence_id": "c79550f7bddd0906a36ffc99", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The benefit figures in this chapter are compiled from the canton's own register, and the Multi-Topic Targeting and Payments Survey (MTPS), the survey conducted in Switzerland in 2023, is quoted only to place those figures in context.\n\nThe survey puts the share of households eligible for a benefit who receive it at 74 per cent, the mean months a household has been receiving before it leaves the roll at 22, and the share of recipients who took up work while receiving at 31 per cent. All three are given as the survey issues them, with its own definition of eligibility stated, and no chapter figure was built from them.\n\nTheir use here is to show how the wider scheme behaves beside the one canton whose register the chapter works from. The chapter's argument rests on that register, which names each recipient rather than sampling households, and the two are kept apart rather than merged.\n\nThe annex lists the survey tables consulted and the day the table was pulled, which is the full extent of the survey's use in this chapter.\n\nMention: \"Multi-Topic Targeting and Payments Survey (MTPS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Multi-Topic Targeting and Payments Survey is quoted as context beside the canton's benefit register.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:a8a349b4d9356a01850a414a", "occurrence_id": "3b9d893c5ea045bd8b94977a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Facility records were catalogued into the catchment file from the Recurrent Service Catchments System (RSCS), the register CCCM Cluster maintains for Moscow, and the work done on them was cataloguing rather than measurement.\n\nEach facility was given the file's own catchment band and its opening dates checked against the file's reference window; where the system carries a facility the file has no band for, it was parked for the analyst rather than entered under the nearest one. Facilities the system lists without a coordinate were left unbanded rather than placed at the district centre, and 620 of them entered the file in that condition. The system's own round date was kept on every facility so a later round can be matched without re-deriving it.\n\nThe catchment file's population weights come from the municipal register, from which the system supplied nothing; it is the facility coding that lets those weights be read against distance for the access table the report draws up.\n\nThe coding sheet, the count of parked facilities and the day of deposit are held in the annex log.\n\nMention: \"Recurrent Service Catchments System (RSCS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The Recurrent Service Catchments System supplies the coded facility records behind the catchment file.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:8ac19126b00feac3c70d88dd", "occurrence_id": "b222c2fed95efeabfdb29cae", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Displacement and return were assembled for this review out of the Rural Forced Population Registry (RFPR), Phase 7, the register RVRN maintains, and what follows is the review's arithmetic, not the register's.\n\nHouseholds were sorted by the months since the register first listed them as displaced and the share that have since returned to their origin computed: 12 per cent of households listed two years or more against 31 per cent of those listed under six months. The register holds a first-listing date for each household and a return flag for the same household and never sets one against the other.\n\nMaking the groups alike on the distance from the site to the origin brings the gap down to 6 points, which leaves distance as part of the gap the opening contrast assigned to the time since listing.\n\nHouseholds the register lists twice under different site codes are dropped rather than counted at the earlier listing, and they are 4 per cent of households. Those references, the extraction day and the denominators are laid out in the annex working papers.\n\nMention: \"Rural Forced Population Registry (RFPR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Rural Forced Population Registry relates months since listing to households returning to their origin.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:4ed2bdf674c023a02d78b98f", "occurrence_id": "0382e918cfa28815e6328daa", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Whether the service's own count of hot days holds was tested against the High-Frequency Temperature Records Registry (HTRR), the register HZZO maintains for 2023, and the registry is the principal source for the test.\n\nThe service's summary gives 46 days above 35 degrees at its stations in the season. The registry gives 53 for the same stations, because it counts a day hot if the reading is exceeded at any hour, while the service's summary uses the daily maximum it publishes once a day. Neither rule was applied to the other's count.\n\nThe registry therefore carries the section's count of hot days, and the service's smaller figure is printed beside it so the reader can see that the distance is the reading rule rather than a disagreement about the season. Where a day appears in both, the service's maximum is kept and the two counts are not averaged.\n\nDays the registry holds a reading for only one of the two required hours are dropped from both rather than counted by the registry alone, 8 per cent of its listings. The reading rule each source works to is recorded in the method note in the annex.\n\nMention: \"High-Frequency Temperature Records Registry (HTRR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The High-Frequency Temperature Records Registry tests the service's count of days above 35 degrees.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:8604e28a488ac03c007c5005", "occurrence_id": "7c46917794880e3ffeff344e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Health workforce the numbers held for the country come from the High-Frequency Health Workforce Registry (HHWR), Edition 6, the register kept for Slovakia in 2022, and the four values are quoted here as printed.\n\nThe register puts the number of physicians per 10 000 population at 62, the share of posts in the poorest districts vacant at 19 per cent, the mean months a vacant post stands unfilled at 8, and the share of physicians working at more than one site at 27 per cent. No arithmetic was applied to them here, and the register's own reading of a filled post, which requires a contract current at the reference date, comes with them.\n\nA physician on unpaid leave was counted as filling no post, and a session given at a second site under one contract was counted as one site. Each figure keeps the district it was drawn from listed with it, rather than smoothed across the country.\n\nThe annex note carries the register line for each value, the day of the pull and, alongside it, the posts each share divides.\n\nMention: \"High-Frequency Health Workforce Registry (HHWR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The High-Frequency Health Workforce Registry supplies physician density, poor-district vacancies, wait and multi-site work.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:4a6ccadeb52063240ff56f0e", "occurrence_id": "92bdf08fce893bbc25df4e32", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Inspection and earnings were reconstructed for this review from the Linked Labour Inspection Survey, which the Planning Commission, GoI conducted in Ethiopia in 2020, and the readings below are the review's, not the survey's.\n\nWorkers were sorted by whether their workplace had been inspected and the weekly earnings they report computed: 640 among workers in inspected workplaces against 520 among workers in workplaces never inspected. The survey holds an inspection item for each workplace and an earnings item for each worker and never sets one against the other.\n\nHolding the hours a worker reports even narrows the gap to 40, so hours carry part of the distance the first comparison ascribed to the inspection.\n\nWorkers in workplaces that closed before the interview are dropped rather than counted at their last earnings, and they are 5 per cent of workers. A note in the annex gives the sorting rule, the pull date as well as the base for each figure.\n\nMention: \"Linked Labour Inspection Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Linked Labour Inspection Survey relates workplace inspection to weekly earnings.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:651f22b4558f2a981f93a6b4", "occurrence_id": "c051cdeae2133304dd53be58", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The reach figure in this chapter is the authority's own, taken from its client list, and the Independent Service Delivery Registry (ISDR), Round 7, the register CBK maintains, is cited to test it.\n\nThe authority's list shows 88 per cent of households in its area holding at least one account. The registry shows 79 per cent for the same areas, because it counts an account only where a transaction passed through it in the year, which the authority's list does not test. Neither rule was applied to the other's list.\n\nThe 9-point distance is smaller than the movement the authority's list shows between two rounds, so the authority's figure is printed as listed and the registry supplies no revision to it and no households for the wards its round did not enter.\n\nExcluded are accounts the registry marks dormant without a closing date, 1 420 of them, since neither count can settle them. Which accounts each source counts is filed in the annex's method note.\n\nMention: \"Independent Service Delivery Registry (ISDR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Independent Service Delivery Registry checks the authority's share of households holding an account.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:0976adf1bdcdee8aab110146", "occurrence_id": "a836a04272ecf73d25a622f2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Time use country figures are assembled from the Clustered Time Use Survey (CTUS), Second Version, the survey conducted in Tajikistan in 2020, and the four values are carried here as tabulated.\n\nThe survey puts the mean hours a woman gives to unpaid care in the week at 41, the mean hours a man gives to the same at 6, the share of adults recording no paid work in the week at 34 per cent, and the share of households where nobody reports a paid activity at 18 per cent. No reweighting was applied to them, and the survey's own reading of an activity, which requires it to have been reported against a time diary, is kept with them.\n\nCare given while supervising a child's homework was counted as unpaid care, and an hour of paid work at home was counted as paid. Each figure keeps the cluster it was drawn from carried with it, rather than mixed across the country.\n\nThe table's footnotes hold the survey line for each value: the day the rows were taken and the adults each share divides.\n\nMention: \"Clustered Time Use Survey (CTUS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Clustered Time Use Survey supplies unpaid care hours by sex, paid-work incidence and households with no earner.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:129aa629f1596f4d24de613b", "occurrence_id": "92af17f2e21f228447e05f2f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two operations lines in the transport annex come from the Metropolitan Regional Operations Index (MROI), Wave 4, the index the International Road Federation compiled for 2024, and both are transcribed without alteration.\n\nThe index puts the share of metropolitan units filing a traffic operations return each month at 66 per cent and the mean days between an incident closing and its entry appearing in the return at 9. Nothing here was recomputed here, and the index's own reading of an entry, which requires the incident's own reference number, travels with the figures.\n\nThe other annex entries are assembled from the authority's own control log, so the index carries two lines while the log carries the rest. A figure given for a single corridor stays with that corridor instead of being averaged across the network.\n\nThe annex log holds the index entry behind each value, the day the extract ran together with the denominators behind each share.\n\nMention: \"Metropolitan Regional Operations Index (MROI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Metropolitan Regional Operations Index supplies monthly filing rates and incident-entry delay.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:d238d716518b3f789338e596", "occurrence_id": "e2f80cb1b5d95f2be3086a9a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Debt and consumption were reconstructed for this review from the Recurrent Debt and Savings Survey (RDSS), Collection 5, the survey IIASA follows, and the readings below are the review's, not the survey's.\n\nHouseholds were sorted by the number of lenders they owe and the share cutting food spending to keep up repayments computed: 46 per cent of households owing three lenders or more against 17 per cent of those owing one. The survey holds a lender count for each household and a food item for the same household and never sets one against the other.\n\nHolding the household's monthly income even narrows the gap to 12 points, so income carries part of what the raw comparison laid at the door of the number of lenders.\n\nHouseholds that owe nothing but report a repayment to a relative are dropped rather than counted as indebted, and they are 5 per cent of households. A note in the annex gives the sorting rule, the pull date plus the base for each share.\n\nMention: \"Recurrent Debt and Savings Survey (RDSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Recurrent Debt and Savings Survey relates number of lenders to cutting food to repay debt.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:2711bd794b8cddf908c39050", "occurrence_id": "ca618ffc206f3183e074a07c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Service delivery figures spanning the regions are taken from the Endline Service Delivery Registry (ESDR), First Release, the register kept for Mali in 2023, and the four values pass into this report as printed.\n\nThe register puts the share of localities with a service point open on the day of the round at 71 per cent, the mean kilometres from a locality to its nearest open point at 9, the share of points reporting a stock-out in the month at 22 per cent, and the share of points whose staff were present for the visit at 64 per cent. Nothing was re-estimated here, and the register's own reading of an open point, which requires a service actually given, accompanies the figures.\n\nA point open only to register a complaint was not counted as open, and a point staffed by a volunteer was counted as staffed. Each figure keeps the region it was collected in with it, rather than pooled across the country.\n\nUnder the table you will find the register line for each value, the pull date, and the localities each share divides.\n\nMention: \"Endline Service Delivery Registry (ESDR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Endline Service Delivery Registry supplies open-point coverage, distance, stock-outs and staff presence.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:9151fbde8fc3ba14498f3b0a", "occurrence_id": "db940aed2e90e4dc951bfc7f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Targeting and payment figures for the rural areas are reported from the Rural Targeting and Payments Survey (RTPS), First Version, the survey WBG follows in the United States, and the four values come into this report unchanged.\n\nThe survey puts the share of households screened as eligible at 29 per cent, the share of those eligible who were enrolled at 74 per cent, the mean months between enrolment and first payment at 3, and the share of enrolled households reporting a move since enrolment at 11 per cent. No figure was recalculated in this report, and the survey's own reading of eligibility follows the programme's formula as at the screening date.\n\nA household that declined enrolment was still counted as eligible, and a household that moved within its own county was counted as having moved. Each figure keeps the county it was drawn from attached, instead of being blended across the areas.\n\nThe annex keeps the survey line for each value, the day of extraction as well as the households each share divides.\n\nMention: \"Rural Targeting and Payments Survey (RTPS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Rural Targeting and Payments Survey supplies eligibility, enrolment, payment delay and post-enrolment moves.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:68b653a30aa812b7c3d867c0", "occurrence_id": "0b78eb358c75fe7483caa19d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Apprenticeship outcomes were derived for this chapter from the Diagnostic Apprenticeship Survey (DAS), Edition 2, the survey ICP follows, and the figures below are the chapter's own, not the survey's.\n\nApprentices were sorted by the number of assessed workplace hours in the year and the share holding a certificate at the end of the period computed: 77 per cent of apprentices above 800 assessed hours against 41 per cent of those below 200. The survey holds an hours figure for each apprentice and a certificate flag for the same apprentice and never sets one against the other.\n\nThat comparison supports one claim, that assessed hours should be logged weekly rather than at the end of the spell. The cost lines the proposal quotes come from the training authority's own budget, from which the survey supplied nothing.\n\nApprentices who left the employer without withdrawing from the course are dropped rather than counted as uncertified, and they are 6 per cent of apprentices. The bands and the counts behind each share are recorded under the table.\n\nMention: \"Diagnostic Apprenticeship Survey (DAS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Diagnostic Apprenticeship Survey supports a weekly-logging claim built from assessed hours and certification.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:1af07c4bbd41d330edefb6ff", "occurrence_id": "5f3b6ab452d12d0d1a79600a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Feeding and staffing were assembled for this review out of the Analytical School Feeding Survey (ASFS), Round 7, the survey conducted in Armenia in 2015, and what follows is the review's arithmetic, not the survey's.\n\nSchools were sorted by the number of staff days the kitchen requires in a week and the share of teaching hours lost to kitchen duties computed: 6 per cent at schools needing three staff days or more against 1 per cent at schools needing none. The survey holds a kitchen requirement for each school and a lost-hours figure for the same school and never sets one against the other.\n\nMaking the groups alike on the number of pupils on roll brings the gap down to 2 points, which leaves roll size as part of the gap the opening comparison put down to the kitchen days.\n\nSchools whose kitchen was closed for the whole term are dropped rather than counted as losing no hours, and they are 4 per cent of schools. Those references, the extraction day and the denominators sit in the annex's method sheet.\n\nMention: \"Analytical School Feeding Survey (ASFS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Analytical School Feeding Survey relates kitchen staff days to teaching hours lost to kitchen duties.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:c4563a694440b6470d3772cb", "occurrence_id": "43fc4234c7f4d6ed33a62dd2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Campaign reach and movement were reconstructed for this review from the Sentinel Vaccination Campaign Survey (SVCS), the survey WBG follows in Senegal, and the readings below are the review's, not the survey's.\n\nSettlements were sorted by the share of households the survey finds arrived in the last year and the share of children the campaign reached computed: 68 per cent of settlements below a tenth arriving against 84 per cent of those above a third. The survey holds an arrival share for each settlement and a reach flag for each child and never sets one against the other.\n\nHolding the distance from the settlement to the nearest post even narrows the gap to 9 points, so distance carries part of the distance the first comparison charged to the arrival share.\n\nSettlements the survey reached only after the campaign closed are dropped rather than counted as unreached, and they are 5 per cent of settlements. A note in the annex gives the sorting rule, the pull date, and the base for each share.\n\nMention: \"Sentinel Vaccination Campaign Survey (SVCS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Sentinel Vaccination Campaign Survey relates recent arrivals in a settlement to campaign reach.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:9872a05599b9ea472e65afb6", "occurrence_id": "40662d96394cb8a5b34b5bed", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two graduation lines in the welfare annex come from the High-Frequency Graduation Programmes System (HGPS), Phase 2, the register CBK maintains for Lebanon, and both are transcribed without alteration.\n\nThe system puts the share of enrolled households still above the exit line twelve months after leaving at 63 per cent and the mean assets a household holds at graduation, scored on the system's own scale, at 7. Nothing here was recalculated for this report, and the system's own reading of graduation, which requires the worker's visit as well as the score, travels with the figures.\n\nThe annex's other lines are taken from the programme's own case file, so the system carries two lines while the file holds the other lines. A figure reported for one area keeps that area attached rather than being averaged across the programme.\n\nThe annex log holds the system entry behind each value, the date it was drawn and the base each share is taken on.\n\nMention: \"High-Frequency Graduation Programmes System (HGPS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The High-Frequency Graduation Programmes System supplies post-exit retention and assets at graduation.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:18225d6cb6595b2d8e291d50", "occurrence_id": "14f290b877ff6d90f48b0081", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Housing and hardship were estimated for this review using the Rural Housing Conditions Survey (RHCS), Version 7, the survey REACH, in the follows, and the figures are the review's own.\n\nDwellings were sorted by the material of the floor the survey records and the share of households below the national adequacy line computed: 58 per cent of households in dwellings with an earth floor against 22 per cent of those with a solid floor. The survey holds a floor material for each dwelling and a consumption aggregate for the same household and never sets one against the other.\n\nMaking the groups alike on the land a household holds brings the gap down to 14 points, which leaves land as part of what the unadjusted contrast attributed to the floor.\n\nDwellings the survey records as unoccupied at the time of the visit are dropped rather than counted as earth-floored, and they are 3 per cent of dwellings. Working notes list the sorting rule, the extraction day, with the denominators each share rests on alongside.\n\nMention: \"Rural Housing Conditions Survey (RHCS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Rural Housing Conditions Survey relates floor material to households below the adequacy line.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:000dd39c245706fc48b74f3c", "occurrence_id": "f0337cb482d500852067192a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The enrolment figures in this chapter are taken from the ministry's own class registers, and the Provincial Enrolment and Retention Registry (PERR), Fifth Edition, the register Badan Pusat Statistik or BPS maintains, is referred to only as background to them.\n\nThe registry puts the share of pupils enrolling in the first year of upper secondary at 61 per cent, the share of those pupils still on roll three years later at 72 per cent, and the mean years a pupil repeats before leaving at 1. All three are given as the registry issues them, with its own roll definition stated, and no chapter figure rests on them.\n\nNo chapter table is built from its entries. The figures the chapter argues from come from the class registers, which list each pupil in a class rather than a sample of schools, and where the two accounts of retention differ the difference is noted rather than reconciled.\n\nThe annex lists the registry tables consulted and the date the file was taken, and the registry is used for nothing beyond that background.\n\nMention: \"Provincial Enrolment and Retention Registry (PERR)\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The Provincial Enrolment and Retention Registry is background to the ministry's class-register enrolment figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:0275fdec3d998f0e105b5757", "occurrence_id": "492faa7e9948738b6b1d0303", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Business demography figures at the country level are taken from the Civil Business Demography Survey (CBDS), the survey conducted in Thailand in 2016, and the four values enter this report as they stand.\n\nThe survey puts the share of enterprises active at the start of the year still active a year later at 84 per cent, the rate at which new enterprises enter in the year at 9 per cent, the mean workers an enterprise holds at entry at 3, and the share of entries registered as a limited company at 27 per cent. They are carried as published, and the survey's own reading of an active enterprise comes down with them.\n\nAn enterprise dormant for the year but not struck off was counted as active, and an unincorporated workshop trading under a trade name was counted as a sole entry. Each figure keeps the province it was drawn from shown with it, instead of combined into one country figure.\n\nBeneath the table sit the survey line for each value, plus the day of the extract and the enterprises each share divides.\n\nMention: \"Civil Business Demography Survey (CBDS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Civil Business Demography Survey supplies survival, entry rates, size at entry and legal form.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:2d34565c57f3f697ab5c8ced", "occurrence_id": "2988b0f56eaa8744f09c7948", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Output area country-level figures come from the Quality-Assured Output Area Dataset (QOAD), the geospatial dataset compiled for Uzbekistan in 2021, and the four values reach this report as published.\n\nThe dataset puts the share of output areas whose boundary was checked on the ground at 55 per cent, the mean population of an output area at 1 420, the share of areas whose boundary follows a mapped feature rather than an administrative line at 38 per cent, and the share of areas that changed between the two editions at 12 per cent. Nothing here was recomputed, and the dataset's own reading of a checked boundary is carried alongside them.\n\nA boundary checked by telephone with the local office was not counted as checked on the ground, and a line redrawn for cartographic clarity was not counted as a change. Each figure keeps the region it was drawn from reported with it, rather than blended into a single country share.\n\nWorking notes record the dataset layer for each value, along with the date of the pull and the areas each share divides.\n\nMention: \"Quality-Assured Output Area Dataset (QOAD)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Quality-Assured Output Area Dataset supplies ground-checked boundaries, area size, basis and inter-edition change.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:54a15ff92dd90bf2fdb0f237", "occurrence_id": "3c1f5c1ff409d60652508170", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Building density and damage were reconstructed for this review from the Pilot Building Footprint Registry (PBFR), Phase 7, the register INFRAS/IWW maintains, and the readings below are the review's, not the registry's.\n\nBlocks were matched to their footprint count and grouped by buildings per hectare, and the share of buildings the damage survey marks as unusable computed: 19 per cent of blocks above forty buildings to the hectare against 7 per cent of those below ten. The registry counts footprints per block and the damage survey grades each building, and neither reports the two together.\n\nHolding the age band of the buildings in the block even narrows the gap to 6 points, so the age band carries part of what the unadjusted difference owed to the density.\n\nBlocks the registry maps without a footprint because of overlap are dropped rather than counted as unbuilt, and they are 5 per cent of blocks. A note in the annex gives the grouping rule, the pull date and the base for each share.\n\nMention: \"Pilot Building Footprint Registry (PBFR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Pilot Building Footprint Registry relates buildings per hectare to buildings marked unusable.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:5c5b30e4dd9840f11d532ea2", "occurrence_id": "4bad7e6553016d3e9e09e0ce", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The records figures in this chapter are drawn from the department's own inventory, and the Departmental Administrative Records System (DARS), Sixth Round, the system GACC maintains for 2021, is quoted only to place those figures in context.\n\nThe system puts the share of record series with a named custodian at 78 per cent, the mean years since a series was last appraised at 6, and the share of series transferred to the archive on schedule at 41 per cent. All three are given as the system issues them, with its own list of series stated, and no chapter figure was built from them.\n\nTheir use here is to show how record-keeping looks across departments beside the one department the chapter's inventory covers. The chapter's argument rests on that inventory, which lists the department's own series rather than a comparable panel, and the two are kept apart rather than merged.\n\nThe annex lists the system tables consulted and the date on the extract, which is the full extent of the system's use in this chapter.\n\nMention: \"Departmental Administrative Records System (DARS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Departmental Administrative Records System is quoted as context beside the department's record inventory.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:3cf238acee683951bafaafbc", "occurrence_id": "1cdcfb72ba330fc4f195a115", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two household lines in the labour annex come from the Linked Household Income and Expenditure Survey (LHIES), Second Collection, the survey CCCM Cluster follows for 2023, and both are transcribed without alteration.\n\nThe survey puts the mean share of household income arriving from wage work at 54 per cent and the share of households whose income fell short of their recorded spending in the month at 31 per cent. Nothing here was recomputed in this report, and the survey's own reading of spending, which counts a purchase on credit at the time of purchase, travels with the figures.\n\nThe annex's other entries are drawn from the office's own beneficiary file, so the survey carries two lines while the file carries the lines the table does not. The report keeps each stratum apart rather than averaging across the sample.\n\nThe annex log holds the survey line behind each value, the day of the pull and, alongside it, the counts the shares rest on.\n\nMention: \"Linked Household Income and Expenditure Survey (LHIES)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Linked Household Income and Expenditure Survey supplies wage-income share and households spending above income.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:485fc5e6c173e9fab04d72d9", "occurrence_id": "c5d8496e9b6406208b7905c6", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Living standards and distance were assembled for this review out of the Regional Living Standards Survey (RLSS), Collection 8, the survey Growth Lab at Harvard follows, and what follows is the review's arithmetic, not the survey's.\n\nHouseholds were sorted by the travel time to the nearest market and the share below the national poverty line computed: 49 per cent of households above an hour against 23 per cent of those under twenty minutes. The survey holds a travel time for each household and a consumption aggregate for the same household and never sets one against the other.\n\nMaking the groups alike on the land a household works brings the gap down to 12 points, which leaves land as part of what the first comparison credits to travel time.\n\nHouseholds the survey visited twice in the collection are counted at their first visit rather than their lower reading, and they are 6 per cent of households. Those references, the extraction day and the denominators are recorded in the annex notes.\n\nMention: \"Regional Living Standards Survey (RLSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Regional Living Standards Survey relates travel time to market and poverty-line incidence.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:567252206e875e7340ef988f", "occurrence_id": "b5a3998438198a6ccc9b6cc4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The operations figures in this chapter are drawn from the office's own returns, and the Record-Based Regional Operations Registry (RROR), the register World Bank/IFC maintains for 2019, is referred to only as background to them.\n\nThe registry puts the share of regions whose returns reach the centre complete at 67 per cent, the mean weeks a return takes to arrive after the period closes at 5, and the share of regions whose return was revised once received at 18 per cent. All three are given as the registry issues them, with its own list of regions stated, and no chapter figure rests on them.\n\nIts entries do not appear in any chapter table. The figures the chapter argues from come from the office's returns, which cover each of its own branches rather than a sample of regions, and where the two accounts of timeliness differ the difference is noted rather than reconciled.\n\nThe annex lists the registry tables consulted and the extraction day, and the registry is used for nothing beyond that background.\n\nMention: \"Record-Based Regional Operations Registry (RROR)\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The Record-Based Regional Operations Registry is background to the office's own operations returns.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:925bef8a66a05e84e423323e", "occurrence_id": "78d396065cd2e14bf500ff0e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Credit and land improvement were reconstructed for this review from the Pilot Rural Credit Survey (PRCS), Version 2, the survey the World Bank follows, and the readings below are the review's, not the survey's.\n\nHoldings were sorted by whether they took a formal loan in the last two seasons and the share that dug or lined a canal since computed: 34 per cent of holdings with a loan against 12 per cent of those without. The survey holds a loan item for each holding and an improvement item for the same holding and never sets one against the other.\n\nHolding the area the holding works even narrows the gap to 9 points, so area carries part of the gap the opening contrast assigned to the loan.\n\nHoldings reporting a loan from a trader rather than an institution are dropped rather than counted as formal borrowers, and they are 7 per cent of holdings. A note in the annex gives the sorting rule, the pull date as well as the base for each share.\n\nMention: \"Pilot Rural Credit Survey (PRCS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Pilot Rural Credit Survey relates formal borrowing to canal digging or lining by holdings.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:0659ba96c15c4fa077695f36", "occurrence_id": "91228ab831178e6de869363f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: School water and attendance were estimated for this review using the Regional Facilities and WASH Survey (RFWS), Edition 5, the survey conducted in Burkina-Faso in 2021, and the figures are the review's own.\n\nSchools were sorted by whether they hold a working water point on the compound and the mean days a pupil attends in the term computed: 96 days at schools with a point against 82 days at those without. The survey records the water point and the attendance count for the same school and never sets one against the other.\n\nMaking the groups alike on the distance a pupil walks to school brings the gap down to 5 days, which leaves walking distance as part of the distance the first comparison ascribed to the water point.\n\nSchools closed for part of the term are dropped rather than counted at the days they sat open, and they are 8 per cent of schools. Working notes list the sorting rule, the extraction day together with the denominators each share rests on.\n\nMention: \"Regional Facilities and WASH Survey (RFWS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Regional Facilities and WASH Survey relates a school water point to days attended in the term.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:66adfb18d9b5934771374baf", "occurrence_id": "95b007e5c040d08d46cb6fea", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Fuel and energy figures for each part of the country are drawn from the Rapid Fuel and Energy Use Survey (RFEUS), Collection 5, the survey conducted in Uganda in 2015, and the four values appear here exactly as printed.\n\nThe survey puts the share of households cooking with charcoal at 62 per cent, the share with any electric connection at 26 per cent, the mean monthly spending on fuel as a share of household spending at 17 per cent, and the share of households collecting wood as their own fuel at 31 per cent. They are reproduced as tabulated, and the survey's own reading of a connection, which requires a meter or a billed supply, is printed beside them.\n\nA household wired to a neighbour's meter was counted as unconnected, and wood bought as well as gathered was counted as bought. Each figure keeps the district it was collected in kept with it, instead of being rolled up to the country.\n\nThe documentation under the table gives the survey line for each value, the extraction date plus the households each share divides.\n\nMention: \"Rapid Fuel and Energy Use Survey (RFEUS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Rapid Fuel and Energy Use Survey supplies charcoal cooking, connections, fuel spending and wood gathering.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:07445925bbccb34057f08f97", "occurrence_id": "6fb33cd207ab11960f754a45", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Skills and earnings were estimated for this review using the Regional Skills and Employment Microdata, the extract oDesk compiled for the Amhara region in 2020, and the figures are the review's own.\n\nWorkers were sorted by the highest qualification they report and the hourly wage the current job returns computed: 12 for workers reporting a completed secondary education against 7 for those reporting none. The extract holds a qualification for each worker and a wage for the same worker and never sets one against the other.\n\nMaking the groups alike on the years of experience the worker reports brings the gap down to 3, which leaves experience as part of what the raw comparison laid at the door of the qualification.\n\nWorkers reporting a job but no hours for it in the reference week are dropped rather than counted at zero hours, and they are 5 per cent of workers. Working notes list the sorting rule, the extraction day, and the denominators each share rests on.\n\nMention: \"Regional Skills and Employment Microdata\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Regional Skills and Employment Microdata relates highest qualification to the hourly wage.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:5aa91799e7ecb485eabe2361", "occurrence_id": "6c7a759182b6c43ce82fb58f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Deprivation and schooling were reconstructed for this review from the Inter-Agency Multidimensional Poverty Microdata (IMPM), the extract Frontex compiled for 2022, and the readings below are the review's, not the extract's.\n\nHouseholds were sorted by the number of deprivation indicators they carry and the share of children aged 10-14 unable to read a short sentence computed: 41 per cent of households with five indicators or more against 9 per cent of those with none. The extract holds an indicator flag for each household and a reading result for each child and never sets one against the other.\n\nHolding the school the child attends even narrows the gap to 14 points, so the school carries part of the gap the opening comparison put down to the count of indicators.\n\nHouseholds whose indicator for schooling was answered by proxy are dropped rather than counted as deprived on that indicator, and they are 4 per cent of households. A note in the annex gives the sorting rule, the pull date, with the base for each share alongside.\n\nMention: \"Inter-Agency Multidimensional Poverty Microdata (IMPM)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Inter-Agency Multidimensional Poverty Microdata relates deprivation indicators to children unable to read.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:b85b21e3a077f5c3c20a7b1c", "occurrence_id": "0193565a1e02e5f752767511", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Records handling and delay were assembled for this review out of the Audited Public Records Index (APRI), First Phase, the index kept for Lebanon in 2015, and what follows is the review's arithmetic, not the index's.\n\nAuthorities were sorted by the number of audit findings recorded against them and the median days before a request is answered computed: 22 days at authorities with three findings or fewer against 58 days at those with ten or more. The index holds a finding count for each authority and an answer date for each request and never sets one against the other.\n\nMaking the groups alike on the volume of requests the authority receives brings the gap down to 12 days, which leaves the volume as part of the distance the first comparison charged to the findings.\n\nAuthorities the index audits once in the period are dropped rather than counted as finding-free, and they are 6 per cent of authorities. Those references, the extraction day and the denominators are stored in the annex file.\n\nMention: \"Audited Public Records Index (APRI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Audited Public Records Index relates audit findings against authorities to request answering delay.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:4fbdf50780bfbe07c3d589b6", "occurrence_id": "c492b9b933361d73eca0091b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Inspection coverage and compliance were reconstructed for this review from the Rapid Labour Inspection Index, the index Russian Nati compiled for Nicaragua in 2020, and the readings below are the review's, not the index's.\n\nRegions were sorted by the inspections the index records per thousand workers and the share of establishments meeting the wage rule at the following visit computed: 71 per cent in regions above four inspections per thousand against 49 per cent in those below one. The index holds an inspection count per region and the follow-up visit records the wage finding per establishment, and neither reports the two together.\n\nHolding the size of the region's workforce even narrows the gap to 9 points, so workforce size carries part of what the unadjusted contrast attributed to the inspection rate.\n\nRegions the index covers for part of the year are dropped rather than counted at a lower rate, and they are 5 per cent of regions. A note in the annex gives the sorting rule, the pull date and the base for each share.\n\nMention: \"Rapid Labour Inspection Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Rapid Labour Inspection Index relates inspections per thousand workers to wage-rule compliance.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:18162431a7710e7d842bb04d", "occurrence_id": "2a04d5bde953ab7407f230d8", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Monitoring coverage and delivery were assembled for this review out of the Experimental Programme Monitoring Index, the index TurkSTAT compiled for Chengdu in 2019, and what follows is the review's arithmetic, not the index's.\n\nProjects were sorted by the number of monitoring visits the index records in the year and the share delivering on its planned schedule computed: 74 per cent of projects with four visits or more against 46 per cent of those with one or none. The index holds a visit count for each project and a delivery flag for the same project and never sets one against the other.\n\nMaking the groups alike on the size of the project's budget brings the gap down to 11 points, which leaves budget size as part of what the unadjusted difference owed to the visits.\n\nProjects terminated before the year's end are dropped rather than counted as late, and they are 7 per cent of projects. Those references, the extraction day and the denominators are held in the papers behind the annex.\n\nMention: \"Experimental Programme Monitoring Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Experimental Programme Monitoring Index relates monitoring visits to delivery on schedule.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:4b5a681974f81a48e9633e6e", "occurrence_id": "d114dd26087ac3a7dc84fb79", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Minimum needs and shortfall were reconstructed for this review from the Periodic Minimum Needs Survey, which UNHCR & WFP conducted in Singapore in 2023, and the readings below are the review's, not the survey's.\n\nHouseholds were sorted by the number of budget items they could not meet and the share reporting they went without a necessary medical item computed: 33 per cent of households missing six items or more against 7 per cent of those missing none. The survey holds an item list for each household and a medical item for the same household and never sets one against the other.\n\nHolding whether the household holds private insurance even narrows the gap to 11 points, so insurance carries part of what the first comparison credits to the count of missing items.\n\nHouseholds that refused the medical item but answered the rest are dropped rather than counted as meeting it, and they are 4 per cent of households. A note in the annex gives the sorting rule, the pull date and, alongside it, the base for each share.\n\nMention: \"Periodic Minimum Needs Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Periodic Minimum Needs Survey relates unmet budget items to going without necessary medical care.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:7c4f5bf634ca23174a1571c5", "occurrence_id": "b86985c7de685802b0ce1aa2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The poverty figures in this chapter are computed from the office's own panel, and the Government Poverty and Vulnerability Survey conducted by Federal Police in Ghana in 2017 is quoted only to place those figures in context.\n\nThe survey puts the share of households below the national poverty line at 24 per cent, the share vulnerable to falling below it at 19 per cent, and the share reporting a shock that pushed them below in the last two years at 11 per cent. All three are given as the survey issues them, with its own line stated, and no chapter figure was built from them.\n\nTheir use here is to show how the wider population stands beside the households the chapter's panel follows. The chapter's argument rests on the panel, which measures the same households across rounds rather than once at a visit, and the two are kept apart rather than merged.\n\nThe annex lists the survey tables consulted and the day the table was pulled, which is the full extent of the survey's use in this chapter.\n\nMention: \"Government Poverty and Vulnerability Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Government Poverty and Vulnerability Survey is quoted as context beside the office's poverty panel.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:33a910090d81157acdc04065", "occurrence_id": "00d2bdb8e50e494e2eece8d7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Remote sensing figures for the metropolitan frame are reported from the Metropolitan Remote Sensing Census (MRSC), Seventh Version, the census IIASA follows, and the four values are reproduced here as issued.\n\nThe census puts the share of the frame built over at 47 per cent, the share of that built area whose extent changed since the previous version at 9 per cent, the mean size of a newly mapped block at 0.6 hectares, and the share of the frame left unobserved for want of a clear image at 6 per cent. They were not re-derived here, and the census's own reading of a mapped block, which requires two clear observations, is given next to them.\n\nA block visible at one date only was not counted as built over, and a block cleared and rebuilt was counted as unchanged. Each figure keeps the metropolitan unit it was drawn from named with it, instead of averaged over the frame.\n\nThe census layer for each value: the day of the pull and the blocks each share divides sit with the annex tables.\n\nMention: \"Metropolitan Remote Sensing Census (MRSC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Metropolitan Remote Sensing Census supplies built-over share, extent change, new block size and unobserved frame.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:b2fdf9dd4b8aaec7e92e4fff", "occurrence_id": "792935e13dd367beaced2131", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Three metropolitan lines in the welfare annex come from the Metropolitan Household Consumption Index, the index WALS compiled for Rwanda in 2020, and all three are transcribed without alteration.\n\nThe index puts the share of metropolitan households below the national adequacy line at 16 per cent, the share of children with a health episode in the month gone untreated at 8 per cent, and the share of households with no child enrolled at 4 per cent. Nothing here was recalculated in this report, and the index's own reading of an untreated episode, which requires the household to have sought no care at all, travels with the figures.\n\nThe remaining annex lines come from the office's own household register, so the index carries three lines while the register carries the balance. A figure reported for one city keeps that city attached rather than being averaged across the metropolitan frame.\n\nThe annex log holds the index line behind each value, the day the rows were taken and the base of each share.\n\nMention: \"Metropolitan Household Consumption Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Metropolitan Household Consumption Index supplies adequacy, untreated episodes and households with no child enrolled.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:aacb335470453fbad83c4688", "occurrence_id": "3094f2643a3dda9466e85cdf", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Household records were catalogued into the operations file from the Monthly Operations Monitoring Registry, the register Freedom House maintains for the CHT and Sylhet divisions in 2019, and the work done on them was cataloguing rather than measurement.\n\nEach household was given the file's own site band and its contact dates checked against the file's month window; where a record carries a contact the file cannot place in a month, it was parked for the analyst rather than pushed into the nearest one. Households whose band the registry left blank were left unfiled rather than assumed to sit at the site median, and 690 of them entered the file in that condition. The registry's own month stamp was kept on every record so a later month can be matched without re-deriving it.\n\nThe operations file's coverage weights come from the site registration census, from which the registry supplied nothing; it is the household coding that lets those weights be read against contact for the delivery table the report draws up.\n\nThe coding sheet, the count of parked records and the day of deposit are listed in the annex log.\n\nMention: \"Monthly Operations Monitoring Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The Monthly Operations Monitoring Registry supplies the coded household records behind the operations file.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:e94dca41070e8e8a1b8ec993", "occurrence_id": "381c23cfbefac0cf4c3ecc78", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Water stress was derived for this chapter from the Diagnostic Water and Climate Registry (DWCR), Fifth Collection, the register kept for Beijing in 2022, and the figures below belong to the chapter rather than to the register's.\n\nSectors were sorted by the mean draw the register records against the available supply and the share of days on which a restriction was applied computed: 21 days at sectors drawing above four-fifths of supply against 4 days at those below half. The register holds a draw figure for each sector and a restriction count for the same sector and never sets one against the other.\n\nThat comparison supports one claim, that the abstraction permit should be tightened before the network is extended. The capital lines the proposal quotes come from the utility's own investment plan, from which the register supplied nothing.\n\nSectors the register lists with a draw but no supply figure are dropped rather than counted at the network average, and they are 5 per cent of sectors. The bands and the counts behind each share are recorded under the table.\n\nMention: \"Diagnostic Water and Climate Registry (DWCR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Diagnostic Water and Climate Registry supports a permit-tightening claim built from draw share and restriction days.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:c2cf9768e5bed3d2c29604fe", "occurrence_id": "3115aef02e2c10cbab64fd05", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Inequality and income were reconstructed for this review from the Combined Inequality Registry (CIR), Release 4, the register SBP maintains, and the readings below are the review's, not the register's.\n\nAreas were sorted by the share of income the top tenth holds and the share of households below the national adequacy line computed: 52 per cent of areas where the top tenth holds above a third against 24 per cent of those where it holds below a fifth. The register holds a top-tenth share for each area and an adequacy share for the same area and never sets one against the other.\n\nHolding the mean household size of the area even narrows the gap to 13 points, so household size carries part of the gap the opening contrast assigned to the top-tenth share.\n\nAreas the register reports on a base of fewer than fifty households are dropped rather than pooled with their neighbours, and they are 9 per cent of areas. A note in the annex gives the sorting rule, the pull date as well as the base for each share.\n\nMention: \"Combined Inequality Registry (CIR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Combined Inequality Registry relates the top tenth's income share to adequacy-line incidence.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:ec25e62ee1ebdf830ae95ae2", "occurrence_id": "b3f1fa35d9eb171bdf5283cb", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Internal movement figures measured across the regions are read from the Multi-Topic Internal Movement Dataset (MIMD), the geospatial dataset compiled for Ethiopia in 2021, and the four values are quoted here as printed.\n\nThe dataset puts the number of movement events recorded in the year at 214 000, the share of events ending in a settlement already hosting displaced households at 61 per cent, the mean distance between origin and destination at 38 kilometres, and the share of events whose destination lies outside the origin's own zone at 29 per cent. No arithmetic was applied to them here, and the dataset's own reading of a movement event, which requires a change of dwelling held for a month, is bound to the figures.\n\nA household that moved twice in a month was counted once, and a move into a camp that closed within the month was still counted as ending in a hosting settlement. Each figure keeps the zone it was drawn from listed with it, rather than smoothed across the regions.\n\nUnder the table are the dataset line for each value, the day the extract ran, and the events each share divides.\n\nMention: \"Multi-Topic Internal Movement Dataset (MIMD)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Multi-Topic Internal Movement Dataset supplies movement events, hosting destinations, distance and zone crossing.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:19b9bc78030e528a18736129", "occurrence_id": "2a30ce409d870b42a66a2330", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The facility figure in this chapter is the ministry's own, taken from its own inspection book, and the Verified Facility Assessment Survey (VFAS), Collection 6, the survey UNHCR, IOM follows, is cited to test it.\n\nThe ministry's book shows 82 per cent of facilities meeting its basic service standard. The survey shows 74 per cent for the same facilities, because it counts a facility as meeting the standard only where every item on its list was seen on the day, while the ministry's book accepts a facility missing one item. Neither rule was applied to the other's book.\n\nThe 8-point distance is smaller than the movement the ministry's book shows between two rounds, so the ministry's figure is printed as returned and the survey supplies no revision to it and no facilities for the districts its teams did not enter.\n\nExcluded are facilities the survey visited twice in the collection, 210 of them, since the two visits give different answers. Which facilities each source counts is listed in the working papers behind the annex.\n\nMention: \"Verified Facility Assessment Survey (VFAS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Verified Facility Assessment Survey checks the ministry's share of facilities meeting its service standard.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:af3d13bd36e9b2c4d67298cb", "occurrence_id": "73977b653fbc598ed5f1f61e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two service lines in the state annex come from the Rapid Service Register Report, the report MRC issued for Lagos State in 2024, and both are transcribed without alteration.\n\nThe report puts the share of registered service points open on the day of the round at 73 per cent and the mean minutes a customer waits at an open point at 55. Nothing here was re-derived for this report, and the report's own reading of an open point, which requires a transaction to have taken place that morning rather than the gate to stand unlocked, travels with the figures.\n\nThe annex's remaining lines are compiled from the ministry's own facility listing, so the report carries two lines while the listing supplies the balance. A figure reported for one local government area keeps that area attached rather than being averaged across the state, and where the report and the listing disagree about whether a point was open the report's figure is kept for these two lines and the listing's for the rest, with the difference noted under the table rather than smoothed away.\n\nPoints under renovation on the day of the round are excluded from both lines, since neither source can say whether they served anyone. The annex log holds the report page behind each value, the pull date together with the counts each share divides.\n\nMention: \"Rapid Service Register Report\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Rapid Service Register Report supplies open-point coverage and customer waiting minutes.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:de5c0516e4217cff7a34076c", "occurrence_id": "a3ab5466def862822e819370", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Unpaid work and labour supply were assembled for this review out of the Integrated Time Use System (ITUS), First Phase, the register PWT maintains, and what follows is the review's arithmetic, not the register's.\n\nAdults were sorted by the hours they record for unpaid household work in the week and the share reporting they took paid work in the same week computed: 38 per cent of adults below ten unpaid hours against 21 per cent of those above twenty-five. The register holds an unpaid-hours item for each adult and a paid-work flag for the same adult and never sets one against the other.\n\nMaking the groups alike on the number of children in the household brings the gap down to 8 points, which leaves the children as part of the distance the first comparison ascribed to the unpaid hours.\n\nAdults the register lists as seeking work but recording no activity are dropped rather than counted as unpaid workers, and they are 4 per cent of adults. Those references, the extraction day and the denominators are set down in the annex working papers.\n\nMention: \"Integrated Time Use System (ITUS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Integrated Time Use System relates unpaid household hours to taking paid work in the week.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:26f81a63e16ce7491956d95d", "occurrence_id": "c4824248becbdd6bbac87bfe", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two cooperative lines in the agricultural annex come from the High-Resolution Cooperative Membership Registry, the register New York Fed maintains for India in 2024, and both are transcribed without alteration.\n\nThe registry puts the share of holdings belonging to at least one cooperative at 31 per cent and the mean number of years a member has held that membership at 8. Nothing here was worked out again here, and the registry's own reading of membership, which requires a contribution paid in the last season, travels with the figures.\n\nThe other lines in the annex are copied from the department's own marketing returns, so the registry carries two lines while the returns carry the balance. A district figure carries its district with it instead of being smoothed across the country.\n\nThe annex log holds the registry entry behind each value, the day of extraction plus the counts under each share.\n\nMention: \"High-Resolution Cooperative Membership Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The High-Resolution Cooperative Membership Registry supplies cooperative membership and years of membership.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:9db840992a54ffd5c3e438d9", "occurrence_id": "cb6a1aef378fd6554c9b230a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Network reach and delay were reconstructed for this review from the Stratified Transport Network Collection (STNC), First Collection, the collection compiled for Cyprus in 2022, and the readings below are the review's, not the collection's.\n\nWards were matched to their network class and grouped by the headway the collection records at the nearest stop, and the share of commuters exceeding 45 minutes each way computed: 29 per cent of wards on a headway above half an hour against 8 per cent of those under ten minutes. The collection grades a headway per stop and the commuter survey records the journey per person, and neither reports the two together.\n\nHolding the ward's car ownership even narrows the gap to 11 points, so car ownership carries part of what the raw comparison laid at the door of the headway.\n\nWards the collection maps a stop for but scores as withdrawn are dropped rather than counted as unserved, and they are 5 per cent of wards. A note in the annex gives the grouping rule, the pull date, and the base for each share.\n\nMention: \"Stratified Transport Network Collection (STNC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Stratified Transport Network Collection relates stop headway to commuters exceeding a 45-minute journey.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:a7bb57075e0b2edfb2ead392", "occurrence_id": "86eec2de029c5f132db55ee1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two cadastral lines in the risk annex come from the Quarterly Cadastral Frame Dataset (QCFD), Collection 7, the geospatial dataset compiled for South Africa in 2023, and both are transcribed without alteration.\n\nThe dataset puts the share of parcels in the frame lying within a declared fire-prone band at 18 per cent and the share whose boundary was redrawn between collections at 6 per cent. Nothing here was computed afresh for this report, and the dataset's own reading of a redrawn boundary, which requires the change to exceed its minimum mapping area, travels with the figures.\n\nThe annex's remaining entries follow the municipality's own valuation roll, so the dataset carries two lines while the roll holds the remainder. No figure is folded across the country; each one names the district it came from.\n\nThe annex log holds the dataset layer behind each value, the date it was drawn and the population each share divides.\n\nMention: \"Quarterly Cadastral Frame Dataset (QCFD)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Quarterly Cadastral Frame Dataset supplies fire-band parcel share and boundary redraws between collections.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:87bbc2612d178ef81a3a35da", "occurrence_id": "649275243161a0049fd566c3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The flooded area in this chapter is the authority's own, taken from its damage schedules, and the Continuous Flood Extent Index (CFEI), Release 6, the index D¨ur et al. maintain, is cited to test it.\n\nThe authority's schedules put 3 400 square kilometres flooded in the season. The index puts 4 050 for the same basins, because it counts standing water visible at the next clear observation, which the schedules record only where a report reaches the office. Each basis keeps its own figures untouched.\n\nThe 650 distance is smaller than the area the authority's schedules show between two seasons, so the authority's figure is printed as scheduled and the index supplies no revision to it and no basins for the reaches its images did not cover.\n\nExcluded are reaches the index dates to a month rather than a day, 22 of them, since the two spells cannot be separated for them. Which area each source counts is noted in the method note in the annex.\n\nMention: \"Continuous Flood Extent Index (CFEI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Continuous Flood Extent Index checks the authority's square kilometres flooded in the season.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:0a856c56fa6622937f693ce1", "occurrence_id": "f1ddfb0cedf4fe99b7b0f908", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The livelihood figures in this chapter are drawn from the district's own household listing, and the District Socioeconomic Survey (DSS), Phase 3, the survey World Bank-sponsored follows, is referred to only as background to them.\n\nThe survey puts the share of households with no member in paid work at 27 per cent, the share receiving any transfer at 34 per cent, and the share reporting a member who moved for work in the last two years at 19 per cent. All three are given as the survey issues them, with its own household definition stated, and no chapter figure rests on them.\n\nIts responses are kept out of the chapter's tables. The figures the chapter argues from come from the district listing, which names each household rather than a sample of them, and where the two accounts of work differ the difference is noted rather than reconciled.\n\nThe annex lists the survey tables consulted and the date the file was taken, and the survey is used for nothing beyond that background.\n\nMention: \"District Socioeconomic Survey (DSS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The District Socioeconomic Survey is background to the district's own household livelihood listing.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:157088745c05a7fcd169fdcd", "occurrence_id": "3f1bf49d0ef97194f2690d35", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Staffing and waiting were estimated for this review using the Stratified Health Workforce Survey (SHWS), Sixth Phase, the survey conducted in Sulaymaniya in 2023, and the figures are the review's own.\n\nFacilities were sorted by the clinical staff rostered on the day of visit and the mean minutes a patient waits before being seen computed: 38 minutes at facilities rostering five staff or more against 96 minutes at those rostering one or two. The survey records a roster and a waiting time for the same facility and never sets one against the other.\n\nMaking the groups alike on the number of patients arriving in the session brings the gap down to 19 minutes, which leaves the arrivals as part of the gap the opening comparison put down to the roster.\n\nFacilities that closed before the visit ended are dropped rather than counted at the patients they saw, and they are 6 per cent of facilities. Working notes list the sorting rule, the extraction day, with the denominators each share rests on alongside.\n\nMention: \"Stratified Health Workforce Survey (SHWS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Stratified Health Workforce Survey relates rostered clinical staff to patient waiting minutes.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:07a921cfeb32324a4ef9aeed", "occurrence_id": "54ce7110076f8d80cf9388f2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Livelihood change and work were reconstructed for this review from the Panel Rural Livelihoods Survey (PRLS), the survey followed in Yemen in 2017, and the readings below are the review's, not the survey's.\n\nHouseholds were sorted by whether they lost their main source of work between the two rounds and the share taking a second occupation by the following round computed: 46 per cent of households that lost one against 18 per cent of those that did not. The survey holds a loss item for each household and an occupation list for the same household and never sets one against the other.\n\nHolding the land the household still holds even narrows the gap to 14 points, so the land carries part of the distance the first comparison charged to the loss.\n\nHouseholds that did not appear in the second round are dropped rather than counted as unchanged, and they are 9 per cent of households. A note in the annex gives the sorting rule, the pull date and the base for each share.\n\nMention: \"Panel Rural Livelihoods Survey (PRLS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Panel Rural Livelihoods Survey relates loss of main work to taking a second occupation.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:163762a6cdb87823538c796f", "occurrence_id": "32e501f03bd89546d4cb68cd", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Building footprint figures for the state are reported from the Nationwide Building Footprint System, the register UNDP maintains for Florida as of 2019, and the four values are carried here as tabulated.\n\nThe register puts the number of footprints held for the state at 7.4 million, the share carrying a roof material attribute at 63 per cent, the share whose outline changed between the two releases at 4 per cent, and the share sitting inside a declared coastal band at 12 per cent. No reweighting was applied to them, and the register's own reading of a footprint, which requires a roof line visible at both passes, goes with the figures.\n\nA structure with its roof removed was not counted as a footprint, and an outline redrawn for cartographic clarity was not counted as a change. Each figure keeps the county it was drawn from carried with it, rather than mixed across the state.\n\nThe register layer for each value, the day of the extract as well as the footprints each share divides are recorded under the table.\n\nMention: \"Nationwide Building Footprint System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Nationwide Building Footprint System supplies footprint counts, roof attributes, outline change and coastal position.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:6c71ee0d8ef33b6a71cb38ed", "occurrence_id": "6f9ab95d89e9761635c4c129", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Record completeness and delay were assembled for this review out of the Civil Administrative Records Index, the index EMES compiled for Yemen in 2022, and what follows is the review's arithmetic, not the index's.\n\nOffices were sorted by the completeness the index scores for their birth register and the median days before a certificate is issued computed: 9 days at offices scoring above nine-tenths against 41 days at those below six-tenths. The index holds a completeness score for each office and an issue date for each certificate and never sets one against the other.\n\nMaking the groups alike on the staff an office has for the work brings the gap down to 11 days, which leaves staffing as part of what the unadjusted contrast attributed to the score.\n\nOffices the index scores on a part-year of returns are dropped rather than counted at that score, and they are 5 per cent of offices. Those references, the extraction day and the denominators are written up in the annex's method note.\n\nMention: \"Civil Administrative Records Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Civil Administrative Records Index relates birth-register completeness to days before a certificate issues.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:304c9e1f79d21985a0139c55", "occurrence_id": "6a3ed063a85589225e089663", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two census lines in the labour annex come from the Combined Socioeconomic Census (CSC), Round 4, the census the World Economic Forum runs, and both are transcribed without alteration.\n\nThe census puts the share of households whose principal income is a wage at 44 per cent and the share of working-age adults with no occupation at all at 16 per cent. Nothing here was recomputed for this report, and the census's own reading of a principal income, which takes the source the household names first, travels with the figures.\n\nThe rest of the annex lines come from the office's own client register, so the census carries two lines while the register carries what is left over. Nothing is averaged across the frame: a figure for one subdivision keeps that subdivision named.\n\nThe annex log holds the census table behind each value, the date of the pull and, alongside it, the denominators each share uses.\n\nMention: \"Combined Socioeconomic Census (CSC)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Combined Socioeconomic Census supplies wage-income households and adults without occupation.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:c2007db8211f90169ffdb459", "occurrence_id": "85775a1deec8c2458032b37b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two household lines in the water annex come from the Household Water and Climate Registry (HWCR), Release 3, the register APRA maintains, and both are transcribed without alteration.\n\nThe registry puts the share of households whose supply failed for more than a day in the driest month at 37 per cent and the mean litres a household reports drawing in a day at 60. Nothing here was recomputed here, and the registry's own reading of a failed supply, which requires the household to have bought water as a result, travels with the figures.\n\nThe other annex entries are assembled from the utility's own outage book, so the registry carries two lines while the book accounts for the remaining lines. A figure reported for one zone keeps that zone attached rather than being averaged across the service area.\n\nThe annex log holds the registry entry behind each value, the date on the extract as well as the denominators behind each share.\n\nMention: \"Household Water and Climate Registry (HWCR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Household Water and Climate Registry supplies supply failures in the dry month and litres drawn per day.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:54e660c83ab6499e40d3db2d", "occurrence_id": "b4cb622f7b914c6161401f0c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Vocational training the year figures are compiled from the Ministerial Vocational Training Survey (MVTS), the survey the UN Population Division conducted in 2019, and the four values pass into this report as printed.\n\nThe survey puts the share of leavers placed in a related job within six months at 58 per cent, the mean months a course runs at 9, the share of places taken by women at 41 per cent, and the share of leavers whose employer helped fund the course at 22 per cent. Nothing was re-estimated here, and the survey's own reading of placement, which requires the job to have begun after the course closed, travels with the values.\n\nA leaver already holding the job before enrolling was not counted as placed, and a course shortened by a closure was counted at its planned length. Each figure keeps the institute it was drawn from with it, rather than pooled across the country.\n\nThe annex note carries the survey line for each value, plus the day of the pull and the leavers each share divides.\n\nMention: \"Ministerial Vocational Training Survey (MVTS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Ministerial Vocational Training Survey supplies placement, course length, women's places and employer funding.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:97e7b391028c152ecefeae39", "occurrence_id": "277ff1d8a532e877abc836ab", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The poverty figure in this chapter is the office's own, taken from its beneficiary roll, and the Civil Poverty and Vulnerability Survey (CPVS), Second Release, the survey conducted in Sudan in 2022, is cited to test it.\n\nThe office's roll shows 34 per cent of its households below the national line. The survey shows 41 per cent for the same areas, because it measures the household as it stands at the visit, while the roll carries the assessment made when the household enrolled. Neither basis was carried across to the other's count.\n\nThe 7-point distance is smaller than the movement the roll shows between two enrolment drives, so the roll's figure is printed as enrolled and the survey supplies no revision to it and no households for the locations its teams did not reach.\n\nExcluded are households the survey lists at a location the roll cannot place, 1 080 of them, since the two cannot be matched. Which households each source counts is laid out in the annex's method sheet.\n\nMention: \"Civil Poverty and Vulnerability Survey (CPVS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Civil Poverty and Vulnerability Survey checks the office's share of households below the national line.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:1b56765b98932e1b03689225", "occurrence_id": "af1ba2bd1fec93ca57863ca4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Work and assets were reconstructed for this review from the Experimental Rural Livelihoods System (ERLS), the register FISE maintains for 2016, and the readings below are the review's, not the register's.\n\nHouseholds were sorted by whether the register records livestock held and the share of adults in the household with off-farm work computed: 29 per cent of households holding livestock against 47 per cent of those holding none. The register holds a livestock item for each household and a work item for each adult and never sets one against the other.\n\nHolding the area the household cultivates even narrows the gap to 9 points, so the cultivated area carries part of what the unadjusted difference owed to the livestock.\n\nHouseholds the register lists with livestock but no head count are dropped rather than counted as holding none, and they are 6 per cent of households. A note in the annex gives the sorting rule, the pull date together with the base for each share.\n\nMention: \"Experimental Rural Livelihoods System (ERLS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Experimental Rural Livelihoods System relates livestock held to adults in off-farm work.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:a671f841cdace5226fe2b073", "occurrence_id": "20930e7780b1306106884abe", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Storm surge figures for the whole coast are sourced from the Enumerative Storm Surge Dataset (ESSD), Fourth Phase, the geospatial dataset compiled for Mauritania in 2018, and the four values come into this report unchanged.\n\nThe dataset puts the length of coastline reached inland by more than a metre at 210 kilometres, the share of that length whose households were warned beforehand at 34 per cent, the number of dwellings listed as damaged within the reached band at 3 900, and the share of the reached band falling inside a declared hazard zone at 58 per cent. No figure was recalculated in this report, and the dataset's own reading of a reached band, which requires water observed inland rather than modelled, follows the figures.\n\nA band reached at one observation only was still counted as reached, and a dwelling damaged by wind inside the band was not counted as damaged by the surge. Each figure keeps the commune it was drawn from attached, instead of being blended across the coast.\n\nThe table's footnotes hold the dataset line for each value, along with the extraction day and the dwellings each share divides.\n\nMention: \"Enumerative Storm Surge Dataset (ESSD)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Enumerative Storm Surge Dataset supplies reached coastline, warnings, damaged dwellings and hazard-zone cover.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:291f47db477c5e7af278d7b8", "occurrence_id": "fe81040282b02d424df67d4d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Road condition and travel were derived for this chapter from the Midline Road Segment Dataset (MRSD), the geospatial dataset MICT/DTO maintains for the Republic of Zambia, and the figures below are counted by the chapter rather than by the dataset's.\n\nSegments were sorted by the surface the dataset records and the mean speed a survey run achieves over them computed: 46 kilometres an hour on sealed segments against 17 kilometres an hour on earth segments. The dataset holds a surface class for each segment and the survey run records a speed for the same segment and neither reports the two together.\n\nThat comparison supports one claim, that grading should follow the sealed network rather than reach the last village first. The cost lines the proposal quotes come from the ministry's own works programme, from which the dataset supplied nothing.\n\nSegments the dataset leaves uncrossed at the midline are dropped rather than counted at their opening condition, and they are 7 per cent of segments. The bands and the counts behind each speed are recorded under the table.\n\nMention: \"Midline Road Segment Dataset (MRSD)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Midline Road Segment Dataset supports a grading-priority claim built from surface class and travel speed.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:65e51d9ace9640f48b98d4a3", "occurrence_id": "160d43404fbf09c2d0b20f1c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Asset ownership the year figures are tabulated by the Subnational Asset Ownership Survey (SAOS), Eighth Wave, the survey DANE follows in 2019, and the four values enter this report as they stand.\n\nThe survey puts the share of households owning their dwelling at 63 per cent, the share holding no livestock at 71 per cent, the share with a head who completed lower secondary at 48 per cent, and the share holding a bank account at 22 per cent. They are carried as published, and the survey's own reading of ownership, which requires the title to name a household member, is reported with them.\n\nA dwelling held under a customary arrangement was not counted as owned, and poultry kept for the market was counted as livestock. Each figure keeps the department it was drawn from shown with it, instead of combined into one country figure.\n\nUnder the table you will find the survey line for each value, the extraction date plus the households each share divides.\n\nMention: \"Subnational Asset Ownership Survey (SAOS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Subnational Asset Ownership Survey supplies dwelling ownership, livestock, schooling of the head and accounts.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:8e4e7230b571119d77d1d22a", "occurrence_id": "088d9e30a546b02eca9def1d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Deprivation and support were reconstructed for this review from the Endline Multidimensional Poverty Registry (EMPR), the register SHIP maintains for Ukraine, and the readings below are the review's, not the register's.\n\nHouseholds were sorted by the number of deprivation indicators they carry and the share receiving any form of support computed: 82 per cent of households with four indicators or more against 36 per cent of those with one or none. The register holds an indicator flag for each household and a support item for the same household and never sets one against the other.\n\nHolding the distance from the household to the nearest payment point even narrows the gap to 21 points, so distance carries part of what the first comparison credits to the count of indicators.\n\nHouseholds whose indicator answers came from a neighbour rather than a member are dropped rather than counted as deprived, and they are 3 per cent of households. A note in the annex gives the sorting rule, the pull date, and the base for each share.\n\nMention: \"Endline Multidimensional Poverty Registry (EMPR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Endline Multidimensional Poverty Registry relates deprivation indicators to receiving support.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:e8dba0876a8c34fd3147e42f", "occurrence_id": "9be424ae6932c4ce7b5e2633", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Two record lines in the annex come from the Standardised Administrative Records Registry (SARR), Collection 9, the register TURKSTAT maintains, and both are transcribed without alteration.\n\nThe registry puts the share of record series held in a format the centre can read at 69 per cent and the mean years a series is kept before it is appraised at 7. Nothing here was recomputed here, and the registry's own reading of a readable format, which requires the series to open without a licensed tool, travels with the figures.\n\nThe annex's remaining entries follow the department's own retention schedule, so the registry carries two lines while the schedule carries the lines the table does not. A figure reported for one series keeps that series attached rather than being averaged across the collection.\n\nThe annex log holds the registry entry behind each value, the day the table was pulled and the counts each share divides.\n\nMention: \"Standardised Administrative Records Registry (SARR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Standardised Administrative Records Registry supplies readable-format series and years before appraisal.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:fb1269e3fe7833d5bb09ed0e", "occurrence_id": "6af8e28bda66b9fb6f42715b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Housing condition the figures for the country are reported by the National Housing Conditions Index, the indicator WIPO compiled for Spain in 2018, and the four values reach this report as published.\n\nThe index puts the share of dwellings lacking central heating at 29 per cent, the share whose households spend above a third of income on housing at 24 per cent, the mean number of rooms per occupant at 2.3, and the share of dwellings classified as deficient in insulation at 37 per cent. Nothing here was recomputed, and the index's own reading of a dwelling, which requires it to be habitable at the reference date, is attached to them.\n\nA dwelling let for part of the year was counted as a dwelling, and a room used only for storage was counted as a room. Each figure keeps the region it was drawn from reported with it, rather than blended into a single country share.\n\nThe annex keeps the index line for each value: the day of the pull and the dwellings each share divides.\n\nMention: \"National Housing Conditions Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The National Housing Conditions Index supplies heating lack, housing cost burden, rooms per occupant and insulation.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:61fef8b99df495b5846bfeab", "occurrence_id": "58b1cd000805784f7d2e0c71", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Mental health and hardship figures are reported from the Audited Mental Health Survey (AMHS), Edition 7, the survey UBOS follows, and the four values appear here exactly as printed.\n\nThe survey puts the share of adults reporting a depressive episode in the last year at 11 per cent, the share of those adults who sought no care at 46 per cent, the share of households where someone's mental health stopped them working at 7 per cent, and the share of those households below the national adequacy line at 52 per cent. They are reproduced as tabulated, and the survey's own reading of an episode, which requires the question set to be answered in full, is stated with them.\n\nAn episode treated with medication bought without a prescription was counted as an episode with no care sought, and a reduced week was counted as stopped from working. Each figure keeps the stratum it was drawn from kept with it, instead of being rolled up to the sample.\n\nBeneath the table sit the survey line for each value, the day the rows were taken, and the adults each share divides.\n\nMention: \"Audited Mental Health Survey (AMHS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Audited Mental Health Survey supplies episodes, untreated share, work stopped and adequacy-line incidence.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:9622ced72da0142a48a61bad", "occurrence_id": "b841423092e638e9449930ea", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The building count in this chapter is the office's own, taken from its permit file, and the High-Frequency Building Footprint Dataset (HBFD), the geospatial dataset WITS compiled for 2021, is cited to test it.\n\nThe office's file shows 12 400 buildings completed in the year. The dataset shows 13 900 appearing for the same areas, because it counts a structure once its roof is visible at a pass, which the permit file records only when the completion notice is returned. One basis was never used to adjust the other's count.\n\nThe 1 500 distance is smaller than the number the file shows in a quarter at the year's peak, so the office's figure is printed as filed and the dataset supplies no revision to it and no buildings for the areas its passes did not cover.\n\nExcluded are footprints the dataset dates to a pass spanning two years, 340 of them, since the year cannot be fixed for them. Which buildings each source counts is set out in the annex notes.\n\nMention: \"High-Frequency Building Footprint Dataset (HBFD)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The High-Frequency Building Footprint Dataset checks the office's count of buildings completed in the year.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:3396230947aaa7d1b1da5fb6", "occurrence_id": "c871dd4849dd5ea6b46043b3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Deprivation and work were reconstructed for this review from the Decennial Multidimensional Poverty Survey (DMPS), the survey conducted in Turkey in 2021, and the readings below are the review's, not the survey's.\n\nHouseholds were sorted by the number of deprivation indicators they carry and the share of adults with no regular work computed: 57 per cent of households with four indicators or more against 18 per cent of those with one or none. The survey holds an indicator flag for each household and a work item for each adult and never sets one against the other.\n\nHolding the education of the household head even narrows the gap to 19 points, so education carries part of the gap the opening contrast assigned to the count of indicators.\n\nHouseholds answering the indicators by proxy are dropped rather than counted as deprived on them, and they are 4 per cent of households. A note in the annex gives the sorting rule, the pull date, with the base for each share alongside.\n\nMention: \"Decennial Multidimensional Poverty Survey (DMPS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Decennial Multidimensional Poverty Survey relates deprivation indicators to adults without regular work.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:705818ede5e3e84e9172997b", "occurrence_id": "6053fd4d31eac060d9254d52", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Whether the agency's own count of crossings holds was tested against the Rapid Border Crossing Survey, which the Banque de France conducted across Africa and South Asia in 2020, and the survey is the principal source for the test.\n\nThe agency's post book gives 214 000 crossings in the year. The survey gives 268 000 for the same posts, because it counts a crossing reported by a household that made it and never presented itself at a post, which the book records only where an officer wrote it down. The two bases were not applied to each other's count.\n\nThe survey therefore carries the chapter's count of crossings, and the agency's smaller figure is printed beside it so the reader can see that the distance is the recording rule rather than a disagreement about the year. Where a crossing appears in both, the agency's date is kept and the two counts are not averaged.\n\nCrossings the survey dates to a month rather than a day are dropped from both rather than counted by the survey alone, 7 per cent of its listings. The recording rule each source works to is written up in the annex file.\n\nMention: \"Rapid Border Crossing Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Rapid Border Crossing Survey tests the agency's count of border crossings in the year.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:cd6fa7fcaa955f48cbdddd6e", "occurrence_id": "0736a411cf9fed4b089f3cf8", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Socioeconomic figures covering the country are drawn from the Analytical Socioeconomic Survey, the survey FEPB conducted in Pakistan in 2021, and the four values are reproduced here as issued.\n\nThe survey puts the share of households below the provincial welfare line at 31 per cent, the share whose nearest all-season road lies beyond five kilometres at 22 per cent, the share with a household member absent for work at 18 per cent, and the share whose dwelling stands in a flood-designated unit at 9 per cent. They were not re-derived here, and the survey's own reading of a household comes with them.\n\nA member working in the same union was not counted as absent, and a dwelling on the edge of a designated unit was counted as inside it. Each figure keeps the union it was drawn from named with it, instead of averaged over the country.\n\nWorking notes record the survey line for each value, the day the extract ran as well as the households each share divides.\n\nMention: \"Analytical Socioeconomic Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Analytical Socioeconomic Survey supplies welfare-line incidence, road distance, absent members and flood-unit position.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:4ce715b44fd64614f9cf7923", "occurrence_id": "b9a3a96c1afeca7e15bf5ad8", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Elevation tiles were catalogued into the analysis frame from the Biennial Elevation Model System (BEMS), Collection 3, the register the International Labor Organization maintains, and the work done on them was cataloguing rather than measurement.\n\nEach tile was given the frame's own vertical datum band and its acquisition dates checked against the frame's reference window; where the system carries a tile the frame has no band for, it was parked for the analyst rather than entered under the nearest one. Tiles the system lists without an acquisition date were left unbanded rather than dated to the collection, and 640 of them entered the frame in that condition. The system's own collection stamp was kept on every tile so a later collection can be matched without re-deriving it.\n\nThe frame's population surface comes from the census, from which the system supplied nothing; it is the tile coding that lets that surface be read against slope for the exposure table the report draws up.\n\nThe coding sheet, the count of parked tiles and the day of deposit are detailed in the annex log.\n\nMention: \"Biennial Elevation Model System (BEMS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The Biennial Elevation Model System supplies the coded elevation tiles behind the analysis frame.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:fbb5e8c267094cda84d9afe1", "occurrence_id": "45bafac5c215b57c064948c0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Employment and commuting were reconstructed for this review from the Record-Based Labour Force System (RLFS), First Round, the register kept for Panama in 2021, and the readings below are the review's, not the register's.\n\nWorkers were sorted by the commuting time the register records and the share holding a second job computed: 14 per cent of workers above an hour each way against 5 per cent of those under fifteen minutes. The register holds a commuting time for each worker and a second-job flag for the same worker and never sets one against the other.\n\nHolding the worker's hours in the main job even narrows the gap to 4 points, so hours carry part of the distance the first comparison ascribed to the commute.\n\nWorkers the register lists with a commute but no main job are dropped rather than counted as second-job holders, and they are 4 per cent of workers. A note in the annex gives the sorting rule, the pull date and the base for each share.\n\nMention: \"Record-Based Labour Force System (RLFS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Record-Based Labour Force System relates commuting time to workers holding a second job.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:470347c8a14b47c6d784554a", "occurrence_id": "ff5969be9883c16149ba1aeb", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The crop figures in this chapter are compiled from the board's own returns, and the Annual Crop and Livestock Registry (ACLR), Eighth Round, the register S&P maintains, is quoted only to place those figures in context.\n\nThe registry puts the share of holdings reporting a cereal crop at 68 per cent, the mean head of cattle per holding at 4, and the share of holdings reporting a fodder shortfall in the last season at 26 per cent. All three are given as the registry issues them, with its own holding definition stated, and no chapter figure was built from them.\n\nTheir use here is to show how the national picture stands beside the three districts whose returns the chapter works from. The chapter's argument rests on those returns, which name each holding rather than sampling them, and the two are kept apart rather than merged.\n\nThe annex lists the registry tables consulted and the pull date, which is the full extent of the registry's use in this chapter.\n\nMention: \"Annual Crop and Livestock Registry (ACLR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Annual Crop and Livestock Registry is quoted as context beside the board's own crop returns.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:a18c7e7ec7979e3eb5fea4ca", "occurrence_id": "e54ffd694e5d6da296855bb9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Apprenticeship and schooling were estimated for this review using the Departmental Apprenticeship Survey, which FISE conducted in Niger in 2018, and the figures are the review's own.\n\nYoung people were sorted by the years of schooling they completed and the share finishing an apprenticeship contract computed: 62 per cent of those with six years or more against 34 per cent of those with three or fewer. The survey holds a schooling item for each young person and a completion flag for the same person and never sets one against the other.\n\nMaking the groups alike on the distance from the young person's home to the workshop brings the gap down to 12 points, which leaves distance as part of what the raw comparison laid at the door of schooling.\n\nYoung people whose master closed mid-contract are dropped rather than counted as dropouts, and they are 7 per cent of the group. Working notes list the sorting rule, the extraction day and, alongside it, the denominators each share rests on.\n\nMention: \"Departmental Apprenticeship Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Departmental Apprenticeship Survey relates years of schooling to finishing an apprenticeship contract.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:ab759a3447330004ca0aad19", "occurrence_id": "11d113b415b09598c48ccf53", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Public sector monitoring figures covering the whole municipality derive from the Stratified Public Sector Monitoring Index, the indicator the Small Arms Survey compiled for Beijing in 2023, and the four values are quoted here as printed.\n\nThe index puts the share of bureaux publishing a performance return each quarter at 71 per cent, the mean indicators a bureau reports at 12, the share of returns checked by an office outside the bureau at 34 per cent, and the share of bureaux whose return changed after being checked at 19 per cent. No arithmetic was applied to them here, and the index's own reading of a published return, which requires it to stand on a public page rather than in an internal file, is kept with them.\n\nA return posted late for the quarter was still counted as published, and a check made by a sister bureau was counted as external. Each figure keeps the bureau it was drawn from listed with it, rather than smoothed across the municipality.\n\nThe documentation under the table gives the index line for each value, plus the day of extraction and the bureaux each share divides.\n\nMention: \"Stratified Public Sector Monitoring Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Stratified Public Sector Monitoring Index supplies published returns, indicator counts, external checks and revisions.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:aa96cfc2cc8dcd6104bb09d6", "occurrence_id": "bf74abfc6055284cc70e6ddd", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The child figure in this chapter is the institute's own, taken from its school health returns, and the Harmonised Child Wellbeing Survey (HCWS), First Round, the survey conducted in Kosovo in 2019, is cited to test it.\n\nThe institute's returns show 12 per cent of children recorded as going without a needed check-up. The survey shows 17 per cent for the same ages, because it asks the household directly and counts a check-up missed for any reason, while the returns record only a child brought to a clinic and found unpaid for. Neither basis reworks the other's count.\n\nThe 5-point distance is smaller than the movement the returns show between two school years, so the institute's figure is printed as returned and the survey supplies no revision to it and no children for the households its teams did not reach.\n\nExcluded are children the survey lists in a household the returns cannot match, 380 of them, since the two records cannot be joined. Which children each source counts is detailed in the papers behind the annex.\n\nMention: \"Harmonised Child Wellbeing Survey (HCWS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Harmonised Child Wellbeing Survey checks the institute's share of children missing a needed check-up.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:1f1e9ba5b63a23c214ce142a", "occurrence_id": "2f69b902b42649057e8d0dc3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Three housing lines in the labour annex come from the Representative Housing Conditions Survey (RHCS), Collection 7, the survey D¨ur et al. follows, and all three are transcribed without alteration.\n\nThe survey puts the share of households renting their dwelling at 38 per cent, the mean share of household earnings spent on rent at 27 per cent, and the share of renting households reporting a rent arrears episode in the last year at 14 per cent. Nothing here was recomputed in this report, and the survey's own reading of rent, which counts a payment to a landlord but not one to a relative, travels with the figures.\n\nThe annex's other entries are drawn from the office's own caseload file, so the survey carries three lines while the file holds the other lines. A figure reported for one stratum keeps that stratum attached rather than being averaged across the sample, and where the survey and the file disagree about who is renting the survey's figure is kept for these three lines and the file's for the rest, with the difference noted under the table rather than smoothed away.\n\nHouseholds living rent-free in an employer's dwelling are excluded from all three lines, since no rent passes for them to report. The annex log holds the survey line behind each value, the date it was drawn and the counts the shares rest on.\n\nMention: \"Representative Housing Conditions Survey (RHCS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Representative Housing Conditions Survey supplies renting households, rent burden and arrears episodes.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:98e39c275d51b6db8ba5787c", "occurrence_id": "750e2eb2dd1649b890cfc60b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The delivery figure in this chapter is the programme's own, taken from its activity returns, and the Triennial Programme Monitoring Registry (TPMR), Release 4, the register the United Nations maintains for Sudan, is cited to test it.\n\nThe programme's returns show 88 per cent of its planned activities delivered in the period. The registry shows 79 per cent for the same sites, because it counts an activity as delivered only where its own monitor saw it, while the programme's returns accept the site manager's report. Neither rule was applied to the other's returns.\n\nThe 9-point distance is smaller than the movement the returns show between two periods, so the programme's figure is printed as returned and the registry supplies no revision to it and no activities for the sites its monitors did not visit.\n\nExcluded are activities the registry records as planned but never dates, 260 of them, since neither count can place them in the period. Which activities each source counts is recorded in the working papers behind the annex.\n\nMention: \"Triennial Programme Monitoring Registry (TPMR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Triennial Programme Monitoring Registry checks the programme's share of planned activities delivered.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:1421ce1034b36dfdeb6a3014", "occurrence_id": "1d0abee0e2e46d48d91785a7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Whether the ministry's own readiness figure holds was tested against the Diagnostic Service Delivery and Readiness Survey, which the Turkish Statistical Institute conducted in Cyprus in 2015, and the survey is the principal source for the test.\n\nThe ministry's summary gives 84 per cent of facilities ready to give its basic package. The survey gives 76 per cent for the same facilities, because it requires every item on the list to be present and in use on the day, while the ministry's summary counts a facility ready if it ordered the items. Neither basis was applied to the other's count.\n\nThe survey therefore carries the section's readiness figure, and the ministry's larger figure is printed beside it so the reader can see that the distance is the readiness rule rather than a disagreement about the facilities. Where a facility appears in both, the ministry's list of items is kept and the two shares are not averaged.\n\nFacilities the survey visited without a clinician present are dropped from both rather than counted by the ministry alone, 5 per cent of its listings. The readiness rule each source works to is carried in the annex working papers.\n\nMention: \"Diagnostic Service Delivery and Readiness Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Diagnostic Service Delivery and Readiness Survey tests the ministry's share of ready facilities.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:8861e31d499a9f20507a7402", "occurrence_id": "b7e48bbfe3fae4dcd7b1de22", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Bulletin quality was derived for this chapter from the Nationwide Statistical Bulletin Index, the index CEPII compiled for Africa in 2023, and the figures below are the chapter's own figures and not the index's.\n\nBulletins were sorted by the number of revisions the index records after publication and the median days before the corrected figures appear computed: 11 days for bulletins with no revision against 39 days for those with two or more. The index holds a revision count for each bulletin and a correction date for each bulletin and never sets one against the other.\n\nThat comparison supports one claim, that a bulletin should be released once with a stated revision policy rather than corrected repeatedly. The staffing lines the proposal quotes come from the institute's own establishment return, from which the index supplied nothing.\n\nBulletins the index lists as withdrawn without replacement are dropped rather than counted as unrevised, and they are 4 per cent of bulletins. The bands and the counts behind each median are recorded under the table.\n\nMention: \"Nationwide Statistical Bulletin Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Nationwide Statistical Bulletin Index supports a revision-policy claim built from revisions and correction delay.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:396309026c8577105e9f4e63", "occurrence_id": "cf08eecbbee971dbbdae7500", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Three operations lines in the annex come from the Comparative Operations Index Report (COIR), Fifth Round, the report MICS issues for Germany, and all three are transcribed without alteration.\n\nThe report puts the share of offices meeting its timeliness standard at 66 per cent, the mean days a case takes from opening to closing at 41, and the share of cases reopened after closing at 6 per cent. Nothing here was recalculated in this report, and the report's own reading of a closed case, which requires the decision letter to have been sent, travels with the figures.\n\nThe remaining annex lines come from the authority's own case system, so the report carries three lines while the system carries the lines the table does not. A figure reported for one office keeps that office attached rather than being averaged across the country, and where the report and the system disagree about a closing date the report's figure is kept for these three lines and the system's for the rest, with the difference noted under the table rather than smoothed away.\n\nCases still open when the round closed are excluded from all three lines, since neither source can give them a closing date. The annex log holds the report page behind each value, the date the file was taken and the base of each share.\n\nMention: \"Comparative Operations Index Report (COIR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Comparative Operations Index Report supplies timeliness, case duration and reopened cases.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:46b3b6faea93ff72b79a6ba1", "occurrence_id": "196aecf824f75079de82c419", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Boundary change and service were reconstructed for this review from the Comparative Administrative Boundary Registry (CABR), Eighth Phase, the register kept for Chile in 2023, and the readings below are the review's, not the register's.\n\nCommunes were sorted by whether their boundary changed between phases and the share of households the service reaches within its own standard computed: 68 per cent of communes with no change against 52 per cent of those whose boundary moved. The register holds a change flag for each commune and the service records a reach figure for the same commune and neither reports the two together.\n\nHolding the commune's population density even narrows the gap to 9 points, so density carries part of the gap the opening comparison put down to the boundary change.\n\nCommunes the register merges into one for the phase are dropped rather than counted as unchanged, and they are 4 per cent of communes. A note in the annex gives the sorting rule, the pull date as well as the base for each share.\n\nMention: \"Comparative Administrative Boundary Registry (CABR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Comparative Administrative Boundary Registry relates boundary change to households reached within standard.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:6e15dea5b53e22ef8734fa7b", "occurrence_id": "ba73c93bdb1d97236a328216", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Non-state schooling was derived for this chapter from the Joint Non-State Education Census (JNEC), the census conducted in Botswana in 2016, and the figures below come from the chapter and not from the census's.\n\nSchools were sorted by the fee the census records for the year and the share of teachers holding a teaching qualification computed: 71 per cent of schools charging above the median fee against 44 per cent of those below it. The census holds a fee for each school and a qualification count for the same school and never sets one against the other.\n\nThat comparison supports one claim, that a registration rule should require a qualified teacher rather than a building inspection alone. The budget lines the proposal quotes come from the ministry's own vote, from which the census supplied nothing.\n\nSchools the census lists as opened but never visited are dropped rather than counted at the fee they register, and they are 6 per cent of schools. The bands and the counts behind each share are recorded under the table.\n\nMention: \"Joint Non-State Education Census (JNEC)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Joint Non-State Education Census supports a teacher-registration claim built from school fees and qualifications.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:d12cf11ce55dd156622f9314", "occurrence_id": "ea14ec68886d7935b3aa1365", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Graduation and movement were assembled for this review out of the Periodic Graduation Programmes Survey, which Gallup conducted in Timor-Leste in 2018, and what follows is the review's arithmetic, not the survey's.\n\nHouseholds were sorted by the distance from the household to the programme office and the share still above the exit line a year after leaving computed: 71 per cent of households within five kilometres against 49 per cent of those beyond twenty. The survey holds a distance for each household and an exit assessment for the same household and never sets one against the other.\n\nMaking the groups alike on whether the household reports a move since enrolment brings the gap down to 12 points, which leaves the move as part of the distance the first comparison charged to distance.\n\nHouseholds the survey found at a location the programme roll cannot place are dropped rather than counted as lost, and they are 5 per cent of households. Those references, the extraction day and the denominators are set down in the method note in the annex.\n\nMention: \"Periodic Graduation Programmes Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Periodic Graduation Programmes Survey relates distance to the office and exit-line retention after moves.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:66559b36e85c2e3ecb701265", "occurrence_id": "d6c1f25070b58ab2e580affb", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Public works and hiring were estimated for this review using the Provincial Public Works Survey, which Refinitiv Eikon conducted in Cairo in 2022, and the figures are the review's own.\n\nProjects were sorted by the number of working days the survey records in the month and the share of labourers hired from the district itself computed: 63 per cent of projects above twenty working days against 38 per cent of those below ten. The survey holds a working-days figure for each project and a hiring place for each labourer and never sets one against the other.\n\nMaking the groups alike on the share of the workforce the survey finds newly arrived in the district brings the gap down to 9 points, which leaves the arrivals as part of what the unadjusted contrast attributed to the working days.\n\nProjects stopped by a permit dispute are dropped rather than counted at the days they ran, and they are 4 per cent of projects. Working notes list the sorting rule, the extraction day together with the denominators each share rests on.\n\nMention: \"Provincial Public Works Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Provincial Public Works Survey relates working days per project to labourers hired from the district.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:66a0430626124ba895e5b97d", "occurrence_id": "87789e3fae5fc2daa2b75c74", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Operations and cost were reconstructed for this review from the Recurrent Operations Index Survey (ROIS), Wave 6, the survey the World Bank GWSP‐SIEF follows, and the readings below are the review's, not the survey's.\n\nUnits were sorted by the number of operations the index credits them with per member of staff and the mean cost the unit reports per operation computed: 21 at units above forty operations per member against 34 at those below fifteen. The survey holds a staff count for each unit and an operation cost for the same unit and never sets one against the other.\n\nHolding the age of the unit's equipment even narrows the gap to 5, so equipment age carries part of what the unadjusted difference owed to operations per member.\n\nUnits the survey finds with no staff recorded for the wave are dropped rather than counted as costing nothing, and they are 5 per cent of units. A note in the annex gives the sorting rule, the pull date plus the base for each figure.\n\nMention: \"Recurrent Operations Index Survey (ROIS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Recurrent Operations Index Survey relates operations per member of staff to cost per operation.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:b67b87155a6c748f5951d886", "occurrence_id": "fbcd28c69c870eac1349c054", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The bulletin figure in this chapter is the institute's own, taken from its publication log, and the Government Statistical Bulletin Registry (GSBR), First Wave, the register the World Inequality Lab maintains for Burkina-Faso, is cited to test it.\n\nThe institute's log shows 42 bulletins issued in the period. The registry shows 37 for the same period, because it counts a bulletin only once its tables stand on a public page, while the institute's log records a bulletin at the day it is signed off. Neither rule was applied to the other's log.\n\nThe 5 distance is smaller than the number the log shows in a quarter at the period's peak, so the institute's figure is printed as logged and the registry supplies no revision to it and no bulletins for the offices its wave did not visit.\n\nExcluded are bulletins the registry lists as prepared without an issue date, 19 of them, since the period cannot be fixed for them. Which bulletins each source counts is kept in the annex's method note.\n\nMention: \"Government Statistical Bulletin Registry (GSBR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Government Statistical Bulletin Registry checks the institute's count of bulletins issued in the period.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:ec2558ac275a365c01127d0b", "occurrence_id": "10c32aaef1faef286a09a5d6", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Population grid the numbers held for the country come from the Comparative Population Grid Report, the report UNDESA issued for Panama in 2019, and the four values are carried here as tabulated.\n\nThe report puts the share of the population living in grid cells above a thousand people per cell at 34 per cent, the mean travel time from a cell to the nearest town at 46 minutes, the share of cells whose estimate changed between editions at 21 per cent, and the share of the land area left unpopulated by the grid at 27 per cent. No reweighting was applied to them, and the report's own reading of a populated cell, which requires an estimate above zero for the reference year, accompanies the figures.\n\nA cell whose people were counted at a neighbouring cell was not counted as populated, and a cell redrawn for the edition was counted as changed. Each figure keeps the province it was drawn from carried with it, rather than mixed across the country.\n\nThe report line for each value, along with the day of the extract and the cells each share divides sit with the annex tables.\n\nMention: \"Comparative Population Grid Report\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Comparative Population Grid Report supplies dense-cell share, travel time, inter-edition change and unpopulated land.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:9f21415795a40268ed6a92e9", "occurrence_id": "4cb814f1f8491d9ffd46677e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Informal work and hardship were reconstructed for this review from the Linked Socioeconomic Survey (LSS), Version 9, the survey conducted in Cape Town in 2015, and the readings below are the review's, not the survey's.\n\nHouseholds were sorted by whether the head works in an informal arrangement and the share below the city's adequacy line computed: 57 per cent of households with an informal head against 26 per cent of those without. The survey holds a work item for each head and a consumption aggregate for the same household and never sets one against the other.\n\nHolding the size of the household even narrows the gap to 18 points, so household size carries part of what the first comparison credits to the informal arrangement.\n\nHouseholds whose head refused the work question are dropped rather than counted as formal, and they are 5 per cent of households. A note in the annex gives the sorting rule, the pull date, and the base for each share.\n\nMention: \"Linked Socioeconomic Survey (LSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Linked Socioeconomic Survey relates an informal head of household to adequacy-line incidence.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:02a4ab6522258c44563c24df", "occurrence_id": "c935b536b5bde06b8d9f3df4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Rainfall anomaly region figures are assembled from the Panel Rainfall Anomaly Registry, the register Lane and Milesi-Ferretti maintain for Central Asia as of 2015, and the four values pass into this report as printed.\n\nThe register puts the share of stations recording a deficit in the season at 44 per cent, the mean size of that deficit where one is recorded at 18 per cent of normal, the share of stations whose record begins before the register's own baseline year at 62 per cent, and the share of the region's cropped area lying over a deficit station at 29 per cent. Nothing was re-estimated here, and the register's own reading of a deficit, which compares the season against the station's own mean, comes down with them.\n\nA station with a broken season was left out of the deficit count rather than counted as normal, and a station shared by two countries was counted once. Each figure keeps the basin it was drawn from with it, rather than pooled across the region.\n\nUnder the table are the register line for each value, the date of the pull, with the stations each share divides alongside.\n\nMention: \"Panel Rainfall Anomaly Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Panel Rainfall Anomaly Registry supplies deficit stations, deficit size, record length and cropped area covered.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:a2a3801b28dccf2a79305a01", "occurrence_id": "743414f91a1799369abcc287", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Deprivation and farm work were estimated for this review using the Representative Deprivation and Hardship Survey (RDHS), Third Edition, the survey conducted in Cambodia in 2016, and the figures are the review's own.\n\nHouseholds were sorted by the number of deprivations the survey scores them with and the share whose holding fell below subsistence in the season computed: 48 per cent of households with four deprivations or more against 17 per cent of those with one or none. The survey holds a deprivation flag for each household and a subsistence item for the same household and never sets one against the other.\n\nMaking the groups alike on the area the household holds brings the gap down to 14 points, which leaves area as part of the gap the opening contrast assigned to the deprivations.\n\nHouseholds that sold all their land before the season are dropped rather than counted as subsistence-short, and they are 3 per cent of households. Working notes list the sorting rule, the extraction day and the denominators each share rests on.\n\nMention: \"Representative Deprivation and Hardship Survey (RDHS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Representative Deprivation and Hardship Survey relates deprivation counts to holdings below subsistence.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:efe7d2fd4e12b54d238d6592", "occurrence_id": "6d321faed1148051b959c930", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Whether the board's own pass rate holds was tested against the Multi-Topic Exam Performance Collection (MEPC), Phase 8, the collection Afristat maintains, and the collection is the principal source for the test.\n\nThe board's summary gives 78 per cent of its candidates passing the language paper. The collection gives 69 per cent for the same centres, because it counts a candidate who sat and was graded, while the board's summary leaves out candidates excluded for malpractice before grading. No basis was used on the other's count.\n\nThe collection therefore carries the chapter's pass rate, and the board's larger figure is printed beside it so the reader can see that the distance is the exclusion rule rather than a disagreement about the paper. Where a candidate appears in both, the board's grade is kept and the two rates are not averaged.\n\nCandidates the collection lists as displaced before the sitting are dropped from both rather than counted by the collection alone, 6 per cent of its listings. The exclusion rule each source works to is held in the annex's method sheet.\n\nMention: \"Multi-Topic Exam Performance Collection (MEPC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Multi-Topic Exam Performance Collection tests the board's language-paper pass rate for displaced candidates.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:40d540a95b33bd90b75a12f4", "occurrence_id": "9b979266ed50936ab374c3f4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Operations figures spanning the city are taken from the Clustered Operations Index Survey (COIS), Version 6, the survey conducted in Rio de Janeiro in 2022, and the four values come into this report unchanged.\n\nThe survey puts the share of units filing their return on the due day at 58 per cent, the mean days a return takes to reach the centre at 6, the share of units whose return was returned for correction at 17 per cent, and the share of units whose staff named the form as their main difficulty at 24 per cent. No figure was recalculated in this report, and the survey's own reading of a filed return, which requires it to arrive complete rather than in part, is carried alongside them.\n\nA return filed late and complete was counted as late, and a return corrected before the deadline was counted as filed on the day. Each figure keeps the cluster it was drawn from attached, instead of being blended across the city.\n\nThe survey line for each value: the date on the extract and the units each share divides are recorded under the table.\n\nMention: \"Clustered Operations Index Survey (COIS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Clustered Operations Index Survey supplies on-time filing, return delay, corrections and reported form difficulty.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:772ec341a9c0f7f200bc68b5", "occurrence_id": "c9e3d8acf432ef4de9896b6a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Local delivery figures at the country level are taken from the Local Service Delivery Index (LSDI), the indicator compiled for Egypt in 2017, and the four values enter this report as they stand.\n\nThe index puts the share of local authorities publishing their budget at 47 per cent, the share of spending released by the end of the third quarter at 62 per cent, the mean score an authority earns on its service standards at 5.4 out of 10, and the share of authorities audited within the year at 31 per cent. They are carried as published, and the index's own reading of released spending, which requires the money to reach the authority's own account, is printed beside them.\n\nA commitment made but not paid was not counted as released, and an audit begun but not reported was not counted as audited. Each figure keeps the governorate it was drawn from shown with it, instead of combined into one country figure.\n\nThe annex note carries the index line for each value, the day of the pull, and the authorities each share divides.\n\nMention: \"Local Service Delivery Index (LSDI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Local Service Delivery Index supplies budget publication, spending release, service scores and audits.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:3b9990e467c86589315439ae", "occurrence_id": "e43956b66cf1353dee984c24", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Border crossing corridor-level figures come from the Representative Border Crossing Report (RBCR), Sixth Round, the report EPFR issues for Niger, and the four values reach this report as published.\n\nThe report puts the number of crossings recorded at its posts in the period at 96 000, the share of crossings arriving from a single neighbouring country at 71 per cent, the mean days a household reports spending in transit at 12, and the share of crossings registering at a post rather than arriving through a gap in the patrol line at 58 per cent. Nothing here was recomputed, and the report's own reading of a crossing, which requires an officer's entry at the post, is given next to them.\n\nA household crossing twice in the period was counted twice, and a crossing registered at a post but not inspected was still counted as registered. Each figure keeps the post it was recorded at reported with it, rather than blended into a single corridor share.\n\nThe table's footnotes hold the report line for each value, the extraction day as well as the crossings each share divides.\n\nMention: \"Representative Border Crossing Report (RBCR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Representative Border Crossing Report supplies crossings recorded, origin concentration, transit days and post registration.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:9ac1c1867b28ca52ae17b107", "occurrence_id": "ff37066ff9f9fda86f42c524", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Informal work and earnings were reconstructed for this review from the Rapid Informal Sector Survey (RISS), the survey the Penn World table follows for 2015, and the readings below are the review's, not the survey's.\n\nWorkers were sorted by whether their unit holds a registration number and the weekly earnings the unit returns computed: 520 for registered units against 310 for unregistered ones. The survey holds a registration item for each unit and an earnings figure for the same unit and never sets one against the other.\n\nHolding the hours the unit reports even narrows the gap to 120, so hours carry part of the distance the first comparison ascribed to the registration.\n\nUnits the survey finds trading only in the season before the visit are dropped rather than counted as unregistered earners, and they are 6 per cent of units. A note in the annex gives the sorting rule, the pull date and, alongside it, the base for each figure.\n\nMention: \"Rapid Informal Sector Survey (RISS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Rapid Informal Sector Survey relates unit registration to weekly earnings in the informal sector.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:29b5ab4933c40e5901bb64c9", "occurrence_id": "f80440f6ff74e809bc9a6704", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Public records figures for each part of the country are drawn from the Inter-Agency Public Records Registry (IPRR), the register kept for Liberia in 2017, and the four values appear here exactly as printed.\n\nThe registry puts the share of series held by an office with a named custodian at 64 per cent, the share stored off the site of the office that created it at 22 per cent, the mean years since a series was last listed at 5, and the share of series the registry finds missing at the office that should hold them at 9 per cent. They are reproduced as tabulated, and the registry's own reading of a series, which requires a listing the office confirms, is bound to the figures.\n\nA series held by two offices was counted once, and a series listed but not seen was counted as missing. Each figure keeps the ministry it was drawn from kept with it, instead of being rolled up to the country.\n\nUnder the table you will find the registry line for each value, plus the extraction date and the series each share divides.\n\nMention: \"Inter-Agency Public Records Registry (IPRR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Inter-Agency Public Records Registry supplies custodians, off-site storage, listing age and missing series.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:68ea96b2fc8d6d4f5cffc38f", "occurrence_id": "0fc93ec36f2a49dce011f70a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Education spending and outcomes were derived for this chapter from the Audited Education Finance Index (AEFI), Wave 8, the index compiled for Bolivia in 2024, and the figures below come from the chapter, not from the index's.\n\nDistricts were sorted by the spending per pupil the index records and the share of pupils reaching its reading standard computed: 61 per cent of districts above the national mean spend against 38 per cent of those below it. The index holds a spending figure for each district and a reading result for the same district and never sets one against the other.\n\nThat comparison supports one claim, that the formula should carry a weight for pupil numbers rather than for schools. The health lines the proposal quotes come from the ministry's own facility returns, from which the index supplied nothing, and the poverty lines come from the household register the ministry keeps alongside them.\n\nDistricts the index audits without a spending figure are dropped rather than counted at the national mean, and they are 5 per cent of districts. The bands and the counts behind each share are recorded under the table.\n\nMention: \"Audited Education Finance Index (AEFI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Audited Education Finance Index supports a funding-formula claim built from spending per pupil and reading results.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:73c2fe65952e1aed84df0091", "occurrence_id": "f467223e32116b225c36fb4a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Operations and backlog were reconstructed for this review from the Joint Operations Index Registry (JOIR), Sixth Phase, the register JFO maintains, and the readings below are the review's, not the register's.\n\nTeams were sorted by the cases the registry closes per member in the period and the share of cases still open beyond its own limit computed: 11 per cent of teams above thirty cases per member against 29 per cent of those below twelve. The register holds a case count for each team and an open date for each case and never sets one against the other.\n\nHolding the age of the cases each team holds even narrows the gap to 7 points, so case age carries part of what the raw comparison laid at the door of the cases per member.\n\nTeams the registry lists with cases but no members are dropped rather than counted as slow, and they are 4 per cent of teams. A note in the annex gives the sorting rule, the pull date as well as the base for each share.\n\nMention: \"Joint Operations Index Registry (JOIR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Joint Operations Index Registry relates cases closed per member to cases open beyond the limit.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:37fe3d59d91a8038e7c73309", "occurrence_id": "ffb393ea3a0da407c0c9c26b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Service register figures for the reporting units are reported from the Statistical Service Register Index (SSRI), Wave 6, the indicator UNFCCC maintains, and the four values are reproduced here as issued.\n\nThe index puts the share of units with a service register in current use at 69 per cent, the mean months between a register's last entry and the wave at 4, the share of registers whose entries carry a responsible officer at 52 per cent, and the share of registers checked by an office outside the unit at 17 per cent. They were not re-derived here, and the index's own reading of a register in use, which requires an entry within the wave's own window, goes with the figures.\n\nA register kept on paper and posted up each month was counted as in use, and an entry naming a post rather than a person was counted as carrying no officer. Each figure keeps the unit it was drawn from named with it, instead of averaged over the reporting frame.\n\nThe annex keeps the index line for each value, along with the day the table was pulled and the registers each share divides.\n\nMention: \"Statistical Service Register Index (SSRI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Statistical Service Register Index supplies registers in use, entry age, named officers and external checks.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:99cd0dfe243984a68cb23d51", "occurrence_id": "680b027d8d909a2d75c2e58b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The regional figure in this chapter is the office's own, taken from its own returns, and the Statistical Regional Operations Registry, the register the Department of Defense maintains for Mauritania in 2019, is cited to test it.\n\nThe office's returns show 91 per cent of its regions reporting on time. The registry shows 83 per cent for the same regions, because it counts a return as on time only where it arrives complete, while the office's returns accept a return completed later against the same reference period. Neither rule was applied to the other's returns.\n\nThe 8-point distance is smaller than the movement the returns show between two rounds, so the office's figure is printed as returned and the registry supplies no revision to it and no regions whose returns it never received.\n\nExcluded are returns the registry dates to a quarter rather than a day, 57 of them, since the day cannot be fixed for them. Which returns each source counts is laid out in the annex notes.\n\nMention: \"Statistical Regional Operations Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Statistical Regional Operations Registry checks the office's share of regions reporting on time.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:a7e09a7cedb6243fb9e9eafa", "occurrence_id": "4c6ab380a410991204e06e08", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The literacy figure in this chapter is the programme's own, taken from its class assessments, and the Government Literacy and Numeracy Survey (GLNS), Edition 4, the survey conducted in Panama in 2022, is cited to test it.\n\nThe programme's assessments show 86 per cent of its adults reaching its own reading band. The survey shows 79 per cent for the same districts, because it applies one test to everyone, while the programme's assessment is graded by the teacher who taught the adult. The bases stay separate: neither is applied to the other's count.\n\nThe 7-point distance is smaller than the movement the assessments show between two cohorts, so the programme's figure is printed as assessed and the survey supplies no revision to it and no adults who sat neither its own nor the survey's test.\n\nExcluded are adults the survey tests but whose district the programme cannot name, 240 of them, since the two cannot be matched. Which adults each source counts is set out in the annex file.\n\nMention: \"Government Literacy and Numeracy Survey (GLNS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Government Literacy and Numeracy Survey checks the programme's share of adults reaching its reading band.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:df44290a2535e011bd099456", "occurrence_id": "fbd245e978ec14bf82a05630", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The poverty figure in this chapter is the corporation's own, taken from its household cards, and the District Multidimensional Poverty System (DMPS), Eighth Round, the register kept for Tamil Nadu in 2019, is cited to test it.\n\nThe corporation's cards show 29 per cent of its households multidimensionally poor on the ward formula. The register shows 35 per cent for the same wards, because it scores a household on the deprivation the register itself records, while the cards carry the score the surveyor entered at the door. Each basis keeps its own count untouched.\n\nThe 6-point distance is smaller than the movement the cards show between two rounds, so the corporation's figure is printed as carded and the register supplies no revision to it and no households for the wards its round did not enter.\n\nExcluded are households the register scores on a part of its indicator list, 1 300 of them, since the two scores are not on the same footing. Which households each source counts is written up in the papers behind the annex.\n\nMention: \"District Multidimensional Poverty System (DMPS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The District Multidimensional Poverty System checks the corporation's multidimensional poverty share.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:8392a6d3211c0a39bcca3e43", "occurrence_id": "deca7eef7f75b8e24ed923a8", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Whether the federation's own count of informal workers holds was tested against the Provincial Informal Sector Survey (PISS), the survey conducted in Korea in 2023, and the survey is the principal source for the test.\n\nThe federation's card file gives 1.4 million workers in its trades. The survey gives 1.9 million for the same provinces, because it counts a worker the federation never carded, which the file records only where a contribution was paid. Neither basis was carried across to the other's count.\n\nThe survey therefore carries the chapter's count of informal workers, and the federation's smaller figure is printed beside it so the reader can see that the distance is the carding rule rather than a disagreement about the trades. Where a worker appears in both, the federation's trade code is kept and the two counts are not summed.\n\nWorkers the survey finds in a province its sample did not enter are dropped from both rather than counted by the survey alone, 8 per cent of its listings. The carding rule each source works to is detailed in the working papers behind the annex.\n\nMention: \"Provincial Informal Sector Survey (PISS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Provincial Informal Sector Survey tests the federation's count of informal workers in its trades.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:ba9a42b2642b340fbdc2667d", "occurrence_id": "f7cba642c23110015a5b125d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The monitoring figure in this chapter is the department's own, taken from its scorecard, and the Annual Public Sector Monitoring Registry (APSMR), Wave 7, the register INSEE maintains, is cited to test it.\n\nThe department's scorecard shows 76 per cent of its units meeting the target. The registry shows 68 per cent for the same units, because it accepts only a target evidenced by a figure the registry itself holds, while the scorecard accepts the unit's own declaration. Neither rule was applied to the other's scorecard.\n\nThe 8-point distance is smaller than the movement the scorecard shows between two waves, so the department's figure is printed as declared and the registry supplies no revision to it and no units it did not visit in the wave.\n\nExcluded are units the registry scores on a part of its target list, 145 of them, since the two scores are not on the same footing. Which units each source counts is laid out in the annex working papers.\n\nMention: \"Annual Public Sector Monitoring Registry (APSMR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Annual Public Sector Monitoring Registry checks the department's share of units meeting target.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:512129942767fa6c2fd24b9c", "occurrence_id": "3d52e28aff3ac337cd5e6def", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Three records lines in the city annex come from the Local Public Records Registry (LPRR), the register kept for Addis Ababa in 2022, and all three are transcribed without alteration.\n\nThe registry puts the share of offices holding their series in a listed form at 58 per cent, the mean years a series waits before transfer to the city archive at 9, and the share of offices whose register was reviewed in the year at 31 per cent. Nothing here was re-derived for this report, and the registry's own reading of a listed form, which requires the series to appear on the office's own list, travels with the figures.\n\nThe annex's remaining lines are compiled from the sub-city's own inventory, so the registry carries three lines while the inventory carries the balance. A figure reported for one woreda keeps that woreda attached rather than being averaged across the city, and where the registry and the inventory disagree about whether a series is listed the registry's figure is kept for these three lines and the inventory's for the rest, with the difference noted under the table rather than smoothed away.\n\nSeries held only on a personal computer are excluded from all three lines, since the registry cannot see them. The annex log holds the registry entry behind each value, the day of the pull together with the counts each share divides.\n\nMention: \"Local Public Records Registry (LPRR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Local Public Records Registry supplies listed series, transfer waiting years and reviewed registers.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:71f4242ca468b58db386d3d2", "occurrence_id": "ee659ca1696e200f0b50ca33", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Three records lines in the annex come from the Periodic Administrative Records Index (PARI), Wave 5, the indicator the National Statistics Office maintains, and all three are transcribed without alteration.\n\nThe index puts the share of sources with a named contact at the producing office at 72 per cent, the mean days between a source's reference period closing and its first release at 84, and the share of sources whose documentation names its own collection method at 46 per cent. Nothing here was worked out again here, and the index's own reading of a named contact, which requires a post rather than a personal mailbox, travels with the figures.\n\nThe other lines in the annex are copied from the office's own release calendar, so the index carries three lines while the calendar supplies the balance. A figure reported for one source keeps that source attached rather than being averaged across the wave, and where the index and the calendar disagree about a release date the index's figure is kept for these three lines and the calendar's for the rest, with the difference noted under the table rather than smoothed away.\n\nSources released only as a regional aggregate are excluded from all three lines, since the index does not document them at this level. The annex log holds the index line behind each value, the day the rows were taken and the counts under each share.\n\nMention: \"Periodic Administrative Records Index (PARI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Periodic Administrative Records Index supplies named contacts, release delay and documented collection methods.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:1e85e8cc2709cce5163dd952", "occurrence_id": "364c61c30eb08587b80236aa", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Record series were catalogued into the review's inventory from the Regional Integrated Records Index (RIRI), Fifth Collection, the index CRU maintains, and the work done on them was cataloguing rather than measurement.\n\nEach series was given the inventory's own retention band and its review dates checked against the index's reference window; where the index carries a series the inventory has no band for, it was parked for the analyst rather than entered under the nearest one. Series the index lists without a custodian were left unbanded rather than assigned to the nearest office, and 410 of them entered the inventory in that condition. The index's own collection stamp was kept on every series so a later collection can be matched without re-deriving it.\n\nThe inventory's volume counts come from the archive's own accession book, from which the index supplied nothing; it is the series coding that lets those counts be read against retention for the appraisal table the report draws up.\n\nThe coding sheet, the count of parked series and the day of deposit are filed in the annex log.\n\nMention: \"Regional Integrated Records Index (RIRI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The Regional Integrated Records Index supplies the coded record series behind the review's inventory.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:0ac58d66bceca43e114b1a29", "occurrence_id": "889e79994568e9724c841e8c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Household health records were catalogued into the study's working file from the Multi-Topic Health and Household Survey, which FRS conducted in Romania in 2020, and the work done on them was cataloguing rather than measurement.\n\nEach household was given the file's own care-use band and its episode dates checked against the survey's reference window; where the survey carries an episode the file has no band for, it was parked for the analyst rather than entered under the nearest one. Households the survey lists without a response to the care item were left unbanded rather than assumed to have sought none, and 780 of them entered the file in that condition. The survey's own visit date was kept on every household so a later round can be matched without re-deriving it.\n\nThe study's population weights come from the census register, from which the survey supplied nothing; it is the household coding that lets those weights be read against care use for the coverage table the report draws up.\n\nThe coding sheet, the count of parked households and the day of deposit are recorded in the annex log.\n\nMention: \"Multi-Topic Health and Household Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The Multi-Topic Health and Household Survey supplies the coded household records behind the study's working file.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:2bb4b15bf70fcfc1ddedbe3f", "occurrence_id": "0016256602635b10f6d132b1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Unit records were catalogued into the operations file from the Nationwide Regional Operations Registry (NROR), Edition 5, the register the Federal Reserve maintains for 2017, and the work done on them was cataloguing rather than measurement.\n\nEach unit was given the file's own reporting band and its return dates checked against the registry's reference window; where the registry carries a unit the file has no band for, it was parked for the analyst rather than entered under the nearest one. Units the registry lists without a return date were left unbanded rather than dated to the edition, and 520 of them entered the file in that condition. The registry's own edition stamp was kept on every unit so a later edition can be matched without re-deriving it.\n\nThe file's staffing counts come from the establishment return, from which the registry supplied nothing; it is the unit coding that lets those counts be read against reporting for the timeliness table the report draws up.\n\nThe coding sheet, the count of parked units and the day of deposit are stored in the annex log.\n\nMention: \"Nationwide Regional Operations Registry (NROR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The Nationwide Regional Operations Registry supplies the coded unit records behind the operations file.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:38d20485de6218b9955c98be", "occurrence_id": "0b4ae4cd1f03da5b10f8afcb", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The bulletin figures in this chapter are drawn from the institute's own publication log, and the Provincial Statistical Bulletin Index, the indicator Polity IV compiled for Thailand in 2016, is referred to only as background to them.\n\nThe index puts the share of provinces issuing a statistical bulletin at 44 per cent, the mean number of tables a province's bulletin carries at 18, and the share of bulletins whose sources are named inside them at 57 per cent. All three are given as the index issues them, with its own list of provinces stated, and no chapter figure rests on them.\n\nNo chapter table rests on its entries. The figures the chapter argues from come from the institute's log, which records each of its own releases rather than a sample of provinces, and where the two accounts of publication differ the difference is noted rather than reconciled.\n\nThe annex lists the index tables consulted and the day the extract ran, and the index is used for nothing beyond that background.\n\nMention: \"Provincial Statistical Bulletin Index\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The Provincial Statistical Bulletin Index is background to the institute's own publication log.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:38bacdbba910cb70cb7b2c82", "occurrence_id": "deabe4513b08f0ffc5000bf1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The inequality figures in this chapter are computed from the office's own household panel, and the Comparative Inequality Survey (CIS), the survey conducted in Thailand in 2017, is quoted only to place those figures in context.\n\nThe survey puts the share of income the top tenth holds at 31 per cent, the share the bottom half holds at 14 per cent, and the mean years a household in the bottom half reports its income unchanged at 4. All three are given as the survey issues them, with its own income definition stated, and no chapter figure was built from them.\n\nTheir use here is to show how the national distribution stands beside the panel the chapter follows. The chapter's argument rests on that panel, which measures the same households across rounds rather than once at a visit, and the two are kept apart rather than merged.\n\nThe annex lists the survey tables consulted and the pull date, which is the full extent of the survey's use in this chapter.\n\nMention: \"Comparative Inequality Survey (CIS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Comparative Inequality Survey is quoted as context beside the office's household inequality panel.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:dded73253b3517dba0345635", "occurrence_id": "4a1911b0120a77db65790f8e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Skills and hardship figures measured across the municipality are read from the Rolling Skills and Employment Survey (RSES), Wave 9, the survey conducted in Chengdu in 2020, and the four values are quoted here as printed.\n\nThe survey puts the share of adults holding a recognised vocational certificate at 42 per cent, the share of certificate holders without paid work at 11 per cent, the share of households below the municipal adequacy line at 9 per cent, and the share of those households with no certificate holder at 34 per cent. No arithmetic was applied to them here, and the survey's own reading of paid work, which requires an hour paid in the reference week, travels with the values.\n\nA person paid in kind was counted as in paid work, and a certificate held by a member living away was counted for the household. Each figure keeps the district it was drawn from listed with it, rather than smoothed across the municipality.\n\nBeneath the table sit the survey line for each value, the day of extraction plus the households each share divides.\n\nMention: \"Rolling Skills and Employment Survey (RSES)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Rolling Skills and Employment Survey supplies certificate holding, workless certificate holders, adequacy incidence and household coverage.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:5b0410f5ef76eb2da35e8ba7", "occurrence_id": "c9e76015779bb09f60542f98", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The health figures in this chapter are drawn from the service's own patient records, and the Experimental Health and Household Survey (EHHS), Round 9, the survey conducted in Korea in 2018, is referred to only as background to them.\n\nThe survey puts the share of adults with a long-standing condition at 21 per cent, the share of those adults whose care is covered at 78 per cent, and the share of households paying for care out of current income at 26 per cent. All three are given as the survey issues them, with its own household definition stated, and no chapter figure rests on them.\n\nNone of the chapter's tables rest on its responses. The figures the chapter argues from come from the service's records, which follow each enrolled patient rather than a sample of households, and where the two accounts of coverage differ the difference is noted rather than reconciled.\n\nThe annex lists the survey tables consulted and the date it was drawn, and the survey is used for nothing beyond that background.\n\nMention: \"Experimental Health and Household Survey (EHHS)\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The Experimental Health and Household Survey is background to the service's own patient record figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:f80704974eacb359edf6250b", "occurrence_id": "346a9658f066c8c0b0dec500", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Protracted situation figures for the sample are reported from the Longitudinal Protracted Situation Survey (LPSS), Edition 9, the survey the World Bank GWSP‐SIEF follows, and the four values are carried here as tabulated.\n\nThe survey puts the mean years a household has been displaced at 7, the share of children enrolled in the year at 64 per cent, the share of households with no document valid for schooling at 18 per cent, and the share reporting they intend to stay where they are at 41 per cent. No reweighting was applied to them, and the survey's own reading of enrolment, which requires attendance recorded by the school, follows the figures.\n\nA child enrolled and then withdrawn mid-year was counted as enrolled, and a household holding only an expired permit was counted as holding no valid document. Each figure keeps the site it was drawn from carried with it, rather than mixed across the sample.\n\nWorking notes record the survey line for each value: the date the file was taken and the children each share divides.\n\nMention: \"Longitudinal Protracted Situation Survey (LPSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Longitudinal Protracted Situation Survey supplies displacement years, enrolment, schooling documents and stay intentions.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:1ef9e1a3f06507781dc8911c", "occurrence_id": "bf2a028787ecdf868269c3b4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Housing condition the sample figures are compiled from the Harmonised Housing Conditions Survey (HHCS), Sixth Collection, the survey WBG/IFC follows, and the four values pass into this report as printed.\n\nThe survey puts the share of households renting at 46 per cent, the share whose dwelling lacks a private cooking space at 27 per cent, the mean occupants per room at 1.8, and the share of renting households below the national adequacy line at 39 per cent. Nothing was re-estimated here, and the survey's own reading of adequacy follows the national line current at the interview.\n\nA household sharing a kitchen with another household was counted as lacking a private space, and a boarder counted as an occupant. Each figure keeps the stratum it was drawn from with it, rather than pooled across the sample.\n\nThe documentation under the table gives the survey line for each value, the day of the extract, and the households each share divides.\n\nMention: \"Harmonised Housing Conditions Survey (HHCS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Harmonised Housing Conditions Survey supplies renting, shared cooking space, occupancy and adequacy incidence.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:91894615946cb0f9098b1530", "occurrence_id": "a3ae0b778415c2c4c92748cb", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Three time use lines in the labour annex come from the Household Time Use Survey (HTUS), Edition 8, the survey UNEP Risoe follows for 2017, and all three are transcribed without alteration.\n\nThe survey puts the mean hours adults give to unpaid care in the week at 19, the share of adults recording no paid activity at 29 per cent, and the mean hours of paid work among those who record some at 43. Nothing here was computed afresh for this report, and the survey's own reading of an activity, which requires it to be logged against the diary, travels with the figures.\n\nThe annex's remaining entries follow the office's own claimant file, so the survey carries three lines while the file holds the remainder. A figure reported for one stratum keeps that stratum attached rather than being averaged across the sample, and where the survey and the file disagree about who is working the survey's figure is kept for these three lines and the file's for the rest, with the difference noted under the table rather than smoothed away.\n\nAdults whose diary covers fewer than three days are excluded from all three lines, since the survey does not weight them. The annex log holds the survey line behind each value, the date of the pull, and the population each share divides.\n\nMention: \"Household Time Use Survey (HTUS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Household Time Use Survey supplies unpaid care hours, adults without paid activity and paid hours.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:124b7a90b468ff66f4c335a9", "occurrence_id": "daa6c9f772c5e4889db720c4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Deprivation and illness cost were reconstructed for this review from the Analytical Multidimensional Poverty Survey (AMPS), First Edition, the survey conducted in Sindh in 2020, and the readings below are the review's, not the survey's.\n\nHouseholds were sorted by the number of deprivations the survey scores them with and the share reporting an illness that pushed them into debt computed: 34 per cent of households with four deprivations or more against 9 per cent of those with one or none. The survey holds a deprivation flag for each household and a debt item for the same household and never sets one against the other.\n\nHolding whether the household holds any health coverage even narrows the gap to 12 points, so coverage carries part of the gap the opening comparison put down to the count of deprivations.\n\nHouseholds whose debt predates the illness the survey records are dropped rather than counted as illness-indebted, and they are 5 per cent of households. A note in the annex gives the sorting rule, the pull date, with the base for each share alongside.\n\nMention: \"Analytical Multidimensional Poverty Survey (AMPS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Analytical Multidimensional Poverty Survey relates deprivation counts to illness that pushes households into debt.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:def1c9e94b2ff24a0e1b0570", "occurrence_id": "9487df2eb91b29ddef6c1ad9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Consumption and farm sales were estimated for this review using the Analytical Household Income and Expenditure Survey (AHIES), Collection 7, the survey followed in Yemen in 2023, and the figures are the review's own.\n\nHouseholds were sorted by the share of their food that comes from their own holding and the share below the national adequacy line computed: 52 per cent of households taking more than half their food from the holding against 28 per cent of those taking less than a tenth. The survey holds a home-produced share for each household and a consumption aggregate for the same household and never sets one against the other.\n\nMaking the groups alike on the area the household works brings the gap down to 12 points, which leaves area as part of the distance the first comparison charged to the home-produced share.\n\nHouseholds reporting a holding but no crop in the season are dropped rather than counted as self-provisioning, and they are 6 per cent of households. Working notes list the sorting rule, the extraction day and the denominators each share rests on.\n\nMention: \"Analytical Household Income and Expenditure Survey (AHIES)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Analytical Household Income and Expenditure Survey relates home-produced food share to adequacy-line incidence.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:3024611a1e35621776e35c1a", "occurrence_id": "a3e15e19848a147f0c132b23", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Graduation and earnings were assembled for this review out of the Midline Graduation Programmes Survey, which TFDD conducted in Korea in 2019, and what follows is the review's arithmetic, not the survey's.\n\nHouseholds were sorted by the months they had been enrolled and the share whose earnings cover their recorded spending computed: 44 per cent of households enrolled two years or more against 21 per cent of those enrolled under a year. The survey holds an enrolment date for each household and a spending aggregate for the same household and never sets one against the other.\n\nMaking the groups alike on the number of earners in the household brings the gap down to 9 points, which leaves earners as part of what the unadjusted contrast attributed to the months enrolled.\n\nHouseholds that left the programme before the midline are dropped rather than counted as not covering, and they are 7 per cent of households. Those references, the extraction day and the denominators are recorded in the method note in the annex.\n\nMention: \"Midline Graduation Programmes Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Midline Graduation Programmes Survey relates months enrolled to earnings covering recorded spending.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:2b4e8597a9e0a5aae1692cbf", "occurrence_id": "723811a01dcd58d08a3faacf", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Whether the ministry's own monitoring figure holds was tested against the Continuous Public Sector Monitoring Index, the indicator UNDESA compiled for Ghana in 2021, and the index is the principal source for the test.\n\nThe ministry's summary gives 77 per cent of its agencies meeting their published target. The index gives 68 per cent for the same agencies, because it counts a target met only where the index itself holds the underlying figure, while the ministry's summary accepts the agency's own declaration. One basis was never used to adjust the other's count.\n\nThe index therefore carries the chapter's monitoring figure, and the ministry's larger figure is printed beside it so the reader can see that the distance is the evidencing rule rather than a disagreement about the targets. Where an agency appears in both, the ministry's target wording is kept and the two shares are not averaged.\n\nAgencies the index scores on part of its target list are dropped from both rather than counted by the ministry alone, 6 per cent of its listings. The evidencing rule each source works to is listed in the annex's method note.\n\nMention: \"Continuous Public Sector Monitoring Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Continuous Public Sector Monitoring Index tests the ministry's share of agencies meeting published targets.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:5d1b9517060dabc39d6348f3", "occurrence_id": "bab9ab56f68ec51433f058c0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Whether the institute's own completion figure holds was tested against the Periodic Tertiary Education System (PTES), Edition 5, the register the Ministry of Health maintains, and the register is the principal source for the test.\n\nThe institute's summary gives 64 per cent of its cohorts completing within the stated duration. The register gives 57 per cent for the same cohorts, because it counts a student as completing only where the award is recorded against the student's own identifier, while the institute's summary counts a student the registrar marks as exited. The two bases were not applied to each other's count.\n\nThe register therefore carries the section's completion figure, and the institute's larger figure is printed beside it so the reader can see that the distance is the counting rule rather than a disagreement about the cohorts. Where a student appears in both, the institute's course code is kept and the two rates are not averaged.\n\nCohorts the register dates to a year rather than an intake are dropped from both rather than counted by the institute alone, 4 per cent of its listings. The counting rule each source works to is set down in the annex's method sheet.\n\nMention: \"Periodic Tertiary Education System (PTES)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Periodic Tertiary Education System tests the institute's cohort completion within the stated duration.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:23c0ff3e7e993bf39a8414cc", "occurrence_id": "216974e76ce815159631faa3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The inspection figure in this chapter is the authority's own, taken from its case file, and the Metropolitan Labour Inspection Survey, the survey CLGF conducted in the United Kingdom in 2020, is cited to test it.\n\nThe authority's file shows 82 per cent of its premises inspected within the cycle. The survey shows 74 per cent for the same premises, because it counts an inspection only where an inspector entered a finding on the day, while the authority's file counts a visit that closed without a finding. Neither rule was applied to the other's file.\n\nThe 8-point distance is smaller than the movement the file shows between two cycles, so the authority's figure is printed as filed and the survey supplies no revision to it and no premises for the boroughs its teams did not enter.\n\nExcluded are premises the survey lists at an address the file cannot match, 310 of them, since the two cannot be joined. Which premises each source counts is written up in the annex notes.\n\nMention: \"Metropolitan Labour Inspection Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Metropolitan Labour Inspection Survey checks the authority's share of premises inspected within the cycle.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:a1220d2e9863fe71a6891288", "occurrence_id": "9ae8a8373a5cb1427110befd", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The apprenticeship figure in this chapter is the service's own, taken from its trainee list, and the Household Apprenticeship Survey, the survey MOH conducted in Fiji in 2024, is cited to test it.\n\nThe service's list shows 69 per cent of its trainees completing the workplace component. The survey shows 61 per cent for the same districts, because it asks the household whether the trainee still attends and counts a trainee who stopped without notice, which the list carries as enrolled until the next review. Neither basis reworks the other's list.\n\nThe 8-point distance is smaller than the movement the list shows between two intakes, so the service's figure is printed as listed and the survey supplies no revision to it and no trainees whose households it could not reach.\n\nExcluded are trainees the survey finds working away from their district, 95 of them, since the two cannot say whether the component continued. Which trainees each source counts is laid out in the annex file.\n\nMention: \"Household Apprenticeship Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Household Apprenticeship Survey checks the service's share of trainees completing the workplace component.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:aa0453b6c6e240ebdeaed156", "occurrence_id": "4478992f279769a26d5a5d84", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The operations figures in this chapter are drawn from the partnership's own returns, and the Integrated Regional Operations Registry, the register the United Nations maintains for the United Kingdom in 2023, is referred to only as background to them.\n\nThe registry puts the share of regions filing their return complete at 71 per cent, the mean weeks between a period closing and its return arriving at 5, and the share of regions whose return was revised after arrival at 16 per cent. All three are given as the registry issues them, with its own region list stated, and no chapter figure rests on them.\n\nNo table in the chapter uses its entries. The figures the chapter argues from come from the partnership's returns, which cover each of its own areas rather than a sample of regions, and where the two accounts of timeliness differ the difference is noted rather than reconciled.\n\nThe annex lists the registry tables consulted and the date on the extract, and the registry is used for nothing beyond that background.\n\nMention: \"Integrated Regional Operations Registry\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The Integrated Regional Operations Registry is background to the partnership's own operations returns.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:3ec634b8fa64c22253d00d02", "occurrence_id": "8a42c4455228df11a90a3078", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Rural livelihood figures for the whole state are sourced from the Quarterly Rural Livelihoods Survey (QRLS), Round 7, the survey conducted in Tamil Nadu in 2017, and the four values come into this report unchanged.\n\nThe survey puts the share of households with at least one member in wage work at 57 per cent, the mean days a household reports wage work in the quarter at 41, the share of households below the state adequacy line at 23 per cent, and the share of those households with nobody in wage work at 36 per cent. No figure was recalculated in this report, and the survey's own reading of wage work, which requires payment in cash at the reference week, is reported with them.\n\nA day worked for a meal was not counted as wage work, and a government scheme payment was counted as wage work. Each figure keeps the block it was drawn from attached, instead of being blended across the state.\n\nThe survey line for each value, the day of the pull as well as the households each share divides sit with the annex tables.\n\nMention: \"Quarterly Rural Livelihoods Survey (QRLS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Quarterly Rural Livelihoods Survey supplies wage-work households, days of work, adequacy incidence and workless poor households.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:195a47a2022b74d15d9c60c6", "occurrence_id": "8e3349cffbeb36dc105d35b7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Records handling was derived for this chapter from the Rolling Public Records Survey (RPRS), the survey conducted in Germany in 2016, and the figures below are the chapter's own rather than the survey's.\n\nOffices were sorted by the number of series the survey finds without a review date and the median days before a request is answered computed: 14 days at offices with fewer than twenty undated series against 33 days at those with sixty or more. The survey holds a series count for each office and an answer date for each request and never sets one against the other.\n\nThat comparison supports one claim, that every series should carry a review date set at its opening rather than at first request. The staffing lines the proposal quotes come from the authority's own establishment return, from which the survey supplied nothing.\n\nOffices the survey visits twice in the year are counted at their second visit rather than their higher count, and they are 5 per cent of offices. The bands and the counts behind each median are recorded under the table.\n\nMention: \"Rolling Public Records Survey (RPRS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Rolling Public Records Survey supports a review-date claim built from undated series and request delay.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:4e5a0709a1f7e8e2b0982fe8", "occurrence_id": "810eb023c1be506549145ba8", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Nutrition and mortality were reconstructed for this review from the Metropolitan Nutrition and Mortality Survey (MNMS), Sixth Round, the survey conducted in Somalia in 2023, and the readings below are the review's, not the survey's.\n\nSettlements were sorted by the months since the survey finds a health post opened and the share of children it scores as acutely malnourished computed: 17 per cent of settlements with a post open two years or more against 29 per cent of those with none. The survey holds an opening date for each settlement and a nutrition result for each child and never sets one against the other.\n\nHolding the distance from the settlement to the nearest town even narrows the gap to 6 points, so distance carries part of what the unadjusted difference owed to the post.\n\nSettlements the survey reaches only in the last week of the round are dropped rather than counted without a post, and they are 4 per cent of settlements. A note in the annex gives the sorting rule, the pull date and, alongside it, the base for each share.\n\nMention: \"Metropolitan Nutrition and Mortality Survey (MNMS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Metropolitan Nutrition and Mortality Survey relates health post age to acute malnutrition in children.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:7dd24c51d812172396e0390a", "occurrence_id": "a5d1f101e3c3dd873545a38c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The referral figure in this chapter is the project's own, taken from its transport log, and the Record-Based Emergency Care Survey, the survey IUCN conducted in Burkina-Faso in 2023, is cited to test it.\n\nThe project's log shows 46 minutes as the median time from a referral call to arrival at the receiving site. The survey shows 63 minutes for the same corridors, because it times the case from the household's first call rather than from the call the log records, which is made once the vehicle has been assigned. Neither basis was applied to the other's log.\n\nThe 17-minute distance is smaller than the movement the log shows between two quarters, so the project's figure is printed as logged and the survey supplies no revision to it and no referrals whose household could not be located.\n\nExcluded are referrals the survey times across a closed road, 74 of them, since neither source can give them a comparable route. Which referrals each source counts is set out in the papers behind the annex.\n\nMention: \"Record-Based Emergency Care Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Record-Based Emergency Care Survey checks the project's median referral arrival time.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:9dcd265b11674ed6755e0cc2", "occurrence_id": "171d77f26a9ba63dc128ff44", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Bargaining coverage and pay were reconstructed for this review from the National Collective Bargaining Survey (NCBS), the survey conducted in Korea in 2022, and the readings below are the review's, not the survey's.\n\nWorkplaces were sorted by whether a bargaining agreement covers them and the mean hourly wage the survey records for the workers in them computed: 14 for covered workplaces against 9 for uncovered ones. The survey holds a coverage item for each workplace and a wage item for each worker and never sets one against the other, so the pairing is the review's own and no figure printed here appears in the survey in that form.\n\nHolding the size of the workplace and the sector it trades in even narrows the gap to 3, so size and sector together carry part of what the first comparison credits to the coverage, and the chapter prints the narrower gap rather than the wider one.\n\nWorkplaces the survey finds with an agreement signed but never applied are dropped rather than counted as covered, and they are 4 per cent of workplaces. Workers paid by output rather than by the hour are converted at the hours the survey records for them. A note in the annex gives the sorting rule, the pull date as well as the base for each figure.\n\nMention: \"National Collective Bargaining Survey (NCBS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The National Collective Bargaining Survey relates agreement coverage to the mean hourly wage.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:3aafbf84ba737d76ccac6610", "occurrence_id": "6dc04069bc7b926ae4a9b87b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Administrative records the state figures are tabulated by the Audited Administrative Records Registry, the register PPVR maintains for Lagos State in 2020, and the four values enter this report as they stand.\n\nThe registry puts the share of offices whose series are listed in a form the audit accepts at 61 per cent, the mean years a series waits before it is appraised at 6, the share of offices whose register was tested by the audit at 44 per cent, and the share of series the audit finds held outside the office that owns them at 12 per cent. They are carried as published, and the registry's own reading of a listed series, which requires the list to be signed by the office head, is attached to them.\n\nA series on a list signed by a predecessor was still counted as listed, and a series held at a shared annex counted as held outside the office. Each figure keeps the local government area it was drawn from shown with it, instead of combined into one state figure.\n\nUnder the table are the registry line for each value, plus the extraction day and the series each share divides.\n\nMention: \"Audited Administrative Records Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Audited Administrative Records Registry supplies listed series, appraisal waiting years, tested registers and off-site holdings.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:cba994695ecc05d60affdd24", "occurrence_id": "1d472e44239df597365f8cf3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: District deprivation figures are reported from the District Deprivation and Hardship Index (DDHI), Wave 3, the indicator VIIRS maintains, and the four values reach this report as published.\n\nThe index puts the share of districts above its own deprivation cut at 28 per cent, the mean number of deprivations a poor household carries at 4, the share of districts whose deprivation worsened between waves at 9 per cent, and the share of districts with a household register the index accepts at 53 per cent. Nothing here was recomputed, and the index's own reading of a deprivation, which requires the indicator to be observed at the dwelling rather than reported, is stated with them.\n\nA household reported poor by its leader but not observed was not counted, and a district observed at two settlements was counted once. Each figure keeps the district it was drawn from reported with it, rather than blended into a single frame share.\n\nThe index line for each value, along with the extraction date and the districts each share divides are recorded under the table.\n\nMention: \"District Deprivation and Hardship Index (DDHI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The District Deprivation and Hardship Index supplies high-deprivation districts, deprivations per household, worsening and accepted registers.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:c950c26f5fe19af0b49e4793", "occurrence_id": "300820cbfc47bb34c41255d9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Land cover and cultivation were reconstructed for this review from the Multi-Topic Land Cover Dataset (MLCD), the geospatial dataset compiled for Sub-Saharan Africa in 2021, and the readings below are the review's, not the dataset's.\n\nUnits were matched to their dominant cover class and grouped by the share of the unit the dataset grades as cultivated, and the share of units the rainfall series records a deficit for computed: 41 per cent of units above a third cultivated against 22 per cent of those below a tenth. The dataset grades cover per unit and the rainfall series records the deficit per station, and neither reports the two together.\n\nHolding the elevation band of the unit even narrows the gap to 8 points, so elevation carries part of the gap the opening contrast assigned to cultivation.\n\nUnits the dataset leaves unclassified for want of a clear observation are dropped rather than counted as uncultivated, and they are 7 per cent of units. A note in the annex gives the grouping rule, the pull date together with the base for each share.\n\nMention: \"Multi-Topic Land Cover Dataset (MLCD)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Multi-Topic Land Cover Dataset relates cultivated share of a unit to rainfall deficit recorded over it.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:01d8a08af231f7cc73caa155", "occurrence_id": "3cd434dcaa5b85cb8d7fac0e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Three settlement lines in the displacement annex come from the Coordinated Settlement Location Report (CSLR), Release 7, the report issued for Lao PDR in 2020, and all three are transcribed without alteration.\n\nThe report puts the number of sites holding displaced households at 148, the share of those sites whose location was confirmed on the ground at 79 per cent, and the mean hectares a site holds at 26. Nothing here was recomputed for this report, and the report's own reading of a site, which requires a coordination meeting to have visited it, travels with the figures.\n\nThe rest of the annex lines come from the province's own hosting list, so the report carries three lines while the list carries what is left over. A figure reported for one district keeps that district attached rather than being averaged across the country, and where the report and the list disagree about whether a site is open the report's figure is kept for these three lines and the list's for the rest, with the difference noted under the table rather than smoothed away.\n\nSites the report lists as closed within the release are excluded from all three lines, since the coordination record stops before the reference date. The annex log holds the report page behind each value, the day the table was pulled and the denominators each share uses.\n\nMention: \"Coordinated Settlement Location Report (CSLR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Coordinated Settlement Location Report supplies site counts, ground-confirmed locations and site area.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:bf2b6517dd16bc1c74a70644", "occurrence_id": "0efc9ec6f003c19cc76fb2e3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Bulletin the figures for the country are reported by the Endline Statistical Bulletin Survey (ESBS), Fourth Release, the survey CEDLAS-World Bank follows in Hungary, and the four values appear here exactly as printed.\n\nThe survey puts the share of offices issuing a bulletin on a stated calendar at 58 per cent, the mean days between a reference period closing and publication at 74, the share of bulletins whose method is described inside them at 41 per cent, and the share of offices that revised a bulletin once at 17 per cent. They are reproduced as tabulated, and the survey's own reading of a bulletin, which requires tables to stand on a public page, comes with them.\n\nA bulletin circulated to departments only was not counted, and a correction issued as a note was counted as a revision. Each figure keeps the office it was drawn from kept with it, instead of being rolled up to the country.\n\nThe annex note carries the survey line for each value, the day of the pull plus the offices each share divides.\n\nMention: \"Endline Statistical Bulletin Survey (ESBS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Endline Statistical Bulletin Survey supplies calendar publication, delay, method description and revisions.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:d091def1a4e62d5877903190", "occurrence_id": "1110ba417052e3e560be6532", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Elevation model figures covering the frame are drawn from the National Elevation Model Census (NEMC), Release 4, the census GACC compiled for 2023, and the four values are reproduced here as issued.\n\nThe census puts the share of the frame covered by a model at a cell size of ten metres or finer at 62 per cent, the mean vertical error the census reports for that cover at 0.9 metres, the share of the frame whose model was acquired in the last two years at 37 per cent, and the share left uncovered at 8 per cent. They were not re-derived here, and the census's own reading of cover, which requires the cell to hold a ground value rather than an interpolated one, is kept with them.\n\nA cell filled from a neighbouring model was not counted as cover, and a model reissued without new acquisition was counted at its original acquisition date. Each figure keeps the tile it was drawn from named with it, instead of averaged over the frame.\n\nThe table's footnotes hold the census layer for each value: the day the rows were taken and the cells each share divides.\n\nMention: \"National Elevation Model Census (NEMC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The National Elevation Model Census supplies fine-cell cover, vertical error, acquisition recency and uncovered frame.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:c6de9ef6d78bb84fc37ea84c", "occurrence_id": "51c1645fa4afea426cf116a1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Minimum needs and housing cost were reconstructed for this review from the Diagnostic Minimum Needs Survey (DMNS), the survey conducted in Riga in 2018, and the readings below are the review's, not the survey's.\n\nHouseholds were sorted by the share of their income that housing takes and the share reporting they could not meet the survey's basket computed: 39 per cent of households above a quarter of income against 16 per cent of those below a tenth. The survey holds a housing share for each household and a basket item for the same household and never sets one against the other.\n\nHolding the number of people in the household even narrows the gap to 12 points, so household size carries part of the distance the first comparison ascribed to the housing share.\n\nHouseholds whose housing cost is nil because they live in an employer's dwelling are dropped rather than counted at the low end, and they are 5 per cent of households. A note in the annex gives the sorting rule, the pull date, and the base for each share.\n\nMention: \"Diagnostic Minimum Needs Survey (DMNS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Diagnostic Minimum Needs Survey relates the housing share of income to households unable to meet the basket.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:724958775d9edde93d3ade77", "occurrence_id": "a4974b2c9c49704e029ed09f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Three agricultural lines in the climate annex come from the Integrated Irrigation and Inputs Registry, the register MOEE maintains for Saint Lucia in 2019, and all three are transcribed without alteration.\n\nThe registry puts the share of holdings drawing irrigation water at 34 per cent, the mean kilograms of fertiliser a holding records in the season at 60, and the share of holdings whose water source the registry lists as a spring at 22 per cent. Nothing here was recomputed here, and the registry's own reading of an irrigated holding, which requires a source used in the season just closed, travels with the figures.\n\nThe other annex entries are assembled from the extension service's own visit book, so the registry carries three lines while the book accounts for the remaining lines. A figure reported for one parish keeps that parish attached rather than being averaged across the island, and where the registry and the visit book disagree about irrigation the registry's figure is kept for these three lines and the book's for the rest, with the difference noted under the table rather than smoothed away.\n\nHoldings the registry lists as abandoned in the season are excluded from all three lines, since they record no use. The annex log holds the registry entry behind each value, the day the extract ran, with the denominators behind each share alongside.\n\nMention: \"Integrated Irrigation and Inputs Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Integrated Irrigation and Inputs Registry supplies irrigated holdings, fertiliser use and spring-fed sources.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:efcfddeb95999f040b0518f1", "occurrence_id": "85dc68a5d6ba34d85049fc14", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Service point records were catalogued into the register file from the Decennial Service Register Collection (DSRC), Phase 5, the collection ZamStats maintains, and the work done on them was cataloguing rather than measurement.\n\nEach point was given the file's own service band and its opening dates checked against the collection's reference window; where the collection carries a point the file has no band for, it was parked for the analyst rather than entered under the nearest one. Points the collection lists without an operator were left unbanded rather than assigned to the district list, and 560 of them entered the file in that condition. The collection's own phase stamp was kept on every point so a later phase can be matched without re-deriving it.\n\nThe file's catchment populations come from the census output, from which the collection supplied nothing; it is the point coding that lets those populations be read against distance for the coverage table the report draws up.\n\nThe coding sheet, the count of parked points and the day of deposit are held in the annex log.\n\nMention: \"Decennial Service Register Collection (DSRC)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The Decennial Service Register Collection supplies the coded service point records behind the register file.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:b775b99154eef95d2bd1db4c", "occurrence_id": "8e38d43fa3fa0b194f8e2725", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Household consumption figures covering the whole country derive from the Decennial Household Consumption Registry (DHCR), the register kept for Uruguay in 2015, and the four values are quoted here as printed.\n\nThe register puts the mean monthly consumption per household at 3 200, the share of households below the national line at 11 per cent, the share of spending going to food at 21 per cent, and the share of households reporting no savings in the year at 38 per cent. No arithmetic was applied to them here, and the register's own reference period of one month accompanies the figures.\n\nA household eating from its own plot was counted at the value of what it ate, and a purchase made on a card was counted at the month of purchase. Each figure keeps the department it was drawn from listed with it, rather than smoothed across the country.\n\nUnder the table you will find the register line for each value, the pull date, and the households each share divides.\n\nMention: \"Decennial Household Consumption Registry (DHCR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Decennial Household Consumption Registry supplies consumption, poverty incidence, food share and households without savings.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:8e9ff7667edc9e846470c0f1", "occurrence_id": "2b2b541a68b590084f960a53", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Place names and delivery were reconstructed for this review from the Combined Place Name Index (CPNI), the indicator compiled for Kosovo in 2020, and the readings below are the review's, not the index's.\n\nSettlements were matched to their index entry and grouped by whether the name they carry in the delivery list agrees with the index, and the share of deliveries the post records as failed computed: 9 per cent of settlements whose name agrees against 23 per cent of those whose name differs. The index holds one name per place and the delivery list records another, and neither reports the two together.\n\nHolding the size of the settlement even narrows the gap to 6 points, so size carries part of what the raw comparison laid at the door of the name.\n\nSettlements the index lists under two entries are dropped rather than counted as agreeing, and they are 4 per cent of settlements. A note in the annex gives the matching rule, the pull date and the base for each share.\n\nMention: \"Combined Place Name Index (CPNI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Combined Place Name Index relates agreement between place names and failed deliveries.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:666443ba882d1501f172bd6a", "occurrence_id": "1558753913068fb9656d6bea", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Three surveillance lines in the health annex come from the Ministerial Disease Surveillance Report (MDSR), Seventh Edition, the report WBG issues for Fiji, and all three are transcribed without alteration.\n\nThe report puts the number of suspected cases notified in the season at 1 830, the share of notifications arriving from a health facility rather than a laboratory at 66 per cent, and the share of the reporting units filing weekly at 74 per cent. Nothing here was recalculated for this report, and the report's own reading of a notification, which requires the unit to have entered it within its own week, travels with the figures.\n\nThe annex's other lines are taken from the district's own clinic register, so the report carries three lines while the register carries the rest. A figure reported for one island group keeps that group attached rather than being averaged across the country, and where the report and the register disagree about a case the report's figure is kept for these three lines and the register's for the rest, with the difference noted under the table rather than smoothed away.\n\nCases notified twice by different units are excluded from all three lines, since the report counts the first entry only. The annex log holds the report page behind each value, the day of extraction and the base each share is taken on.\n\nMention: \"Ministerial Disease Surveillance Report (MDSR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Ministerial Disease Surveillance Report supplies notified cases, facility notifications and weekly filing.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:d852747e06b510081c64aa3d", "occurrence_id": "583957b3fbadeda662fed511", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Monitoring and responsiveness were reconstructed for this review from the Audited Public Sector Monitoring Survey (APSMS), Phase 7, the survey Doing Business follows, and the readings below are the review's, not the survey's.\n\nUnits were sorted by the number of monitoring visits the survey records in the year and the median days before a complaint receives a substantive reply computed: 12 days at units with four visits or more against 31 days at those with one or none. The survey holds a visit count for each unit and a reply date for each complaint and never sets one against the other.\n\nHolding the volume of complaints the unit receives even narrows the gap to 9 days, so the volume carries part of the gap the opening comparison put down to the visits.\n\nUnits the survey finds merged partway through the phase are dropped rather than counted at their predecessor's visits, and they are 5 per cent of units. A note in the annex gives the sorting rule, the pull date and, alongside it, the base for each figure.\n\nMention: \"Audited Public Sector Monitoring Survey (APSMS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Audited Public Sector Monitoring Survey relates monitoring visits to days before a complaint is answered.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:84eac5055bbfa27722ebb91c", "occurrence_id": "be6ce6046a5a3cbfc6f04ace", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Business entry and hiring were estimated for this review using the Harmonised Business Demography Survey (HBDS), Version 5, the survey Lederman y Saenz follows in Liberia, and the figures are the review's own.\n\nEnterprises were sorted by the months since they first traded and the share employing anyone beyond the owner computed: 27 per cent of enterprises trading three years or more against 9 per cent of those trading under a year. The survey holds a starting month for each enterprise and a worker count for the same enterprise and never sets one against the other.\n\nMaking the groups alike on the sector the enterprise trades in brings the gap down to 7 points, which leaves sector as part of the distance the first comparison charged to the months trading.\n\nEnterprises the survey finds trading only at a market stall are dropped rather than counted as employer units, and they are 6 per cent of enterprises. Working notes list the sorting rule, the extraction day as well as the denominators each share rests on.\n\nMention: \"Harmonised Business Demography Survey (HBDS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Harmonised Business Demography Survey relates months trading to enterprises employing beyond the owner.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:eab48425c192ac3666523306", "occurrence_id": "447aa8dd1ec4a17c12aebc4b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Deprivation and hardship the numbers held for the sample come from the Diagnostic Deprivation and Hardship Survey (DDHS), the survey Rhodes University follows for 2023, and the four values are carried here as tabulated.\n\nThe survey puts the share of households below the national adequacy line at 33 per cent, the share carrying three deprivations or more at 41 per cent, the share of households that skipped a meal in the last month at 22 per cent, and the share reporting no member with a regular income at 28 per cent. No reweighting was applied to them, and the survey's own reading of a deprivation, which requires the condition to hold at the visit, comes down with them.\n\nA household deprived of storage because it shares a facility was counted as deprived, and a seasonally worked day was counted as no regular income. Each figure keeps the settlement it was drawn from carried with it, rather than mixed across the sample.\n\nThe annex keeps the survey line for each value, the date it was drawn as well as the households each share divides.\n\nMention: \"Diagnostic Deprivation and Hardship Survey (DDHS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Diagnostic Deprivation and Hardship Survey supplies adequacy incidence, deprivation counts, skipped meals and workless households.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:30bf76293ec3f8f375530fc4", "occurrence_id": "9a66c612159b28a45e2c4d56", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Emission records were catalogued into the inventory file from the Combined Emission Inventory Survey (CEIS), Second Collection, the survey ENIA follows for 2017, and the work done on them was cataloguing rather than measurement.\n\nEach record was given the file's own sector band and its reporting dates checked against the collection's reference window; where the survey carries a record the file has no band for, it was parked for the analyst rather than entered under the nearest one. Records the survey lists without a stated method were left unbanded rather than assigned to the sector default, and 480 of them entered the file in that condition. The collection's own stamp was kept on every record so a later collection can be matched without re-deriving it.\n\nThe file's activity counts come from the licensing register, from which the survey supplied nothing; it is the record coding that lets those counts be read against a factor for the emissions table the report draws up.\n\nThe coding sheet, the count of parked records and the day of deposit are detailed in the annex log.\n\nMention: \"Combined Emission Inventory Survey (CEIS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The Combined Emission Inventory Survey supplies the coded emission records behind the inventory file.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:2b64c750d953700d459bde5d", "occurrence_id": "68b1877739ea10d1b1450aac", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Records completeness and response were reconstructed for this review from the Local Integrated Records Registry (LIRR), Wave 2, the register SEP maintains for Sudan, and the readings below are the review's, not the register's.\n\nOffices were sorted by the completeness the wave scores for their birth register and the median days before a certificate is issued computed: 11 days at offices scoring above nine-tenths against 44 days at those below six-tenths. The register holds a completeness score for each office and an issue date for each certificate and never sets one against the other, so the pairing is the review's own and no figure printed here appears in the register in that form.\n\nMaking the groups alike on the staff an office holds for the registration work and on the volume of applications it receives brings the gap down to 13 days, which leaves staffing and volume together as part of what the unadjusted contrast attributed to the completeness score. The chapter prints the narrower gap rather than the wider one.\n\nOffices the wave scores on a part of its indicator list are dropped rather than counted at that score, and they are 5 per cent of offices. Certificates issued at a mobile session are counted at the office that holds the register rather than at the session. A note in the annex gives the sorting rule, the extraction day together with the base for each figure.\n\nMention: \"Local Integrated Records Registry (LIRR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Local Integrated Records Registry relates birth-register completeness to days before a certificate issues.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:fcdcb25c79e114fc9533dde6", "occurrence_id": "f7ef880d47c599793794345a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Transfer receipt and hardship were assembled for this review out of the High-Resolution Social Protection Microdata (HSPM), Collection 9, the extract kept for Chad in 2020, and what follows is the review's arithmetic, not the extract's.\n\nHouseholds were sorted by the months since they first received a transfer and the share below the national adequacy line computed: 41 per cent of households receiving under a year against 26 per cent of those receiving two years or more. The extract holds a first-receipt date for each household and a consumption aggregate for the same household and never sets one against the other.\n\nMaking the groups alike on the number of dependants in the household brings the gap down to 9 points, which leaves dependants as part of what the unadjusted difference owed to the months received.\n\nHouseholds the extract lists as enrolled but never paid are dropped rather than counted as recipients, and they are 6 per cent of households. Those references, the extraction day and the denominators sit in the working papers behind the annex.\n\nMention: \"High-Resolution Social Protection Microdata (HSPM)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The High-Resolution Social Protection Microdata relates months of transfer receipt to adequacy-line incidence.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:beee8af42213775025382bce", "occurrence_id": "2ec5139350416b19b4a5bafe", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Statistical service delivery country figures are assembled from the Statistical Service Delivery Index (SSDI), the indicator BVC maintains for Liberia, and the four values pass into this report as printed.\n\nThe index puts the share of offices publishing on a stated calendar at 52 per cent, the mean days between a reference period closing and publication at 88, the share of offices whose releases carry a method note at 34 per cent, and the share whose budget for the statistical work is named in the national plan at 21 per cent. Nothing was re-estimated here, and the index's own reading of a publication, which requires tables on a public page, is carried alongside them.\n\nA release circulated to departments only was not counted, and a method note carried on a separate page of the same bulletin was counted. Each figure keeps the county it was drawn from with it, rather than pooled across the country.\n\nBeneath the table sit the index line for each value, plus the date the file was taken and the offices each share divides.\n\nMention: \"Statistical Service Delivery Index (SSDI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Statistical Service Delivery Index supplies calendar publication, delay, method notes and named budgets.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:05940c055a67051ee9333863", "occurrence_id": "3fbeb801d188b45d9e8a5193", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Apprenticeship figures spanning the city are taken from the Baseline Apprenticeship Survey (BAS), Seventh Collection, the survey conducted in Rio de Janeiro in 2023, and the four values come into this report unchanged.\n\nThe survey puts the share of establishments taking an apprentice at 18 per cent, the share of apprentices completing their spell at 63 per cent, the mean months a spell runs at 11, and the share of completers held in the same trade six months on at 47 per cent. No figure was recalculated in this report, and the survey's own reading of completion, which requires the employer's signed record, is printed beside them.\n\nAn apprentice who moved employer without leaving the course was counted as continuing, and a spell ended by a closure was not counted as a completion. Each figure keeps the district it was drawn from attached, instead of being blended across the city.\n\nWorking notes record the survey line for each value, along with the day of the extract and the establishments each share divides.\n\nMention: \"Baseline Apprenticeship Survey (BAS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Baseline Apprenticeship Survey supplies establishments taking apprentices, completion, spell length and trade retention.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:b8924c62e11c72afe70f6e87", "occurrence_id": "cd975483705154b00249fdf7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Vaccination campaign figures at the round level are taken from the Rapid Vaccination Campaign Survey (RVCS), Fifth Round, the survey NISR follows, and the four values enter this report as they stand.\n\nThe survey puts the share of children reached by the campaign at 84 per cent, the share of caregivers who knew the post's opening day at 58 per cent, the mean kilometres a caregiver reports walking to the post at 4, and the share of unreached children whose caregiver named distance as the reason at 31 per cent. They are carried as published, and the survey's own reading of a reached child, which requires a card entry or an observed mark, is given next to them.\n\nA child reached at a mobile visit was counted as reached, and a child turned away for illness was not. Each figure keeps the settlement it was drawn from shown with it, instead of combined into one round figure.\n\nThe documentation under the table gives the survey line for each value, the date of the pull plus the children each share divides.\n\nMention: \"Rapid Vaccination Campaign Survey (RVCS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Rapid Vaccination Campaign Survey supplies reach, caregiver awareness, walking distance and distance as a reason.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:645552d42d086b67593eef57", "occurrence_id": "6c3e5021c5b2bedd6d00e82f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Movement and work were reconstructed for this review from the National Mixed Movement Registry (NMMR), First Wave, the register kept for Burkina Faso in 2015, and the readings below are the review's, not the register's.\n\nHouseholds were sorted by the months since the register first listed them as moved and the share whose head reports steady work computed: 39 per cent of households listed two years or more against 21 per cent of those listed under six months. The register holds a first-listing date for each household and a work item for the same household and never sets one against the other.\n\nMaking the groups alike on the distance from the site to the nearest town brings the gap down to 9 points, which leaves distance as part of what the first comparison credits to the months since listing.\n\nHouseholds the register lists twice under different site codes are dropped rather than counted at the earlier listing, and they are 4 per cent of households. A note in the annex gives the sorting rule, the pull date, and the base for each share.\n\nMention: \"National Mixed Movement Registry (NMMR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The National Mixed Movement Registry relates months since listing to households whose head holds steady work.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:e407ab9ed4fa50d778e08298", "occurrence_id": "7bf8fb301228778efa15e519", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Assets and hardship were estimated for this review using the Household Asset Ownership Survey (HAOS), Fourth Edition, the survey Bureau Van Dijk follows in Kenya, and the figures are the review's own.\n\nHouseholds were sorted by the number of asset classes the survey finds them holding and the share below the national adequacy line computed: 18 per cent of households holding six classes or more against 54 per cent of those holding two or fewer. The survey holds an asset flag for each household and a consumption aggregate for the same household and never sets one against the other.\n\nMaking the groups alike on the land the household works brings the gap down to 16 points, which leaves land as part of the gap the opening contrast assigned to the asset count.\n\nHouseholds reporting an asset they no longer hold are dropped rather than counted at their older list, and they are 5 per cent of households. Working notes list the sorting rule, the extraction day, with the denominators each share rests on alongside.\n\nMention: \"Household Asset Ownership Survey (HAOS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Household Asset Ownership Survey relates asset classes held to households below the adequacy line.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:80b5eaf4b0de2baad1e06fa8", "occurrence_id": "4c2da39c454b0836e9927d55", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Youth transition figures are reported from the Quarterly Youth Transition Index (QYTI), Edition 5, the indicator INCAJU maintains, and the four values reach this report as published.\n\nThe index puts the share of school leavers in education or training three months after leaving at 46 per cent, the share with no recorded activity at 22 per cent, the mean months a first spell of work lasts at 7, and the share of leavers whose first work matches their field at 34 per cent. Nothing here was recomputed, and the index's own reading of an activity, which requires the centre to have recorded it in the quarter, is bound to the figures.\n\nA short unpaid placement was counted as training, and a season of farm work at home was counted as no recorded activity. Each figure keeps the region it was drawn from reported with it, rather than blended into a single frame share.\n\nThe index line for each value: the date on the extract and the leavers each share divides sit with the annex tables.\n\nMention: \"Quarterly Youth Transition Index (QYTI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Quarterly Youth Transition Index supplies study or training after leaving, inactivity, first job length and field match.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:7cf772728ebf070f92280b18", "occurrence_id": "7edd3b4bfe4ab999748c689b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Service coverage and complaint handling were reconstructed for this review from the Municipal Operations Index System (MOIS), Third Collection, the register kept for Syria in 2021, and the readings below are the review's, not the register's.\n\nMunicipalities were sorted by the share of households the register shows on the water network and the median days before a network complaint is closed computed: 9 days at municipalities above nine-tenths on the network against 34 days at those below half. The register holds a network share for each municipality and a closing date for each complaint and never sets one against the other.\n\nHolding the number of maintenance crews the municipality reports even narrows the gap to 11 days, so crews carry part of the distance the first comparison ascribed to the network share.\n\nMunicipalities the register covers for part of the collection are dropped rather than counted at a lower coverage, and they are 6 per cent of municipalities. A note in the annex gives the sorting rule, the pull date and the base for each figure.\n\nMention: \"Municipal Operations Index System (MOIS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Municipal Operations Index System relates network coverage to days before a network complaint closes.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:e2ff3b3a0dfd9410ddba4a2e", "occurrence_id": "61fa2dd6acc8d279685c88e3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The programme figures in this chapter are drawn from the unit's own delivery log, and the Monthly Programme Monitoring Microdata (MPMM), the extract ICAO compiled for 2018, is quoted only to place those figures in context.\n\nThe extract puts the share of sites reporting to the centre each month at 77 per cent, the mean months a site reports late at 2, and the share of sites whose report the centre accepted without query at 61 per cent. All three are given as the extract issues them, with its own site list stated, and no chapter figure was built from them.\n\nTheir use here is to show how reporting looks across the programme beside the eleven sites whose log the chapter works from. The chapter's argument rests on that log, which names each delivery rather than sampling sites at a visit, and the two are kept apart rather than merged.\n\nThe annex lists the extract tables consulted and the day of the pull, which is the full extent of the extract's use in this chapter.\n\nMention: \"Monthly Programme Monitoring Microdata (MPMM)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Monthly Programme Monitoring Microdata is quoted as context beside the unit's delivery log.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:3cb3ba273b174d7e3ef71197", "occurrence_id": "4d7dba8260bc654c1f6f36b9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Training placement was derived for this chapter from the High-Resolution Vocational Training Index (HVTI), Edition 8, the index HLP AoR maintains, and the figures below are the chapter's rather than the index's.\n\nCourses were sorted by the number of work-placement hours the index records and the share of leavers in related work six months on computed: 69 per cent of courses above 300 placed hours against 38 per cent of those below 100. The index holds a placed-hours figure for each course and a work flag for each leaver and never sets one against the other.\n\nThat comparison supports one claim, that a placement hour should be logged as it happens rather than declared at the close of the course. The cost lines the proposal quotes come from the programme's own budget sheet, from which the index supplied nothing.\n\nCourses the index lists with no leavers traced are dropped rather than counted as placing nobody, and they are 7 per cent of courses. The bands and the counts behind each share are recorded under the table.\n\nMention: \"High-Resolution Vocational Training Index (HVTI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The High-Resolution Vocational Training Index supports a placement-logging claim built from placed hours and related work.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:533e6ca5b53249915506938d", "occurrence_id": "c88e5eb2f4940f0c56be0a78", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Three grid lines in the annex come from the Baseline Population Grid Report (BPGR), Third Release, the report GBIF issues, and all three are transcribed without alteration.\n\nThe report puts the share of the frame covered by a grid cell holding a population estimate at 88 per cent, the mean cell size at one kilometre, and the share of cells whose estimate changed between releases at 13 per cent. Nothing here was recomputed in this report, and the report's own reading of a covered cell, which requires an estimate above zero rather than an interpolated one, travels with the figures.\n\nThe annex's other entries are drawn from the census office's own small-area output, so the report carries three lines while the output holds the other lines. A figure reported for one region keeps that region attached rather than being averaged across the frame, and where the report and the output disagree about a cell the report's figure is kept for these three lines and the output's for the rest, with the difference noted under the table rather than smoothed away.\n\nCells the report marks as derived from a neighbouring cell are excluded from all three lines, since the report does not treat them as estimated. The annex log holds the report layer behind each value, the extraction day and, alongside it, the counts the shares rest on.\n\nMention: \"Baseline Population Grid Report (BPGR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Baseline Population Grid Report supplies covered cells, cell size and inter-release estimate change.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:e77f8102ef6a766180d7eebf", "occurrence_id": "a439f4d3f60d7ce749d24462", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Targeting and access were reconstructed for this review from the Follow-Up Targeting and Payments Survey, which The World Bank conducted in Rio de Janeiro in 2019, and the readings below are the review's, not the survey's.\n\nHouseholds were sorted by the travel time from the dwelling to the payment point and the share reporting a payment received late computed: 26 per cent of households above half an hour against 12 per cent of those under ten minutes. The survey holds a travel time for each household and a payment date for the same household and never sets one against the other.\n\nHolding the age of the household head even narrows the gap to 7 points, so age carries part of what the raw comparison laid at the door of the travel time.\n\nHouseholds the survey finds at an address the programme roll cannot place are dropped rather than counted as far, and they are 5 per cent of households. A note in the annex gives the sorting rule, the pull date as well as the base for each share.\n\nMention: \"Follow-Up Targeting and Payments Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Follow-Up Targeting and Payments Survey relates travel time to the payment point and late payments.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:7d9cc4deca5dbf2fa25d0ae2", "occurrence_id": "2937559b9507d57e51dfb2f5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Elevation model figures for the rural frame are reported from the Rural Elevation Model Index, the indicator Badan Pusat Statistik or BPS compiled for Russia in 2022, and the four values appear here exactly as printed.\n\nThe index puts the share of the frame covered at a cell size of thirty metres or finer at 57 per cent, the mean vertical error it reports for that cover at 1.4 metres, the share of the cover acquired before the last flood season at 44 per cent, and the share of the frame left uncovered at 11 per cent. They are reproduced as tabulated, and the index's own reading of cover, which requires a ground value in the cell rather than an interpolated one, goes with the figures.\n\nA cell filled from a coarser neighbouring model was not counted as cover, and a model reissued without new acquisition kept its original date. Each figure keeps the federal subject it was drawn from kept with it, instead of being rolled up to the frame.\n\nUnder the table are the index line for each value, the extraction date, and the cells each share divides.\n\nMention: \"Rural Elevation Model Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Rural Elevation Model Index supplies fine-cell cover, vertical error, acquisition recency and uncovered frame.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:1e0bbf30a548c53cb0cc4bbd", "occurrence_id": "9992d97e9972b49bd56c15b7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Public records country-level figures come from the High-Frequency Public Records System, the register ZamStats maintains for Belgium in 2021, and the four values are reproduced here as issued.\n\nThe register puts the share of series held in a format the reading room can open at 74 per cent, the mean months between a series closing and its deposit at the archive at 14, the share of series with a review date recorded at 61 per cent, and the share of deposits the archive accepted without a query at 82 per cent. They were not re-derived here, and the register's own reading of a deposit, which requires the archive to have logged the transfer, travels with the values.\n\nA series left on the office's own server was not counted as deposited, and a review date set by a predecessor was counted. Each figure keeps the service it was drawn from named with it, instead of averaged over the country.\n\nThe register line for each value, the day the table was pulled as well as the series each share divides are recorded under the table.\n\nMention: \"High-Frequency Public Records System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The High-Frequency Public Records System supplies readable series, deposit delay, review dates and clean deposits.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:f36ab6373dde138cb638161e", "occurrence_id": "64d996c12554ef4471ff8098", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The bulletin figure in this chapter is the office's own, taken from its release calendar, and the Regional Statistical Bulletin Registry (RSBR), Collection 4, the register NRC-IDMC maintains, is cited to test it.\n\nThe office's calendar shows 39 bulletins released in the period. The registry shows 34 for the same offices, because it counts a bulletin only once its tables stand on a public page, while the calendar records a bulletin at the day it is scheduled. Neither rule was applied to the other's calendar.\n\nThe 5 distance is smaller than the number the calendar shows in a quarter at the period's peak, so the office's figure is printed as scheduled and the registry supplies no revision to it and no bulletins for the offices its collection did not reach.\n\nExcluded are bulletins the registry lists as withdrawn without replacement, 12 of them, since neither count can settle them. Which bulletins each source counts is carried in the annex working papers.\n\nMention: \"Regional Statistical Bulletin Registry (RSBR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Regional Statistical Bulletin Registry checks the office's count of bulletins released in the period.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:b329fe497414834736545f67", "occurrence_id": "75b1196e392eb9e58e5a37ee", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The operations figure in this chapter is the union's own, taken from its member returns, and the Multi-Topic Operations Index Collection (MOIC), Second Version, the collection FAPRI compiled for 2016, is cited to test it.\n\nThe union's returns show 72 per cent of its members meeting the volume standard. The collection shows 65 per cent for the same members, because it counts a member as meeting the standard only where its own surveyor recorded the volume, while the union's returns accept a figure the member declares. Neither rule was applied to the other's returns.\n\nThe 7-point distance is smaller than the movement the returns show between two rounds, so the union's figure is printed as returned and the collection supplies no revision to it and no members its survey did not visit.\n\nExcluded are members the collection scores on a part of its standard, 88 of them, since the two scores are not on the same footing. Which members each source counts is detailed in the method note in the annex.\n\nMention: \"Multi-Topic Operations Index Collection (MOIC)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Multi-Topic Operations Index Collection checks the union's share of members meeting the volume standard.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:a36ef92115c1078b5df50f7b", "occurrence_id": "d2d21f5500cda137879d9b96", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Water body figures for each part of the frame are drawn from the Triennial Water Body Registry (TWBR), Version 5, the register APIS maintains, and the four values are quoted here as printed.\n\nThe registry puts the number of water bodies held in the frame at 41 000, the share whose boundary was drawn from a single observation at 29 per cent, the share whose area changed between versions at 7 per cent, and the share carrying a named local identifier at 63 per cent. No arithmetic was applied to them here, and the registry's own reading of a water body, which requires open water at the reference pass, follows the figures.\n\nA body dry at the reference pass was not counted, and a body split by a bridge was counted once. Each figure keeps the basin it was drawn from listed with it, rather than smoothed across the frame.\n\nThe annex note carries the registry layer for each value, plus the day of the pull and the bodies each share divides.\n\nMention: \"Triennial Water Body Registry (TWBR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Triennial Water Body Registry supplies body counts, single-observation boundaries, area change and local identifiers.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:6c60647ddefacf9f0c93d702", "occurrence_id": "45fc9edc1524ecf253534f6b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The administrative figures in this chapter are drawn from the assembly's own returns, and the District Administrative Survey Registry (DASR), Phase 2, the register kept for Accra in 2019, is quoted only to place those figures in context.\n\nThe registry puts the share of districts publishing their decisions at 54 per cent, the mean months between a decision and its publication at 3, and the share of districts whose staff roster stands current at 47 per cent. All three are given as the registry issues them, with its own district list stated, and no chapter figure was built from them.\n\nTheir use here is to show how district administration looks across the region beside the four whose returns the chapter works from. The chapter's argument rests on those returns, which name each decision rather than sampling districts at a visit, and the two are kept apart rather than merged.\n\nThe annex lists the registry tables consulted and the day the rows were taken, which is the full extent of the registry's use in this chapter.\n\nMention: \"District Administrative Survey Registry (DASR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The District Administrative Survey Registry is quoted as context beside the assembly's own returns.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:9b5c2227a23594691bc576e5", "occurrence_id": "d7ccc28029c3f2106f577353", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Unregistered work and distance were reconstructed for this review from the Government Informal Sector Survey (GISS), the survey TurkSTAT follows in France, and the readings below are the review's, not the survey's.\n\nWorkers were sorted by the travel time from home to the place they work and the share reporting no written contract computed: 71 per cent of workers above forty minutes against 48 per cent of those under fifteen. The survey holds a travel time for each worker and a contract item for the same worker and never sets one against the other.\n\nHolding the hours the worker reports even narrows the gap to 11 points, so hours carry part of the gap the opening comparison put down to the travel time.\n\nWorkers the survey finds working only in the household's own business are dropped rather than counted as uncontracted, and they are 6 per cent of workers. A note in the annex gives the sorting rule, the pull date together with the base for each share.\n\nMention: \"Government Informal Sector Survey (GISS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Government Informal Sector Survey relates travel time to work and workers without a written contract.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:134972f7d1172dcd3c010057", "occurrence_id": "b7c04e71aa7ab5d5cca83dbc", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Urban hiring figures measured across the region are read from the Urban Hiring and Vacancies Survey (UHVS), the survey conducted in Sub-Saharan Africa in 2016, and the four values are carried here as tabulated.\n\nThe survey puts the share of establishments with a vacancy open at the visit at 22 per cent, the mean days a vacancy stands unfilled at 46, the share of establishments that filled a vacancy from outside the city at 34 per cent, and the share advertising the vacancy in writing at 41 per cent. No reweighting was applied to them, and the survey's own reading of a vacancy, which requires the post to be open on the day of the visit, is reported with them.\n\nA post held for a returning employee was not counted as vacant, and a vacancy filled the same week was counted as open. Each figure keeps the city it was drawn from carried with it, rather than mixed across the region.\n\nThe table's footnotes hold the survey line for each value, along with the day the extract ran and the establishments each share divides.\n\nMention: \"Urban Hiring and Vacancies Survey (UHVS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Urban Hiring and Vacancies Survey supplies open vacancies, time unfilled, external hiring and written advertising.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:91732a95d4f127b94317c695", "occurrence_id": "f066a74dc9502066a2fbe47c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Apprenticeship figures for the province are reported from the High-Resolution Apprenticeship Survey, the survey CEDLAS-World Bank conducted in Punjab in 2023, and the four values pass into this report as printed.\n\nThe survey puts the share of young people who held an apprenticeship at 14 per cent, the share of those placed in a related job within three months at 52 per cent, the mean monthly stipend an apprentice receives at 9 500, and the share of apprentices whose household falls below the provincial adequacy line at 31 per cent. Nothing was re-estimated here, and the survey's own reading of an apprenticeship, which requires a written workplace agreement, is attached to them.\n\nA short unpaid placement at a relative's shop was not counted, and an apprentice working without a stipend was counted. Each figure keeps the district it was drawn from with it, rather than pooled across the province.\n\nUnder the table you will find the survey line for each value, the pull date plus the young people each share divides.\n\nMention: \"High-Resolution Apprenticeship Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The High-Resolution Apprenticeship Survey supplies apprenticeship incidence, related placement, stipend and adequacy incidence.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:9c1de70c8703da4a296ffe27", "occurrence_id": "487cbb856d845a2ec86dddfe", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Whether the ministry's own count of returns holds was tested against the Subnational Return and Relocation Survey, which Fitch conducted in the Russian Federation in 2018, and the survey is the principal source for the test.\n\nThe ministry's book gives 84 000 households returned to their origin. The survey gives 97 000 for the same regions, because it counts a household the survey finds back in its origin dwelling, which the book records only where the household registered the return with an office. No basis was used on the other's count.\n\nThe survey therefore carries the chapter's count of returns, and the ministry's smaller figure is printed beside it so the reader can see that the distance is the registration rule rather than a disagreement about the year. Where a household appears in both, the ministry's date is kept and the two counts are not added.\n\nHouseholds the survey finds at an origin dwelling the book cannot place are dropped from both rather than counted by the survey alone, 6 per cent of its listings. The registration rule each source works to is kept in the annex's method note.\n\nMention: \"Subnational Return and Relocation Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Subnational Return and Relocation Survey tests the ministry's count of households returned to their origin.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:b3a8b7d981bd56f6a6463232", "occurrence_id": "bea742a8b761e0c70a437be5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Whether the council's own pay figure holds was tested against the Longitudinal Wage Structure Survey (LWSS), Round 8, the survey DESTA follows for 2018, and the survey is the principal source for the test.\n\nThe council's return gives 24 as the mean hourly wage in the trades it covers. The survey gives 21 for the same trades, because it counts a worker paid by output at the hours the survey records rather than the hours the employer books, which the council's return takes from the employer. The bases stay separate: neither is applied to the other's count.\n\nThe survey therefore carries the chapter's mean wage, and the council's larger figure is printed beside it so the reader can see that the distance is the hours rule rather than a disagreement about the trades. Where a worker appears in both, the council's trade code is kept and the two means are not averaged.\n\nWorkers the survey traces to a second employer mid-round are dropped from both rather than counted by the survey alone, 5 per cent of its listings. The hours rule each source works to is held in the annex's method sheet.\n\nMention: \"Longitudinal Wage Structure Survey (LWSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Longitudinal Wage Structure Survey tests the council's mean hourly wage in the trades it covers.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:c9465600b925df57f426c232", "occurrence_id": "bd59fd724012d6c536653e71", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Cash transfer the country figures are compiled from the Local Cash Transfer Registry Survey (LCTRS), the survey conducted in Chad in 2017, and the four values come into this report unchanged.\n\nThe survey puts the share of households receiving any transfer at 28 per cent, the mean amount a receiving household takes in a month at 12 000, the share of receiving households reporting a health episode paid for from the transfer at 44 per cent, and the share reporting the transfer arrived late at 19 per cent. No figure was recalculated in this report, and the survey's own reading of receipt, which requires the household to have collected the payment, is stated with them.\n\nA household enrolled but never paid was not counted as receiving, and an episode treated at home was counted as paid for. Each figure keeps the department it was drawn from attached, instead of being blended across the country.\n\nThe annex keeps the survey line for each value: the day of extraction and the households each share divides.\n\nMention: \"Local Cash Transfer Registry Survey (LCTRS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Local Cash Transfer Registry Survey supplies transfer receipt, amount, health episodes paid from it and late payments.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:c1e249682d364e20df5f44e8", "occurrence_id": "6dadf6e25d9ee75d2bb99dc0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Coverage and hardship were reconstructed for this review from the Subnational Social Protection Survey (SSPS), Phase 2, the survey conducted in Lao PDR in 2020, and the readings below are the review's, not the survey's.\n\nVillages were sorted by the share of poor households the survey finds enrolled in the scheme and the share of those households below the national adequacy line computed: 46 per cent of villages with fewer than half enrolled against 27 per cent of those with nine-tenths enrolled. The survey holds an enrolment flag for each household and a consumption aggregate for the same household and never sets one against the other.\n\nHolding the distance from the village to the district centre even narrows the gap to 9 points, so distance carries part of the distance the first comparison charged to enrolment.\n\nHouseholds enrolled under a name the survey cannot match are dropped rather than counted as covered, and they are 5 per cent of households. A note in the annex gives the sorting rule, the pull date, and the base for each share.\n\nMention: \"Subnational Social Protection Survey (SSPS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Subnational Social Protection Survey relates scheme enrolment of poor households to adequacy-line incidence.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:84e30d66b15e5cb7392afe03", "occurrence_id": "8f3257bce3f448f1a4587c30", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Consumption and heating were put together for this review from the Experimental Living Standards Survey (ELSS), Round 9, the survey EGSII follows, and the readings below are the review's own rather than the survey's.\n\nDwellings were grouped by the fuel they warm with and the share of households below the national adequacy line computed: 44 per cent of households warming with coal against 19 per cent of those warming with gas. The survey carries a heating fuel item for each dwelling and a consumption aggregate for the same dwelling without ever joining the two, so the pairing printed here exists only in this review.\n\nMaking the groups alike on the age of the dwelling and on the number of rooms it holds brings the distance down to 12 points, which leaves age and rooms together as part of the gap the opening comparison puts down to the heating fuel. The chapter prints the narrower of the two.\n\nHouseholds the survey finds empty for the winter are dropped rather than counted at their last recorded fuel, and they are 4 per cent of households. Households warming with paraffin are grouped with coal rather than with gas, since the survey records both under a solid-or-oil heading. The annex note carries the grouping rule, the pull date, with the base of each figure alongside.\n\nMention: \"Experimental Living Standards Survey (ELSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Experimental Living Standards Survey relates heating fuel to households below the adequacy line.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:e965d0e21d52e83d18ee582d", "occurrence_id": "1824350cbf822b4e27f61fd8", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Enterprise births and job creation were assembled for this review from the Baseline Enterprise and Jobs System (BEJS), Round 7, the register kept for Xinjiang province in 2017, and the figures below are the review's, not the register's.\n\nDistricts were grouped by the number of new enterprises the register records per thousand residents and the share of jobs the district added in the year computed: 6.4 per cent of districts above four per thousand against 2.1 per cent of those below one. The register carries a birth count for each district and an employment series for the same district without ever joining the two.\n\nHolding the size of the district's existing workforce even leaves 1.6 points of that distance, so workforce size accounts for part of what the opening comparison puts down to enterprise births.\n\nDistricts the register merges during the round are dropped rather than counted at their predecessor's births, and they are 5 per cent of districts. The annex note carries the grouping rule, the pull date and the base of each figure.\n\nMention: \"Baseline Enterprise and Jobs System (BEJS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Baseline Enterprise and Jobs System relates enterprise births per thousand residents to jobs added in the year.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:5c85dffa1fa2c16605aa4e3e", "occurrence_id": "24e8ac541afe4f265a038be3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Poverty and land tenure were reconstructed for this review from the Record-Based Socioeconomic Microdata, the extract STEP compiled for Botswana in 2023, and the readings here are the review's, not the extract's.\n\nHouseholds were grouped by whether the extract records a registered title in the household's name and the share below the national adequacy line computed: 21 per cent of households with a registered title against 49 per cent of those without one. The extract carries a title item for each household and a consumption aggregate for the same household without ever joining the two.\n\nHolding the herd size the household reports even leaves 14 points of that distance, so livestock accounts for part of what the opening comparison puts down to the title.\n\nHouseholds whose title the extract records as applied for but not granted are dropped rather than counted as titled, and they are 6 per cent of households. The annex note carries the grouping rule, the pull date and, alongside it, the base of each share.\n\nMention: \"Record-Based Socioeconomic Microdata\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Record-Based Socioeconomic Microdata relates a registered land title to adequacy-line incidence.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:d1d2c1b0139c5418aac55926", "occurrence_id": "212e1ff3a7f3bfbb4539ee63", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Three operations lines in the response annex come from the Coordinated Operations Monitoring Report (COMR), Second Version, the report issued for the CHT and Sylhet divisions in 2020, and all three are copied across unchanged.\n\nThe report puts the share of partner sites filing a monthly situation return at 81 per cent, the mean days a filed return takes to reach the coordination cell at 4, and the share of returns carrying a beneficiary count at 66 per cent. Nothing was recomputed for this report, and the report's own reading of a filed return, which requires the coordination cell to have logged it, travels with the figures.\n\nThe annex's other lines are compiled from the camp's own activity sheet, so the report carries three lines while the sheet carries the lines the table does not. A figure reported for one camp keeps that camp attached rather than being averaged across the divisions, and where report and sheet disagree about a filing the report's figure is kept for these three lines and the sheet's for the rest, the difference being noted beneath the table rather than smoothed away.\n\nSites opened after the version closed are excluded from all three lines, since the report holds no window for them. The annex log carries the report page behind each value, the date it was drawn and the base of each share.\n\nMention: \"Coordinated Operations Monitoring Report (COMR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Coordinated Operations Monitoring Report supplies site filing rates, return delay and returns carrying beneficiary counts.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:7986ceae0e69a5e74d1c9eb1", "occurrence_id": "5c4699f5037d4e4757359932", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Records handling was derived for this chapter from the Monthly Public Records Index, the indicator the Fiscal Affairs Department maintains for Nairobi County in 2015, and the figures below are the chapter's own, not the index's.\n\nWards were grouped by the number of series the index finds without a custody note and the median days before a resident request is answered computed: 13 days at wards with fewer than twenty unnoted series against 36 days at those with sixty or more. The index carries a series count for each ward and an answer date for each request without ever joining the two.\n\nThat comparison supports one claim, that a custody note should be written when a series opens rather than when it is first asked for. The staffing lines the proposal quotes come from the county's own establishment return, from which the index supplied nothing.\n\nWards the index visits twice in the year are counted at their second visit rather than their higher count, and they are 5 per cent of wards. The bands and the counts behind each median are recorded beneath the table.\n\nMention: \"Monthly Public Records Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Monthly Public Records Index supports a custody-note claim built from unnoted series and request delay.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:077435bdc5ca37cda72fc80c", "occurrence_id": "44aae1a8bf5901aa4a645f56", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Disability and support figures for the whole state are sourced from the Continuous Disability and Support Survey (CDSS), the survey APRA follows in Florida, and the four values reach this report unchanged.\n\nThe survey puts the share of adults living with a disability at 12 per cent, the share of those adults receiving a support payment at 39 per cent, the mean months a household waits for a decision on an application at 7, and the share of households below the state adequacy line where an adult lives with a disability at 34 per cent. None of them was recalculated here, and the survey's own reading of a disability, which requires the respondent to answer yes to the activity set, comes with them.\n\nAn adult whose condition the survey records as temporary was still counted, and a carer's allowance was counted as a support payment. Each figure keeps the county it was drawn from shown with it, instead of combined into one state figure.\n\nBeneath the table sit the survey row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Continuous Disability and Support Survey (CDSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Continuous Disability and Support Survey supplies disability incidence, payments, decision waits and adequacy incidence.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:3566af47b80d1ecd6aa6ff8a", "occurrence_id": "5554aa5aafe9193ec8c94043", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Unpaid work and job search were put together for this review from the Independent Time Use Survey (ITUS), the survey the German Socioeconomic Panel follows in Bhopal, and the readings below are the review's own.\n\nAdults were grouped by the hours they record for unpaid care in the week and the share reporting a job search in the same week computed: 33 per cent of adults under ten care hours against 14 per cent of those above twenty-five. The survey carries a care-hours item for each adult and a search item for the same adult without ever joining the two.\n\nMaking the groups alike on the number of children in the household leaves 6 points of that distance, which leaves children as part of the gap the opening comparison puts down to care hours.\n\nAdults the survey lists as students are dropped rather than counted as searching, and they are 5 per cent of adults. The annex note carries the grouping rule, the pull date as well as the base of each share.\n\nMention: \"Independent Time Use Survey (ITUS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Independent Time Use Survey relates unpaid care hours to adults searching for work in the same week.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:4ed1ab60cf9df5a74cc862d9", "occurrence_id": "79b26b4c85aaa836ab60b08c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Three carbon lines in the climate annex come from the Experimental Ecosystem Carbon Registry (EECR), Fifth Edition, the register kept for Burkina-Faso in 2019, and all three are copied across unchanged.\n\nThe registry puts the share of the frame whose woodland the registry holds a carbon estimate for at 58 per cent, the mean tonnes of carbon per hectare it records for that woodland at 41, and the share of the frame whose estimate changed between editions at 9 per cent. Nothing was recomputed for this report, and the registry's own reading of an estimate, which requires a plot measured in the field rather than a modelled surface, travels with the figures.\n\nThe annex's other lines are compiled from the service's own land-use return, so the registry carries three lines while the return carries the balance. A figure reported for one province keeps that province attached rather than being averaged across the country, and where registry and return disagree about a boundary the registry's figure is kept for these three lines and the return's for the rest, the difference being noted beneath the table rather than smoothed away.\n\nWoodland the registry marks as burned in the edition is excluded from all three lines, since the registry re-measures it in the next one. The annex log carries the registry layer behind each value, the date the file was taken and the counts each share divides.\n\nMention: \"Experimental Ecosystem Carbon Registry (EECR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Experimental Ecosystem Carbon Registry supplies wooded cover with estimates, carbon density and inter-edition change.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:aeec2583641723f59fe75263", "occurrence_id": "f790423b0aecc7b6e5aa9cee", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The loss figure in this chapter is the board's own, taken from its own purchase books, and the Joint Storage and Loss Index (JSLI), Release 2, the indicator Germanwatch e.V. maintains, is cited to test it.\n\nThe board's books show 14 per cent of delivered grain lost between threshing and sale. The index shows 21 per cent for the same districts, because it asks the household where the grain went, while the board's books see only the grain that reaches the gate. Each basis keeps its own books untouched.\n\nThe 7-point distance is smaller than the movement the books show between two seasons, so the board's figure is printed as booked and the index supplies no revision to it and no grain for the districts its round did not visit.\n\nExcluded are holdings the index records storing grain in two structures, 940 of them, since the two records cannot be settled against one another. Which grain each source counts is laid out in the annex notes.\n\nMention: \"Joint Storage and Loss Index (JSLI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Joint Storage and Loss Index checks the board's share of delivered grain lost before sale.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:78b88856fcf0f59a15842de2", "occurrence_id": "77e077b42d184c7e60b02ff6", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Whether the utility's own count of dry weeks holds was tested against the Endline Water and Climate Report (EWCR), the report the Brazilian Central Bank issued for 2017, and the report is the principal source for the test.\n\nThe utility's summary gives 9 weeks in the year with supply below its own rationing threshold. The report gives 13 for the same basins, because it counts a week below the threshold when the reservoir level falls there at any point, which the utility's summary records only when a rationing order was issued. Neither basis was carried across to the other's count.\n\nThe report therefore carries the section's count of dry weeks, and the utility's smaller figure is printed beside it so the reader can see that the distance is the triggering rule rather than a disagreement about the year. Where a week appears in both, the utility's order date is kept and the two counts are not averaged.\n\nWeeks the report dates to a month rather than a week were dropped from both rather than counted by the report alone, 8 per cent of its listings. The triggering rule each source works to is filed in the annex file.\n\nMention: \"Endline Water and Climate Report (EWCR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Endline Water and Climate Report tests the utility's count of weeks below its rationing threshold.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:e699cdff66c762fc02de2f23", "occurrence_id": "07d11a2c2cb52b51236a4d1e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The consumption figure in this chapter is the office's own, taken from its price collection, and the Standardised Household Income and Expenditure Survey (SHIES), Collection 7, the survey conducted in Xinjiang province in 2022, is cited to test it.\n\nThe office's collection shows spending for the basket rising 6 per cent in the year. The survey shows 9 per cent for the same areas, because it records what the household paid where it shopped, while the office's collection prices the same basket at the town market. One basis was never used to adjust the other's collection.\n\nThe 3-point distance is smaller than the movement the collection shows between two quarters, so the office's figure is printed as collected and the survey supplies no revision to it and no households for the settlements its teams did not reach.\n\nExcluded are households the survey prices at two markets in the same month, 260 of them, since the two cannot be settled against one another. Which spending each source counts is laid out in the papers behind the annex.\n\nMention: \"Standardised Household Income and Expenditure Survey (SHIES)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Standardised Household Income and Expenditure Survey checks the office's basket spending increase.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:f56897eb8b594d33c288b63f", "occurrence_id": "b30e8ca721f6dc08cad65ea1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The insurance figure in this chapter is the agency's own, taken from its claim ledger, and the Midline Unemployment Insurance Survey (MUIS), Release 4, the survey conducted in Florida in 2016, is cited to test it.\n\nThe agency's ledger shows 68 per cent of claimants receiving a payment within its stated fortnight. The survey shows 59 per cent for the same claims, because it dates a claim from the day the claimant says the work ended, while the ledger dates it from the day the employer filed the separation. The two bases were not applied to each other's ledger.\n\nThe 9-point distance is smaller than the movement the ledger shows between two quarters, so the agency's figure is printed as ledged and the survey supplies no revision to it and no claims whose claimant could not be reached.\n\nExcluded are claims the survey finds reopened in the same quarter, 310 of them, since the ledger carries them once and the survey twice. Which claims each source counts is set out in the working papers behind the annex.\n\nMention: \"Midline Unemployment Insurance Survey (MUIS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Midline Unemployment Insurance Survey checks the agency's share of claims paid within the stated fortnight.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:f9706fa9b11c331db348baac", "occurrence_id": "c3d693eb1b7df70aa9a87d9a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Road segment figures for the metropolitan frame are reported from the Metropolitan Road Segment Census (MRSC), Fourth Round, the census the Federal Bureau of Statistics compiled for 2017, and the four values reach this report unchanged.\n\nThe census puts the number of segments held in the frame at 128 000, the share whose surface class was recorded on the ground at 64 per cent, the share whose length changed between rounds at 5 per cent, and the share lying within a kilometre of a school the census names at 22 per cent. None of them was recalculated here, and the census's own reading of a segment, which requires both ends to be fixed at the round, is kept with them.\n\nA segment under works at the round was still counted, and a track usable only in the dry season was counted as unsealed. Each figure keeps the metropolitan unit it was drawn from reported with it, rather than blended into a single frame share.\n\nBeneath the table sit the census row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Metropolitan Road Segment Census (MRSC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Metropolitan Road Segment Census supplies segment counts, ground-recorded surfaces, length change and proximity to schools.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:be0e3cff7ea412325e020c56", "occurrence_id": "b7291e7471fe1fa3057c427d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Learning and household means were put together for this review from the Analytical Learning and Equity Census (ALEC), the census conducted in Central Asia in 2015, and the readings below are the review's own.\n\nChildren were grouped by the highest schooling of the household head and the share able to read a short sentence at the census's own standard computed: 78 per cent of children of a head with upper secondary or more against 46 per cent of those whose head completed none. The census carries a schooling item for each head and a reading result for each child without ever joining the two.\n\nMaking the groups alike on the language the census records for the household leaves 14 points of that distance, which leaves language as part of the gap the opening comparison puts down to the head's schooling.\n\nChildren the census tests but records no result for are dropped rather than counted as unable to read, and they are 4 per cent of children. The annex note carries the grouping rule, the pull date together with the base of each share.\n\nMention: \"Analytical Learning and Equity Census (ALEC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Analytical Learning and Equity Census relates the head's schooling to children able to read a sentence.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:3358f1d093af0169752e9189", "occurrence_id": "5a90a8149cc2d211bbbc413a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Three informal sector lines in the labour annex come from the Multi-Topic Informal Sector Census (MISC), Round 6, the census DOLISA follows, and all three are copied across unchanged.\n\nThe census puts the share of working adults trading without a registration at 57 per cent, the share of those adults who left school before finishing lower secondary at 49 per cent, and the mean weeks in the year an unregistered unit reports trading at 40. Nothing was recomputed for this report, and the census's own reading of an unregistered unit, which requires no number held at the visit, travels with the figures.\n\nThe annex's other lines are compiled from the office's own permit file, so the census carries three lines while the file supplies the balance. A figure reported for one stratum keeps that stratum attached rather than being averaged across the frame, and where census and file disagree about registration the census's figure is kept for these three lines and the file's for the rest, the difference being noted beneath the table rather than smoothed away.\n\nUnits trading only in the season before the round are excluded from all three lines, since the census does not count them as trading. The annex log carries the census row behind each value, the day of the extract plus the counts under each share.\n\nMention: \"Multi-Topic Informal Sector Census (MISC)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Multi-Topic Informal Sector Census supplies unregistered trading, schooling left early and weeks traded.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:7e02fb85950de03ac2b76da5", "occurrence_id": "03a5ab6b76be1ab68f1d3e23", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Insurance receipt and training were reconstructed for this review from the Statistical Unemployment Insurance Survey (SUIS), Collection 8, the survey conducted in Myanmar in 2023, and the readings here are the review's, not the survey's.\n\nClaimants were grouped by the months they had been out of work at the interview and the share taking a training course offered by the scheme computed: 29 per cent of claimants under three months out of work against 11 per cent of those above twelve. The survey carries an out-of-work duration for each claimant and a course item for the same claimant without ever joining the two.\n\nHolding the qualification the claimant last recorded even leaves 6 points of that distance, so the qualification accounts for part of what the opening comparison puts down to the duration.\n\nClaimants the survey finds working informally during their claim are dropped rather than counted as still out of work, and they are 5 per cent of claimants. The annex note carries the grouping rule, the pull date, and the base of each share.\n\nMention: \"Statistical Unemployment Insurance Survey (SUIS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Statistical Unemployment Insurance Survey relates months out of work to claimants taking a scheme course.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:054e1bb3265842ae53e66870", "occurrence_id": "a4fa749c15524303efad7144", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Living standards the country figures are tabulated by the Joint Living Standards Survey (JLSS), the survey conducted in the United Kingdom in 2022, and the four values reach this report unchanged.\n\nThe survey puts the share of households below the national adequacy line after housing costs at 21 per cent, the share of adults in a household with no paid work at 26 per cent, the mean weekly income of the lowest decile of households at 320, and the share of households unable to cover an unexpected cost of 400 at 34 per cent. None of them was recalculated here, and the survey's own reading of adequacy, which is measured after housing costs, accompanies the figures.\n\nA household with a member on unpaid leave was counted as having that member out of paid work, and a cost met by borrowing was counted as not covered. Each figure keeps the region it was drawn from kept with it, instead of being rolled up to the country.\n\nBeneath the table sit the survey row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Joint Living Standards Survey (JLSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Joint Living Standards Survey supplies after-housing adequacy incidence, workless households, lowest-decile income and shock coverage.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:2ecf69e8ecaa2119f8735a1a", "occurrence_id": "f0e52cb11c5c48d6f7634cf3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Ward boundaries and service were reconstructed for this review from the Pilot Ward Outline Dataset (PWOD), Sixth Version, the geospatial dataset Global Financial Data maintains, and the readings here are the review's, not the dataset's.\n\nHouseholds were matched to the ward whose outline the dataset places them in and grouped by whether that outline changed between versions, and the share the service records as reached within its own standard computed: 63 per cent of households in unchanged wards against 51 per cent of those in wards whose outline moved. The dataset holds an outline per ward and the service records a reach figure per household, and neither reports the two together.\n\nHolding the density of the ward even leaves 5 points of that distance, so density accounts for part of what the opening comparison puts down to the boundary change.\n\nWards the dataset splits into two outlines between versions are dropped rather than counted as changed, and they are 4 per cent of wards. The annex note carries the matching rule, the pull date, with the base of each share alongside.\n\nMention: \"Pilot Ward Outline Dataset (PWOD)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Pilot Ward Outline Dataset relates ward outline change to households reached within the service standard.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:bb3f7a6a0f3b0f5198ea0f25", "occurrence_id": "c0938d529a36bbcd68a064cf", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Register upkeep was derived for this chapter from the Panel Service Register Index (PSRI), Release 8, the indicator the Burundi Statistics Institute maintains, and the figures below belong to the chapter rather than to the index's.\n\nOffices were grouped by the months the index records since a register's last entry and the share of registers whose entries carry a responsible officer computed: 71 per cent of registers updated within the release against 34 per cent of those older than a year. The index carries an entry date for each register and an officer item for the same register without ever joining the two.\n\nThat comparison supports one claim, that a register should be checked monthly rather than at the audit. The staffing lines the proposal quotes come from the institute's own establishment return, from which the index supplied nothing.\n\nRegisters the index lists as replaced during the release are dropped rather than counted at their older entry, and they are 5 per cent of registers. The bands and the counts behind each share are recorded beneath the table.\n\nMention: \"Panel Service Register Index (PSRI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Panel Service Register Index supports a monthly-check claim built from entry age and named officers.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:20a425d19b0b56e5a4e532c6", "occurrence_id": "1755dfb404731f3d90aa7cb3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Inspection and heat exposure were put together for this review from the Audited Labour Inspection Survey (ALIS), Second Collection, the survey conducted in Central Asia in 2017, and the readings below are the review's own.\n\nWorkers were grouped by whether their workplace had been inspected in the year and the share reporting no shade or drinking water at the hottest shift computed: 22 per cent of workers in inspected workplaces against 47 per cent of those in workplaces never inspected. The survey carries an inspection item for each workplace and a heat item for the same workplace without ever joining the two.\n\nMaking the groups alike on the hours a worker spends outdoors leaves 11 points of that distance, which leaves outdoor hours as part of the gap the opening comparison puts down to the inspection.\n\nWorkers in workplaces that closed before the interview are dropped rather than counted at their last site, and they are 6 per cent of workers. The annex note carries the grouping rule, the pull date and the base of each share.\n\nMention: \"Audited Labour Inspection Survey (ALIS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Audited Labour Inspection Survey relates workplace inspection to shade and drinking water at the hottest shift.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:eecd529b7584240da3a82eba", "occurrence_id": "3bf939d0884dd291ce4bf86a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Whether the ministry's own count of open service points holds was tested against the Monthly Service Register Registry (MSRR), Fifth Phase, the register PCBS maintains, and the register is the principal source for the test.\n\nThe ministry's return gives 1 240 points open in the month. The register gives 1 385 for the same districts, because it counts a point open where a service was given at any hour in the month, which the ministry's return records only where the point filed its monthly sheet. Neither basis reworks the other's count.\n\nThe register therefore carries the chapter's count of open points, and the ministry's smaller figure is printed beside it so the reader can see that the distance is the filing rule rather than a disagreement about the month. Where a point appears in both, the ministry's opening date is kept and the two counts are not added.\n\nPoints the register holds an address for but no district were dropped from both rather than counted by the register alone, 7 per cent of its listings. The filing rule each source works to is recorded in the annex working papers.\n\nMention: \"Monthly Service Register Registry (MSRR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Monthly Service Register Registry tests the ministry's count of service points open in the month.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:71fd584d4091f4cb053c2b89", "occurrence_id": "1f94d63b13ec03a0b0d79c38", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Asset ownership the figures for the country are reported by the Coordinated Asset Ownership Survey, the survey the WTO conducted in Switzerland in 2019, and the four values reach this report unchanged.\n\nThe survey puts the share of households owning their dwelling at 42 per cent, the share holding financial assets above a year's median income at 31 per cent, the mean number of asset classes a household in the lowest fifth holds at 2, and the share of households holding no asset at all at 6 per cent. None of them was recalculated here, and the survey's own reading of ownership, which requires the title or account to name a household member, comes down with them.\n\nA dwelling held jointly with a sibling was counted as owned, and a pension right not yet drawn was not counted as a financial asset. Each figure keeps the canton it was drawn from named with it, instead of averaged over the country.\n\nBeneath the table sit the survey row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Coordinated Asset Ownership Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Coordinated Asset Ownership Survey supplies dwelling ownership, financial assets, asset classes of the lowest fifth and assetless households.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:7f6870148a4bbe202ede772c", "occurrence_id": "603cea2cf6c95dc9720e1aac", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Works hiring and distance were put together for this review from the Monthly Public Works Survey (MPWS), Seventh Version, the survey Statistics Indonesia follows, and the readings below are the review's own.\n\nProjects were grouped by the travel time from the worksite to the nearest settlement and the share of labourers the project hired within that settlement computed: 71 per cent of projects under fifteen minutes against 34 per cent of those above an hour. The survey carries a travel time for each project and a hiring place for each labourer without ever joining the two.\n\nMaking the groups alike on the number of labourers the project employs leaves 12 points of that distance, which leaves the labourers as part of the gap the opening comparison puts down to the travel time.\n\nProjects the survey finds halted by a permit dispute are dropped rather than counted at the days they ran, and they are 5 per cent of projects. The annex note carries the grouping rule, the pull date and, alongside it, the base of each share.\n\nMention: \"Monthly Public Works Survey (MPWS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Monthly Public Works Survey relates travel time from a worksite to labourers hired from the nearest settlement.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:8f2375a0ae81f6919ed6d75b", "occurrence_id": "56d573972c3d640a551abf01", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Transfer receipt and schooling were reconstructed for this review from the High-Resolution Social Protection Survey (HSPS), the survey CONARE follows for 2016, and the readings here are the review's, not the survey's.\n\nHouseholds were grouped by the months they had been receiving a transfer and the share of their children enrolled at the survey's visit computed: 88 per cent of households receiving two years or more against 69 per cent of those receiving under a year. The survey carries a first-receipt date for each household and an enrolment item for each child without ever joining the two.\n\nHolding the distance from the household to the nearest school even leaves 9 points of that distance, so the school run accounts for part of what the opening comparison puts down to the months received.\n\nHouseholds the survey lists as enrolled in the scheme but never paid are dropped rather than counted as recipients, and they are 6 per cent of households. The annex note carries the grouping rule, the pull date as well as the base of each share.\n\nMention: \"High-Resolution Social Protection Survey (HSPS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The High-Resolution Social Protection Survey relates months of transfer receipt to children enrolled at school.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:35ef6ca241ac7a4ae1c30c1f", "occurrence_id": "bbe3568d31346241da7b5579", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Three minimum needs lines in the welfare annex come from the Biennial Minimum Needs Registry (BMNR), Seventh Phase, the register Reinhart and Rogoff maintains, and all three are copied across unchanged.\n\nThe registry puts the share of households unable to meet the registry's own basket at 29 per cent, the mean number of basket items a household reports missing at 3, and the share of households missing a medical item at 17 per cent. Nothing was recomputed for this report, and the registry's own reading of a missing item, which requires the household to have gone without it in the reference month, travels with the figures.\n\nThe annex's other lines are compiled from the office's own caseload file, so the registry carries three lines while the file holds the remainder. A figure reported for one stratum keeps that stratum attached rather than being averaged across the frame, and where registry and file disagree about who is short the registry's figure is kept for these three lines and the file's for the rest, the difference being noted beneath the table rather than smoothed away.\n\nHouseholds answering the basket list by proxy are excluded from all three lines, since the registry does not weight them. The annex log carries the registry row behind each value, the date of the pull together with the population each share divides.\n\nMention: \"Biennial Minimum Needs Registry (BMNR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Biennial Minimum Needs Registry supplies households unable to meet the basket, items missed and medical shortfalls.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:8ea07d7e8b19cb7201608ecd", "occurrence_id": "a654de968ddb5f856cc20da1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Three arrival lines in the displacement annex come from the Audited Arrival and Registration Registry, the register ICRG maintains for Iraq in 2018, and all three are copied across unchanged.\n\nThe registry puts the number of households registering at a reception point in the year at 63 000, the share registering within a week of arrival at 41 per cent, and the share whose registration the audit could match to a dwelling record at 72 per cent. Nothing was recomputed for this report, and the registry's own reading of an arrival, which requires the household's own statement of the date, travels with the figures.\n\nThe annex's other lines are compiled from the camp's own roll, so the registry carries three lines while the roll carries what is left over. A figure reported for one governorate keeps that governorate attached rather than being averaged across the country, and where registry and roll disagree about a household the registry's figure is kept for these three lines and the roll's for the rest, the difference being noted beneath the table rather than smoothed away.\n\nHouseholds registering twice under different points are excluded from all three lines, since the audit counts the first entry only. The annex log carries the registry row behind each value, the date on the extract plus the denominators each share uses.\n\nMention: \"Audited Arrival and Registration Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Audited Arrival and Registration Registry supplies registrations, timely registration and matched dwelling records.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:61b21d6d6ca249907b0ce5ea", "occurrence_id": "ee3a4c222e8321558e710b7b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Public works figures covering the state are drawn from the Regional Public Works Index (RPWI), Edition 2, the indicator kept for Gujarat in 2021, and the four values reach this report unchanged.\n\nThe index puts the share of works sanctioned in the year actually begun at 66 per cent, the mean months a work runs beyond its planned close at 7, the share of works with a labour register available for inspection at 58 per cent, and the share of works whose payment to labourers was made within the stated fortnight at 49 per cent. None of them was recalculated here, and the index's own reading of a begun work, which requires earthmoving or structure to have started, is carried alongside them.\n\nA work cleared of site only was not counted as begun, and a work paid partly in the following quarter was counted as late. Each figure keeps the district it was drawn from listed with it, rather than smoothed across the state.\n\nBeneath the table sit the index row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Regional Public Works Index (RPWI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Regional Public Works Index supplies works begun, overrun months, labour registers and timely labour payments.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:dee01b11af2fc774cba78236", "occurrence_id": "7817f7f10c960927e9aca99a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Heat and cold extreme figures covering the whole frame derive from the Subnational Heat and Cold Extremes Dataset (SHCED), Edition 4, the geospatial dataset Jacobs et al maintain, and the four values reach this report unchanged.\n\nThe dataset puts the share of the frame recording at least five hot days in the season at 47 per cent, the mean number of cold days it records where any are recorded at 12, the share of the frame whose readings come from a station rather than an interpolated surface at 63 per cent, and the share of the frame whose extreme count changed between editions at 8 per cent. None of them was recalculated here, and the dataset's own reading of a hot day, which requires the threshold to be exceeded at the station's own hour, is printed beside them.\n\nA day interpolated from a neighbouring cell was not counted, and a station reporting both heat and cold in the season was counted in both. Each figure keeps the subnational unit it was drawn from carried with it, rather than mixed across the frame.\n\nBeneath the table sit the dataset layer behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Subnational Heat and Cold Extremes Dataset (SHCED)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Subnational Heat and Cold Extremes Dataset supplies hot-day cover, cold days, station readings and inter-edition change.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:cc84b35323a184a078543d04", "occurrence_id": "58bb2f92780162e0fceabbfd", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The coverage figure in this chapter is the authority's own, taken from its facility list, and the Rapid Service Catchments System (RSCS), Release 6, the register the Growth Lab at Harvard maintains, is cited to test it.\n\nThe authority's list shows 74 per cent of households standing within its own distance standard of a facility. The system shows 66 per cent for the same areas, because it draws the catchment from the road the household actually travels, while the authority's list measures a straight line from the facility gate. Neither basis was applied to the other's list.\n\nThe 8-point distance is smaller than the movement the list shows between two releases, so the authority's figure is printed as listed and the system supplies no revision to it and no households for the wards its release did not cover.\n\nExcluded are households the system places on a road it marks closed, 210 of them, since neither count can settle the trip. Which households each source counts is detailed in the method note in the annex.\n\nMention: \"Rapid Service Catchments System (RSCS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Rapid Service Catchments System checks the authority's share of households within its distance standard.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:8c8c4a0a0d20c21557fd91ef", "occurrence_id": "209be50a825366b96b552dc1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The operations figures in this chapter are drawn from the authority's own control log, and the Integrated Operations Monitoring System (IOMS), Edition 6, the register kept for Cairo in 2016, is quoted only to place those figures in context.\n\nThe register puts the share of units logging their incidents the same day at 62 per cent, the mean hours an incident stays open at 19, and the share of units whose log the system could audit at 48 per cent. All three are given as the register issues them, with its own unit list stated, and no chapter figure was built from them.\n\nTheir use here is to show how logging looks across the governorate beside the nine units whose log the chapter works from. The chapter's argument rests on that log, which names each incident rather than sampling units at a visit, and the two are kept apart rather than merged.\n\nThe annex lists the register tables consulted and the day of the pull, which is the full extent of the register's use in this chapter.\n\nMention: \"Integrated Operations Monitoring System (IOMS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Integrated Operations Monitoring System is quoted as context beside the authority's incident control log.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:7a7ae5d656bf53a30e7a2c95", "occurrence_id": "e4e48ebca07dcb15783b8a94", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The delivery figure in this chapter is the programme's own, taken from its site returns, and the Verified Programme Monitoring Registry (VPMR), Release 8, the register kept for Moscow in 2020, is cited to test it.\n\nThe programme's returns show 92 per cent of planned deliveries recorded complete. The registry shows 85 per cent for the same sites, because it counts a delivery complete only where its verifier signed the site sheet, while the programme's returns accept the contractor's own note. Neither rule was applied to the other's returns.\n\nThe 7-point distance is smaller than the movement the returns show between two releases, so the programme's figure is printed as returned and the registry supplies no revision to it and no sites its verifier did not reach.\n\nExcluded are sites the registry dates to a quarter rather than a visit, 96 of them, since the release window cannot be fixed for them. Which deliveries each source counts is stored in the annex's method note.\n\nMention: \"Verified Programme Monitoring Registry (VPMR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Verified Programme Monitoring Registry checks the programme's share of planned deliveries recorded complete.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:378ef9487605b109e1a86b13", "occurrence_id": "7aeed1e598f792bba2a5d0eb", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The household figure in this chapter is the office's own, taken from its claimant register, and the Household Administrative Survey Registry, the register WB/Makerere/FASID maintains for the United Kingdom in 2015, is cited to test it.\n\nThe office's register shows 31 per cent of its claimants in a household with no earned income. The registry shows 37 per cent for the same areas, because it takes the household's income from the survey interview, while the office's register carries only income reported to it on the claim form. No basis was used on the other's register.\n\nThe 6-point distance is smaller than the movement the register shows between two quarters, so the office's figure is printed as claimed and the registry supplies no revision to it and no claimants whose household it did not interview.\n\nExcluded are claimants the registry matches to two households, 180 of them, since the two records cannot be joined. Which claimants each source counts is listed in the annex's method sheet.\n\nMention: \"Household Administrative Survey Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Household Administrative Survey Registry checks the office's share of claimants without earned income.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:e7567ca6f8b6da5844b22aa4", "occurrence_id": "3bda636a6aaea17fdc34e307", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Record series were catalogued into the review's inventory from the Biennial Public Records Index (BPRI), the indicator INCAJU compiled for 2017, and the work done on them was cataloguing rather than measurement.\n\nEach series was given the inventory's own retention band and its review dates checked against the index's reference window; where the index carries a series the inventory has no band for, it was parked for the analyst rather than entered under the nearest one. Series the index lists without a holding office were left unbanded rather than assigned to the nearest archive, and 390 of them entered the inventory in that condition. The index's own release stamp was kept on every series so a later release can be matched without re-deriving it.\n\nThe inventory's volume counts come from the archive's own accession book, from which the index supplied nothing; it is the series coding that lets those counts be read against retention for the appraisal table the report draws up.\n\nThe coding sheet, the count of parked series and the day of deposit are set down in the annex log.\n\nMention: \"Biennial Public Records Index (BPRI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The Biennial Public Records Index supplies the coded record series behind the review's inventory.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:791262dc4c6f43f0c04ec67f", "occurrence_id": "709651f3cf23743e20fa66f8", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Statistical bulletin the numbers held for the city come from the Triennial Statistical Bulletin Index (TSBI), First Release, the indicator kept for Moscow in 2015, and the four values reach this report unchanged.\n\nThe index puts the share of bureaux issuing a bulletin on a stated calendar at 57 per cent, the mean days between a reference period closing and publication at 68, the share of bulletins carrying a method note inside them at 39 per cent, and the share of bureaux revising a bulletin once at 14 per cent. None of them was recalculated here, and the index's own reading of a bulletin, which requires its tables to stand on a public page, is given next to them.\n\nA bulletin circulated to departments only was not counted, and a correction issued as a note on the same page was counted as a revision. Each figure keeps the bureau it was drawn from with it, rather than pooled across the city.\n\nBeneath the table sit the index row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Triennial Statistical Bulletin Index (TSBI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Triennial Statistical Bulletin Index supplies calendar issue, publication delay, method notes and revisions.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:ee2c13edcd0fab52a0527a3d", "occurrence_id": "831e538ff359538b5e58f4d0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The monitoring figure in this chapter is the bureau's own, taken from its site sheets, and the Urban Programme Monitoring Index, the indicator APRA compiled for Addis Ababa in 2019, is cited to test it.\n\nThe bureau's sheets show 88 per cent of its planned outputs recorded on time. The index shows 80 per cent for the same woredas, because it counts an output on time only where the index's own field team saw it finished, while the bureau's sheets accept the contractor's completion note. Neither rule was applied to the other's sheets.\n\nThe 8-point distance is smaller than the movement the sheets show between two rounds, so the bureau's figure is printed as sheeted and the index supplies no revision to it and no outputs for the woredas its team did not enter.\n\nExcluded are outputs the index records as started but never dated, 130 of them, since neither count can place them in the round. Which outputs each source counts is written up in the annex notes.\n\nMention: \"Urban Programme Monitoring Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Urban Programme Monitoring Index checks the bureau's share of planned outputs recorded on time.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:0797f49f6d51d131d0da46b8", "occurrence_id": "d35a3fce09663b1cf8098f57", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Municipal operations country figures are assembled from the Municipal Operations Index Registry (MOIR), the register INVALSI maintains for Italy, and the four values reach this report unchanged.\n\nThe register puts the share of municipalities publishing their service standards at 49 per cent, the mean days a routine permit takes to be decided at 34, the share of municipalities whose complaints are logged in a system the register can read at 62 per cent, and the share whose decisions were reviewed by an outside office at 18 per cent. None of them was recalculated here, and the register's own reading of a decided permit, which requires the decision to have reached the applicant, is bound to the figures.\n\nA permit refused after the stated limit was counted as late rather than as undecided, and a review made by a sister municipality was counted as outside. Each figure keeps the province it was drawn from attached, instead of being blended across the country.\n\nBeneath the table sit the register row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Municipal Operations Index Registry (MOIR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Municipal Operations Index Registry supplies published standards, permit decision days, readable complaint logs and outside reviews.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:c3f2fdaa6a461aa118450a3c", "occurrence_id": "5d7a456fc5cd1ce68c10bb0c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Labour force figures for the metropolitan area are reported from the Metropolitan Labour Force Registry (MLFR), Second Version, the register kept for Saint Lucia in 2022, and the four values reach this report unchanged.\n\nThe register puts the share of working-age adults in paid work at 63 per cent, the share working in agriculture at 8 per cent, the mean hours a week an agricultural worker reports at 44, and the share of agricultural workers holding a second job at 21 per cent. None of them was recalculated here, and the register's own reading of paid work, which requires an hour worked for pay in the reference week, goes with the figures.\n\nA person working on the family holding for pay was counted as paid work, and a season of unpaid help was not. Each figure keeps the district it was drawn from shown with it, instead of combined into one area figure.\n\nBeneath the table sit the register row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Metropolitan Labour Force Registry (MLFR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Metropolitan Labour Force Registry supplies paid-work incidence, agricultural share, hours and second jobs.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:e0bb0c7df71bc940f3370a0d", "occurrence_id": "9da8b58b05759b37dc253016", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Three operations lines in the annex come from the Local Regional Operations Index, the indicator Bureau Van Dijk compiled for Beijing in 2016, and all three are copied across unchanged.\n\nThe index puts the share of units filing their monthly return complete at 79 per cent, the mean days a return takes to reach the centre at 5, and the share of units whose return was queried before acceptance at 23 per cent. Nothing was recomputed for this report, and the index's own reading of a complete return, which requires every table to arrive together, travels with the figures.\n\nThe annex's other lines are compiled from the office's own control sheet, so the index carries three lines while the sheet accounts for the remaining lines. A figure reported for one district keeps that district attached rather than being averaged across the municipality, and where index and sheet disagree about a filing the index's figure is kept for these three lines and the sheet's for the rest, the difference being noted beneath the table rather than smoothed away.\n\nUnits that filed no return at all are excluded from all three lines, since the index lists them separately. The annex log carries the index row behind each value, the extraction day, and the denominators behind each share.\n\nMention: \"Local Regional Operations Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Local Regional Operations Index supplies complete filings, return delay and queried returns.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:3d17d2a6247001ffd32f0242", "occurrence_id": "c3795fa26e95e7681e9464b4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Household shock records were catalogued into the study's working file from the Urban Shock Coping Survey (USCS), the survey Syrian Customs follows for 2018, and the work done on them was cataloguing rather than measurement.\n\nEach household was given the file's own coping band and its shock dates checked against the survey's reference window; where the survey carries a shock the file has no band for, it was parked for the analyst rather than entered under the nearest one. Households the survey lists without a response to the coping item were left unbanded rather than assumed to have sold nothing, and 720 of them entered the file in that condition. The survey's own interview date was kept on every household so a later round can be matched without re-deriving it.\n\nThe study's stratum weights come from the census frame, from which the survey supplied nothing; it is the household coding that lets those weights be read against work loss for the coping table the report draws up.\n\nThe coding sheet, the count of parked households and the day of deposit are held in the annex log.\n\nMention: \"Urban Shock Coping Survey (USCS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The Urban Shock Coping Survey supplies the coded household shock records behind the study's working file.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:4931eb7cd1c862d501e5e291", "occurrence_id": "2f5b97031e72ee2898ee4eca", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The graduation figure in this chapter is the scheme's own, taken from its case files, and the Household Graduation Programmes Survey (HGPS), Round 2, the register Lagos State Government maintains, is cited to test it.\n\nThe scheme's files show 71 per cent of its leavers holding an enterprise twelve months after leaving. The survey shows 62 per cent for the same areas, because it visits the household and asks whether the enterprise still trades, while the scheme's files carry the record the worker left at exit. The bases stay separate: neither is applied to the other's files.\n\nThe 9-point distance is smaller than the movement the files show between two rounds, so the scheme's figure is printed as filed and the survey supplies no revision to it and no households its teams could not locate.\n\nExcluded are households the survey finds enrolled twice under different names, 240 of them, since the two records cannot be settled. Which households each source counts is detailed in the annex file.\n\nMention: \"Household Graduation Programmes Survey (HGPS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Household Graduation Programmes Survey checks the scheme's share of leavers holding an enterprise after a year.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:1f04b5ede303aa519fdd4943", "occurrence_id": "51e79453b3f423f8e8afa362", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Three apprenticeship lines in the labour annex come from the Rapid Apprenticeship Report (RAR), the report issued for Morocco in 2017, and all three are copied across unchanged.\n\nThe report puts the share of establishments that took an apprentice in the year at 16 per cent, the mean months a spell runs before it closes at 10, and the share of apprentices placed in the trade they trained for at 43 per cent. Nothing was recomputed for this report, and the report's own reading of a spell, which requires a signed workplace agreement, travels with the figures.\n\nThe annex's other lines are compiled from the ministry's own course register, so the report carries three lines while the register carries the rest. A figure reported for one region keeps that region attached rather than being averaged across the country, and where report and register disagree about a spell the report's figure is kept for these three lines and the register's for the rest, the difference being noted beneath the table rather than smoothed away.\n\nSpells closed by an employer's failure are excluded from all three lines, since the report dates them to the closure rather than the trade. The annex log carries the report page behind each value, the extraction date and the base each share is taken on.\n\nMention: \"Rapid Apprenticeship Report (RAR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Rapid Apprenticeship Report supplies establishments taking apprentices, spell length and placement in the trained trade.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:07599687fbd9ebb5ae94fff3", "occurrence_id": "0c81d51137e41e5445612db5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Displacement and schooling were put together for this review from the Verified Protracted Situation Survey, which the International Road Federation conducted in France in 2024, and the readings below are the review's own.\n\nHouseholds were grouped by the years they had been displaced at the interview and the share of their children enrolled at the survey's visit computed: 82 per cent of households displaced under two years against 57 per cent of those displaced five years or more. The survey carries a displacement length for each household and an enrolment item for each child without ever joining the two.\n\nMaking the groups alike on the distance from the household's lodging to the nearest school leaves 11 points of that distance, which leaves the school run as part of the gap the opening comparison puts down to the years displaced.\n\nHouseholds the survey finds moved twice during the school year are dropped rather than counted as enrolled, and they are 5 per cent of households. The annex note carries the grouping rule, the pull date, with the base of each share alongside.\n\nMention: \"Verified Protracted Situation Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Verified Protracted Situation Survey relates years displaced to children enrolled at school.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:9411400f442cabd05b219faa", "occurrence_id": "81b4aec3b429246d9f9bd9a0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Cadastral coverage and dispute were derived for this chapter from the Combined Cadastral Frame Census, the census ECLAC conducted in Punjab in 2021, and the figures below are counted by the chapter rather than by the census's.\n\nTehsils were grouped by the share of parcels the census holds a digitised boundary for and the median months a land dispute stands open computed: 7 months at tehsils above nine-tenths digitised against 26 months at those below half. The census carries a boundary for each parcel and the tribunal records an opening and closing date for each dispute, and neither reports the two together.\n\nThat comparison supports one claim, that digitisation should proceed tehsil by tehsil rather than parcel by parcel. The staffing lines the proposal quotes come from the tribunal's own establishment return, from which the census supplied nothing.\n\nTehsils the census maps without a parcel count are dropped rather than counted at the division average, and they are 6 per cent of tehsils. The bands and the counts behind each median are recorded beneath the table.\n\nMention: \"Combined Cadastral Frame Census\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Combined Cadastral Frame Census supports a tehsil-first digitisation claim built from digitised parcels and dispute duration.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:41617277da376421f913a0ce", "occurrence_id": "1ad3413b8c418048901b98b8", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Poverty and phone access were reconstructed for this review from the Record-Based Socioeconomic Survey (RSS), the survey Research ICT Africa follows for 2015, and the readings here are the review's, not the survey's.\n\nHouseholds were grouped by whether the survey records a mobile phone held by any member and the share below the national adequacy line computed: 24 per cent of households holding a phone against 51 per cent of those holding none. The survey carries a phone item for each household and a consumption aggregate for the same household without ever joining the two.\n\nHolding the education of the household head even leaves 13 points of that distance, so schooling accounts for part of what the opening comparison puts down to the phone.\n\nHouseholds whose only phone the survey records as broken at the visit are dropped rather than counted as holding one, and they are 4 per cent of households. The annex note carries the grouping rule, the pull date and the base of each share.\n\nMention: \"Record-Based Socioeconomic Survey (RSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Record-Based Socioeconomic Survey relates mobile phone holding to households below the adequacy line.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:e066a6088f3c93ae91d711d3", "occurrence_id": "c9b123ca8a4be3a892dab283", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Occupational safety figures spanning the country are taken from the Representative Occupational Safety Survey, the survey the Fraser Institute conducted in Mexico in 2022, and the four values reach this report unchanged.\n\nThe survey puts the share of workers reporting a safety incident at the workplace in the year at 9 per cent, the share of workplaces with a posted safety notice at 61 per cent, the mean days a reported injury kept a worker away at 11, and the share of injured workers whose household fell below the national adequacy line during that absence at 28 per cent. None of them was recalculated here, and the survey's own reading of an incident, which requires the worker to name the date and the task, travels with the values.\n\nAn injury on the way to work was not counted, and an absence shorter than a full day was counted as no absence. Each figure keeps the state it was drawn from reported with it, rather than blended into a single country share.\n\nBeneath the table sit the survey row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Representative Occupational Safety Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Representative Occupational Safety Survey supplies incidents, posted notices, injury absence and adequacy incidence during absence.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:f470a065d5467678112a8c97", "occurrence_id": "f4624997b219df7d042ecd14", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Monitoring and delivery were put together for this review from the Independent Programme Monitoring Survey, which APIS conducted in Malaysia in 2015, and the readings below are the review's own.\n\nProjects were grouped by the number of monitoring visits the survey records in the year and the share delivering within their planned window computed: 72 per cent of projects with four visits or more against 44 per cent of those with one or none. The survey carries a visit count for each project and a delivery flag for the same project without ever joining the two.\n\nHolding the size of the project's budget even leaves 12 points of that distance, which leaves budget size as part of the gap the opening comparison puts down to the visits.\n\nProjects the survey finds terminated before the year closed are dropped rather than counted as late, and they are 7 per cent of projects. The annex note carries the grouping rule, the pull date and, alongside it, the base of each share.\n\nMention: \"Independent Programme Monitoring Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Independent Programme Monitoring Survey relates monitoring visits to delivery within the planned window.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:7b7898409ad6c7d8e4762138", "occurrence_id": "d983d580eafbdc5736c26a95", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Population grid estimates and service distance were reconstructed for this review from the Monthly Population Grid Dataset, the geospatial dataset FEPB compiled for the Russian Federation in 2017, and the readings here are the review's, not the dataset's.\n\nCells were matched to their estimate and grouped by the density the dataset places them in, and the share of households the service records as reached within its own standard computed: 74 per cent of cells above a thousand people, which the dataset prints as a density per cell, against 38 per cent of those below a hundred. The dataset holds an estimate per cell and the service records a reach figure per household, and neither reports the two together.\n\nHolding the road class nearest the cell even leaves 9 points of that distance, so the road accounts for part of what the opening comparison puts down to the density.\n\nCells the dataset leaves without an estimate for want of a source were dropped rather than counted as empty, and they are 8 per cent of cells. The annex note carries the grouping rule, the pull date as well as the base of each share.\n\nMention: \"Monthly Population Grid Dataset\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Monthly Population Grid Dataset relates grid cell density to households reached within the service standard.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:b2c6144c4a794987bc85271d", "occurrence_id": "e835db099acadbec4ecd50ea", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Labour force figures at the country level are taken from the Pilot Labour Force Survey, the survey WIDER conducted in Afghanistan in 2020, and the four values reach this report unchanged.\n\nThe survey puts the share of working-age men in any work at 68 per cent, the share of working-age women in any work at 14 per cent, the mean hours a week a worker reports at 46, and the share of workers paid in kind rather than in cash at 23 per cent. None of them was recalculated here, and the survey's own reading of work, which requires an hour worked for pay or profit in the reference week, follows the figures.\n\nWork on the family holding without pay was counted as work, and seasonal migration shorter than a week was counted as work at the destination. Each figure keeps the province it was drawn from kept with it, instead of being rolled up to the country.\n\nBeneath the table sit the survey row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Pilot Labour Force Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Pilot Labour Force Survey supplies work incidence by sex, weekly hours and payment in kind.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:768fd74781ef8ac6dddff506", "occurrence_id": "464afb905b613831d4ef6a1e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Three operations lines in the response annex come from the Local Operations Monitoring Registry, the register UNDYB maintains for Mali in 2017, and all three are copied across unchanged.\n\nThe registry puts the share of partner sites filing a monthly return at 76 per cent, the mean days a filed return takes to reach the coordination cell at 6, and the share of returns carrying a coverage figure at 58 per cent. Nothing was recomputed for this report, and the registry's own reading of a filed return, which requires the cell to have logged it, travels with the figures.\n\nThe annex's other lines are compiled from the region's own activity sheet, so the registry carries three lines while the sheet holds the other lines. A figure reported for one circle keeps that circle attached rather than being averaged across the country, and where registry and sheet disagree about a filing the registry's figure is kept for these three lines and the sheet's for the rest, the difference being noted beneath the table rather than smoothed away.\n\nSites opened after the registry's window closed are excluded from all three lines, since the registry holds no period for them. The annex log carries the registry row behind each value, the day the table was pulled and the counts the shares rest on.\n\nMention: \"Local Operations Monitoring Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Local Operations Monitoring Registry supplies site filing rates, return delay and returns carrying coverage figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:b2d5b91f66505fc0c8707fa2", "occurrence_id": "d16881fc0381ea885d827332", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Three monitoring lines in the rural annex come from the Rural Public Sector Monitoring System, the register IRS maintains for Lebanon in 2022, and all three are copied across unchanged.\n\nThe register puts the share of cazas publishing a service standard at 41 per cent, the mean months between a standard's publication and its review at 20, and the share of cazas whose performance figure the register could verify at 55 per cent. Nothing was recomputed for this report, and the register's own reading of a verified figure, which requires the register to hold the underlying count, travels with the figures.\n\nThe annex's other lines are compiled from the ministry's own returns, so the register carries three lines while the returns carry the balance. A figure reported for one caza keeps that caza attached rather than being averaged across the country, and where register and returns disagree about a figure the register's is kept for these three lines and the returns' for the rest, the difference being noted beneath the table rather than smoothed away.\n\nCazas the register lists without a published standard are excluded from all three lines, since the register scores them separately. The annex log carries the register row behind each value, the day of the pull together with the base of each share.\n\nMention: \"Rural Public Sector Monitoring System\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Rural Public Sector Monitoring System supplies published standards, review intervals and verified performance figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:76d1d17e7fd07d69f5659c38", "occurrence_id": "5067a4e434bc0550311a9c96", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Integrated records country-level figures come from the Comparative Integrated Records Registry (CIRR), Third Edition, the register kept for Honduras in 2017, and the four values reach this report unchanged.\n\nThe register puts the share of offices holding their series on a list the register accepts at 59 per cent, the mean years a series waits before appraisal at 8, the share of series held in a format the reading room can open at 47 per cent, and the share of offices whose register was reviewed within the edition at 33 per cent. None of them was recalculated here, and the register's own reading of an accepted list, which requires the signature of the office head, is reported with them.\n\nA list signed by a predecessor was still counted, and a series held only on a personal machine was counted as unlisted. Each figure keeps the department it was drawn from named with it, instead of averaged over the country.\n\nBeneath the table sit the register row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Comparative Integrated Records Registry (CIRR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Comparative Integrated Records Registry supplies accepted series lists, appraisal waits, readable formats and reviewed registers.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:4cb77fd3fa46d395f3625503", "occurrence_id": "a152ce53e15c39c5fa16b16a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Inequality figures for each part of the country are drawn from the High-Resolution Inequality Survey, the survey Bureau van Dijk conducted in Panama in 2021, and the four values reach this report unchanged.\n\nThe survey puts the share of income the top tenth holds at 36 per cent, the share the bottom half holds at 12 per cent, the mean monthly income of a household in the lowest fifth at 420, and the share of households below the national adequacy line at 19 per cent. None of them was recalculated here, and the survey's own reading of income, which is measured before taxes and after transfers, is attached to them.\n\nIncome from rent of a second dwelling was counted, and a transfer received only once in the year was counted at its annualised value. Each figure keeps the province it was drawn from listed with it, rather than smoothed across the country.\n\nBeneath the table sit the survey row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"High-Resolution Inequality Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The High-Resolution Inequality Survey supplies top-tenth and bottom-half income shares, lowest-fifth income and adequacy incidence.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:b569609f70cd11a8834d2ca8", "occurrence_id": "f29c56c97fa264c1823ccff9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Apprenticeship completion and firm size were reconstructed for this review from the Quality-Assured Apprenticeship Registry (QAR), Second Edition, the register Landsat compiled for 2017, and the readings here are the review's, not the register's.\n\nEstablishments were grouped by the number of workers the register records for them and the share of their apprentices completing the spell computed: 74 per cent of establishments above fifty workers against 48 per cent of those below ten. The register carries a worker count for each establishment and a completion flag for each apprentice without ever joining the two.\n\nHolding the trade the establishment works in even leaves 9 points of that distance, so the trade accounts for part of what the opening comparison puts down to size.\n\nEstablishments the register lists with apprentices but no worker count are dropped rather than counted as small, and they are 5 per cent of establishments. The annex note carries the grouping rule, the pull date plus the base of each share.\n\nMention: \"Quality-Assured Apprenticeship Registry (QAR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Quality-Assured Apprenticeship Registry relates establishment size to apprentices completing their spell.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:0b2d1886a5465be7d9683d13", "occurrence_id": "5b5e4d148d43a292b7eca080", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Facility readiness was derived for this chapter from the Municipal Service Delivery and Readiness Survey (MSDRS), the survey conducted in Lusaka in 2015, and the figures below are the chapter's own figures and not the survey's.\n\nFacilities were grouped by the number of clinical staff the survey records on the day of visit and the share of poor-ward patients the facility reports turning away for lack of capacity computed: 26 per cent of facilities rostering two staff or fewer against 7 per cent of those rostering six or more. The survey carries a roster for each facility and a referral figure for the same facility without ever joining the two.\n\nThat comparison supports one claim, that the staffing norm should be set by ward poverty rather than by facility size. The budget lines the proposal quotes come from the council's own vote, from which the survey supplied nothing.\n\nFacilities the survey visits during a closure are dropped rather than counted at the staff they roster, and they are 6 per cent of facilities. The bands and the counts behind each share are recorded beneath the table.\n\nMention: \"Municipal Service Delivery and Readiness Survey (MSDRS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Municipal Service Delivery and Readiness Survey supports a staffing-norm claim built from rostered staff and patients turned away.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:7bb0c69167b4293c827e4eed", "occurrence_id": "2825b34f6758489940c71bd1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The training figures in this chapter are drawn from the board's own institute returns, and the Panel Vocational Training Survey (PVTS), the survey conducted in Delhi in 2022, is referred to only as background to them.\n\nThe survey puts the share of leavers placed in a related job within six months at 55 per cent, the mean months a course runs at 8, and the share of courses carrying an employer on their governing committee at 37 per cent. All three are given as the survey issues them, with its own institute list stated, and no chapter figure rests on them.\n\nThe chapter's tables draw on none of its responses. The figures the chapter argues from come from the board's returns, which cover each of its own institutes rather than a sample across the city, and where the two accounts of placement differ the difference is noted rather than reconciled.\n\nThe annex lists the survey tables consulted and the day the rows were taken, and the survey is used for nothing beyond that background.\n\nMention: \"Panel Vocational Training Survey (PVTS)\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"inform\", \"summary\": \"The Panel Vocational Training Survey is background to the board's own institute training returns.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:942807e6111ba4ed671fd721", "occurrence_id": "e915ea7e7e09b41a22efe165", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Enterprise births and hiring were put together for this review from the Combined Enterprise and Jobs System (CEJS), Fifth Release, the register DAC maintains for Ile de France, and the readings below are the review's own.\n\nArrondissements were grouped by the number of new enterprises the register records per thousand residents and the share of jobs added in the year computed: 5.2 per cent of arrondissements above four per thousand against 1.8 per cent of those below one. The register carries a birth count for each arrondissement and an employment series for the same arrondissement without ever joining the two.\n\nMaking the groups alike on the size of the arrondissement's existing workforce leaves 1.4 points of that distance, which leaves the workforce as part of the gap the opening comparison puts down to enterprise births.\n\nArrondissements the register merges during the release are dropped rather than counted at their predecessor's births, and they are 4 per cent of arrondissements. The annex note carries the grouping rule, the pull date, and the base of each figure.\n\nMention: \"Combined Enterprise and Jobs System (CEJS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Combined Enterprise and Jobs System relates enterprise births per thousand residents to jobs added in the year.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:203ed3a96a1bedf1c9fbdde3", "occurrence_id": "f6a46c6fbd7f60fd67dc2004", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Consumption and distance were derived for this chapter from the Annual Household Income and Expenditure Survey, which Social Watch conducted in Lusaka in 2015, and the figures below come from the chapter and not from the survey's.\n\nHouseholds were grouped by the walk to the nearest water point and the share below the city's adequacy line computed: 47 per cent of households above thirty minutes against 25 per cent of those under ten. The survey carries a walk for each household and a consumption aggregate for the same household without ever joining the two.\n\nThat comparison supports one claim, that the tariff subsidy should follow the water point rather than the ward boundary. The capital lines the proposal quotes come from the council's own works programme, from which the survey supplied nothing.\n\nHouseholds the survey finds with a point under repair at the visit are dropped rather than counted at their usual walk, and they are 5 per cent of households. The bands and the counts behind each share are recorded beneath the table.\n\nMention: \"Annual Household Income and Expenditure Survey\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Annual Household Income and Expenditure Survey supports a tariff-subsidy claim built from water-point walk and adequacy incidence.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:0c0752f42f6524af36ef44f8", "occurrence_id": "d248b853fbaf585f727a8f5a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Terrain and clinic access were reconstructed for this review from the Quality-Assured Terrain and Elevation Dataset (QTED), Sixth Wave, the geospatial dataset UNIDO holds for 2017, and the readings here are the review's, not the dataset's.\n\nSettlements were matched to their slope and grouped by the rise per kilometre the dataset records between the settlement and the nearest clinic, and the share of births the health service records at a facility computed: 81 per cent of settlements under thirty metres per kilometre against 52 per cent of those above ninety. The dataset grades a slope per settlement and the service records a place of birth per delivery, and neither reports the two together.\n\nHolding the distance from the settlement to the clinic even leaves 8 points of that distance, so the slope accounts for part of what the opening comparison puts down to the distance.\n\nSettlements the dataset leaves ungraded for want of a clear surface were dropped rather than counted as flat, and they are 6 per cent of settlements. The annex note carries the grouping rule, the pull date, with the base of each share alongside.\n\nMention: \"Quality-Assured Terrain and Elevation Dataset (QTED)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Quality-Assured Terrain and Elevation Dataset relates slope to the clinic to facility births.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:7331fa2f9a2cc6adb182090e", "occurrence_id": "b226da7f7675e4d5b7050db7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The deprivation figure in this chapter is the office's own, taken from its household file, and the Cross-Sectoral Deprivation and Hardship Survey (CDHS), Version 2, the survey conducted in Ile de France in 2024, is cited to test it.\n\nThe office's file shows 17 per cent of its households carrying three deprivations or more. The survey shows 23 per cent for the same departments, because it scores a dwelling on what the surveyor observes, while the office's file carries the score entered when the household registered. Each basis keeps its own file untouched.\n\nThe 6-point distance is smaller than the movement the file shows between two versions, so the office's figure is printed as filed and the survey supplies no revision to it and no households for the communes its teams did not reach.\n\nExcluded are households the survey scores on part of its indicator list, 420 of them, since the two scores are not on the same footing. Which households each source counts is laid out in the papers behind the annex.\n\nMention: \"Cross-Sectoral Deprivation and Hardship Survey (CDHS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Cross-Sectoral Deprivation and Hardship Survey checks the office's share of households with three deprivations or more.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:ea758fe910ce35e52aece075", "occurrence_id": "44a8df96c4d45127d90289a3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The wage figures in this chapter are compiled from the bureau's own payroll returns, and the Pilot Wage Structure Survey (PWSS), the survey conducted in Buenos Aires in 2019, is quoted only to place those figures in context.\n\nThe survey puts the median monthly wage in the city's registered jobs at 38 000, the share of workers earning below a set minimum at 12 per cent, and the mean gap between the highest and lowest paid in the same occupation at 2.4 times. All three are given as the survey issues them, with its own definition of a registered job stated, and no chapter figure was built from them.\n\nTheir use here is to show how the registered labour market stands beside the eleven occupations whose payroll the chapter works from. The chapter's argument rests on those payrolls, which name each worker rather than sampling establishments, and the two are kept apart rather than merged.\n\nThe annex lists the survey tables consulted and the day the extract ran, which is the full extent of the survey's use in this chapter.\n\nMention: \"Pilot Wage Structure Survey (PWSS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Pilot Wage Structure Survey is quoted as context beside the bureau's payroll wage returns.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:d05b20fdcbbf9f66b50e160f", "occurrence_id": "923ff897a5ad41b05e6aed63", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Debt and savings figures measured across the panel are read from the Monthly Debt and Savings Microdata (MDSM), Fourth Round, the microdata APRA follows for 2024, and the four values reach this report unchanged.\n\nThe microdata puts the share of households carrying a repayment to a moneylender at 27 per cent, the mean months a household's debt has stood at 22, the share with any saving held outside the home at 38 per cent, and the share of households whose repayment took them below the panel's adequacy line at 11 per cent. None of them was recalculated here, and the microdata's own reading of a saving, which requires the balance to be named at the round, is stated with them.\n\nA debt owed to a relative was counted, and a saving held as livestock was not. Each figure keeps the stratum it was drawn from carried with it, rather than mixed across the panel.\n\nBeneath the table sit the microdata row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Monthly Debt and Savings Microdata (MDSM)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Monthly Debt and Savings Microdata supplies moneylender repayments, debt age, outside savings and repayments pushing households below adequacy.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:0dbd1919b59a5fb308993cf8", "occurrence_id": "69738748f8420105f9bda4c3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Monitoring practice was derived for this chapter from the Rapid Public Sector Monitoring Index (RPSMI), the indicator Census compiled for 2020, and the figures below come from the chapter, not from the index's.\n\nDepartments were grouped by the months the index records since their last monitoring return and the share of their returns carrying a figure the index could check computed: 66 per cent of departments returning within the year against 29 per cent of those returning later. The index carries a return date for each department and a checkable flag for the same return without ever joining the two.\n\nThat comparison supports one claim, that a return should be filed quarterly rather than at the year's end. The staffing lines the proposal quotes come from the ministry's own establishment return, from which the index supplied nothing.\n\nDepartments the index lists as merged during the reference year are dropped rather than counted at their predecessor's return, and they are 5 per cent of departments. The bands and the counts behind each share are recorded beneath the table.\n\nMention: \"Rapid Public Sector Monitoring Index (RPSMI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Rapid Public Sector Monitoring Index supports a quarterly-filing claim built from return age and checkable figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:91dda0fed7840d7ff2475aac", "occurrence_id": "55ebf9ddd5020bb49094283c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The transfer figure in this chapter is the office's own, taken from its payment ledger, and the Triennial Social Protection Survey (TSPS), Version 6, the survey conducted in Formosa in 2018, is cited to test it.\n\nThe office's ledger shows 84 per cent of eligible households receiving a transfer in the period. The survey shows 76 per cent for the same departments, because it asks the household whether a payment reached it, while the ledger records the payment the office issued. Neither basis was carried across to the other's ledger.\n\nThe 8-point distance is smaller than the movement the ledger shows between two versions, so the office's figure is printed as ledged and the survey supplies no revision to it and no households whose interviewer never completed the visit.\n\nExcluded are households the survey finds registered under two offices, 350 of them, since the ledger carries one entry and the survey two. Which households each source counts is set out in the working papers behind the annex.\n\nMention: \"Triennial Social Protection Survey (TSPS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Triennial Social Protection Survey checks the office's share of eligible households paid in the period.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:2c8ec5b13826231c85721182", "occurrence_id": "1bf9daa94303d79f1045505f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Debt and adequacy were put together for this review from the Harmonised Debt and Savings Survey (HDSS), Wave 2, the survey IDMC follows, and the readings below are the review's own.\n\nHouseholds were grouped by the share of their income the survey records going to repayments and the share below the survey's adequacy line computed: 44 per cent of households above a fifth of income going to repayments against 16 per cent of those under a tenth. The survey carries a repayment share for each household and an adequacy measure for the same household without ever joining the two.\n\nMaking the groups alike on the household's income level leaves 9 points of that distance, which leaves income as part of the gap the opening comparison puts down to the repayments.\n\nHouseholds the survey records with no income at all are dropped rather than counted as repaying nothing, and they are 6 per cent of households. The annex note carries the grouping rule, the pull date and the base of each share.\n\nMention: \"Harmonised Debt and Savings Survey (HDSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Harmonised Debt and Savings Survey relates repayment share of income to households below the adequacy line.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:5ecf47a6467cbf859106ed2c", "occurrence_id": "0f8e0dada4b065e404a9d083", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Seed use figures for the frame are reported from the Endline Seed Use Survey (ESUS), Fifth Edition, the survey Chile’s Internal Revenue Service follows, and the four values reach this report unchanged.\n\nThe survey puts the share of holdings sowing purchased seed at 63 per cent, the mean kilograms of seed an acre sown with purchased seed takes at 14, the share of holdings keeping seed from their own harvest for the next season at 41 per cent, and the share of the frame whose sown area the survey could match to a parcel boundary at 72 per cent. None of them was recalculated here, and the survey's own reading of purchased seed, which requires the household to name where it was bought, comes with them.\n\nSeed given by a programme was counted as purchased, and seed exchanged between neighbours was not. Each figure keeps the ward it was drawn from with it, rather than pooled across the frame.\n\nBeneath the table sit the survey row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Endline Seed Use Survey (ESUS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Endline Seed Use Survey supplies purchased-seed sowing, seed rate, own-seed saving and parcel-matched sown area.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:7f1f86522ede08ca87122291", "occurrence_id": "6739a0f9e99ce0ae01dd8f9c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Monitoring series were catalogued into the review's inventory from the Verified Public Sector Monitoring Index (VPSMI), Sixth Phase, the indicator Latinobarómetro Corporation maintains, and the work done on them was cataloguing rather than measurement.\n\nEach series was given the inventory's own retention band and its review dates checked against the index's reference window; where the index carries a series the inventory has no band for, it was parked for the analyst rather than entered under the nearest one. Series the index lists without a holding office were left unbanded rather than assigned to the nearest archive, and 260 of them entered the inventory in that condition. The index's own phase stamp was kept on every series so a later phase can be matched without re-deriving it.\n\nThe inventory's volume counts come from the archive's own accession book, from which the index supplied nothing; it is the series coding that lets those counts be read against retention for the appraisal table the report draws up.\n\nThe coding sheet, the count of parked series and the day of deposit are carried in the annex log.\n\nMention: \"Verified Public Sector Monitoring Index (VPSMI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The Verified Public Sector Monitoring Index supplies the coded monitoring series behind the review's inventory.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:e91427c26453df956e8c9fad", "occurrence_id": "06ff7553b719279b2e76cca9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The household figures in this chapter are drawn from the programme's own payment file, and the Household Programme Monitoring Registry, the register D¨ur et al. maintains for Gabon in 2015, is quoted only to place those figures in context.\n\nThe register puts the share of enrolled households receiving a payment in the period at 79 per cent, the mean months a household stays enrolled at 14, and the share of households the register could match to a local office record at 68 per cent. All three are given as the register issues them, with its own enrolment list stated, and no chapter figure was built from them.\n\nTheir use here is to show how enrolment looks across the province beside the four hundred households whose payment file the chapter works from. The chapter's argument rests on that file, which names each payment rather than sampling households at a visit, and the two are kept apart rather than merged.\n\nThe annex lists the register tables consulted and the pull date, which is the full extent of the register's use in this chapter.\n\nMention: \"Household Programme Monitoring Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Household Programme Monitoring Registry is quoted as context beside the programme's payment file.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:db157dcf77905e5a2e3ce238", "occurrence_id": "dbeabbcf13b51508c0afff53", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Camp management the settlement figures are compiled from the Midline Camp Management Report (MCMR), Third Phase, the report issued for Cyprus in 2015, and the four values reach this report unchanged.\n\nThe report puts the number of households the report counts in the settlement at 4 300, the share of households holding a plot assigned by the camp office at 88 per cent, the mean days a household waits for that assignment at 26, and the share of households whose registration the report could match to the arrivals list at 74 per cent. None of them was recalculated here, and the report's own reading of a household, which requires one person to be named as its head, is kept with them.\n\nA household sharing a plot with another was still counted once, and a household lodging outside the fence was not counted. Each figure keeps the block it was drawn from attached, instead of being blended across the settlement.\n\nBeneath the table sit the report page behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Midline Camp Management Report (MCMR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Midline Camp Management Report supplies household counts, plot assignments, waiting days and matched registrations.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:815b45dc7ecf0da7b330b173", "occurrence_id": "483f1dbe2311fdb07eed045c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Delivery and complaint were put together for this review from the Harmonised Service Delivery Registry, which UNU-WIDER maintains for Maputo in 2019, and the readings below are the review's own.\n\nWards were grouped by the share of its requests the registry records closed within the stated limit and the share of its households filing a complaint in the year computed: 9 per cent of wards above nine-tenths closed in time against 24 per cent of those below half. The registry carries a request with a closing date for each ward and a complaint count for the same ward without ever joining the two.\n\nMaking the groups alike on the number of requests a ward receives leaves 5 points of that distance, which leaves the request load as part of the gap the opening comparison puts down to the closing time.\n\nWards the registry lists with no requests at all are dropped rather than counted as fully closed, and they are 7 per cent of wards. The annex note carries the grouping rule, the pull date and, alongside it, the base of each share.\n\nMention: \"Harmonised Service Delivery Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Harmonised Service Delivery Registry relates requests closed in time to households filing a complaint.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:b110d12a3dc4e73398d87fcc", "occurrence_id": "c158a8b75584a4c3e6d49c0b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Three migration lines in the labour annex come from the Coordinated Migration and Remittance Registry (CMRR), Version 8, the register DAC maintains, and all three are copied across unchanged.\n\nThe registry puts the share of households with a member working outside the area at 22 per cent, the mean months a member stays away before returning at 9, and the share of those households receiving a remittance in the reference month at 61 per cent. Nothing was recomputed for this report, and the registry's own reading of a remittance, which requires the household to name the channel it came through, travels with the figures.\n\nThe annex's other lines are compiled from the office's own permit file, so the registry carries three lines while the file carries the lines the table does not. A figure reported for one district keeps that district attached rather than being averaged across the frame, and where registry and file disagree about a departure the registry's figure is kept for these three lines and the file's for the rest, the difference being noted beneath the table rather than smoothed away.\n\nHouseholds whose member left for schooling are excluded from all three lines, since the registry does not count a term as work. The annex log carries the registry row behind each value, the day of extraction as well as the counts each share divides.\n\nMention: \"Coordinated Migration and Remittance Registry (CMRR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Coordinated Migration and Remittance Registry supplies households with a migrant, months away and remittance receipt.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:cd90745d4c0a20817a0c2266", "occurrence_id": "b0675f63797253c713860d8e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Parcel records were catalogued into the study's working file from the Small-Area Parcel Footprint Dataset (SPFD), the geospatial dataset compiled for Lusaka in 2020, and the work done on them was cataloguing rather than measurement.\n\nEach parcel was given the file's own footprint band and its boundary checked against the dataset's reference window; where the dataset carries a parcel the file has no band for, it was parked for the analyst rather than entered under the nearest one. Parcels the dataset traces without an owner record were left unbanded rather than assumed unowned, and 1 400 of them entered the file in that condition. The dataset's own trace date was kept on every parcel so a later trace can be matched without re-deriving it.\n\nThe study's structure counts come from the council's own valuation roll, from which the dataset supplied nothing; it is the parcel coding that lets those counts be read against footprint for the built-up area table the report draws up.\n\nThe coding sheet, the count of parked parcels and the day of deposit are kept in the annex log.\n\nMention: \"Small-Area Parcel Footprint Dataset (SPFD)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The Small-Area Parcel Footprint Dataset supplies the coded parcel records behind the study's working file.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:0092becbdb956b583958dd40", "occurrence_id": "d29426f23658985286f3999c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Vegetation condition was derived for this chapter from the Diagnostic Vegetation Condition Index (DVCI), Edition 3, the indicator Latinobarómetro maintains, and the figures below are the chapter's own rather than the index's.\n\nWatersheds were grouped by the share of their area the index records as stressed in the dry season and the mean hectares of grazing the index reports lost between editions computed: 410 hectares at watersheds above a third stressed against 90 hectares at those below a tenth. The index carries a stress share for each watershed and a grazing figure for the same watershed without ever joining the two.\n\nThat comparison supports one claim, that stocking limits should be set watershed by watershed rather than by the provincial boundary. The extension lines the proposal quotes come from the service's own field return, from which the index supplied nothing.\n\nWatersheds the index maps without a grazing figure are dropped rather than counted at the provincial mean, and they are 6 per cent of watersheds. The bands and the counts behind each mean are recorded beneath the table.\n\nMention: \"Diagnostic Vegetation Condition Index (DVCI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Diagnostic Vegetation Condition Index supports a watershed-level stocking claim built from stressed area and grazing lost.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:79e5635610d92e180b331dcb", "occurrence_id": "ba315a90962530046633719c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Training completion and placement were reconstructed for this review from the Ministerial Vocational Training Microdata (MVTM), Fourth Edition, the microdata ENIA follows, and the readings here are the review's, not the microdata's.\n\nLeavers were grouped by the months of prior work the microdata records for them and the share placed in a related job within six months computed: 69 per cent of leavers with prior work against 43 per cent of those with none. The microdata carries a prior-work item for each leaver and a placement flag for the same leaver without ever joining the two.\n\nHolding the trade they trained in even leaves 8 points of that distance, so the trade accounts for part of what the opening comparison puts down to prior work.\n\nLeavers the microdata lists as enrolled elsewhere at the follow-up are dropped rather than counted as unplaced, and they are 5 per cent of leavers. The annex note carries the grouping rule, the pull date together with the base of each share.\n\nMention: \"Ministerial Vocational Training Microdata (MVTM)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Ministerial Vocational Training Microdata relates prior work to leavers placed in a related job.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:e86b0ca93759313a3579882e", "occurrence_id": "d486aee4be85aeb10ed7899b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Disability and adequacy were put together for this review from the Longitudinal Disability and Support Survey (LDSS), Second Edition, the survey Transparency International follows, and the readings below are the review's own.\n\nHouseholds were grouped by whether the survey records an adult with a disability and the share below the survey's adequacy line computed: 38 per cent of households with such an adult against 19 per cent of those without. The survey carries a disability item for each adult and an adequacy measure for the same household without ever joining the two.\n\nMaking the groups alike on the number of dependants in the household leaves 11 points of that distance, which leaves dependants as part of the gap the opening comparison puts down to the disability.\n\nHouseholds the survey records an adult newly acquiring a disability for are dropped rather than counted at their earlier response, and they are 4 per cent of households. The annex note carries the grouping rule, the pull date plus the base of each share.\n\nMention: \"Longitudinal Disability and Support Survey (LDSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Longitudinal Disability and Support Survey relates an adult with a disability to households below adequacy.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:2dbd303e07d9d3d3b9d04d4d", "occurrence_id": "d7a9a0053459a89eebd7440b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Immunisation figures for the whole frame are sourced from the Periodic Immunisation Coverage Survey (PICS), Phase 5, the survey Romania MoF follows, and the four values reach this report unchanged.\n\nThe survey puts the share of children completing the schedule by their first birthday at 82 per cent, the share receiving no dose at all at 5 per cent, the mean months between a dose due and a dose given at 3, and the share of children in the lowest wealth fifth completing the schedule at 68 per cent. None of them was recalculated here, and the survey's own reading of completion, which requires the card to be seen or the register entry read, accompanies the figures.\n\nA dose given outside the programme was counted, and a dose recorded on the card but not dated was not. Each figure keeps the district it was drawn from shown with it, instead of combined into one frame figure.\n\nBeneath the table sit the survey row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Periodic Immunisation Coverage Survey (PICS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Periodic Immunisation Coverage Survey supplies schedule completion, undosed children, dose delay and lowest-fifth completion.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:6555e29ee1f0f7456a2474a7", "occurrence_id": "643c801af63b7f93129531a3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Cooperative membership the state figures are tabulated by the Audited Cooperative Membership Survey, the survey ITESM conducted in Gujarat in 2023, and the four values reach this report unchanged.\n\nThe survey puts the share of holdings belonging to a marketing cooperative at 46 per cent, the mean number of years a holding has held its membership at 11, the share of members selling most of their crop through the cooperative at 57 per cent, and the share of holdings whose parcel the survey could match to the cooperative's list at 71 per cent. None of them was recalculated here, and the survey's own reading of membership, which requires the audit to see the share certificate, comes down with them.\n\nA membership held jointly by two brothers was counted once, and a lapsed membership was not counted. Each figure keeps the taluka it was drawn from reported with it, rather than blended into a single state share.\n\nBeneath the table sit the survey row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Audited Cooperative Membership Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Audited Cooperative Membership Survey supplies membership incidence, years held, crop sold through the cooperative and parcel matches.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:fdbadb5b26d782aad5abc3ce", "occurrence_id": "eb7d1cf58324b66223e28968", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Wages and adequacy were reconstructed for this review from the Nationwide Wage Structure Survey (NWSS), Third Version, the survey conducted in Accra in 2016, and the readings here are the review's, not the survey's.\n\nWorkers were grouped by the hourly wage the survey records for them and the share whose household stands below the national adequacy line computed: 31 per cent of workers below the set minimum against 7 per cent of those above twice it. The survey carries a wage for each worker and a consumption aggregate for the same household without ever joining the two.\n\nHolding the hours a worker reports even leaves 9 points of that distance, so the hours account for part of what the opening comparison puts down to the wage.\n\nWorkers the survey lists as unpaid family labour are dropped rather than counted at a zero wage, and they are 6 per cent of workers. The annex note carries the grouping rule, the pull date, and the base of each share.\n\nMention: \"Nationwide Wage Structure Survey (NWSS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Nationwide Wage Structure Survey relates the hourly wage to workers whose household stands below adequacy.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:2d45db822f89a9efa4ab06d8", "occurrence_id": "f032044340d9fdae640241c4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Protection incident the figures for the city are reported by the High-Resolution Protection Incident Survey (HPIS), the survey conducted in Buenos Aires in 2023, and the four values reach this report unchanged.\n\nThe survey puts the share of displaced households reporting an incident since arrival at 17 per cent, the share reporting an incident to the police at 34 per cent, the mean days between an incident and its being reported at 12, and the share of reporting households standing below the city's adequacy line at 41 per cent. None of them was recalculated here, and the survey's own reading of an incident, which requires the household to name the date and the place, is carried alongside them.\n\nAn incident in the country of origin was not counted, and an incident the household dated only to a season was. Each figure keeps the commune it was drawn from kept with it, instead of being rolled up to the city.\n\nBeneath the table sit the survey row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"High-Resolution Protection Incident Survey (HPIS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The High-Resolution Protection Incident Survey supplies incident incidence, reporting to police, reporting delay and adequacy incidence.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:956821037d59b2ce4d7f33cf", "occurrence_id": "34ebdea741eb8fa9ebb30e23", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Labour inspection figures covering the country are drawn from the Subnational Labour Inspection Registry, the register Polity IV maintains for Gabon in 2024, and the four values reach this report unchanged.\n\nThe register puts the share of registered workplaces inspected in the year at 12 per cent, the mean years between two inspections of the same workplace at 8, the share of inspections resulting in a written notice at 23 per cent, and the share of notices the register records as closed at 55 per cent. None of them was recalculated here, and the register's own reading of an inspection, which requires an inspector's signature at the site, is printed beside them.\n\nA desk review of papers at the office was not counted, and a notice withdrawn on appeal was counted as issued. Each figure keeps the province it was drawn from named with it, instead of averaged over the country.\n\nBeneath the table sit the register row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Subnational Labour Inspection Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Subnational Labour Inspection Registry supplies workplaces inspected, inspection intervals, notices issued and notices closed.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:c85a5db60ca24217dd4bc655", "occurrence_id": "3d2931a3906fa09193939d2b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Rural transport figures covering the whole country derive from the Rural Transport Network Dataset (RTND), Collection 2, the geospatial dataset compiled for Nicaragua in 2016, and the four values reach this report unchanged.\n\nThe dataset puts the share of settlements the dataset places within five kilometres of a passable track at 74 per cent, the mean kilometres of track per settlement it records at 6, the share of the network whose surface class was confirmed on the ground at 48 per cent, and the share of track that changed class between collections at 7 per cent. None of them was recalculated here, and the dataset's own reading of a passable track, which requires passage recorded in the wet season, is given next to them.\n\nA track usable only in the dry season was not counted as passable, and a bridge missing from the field record was counted as a break. Each figure keeps the municipality it was drawn from listed with it, rather than smoothed across the country.\n\nBeneath the table sit the dataset layer behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Rural Transport Network Dataset (RTND)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Rural Transport Network Dataset supplies settlement proximity, track per settlement, ground-confirmed surfaces and class change.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:22a6e78028f56757ee8e08cb", "occurrence_id": "e33ede6f956a74b1fee2a4a4", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Three vegetation lines in the urban annex come from the Urban Vegetation Condition Registry, the register CIA maintains for Chile in 2024, and all three are copied across unchanged.\n\nThe register puts the share of the city's built area carrying tree cover at 21 per cent, the mean canopy loss per district the register records over three years at 3 per cent, and the share of districts whose cover the register rates as stressed at 27 per cent. Nothing was recomputed for this report, and the register's own reading of tree cover, which requires the canopy to close over a planted row, travels with the figures.\n\nThe annex's other lines are compiled from the municipality's own planting return, so the register carries three lines while the return carries the balance. A figure reported for one district keeps that district attached rather than being averaged across the city, and where register and return disagree about a stand the register's figure is kept for these three lines and the return's for the rest, the difference being noted beneath the table rather than smoothed away.\n\nStands the register marks as felled for works are excluded from all three lines, since the return counts them as planted. The annex log carries the register row behind each value, the date it was drawn and the counts under each share.\n\nMention: \"Urban Vegetation Condition Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Urban Vegetation Condition Registry supplies tree cover of built area, canopy loss and stressed districts.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:57b7e361127f5881d00b48ff", "occurrence_id": "70d1c39bc8767a6c240c247c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The records figure in this chapter is the archive's own, taken from its own accessions list, and the Subnational Public Records Registry (SPRR), the register Central Electricity Boards maintains for Peru, is cited to test it.\n\nThe archive's list shows 82 per cent of its series held under a retention band. The registry shows 73 per cent for the same institutions, because it scores a series banded only where the registry can read the band from the institution's own schedule, while the archive's list counts a band written on the box. One basis was never used to adjust the other's list.\n\nThe 9-point distance is smaller than the movement the list shows between two rounds, so the archive's figure is printed as listed and the registry supplies no revision to it and no series for the institutions its round did not visit.\n\nExcluded are series the registry finds held by two institutions, 210 of them, since the two counts cannot be settled. Which series each source counts is held in the annex working papers.\n\nMention: \"Subnational Public Records Registry (SPRR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Subnational Public Records Registry checks the archive's share of series held under a retention band.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:2cdecd7f4d1aea47896beea5", "occurrence_id": "e7dcdbc8decae72e8b0ab1b9", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Transfer receipt and schooling were put together for this review from the Baseline Social Protection Survey, which UNECA conducted in the Arab Republic of Egypt in 2019, and the readings below are the review's own.\n\nHouseholds were grouped by the months the survey records them receiving a transfer and the share of their children enrolled at the visit computed: 91 per cent of households receiving three years or more against 73 per cent of those receiving under a year. The survey carries a first-receipt date for each household and an enrolment item for each child without ever joining the two.\n\nMaking the groups alike on the distance from the household to the nearest school leaves 8 points of that distance, which leaves the school run as part of the gap the opening comparison puts down to the months received.\n\nHouseholds the survey lists as registered but never paid are dropped rather than counted as recipients, and they are 5 per cent of households. The annex note carries the grouping rule, the pull date, with the base of each share alongside.\n\nMention: \"Baseline Social Protection Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Baseline Social Protection Survey relates months of transfer receipt to children enrolled at school.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:af94c6fb6a47e7619b053b8b", "occurrence_id": "addd9376e8391ac1622180bb", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Elevation and clinic access were derived for this chapter from the Record-Based Elevation Model Dataset (REMD), the geospatial dataset compiled for Bolivia in 2019, and the figures below are the chapter's rather than the dataset's.\n\nSettlements were grouped by the altitude band the dataset places them in and the mean hours a household spends reaching a clinic and returning computed: 2.5 hours at settlements above three thousand metres against 1.1 hours at those below one thousand. The dataset holds an elevation for each settlement and the service records a trip for each attendance, and neither reports the two together.\n\nThat comparison supports one claim, that the mobile clinic schedule should be set by altitude rather than by municipal boundary. The fuel lines the proposal quotes come from the service's own vehicle log, from which the dataset supplied nothing.\n\nSettlements the dataset leaves without an elevation for want of a clear surface are dropped rather than counted at the department mean, and they are 7 per cent of settlements. The bands and the counts behind each mean are recorded beneath the table.\n\nMention: \"Record-Based Elevation Model Dataset (REMD)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Record-Based Elevation Model Dataset supports an altitude-based mobile clinic schedule built from altitude band and clinic trip hours.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:ce037024c3a5c8999a83e1d2", "occurrence_id": "b1db060f532370587a07ed91", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Three monitoring lines in the annex come from the Recurrent Public Sector Monitoring Index, the indicator Enerdata compiled for the Arab Republic of Egypt in 2016, and all three are copied across unchanged.\n\nThe index puts the share of directorates filing their quarterly return complete at 68 per cent, the mean days a return waits before the centre logs it at 11, and the share of returns whose figures the index could check at 44 per cent. Nothing was recomputed for this report, and the index's own reading of a logged return, which requires the centre's own date stamp, travels with the figures.\n\nThe annex's other lines are compiled from the ministry's own control sheet, so the index carries three lines while the sheet supplies the balance. A figure reported for one governorate keeps that governorate attached rather than being averaged across the country, and where index and sheet disagree about a filing the index's figure is kept for these three lines and the sheet's for the rest, the difference being noted beneath the table rather than smoothed away.\n\nDirectorates that filed nothing in the year are excluded from all three lines, since the index carries them in a separate list. The annex log carries the index row behind each value, the date the file was taken and the population each share divides.\n\nMention: \"Recurrent Public Sector Monitoring Index\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Recurrent Public Sector Monitoring Index supplies complete quarterly filings, logging delay and checkable returns.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:17df63d62af28fdb2b6bd972", "occurrence_id": "eb127bccc6f7c7bb3719282f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Public sector monitoring the numbers held for the country come from the Independent Public Sector Monitoring Registry (IPSMR), Third Collection, the register kept for Albania in 2024, and the four values reach this report unchanged.\n\nThe register puts the share of institutions publishing their performance figure at 52 per cent, the mean months between a figure's publication and the next revision at 9, the share of institutions whose figure the register could trace to a source at 61 per cent, and the share of institutions audited within the collection at 28 per cent. None of them was recalculated here, and the register's own reading of a published figure, which requires it to stand on the institution's own page, is bound to the figures.\n\nA figure released to journalists only was not counted, and a figure corrected within the same collection was counted once. Each figure keeps the district it was drawn from carried with it, rather than mixed across the country.\n\nBeneath the table sit the register row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Independent Public Sector Monitoring Registry (IPSMR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Independent Public Sector Monitoring Registry supplies published figures, revision intervals, traceable figures and audits.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:8baf9a094ba3a1ed2ddd4a5e", "occurrence_id": "95dfaff0093eba7b27c98c9d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Staffing and attendance were reconstructed for this review from the Experimental Health Workforce Survey, the survey NRS conducted in Sindh in 2017, and the readings here are the review's, not the survey's.\n\nFacilities were grouped by the number of clinical staff the survey counts on the roll and the share of scheduled sessions the survey finds actually held computed: 84 per cent of facilities with six staff or more against 57 per cent of those with two or fewer. The survey carries a roll for each facility and a session count for the same facility without ever joining the two.\n\nHolding the distance from the facility to the nearest town even leaves 12 points of that distance, so the distance accounts for part of what the opening comparison puts down to the roll.\n\nFacilities the survey finds closed on the day of visit are dropped rather than counted as holding no session, and they are 8 per cent of facilities. The annex note carries the grouping rule, the pull date and the base of each share.\n\nMention: \"Experimental Health Workforce Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Experimental Health Workforce Survey relates rostered clinical staff to scheduled sessions actually held.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:f0ca82cbceffadccfcb1e413", "occurrence_id": "de3312f3bfe1f98cf6db0cd7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Maternal and newborn country figures are assembled from the Baseline Maternal and Newborn Survey, the survey CLGF conducted in Cyprus in 2015, and the four values reach this report unchanged.\n\nThe survey puts the share of births attended at a facility at 93 per cent, the share of newborns weighed within a day of birth at 88 per cent, the mean nights a mother stays at the facility after a birth at 2, and the share of mothers in the lowest wealth fifth staying under a night at 19 per cent. None of them was recalculated here, and the survey's own reading of a facility birth, which requires the facility's register to carry the entry, goes with the figures.\n\nA birth attended at home by a nurse was not counted as a facility birth, and a weight taken the following morning was counted. Each figure keeps the district it was drawn from with it, rather than pooled across the country.\n\nBeneath the table sit the survey row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Baseline Maternal and Newborn Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Baseline Maternal and Newborn Survey supplies facility births, newborn weighing, postnatal stay and shortest stay in the lowest fifth.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:31ca25a686b7ac4bbea1898b", "occurrence_id": "28cf6a66ddf6f3fe581bca03", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Light series were catalogued into the study's inventory from the Coordinated Nighttime Lights Registry (CNLR), Wave 6, the register Bank of Spain maintains for 2015, and the work done on them was cataloguing rather than measurement.\n\nEach series was given the inventory's own radiance band and its dates checked against the registry's reference window; where the registry carries a series the inventory has no band for, it was parked for the analyst rather than entered under the nearest one. Series the registry lists without a cloud-free count were left unbanded rather than assumed clear, and 340 of them entered the inventory in that condition. The registry's own wave stamp was kept on every series so a later wave can be matched without re-deriving it.\n\nThe study's settlement counts come from the statistical office's own register, from which the registry supplied nothing; it is the series coding that lets those counts be read against radiance for the lit-area table the report draws up.\n\nThe coding sheet, the count of parked series and the day of deposit are laid out in the annex log.\n\nMention: \"Coordinated Nighttime Lights Registry (CNLR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The Coordinated Nighttime Lights Registry supplies the coded light series behind the study's inventory.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:cd9f59703ecb363be95fd6e1", "occurrence_id": "022e9558add0cba594d4b10d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The parcel figures in this chapter are drawn from the authority's own inspection notes, and the High-Frequency Remote Sensing Dataset, the geospatial dataset UNESCO-IBE compiled for Sierra Leone in 2020, is referred to only as background to them.\n\nThe dataset puts the share of the frame it observes cloud-free in a month at 46 per cent, the mean days between two usable observations at 9, and the share of the frame whose class it holds stable across the year at 69 per cent. All three are given as the dataset issues them, with its own cell size stated, and no chapter figure rests on them.\n\nNo chapter table is built from its observations. The figures the chapter argues from come from the authority's inspection notes, which cover each of its own wards rather than a sampled grid, and where the two accounts of a boundary differ the difference is noted rather than reconciled.\n\nThe annex lists the dataset layers consulted and the day of the extract, and the dataset is used for nothing beyond that background.\n\nMention: \"High-Frequency Remote Sensing Dataset\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The High-Frequency Remote Sensing Dataset is background to the authority's parcel inspection notes.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:d6ec4c44f1122d9681930cf4", "occurrence_id": "a79b36aa5e88494740c7f0e6", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Programme monitoring figures spanning the area are taken from the Provincial Programme Monitoring Registry, the register EBRD maintains for the West Bank and Gaza in 2017, and the four values reach this report unchanged.\n\nThe register puts the share of project sites reporting to the register in the period at 81 per cent, the mean weeks a report waits before the register logs it at 4, the share of reports carrying a spending figure the register could check at 57 per cent, and the share of sites visited by the register's own team at 34 per cent. None of them was recalculated here, and the register's own reading of a logged report, which requires the register's date stamp, travels with the values.\n\nA report handed in at a crossing was counted, and a report withdrawn by its project was not. Each figure keeps the governorate it was drawn from attached, instead of being blended across the area.\n\nBeneath the table sit the register row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Provincial Programme Monitoring Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Provincial Programme Monitoring Registry supplies reporting rates, logging delay, checkable spending and site visits.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:8fb744079a6741faa2c98769", "occurrence_id": "1c67aeb452f82beff5a9f140", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Road segment records were catalogued into the fund's asset file from the Audited Road Segment Census (ARSC), the census NCAER follows for Kosovo, and the work done on them was cataloguing rather than measurement.\n\nEach segment was given the file's own maintenance band and its audit date checked against the census's reference window; where the census carries a segment the file has no band for, it was parked for the engineer rather than entered under the nearest one. Segments the census audits without a surface class were left unbanded rather than assumed unsealed, and 190 of them entered the file in that condition. The census's own audit stamp was kept on every segment so a later round can be matched without re-deriving it.\n\nThe fund's length totals come from the authority's own quantity records, from which the census supplied nothing; it is the segment coding that lets those totals be read against maintenance for the resurfacing list the report draws up.\n\nThe coding sheet, the count of parked segments and the day of deposit are filed in the annex log.\n\nMention: \"Audited Road Segment Census (ARSC)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The Audited Road Segment Census supplies the coded segment records behind the fund's asset file.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:75013d259bf407674646a684", "occurrence_id": "d5ad02db6e003afce3e1ea8b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The targeting figure in this chapter is the office's own, taken from its beneficiary list, and the Local Targeting and Payments Survey (LTPS), Wave 5, the survey Univer sity of Minnesota follows, is cited to test it.\n\nThe office's list shows 78 per cent of its payments reaching a household the office's own score ranks in the bottom two fifths. The survey shows 69 per cent for the same areas, because it ranks the household on what the interview records, while the office's list ranks it on the form the household filed. The two bases were not applied to each other's list.\n\nThe 9-point distance is smaller than the movement the list shows between two waves, so the office's figure is printed as listed and the survey supplies no revision to it and no households whose interview was never completed.\n\nExcluded are households the survey finds listed twice under different heads, 280 of them, since the two records cannot be settled. Which households each source counts is detailed in the method note in the annex.\n\nMention: \"Local Targeting and Payments Survey (LTPS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Local Targeting and Payments Survey checks the office's share of payments reaching the bottom two fifths.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:8362c5f44a511a8abea8ce1d", "occurrence_id": "94205a8a37c2099e26ca57a5", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Three water body lines in the annex come from the Record-Based Water Body Dataset (RWBD), Eighth Phase, the geospatial dataset compiled for Yemen in 2020, and all three are copied across unchanged.\n\nThe dataset puts the share of the frame's reservoirs the dataset holds an outline for at 64 per cent, the mean hectares a mapped reservoir covers at 22, and the share of mapped reservoirs whose outline changed between phases at 11 per cent. Nothing was recomputed for this report, and the dataset's own reading of a reservoir, which requires standing water at the phase's observation, travels with the figures.\n\nThe annex's other lines are compiled from the authority's own gauge return, so the dataset carries three lines while the return holds the remainder. A figure reported for one basin keeps that basin attached rather than being averaged across the country, and where dataset and return disagree about a shoreline the dataset's figure is kept for these three lines and the return's for the rest, the difference being noted beneath the table rather than smoothed away.\n\nReservoirs the dataset marks dry at the phase are excluded from all three lines, since the return still gauges them. The annex log carries the dataset layer behind each value, the date of the pull and, alongside it, the denominators each share uses.\n\nMention: \"Record-Based Water Body Dataset (RWBD)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Record-Based Water Body Dataset supplies mapped reservoir cover, mean surface and outline change.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:816e837973f2d34938ffa8bd", "occurrence_id": "301256029d3b3c3d5a86a4c6", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Three delivery lines in the service annex come from the Baseline Service Delivery System, the register Townsend maintains for Tanzania in 2020, and all three are copied across unchanged.\n\nThe register puts the share of facilities reporting their monthly tally at 86 per cent, the mean days a tally takes to reach the district office at 7, and the share of tallies the office accepted without a query at 63 per cent. Nothing was recomputed for this report, and the register's own reading of an accepted tally, which requires the office's own sign-off, travels with the figures.\n\nThe annex's other lines are compiled from the facility's own ledger, so the register carries three lines while the ledger carries what is left over. A figure reported for one district keeps that district attached rather than being averaged across the country, and where register and ledger disagree about a tally the register's figure is kept for these three lines and the ledger's for the rest, the difference being noted beneath the table rather than smoothed away.\n\nFacilities opened after the register's window are excluded from all three lines, since the register holds no baseline for them. The annex log carries the register row behind each value, the date on the extract as well as the denominators behind each share.\n\nMention: \"Baseline Service Delivery System\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Baseline Service Delivery System supplies facility reporting rates, tally delay and tallies accepted without query.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:d8fd9fe3864ba1b9b088e762", "occurrence_id": "ccc2dbd6293f9890596329ec", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Health and household figures at the country level are taken from the Inter-Agency Health and Household Survey (IHHS), the survey conducted in Egypt in 2017, and the four values reach this report unchanged.\n\nThe survey puts the share of households holding a member with a long-standing illness at 19 per cent, the share of those households unable to pay for a visit when last needed at 27 per cent, the mean distance a household walks to the nearest facility at 4 kilometres, and the share of those households standing below the national adequacy line at 36 per cent. None of them was recalculated here, and the survey's own reading of a long-standing illness, which requires the household to name the condition and the year it began, follows the figures.\n\nAn illness the household attributed to age was counted, and a condition treated only once in the year was not. Each figure keeps the governorate it was drawn from shown with it, instead of combined into one country figure.\n\nBeneath the table sit the survey row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Inter-Agency Health and Household Survey (IHHS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Inter-Agency Health and Household Survey supplies long-standing illness incidence, unmet payment, walk to facility and adequacy incidence.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:f5bf1f614fe26346f1a370af", "occurrence_id": "9119b7f8e6095fd7119918ca", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Arrival and registration were put together for this review from the Ministerial Arrival and Registration Census (MARC), First Wave, the census D¨ur et al. follows, and the readings below are the review's own.\n\nHouseholds were grouped by the days between arrival and registration the census records and the share the census finds holding a plot assigned within a month computed: 77 per cent of households registering within a week against 31 per cent of those taking a month or more. The census carries an arrival date and a registration date for each household and an assignment flag for the same household without ever joining the two.\n\nMaking the groups alike on the size of the household leaves 9 points of that distance, which leaves household size as part of the gap the opening comparison puts down to the registration delay.\n\nHouseholds the census finds registering twice are dropped rather than counted at their first entry, and they are 4 per cent of households. The annex note carries the grouping rule, the pull date together with the base of each share.\n\nMention: \"Ministerial Arrival and Registration Census (MARC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Ministerial Arrival and Registration Census relates registration delay to households assigned a plot within a month.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:55e6ab05facf24f964157c09", "occurrence_id": "2511f42e1742a743cb216b1e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Consumption frame-level figures come from the Multi-Topic Household Consumption Index (MHCI), Collection 5, the indicator Hungarian NHIF maintains, and the four values reach this report unchanged.\n\nThe index puts the mean monthly consumption of a household at 1 240, the share of households below the national adequacy line at 23 per cent, the share of consumption a household spends on food at 54 per cent, and the share of households reporting a shortfall in the reference month at 17 per cent. None of them was recalculated here, and the index's own reading of consumption, which is measured over four weeks and includes home production at the price the household names, is reported with them.\n\nA gift received in kind was counted, and a loan taken to meet spending was not. Each figure keeps the region it was drawn from reported with it, rather than blended into a single frame share.\n\nBeneath the table sit the index row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Multi-Topic Household Consumption Index (MHCI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Multi-Topic Household Consumption Index supplies mean consumption, adequacy incidence, food share and reported shortfalls.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:efa47f8eeeba862c9fe15ea7", "occurrence_id": "dccedf3dc10dd38f91431994", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Monitoring and backlog were reconstructed for this review from the Rolling Operations Monitoring Registry (ROMR), Phase 6, the register kept for Eastern Europe in 2017, and the readings here are the review's, not the register's.\n\nUnits were grouped by the number of reports the register records missing in the phase and the share of its cases the unit closes within its own limit computed: 73 per cent of units missing no report against 41 per cent of those missing three or more. The register carries a reporting flag for each unit and a closing figure for the same unit without ever joining the two.\n\nHolding the number of staff the register records for the unit even leaves 11 points of that distance, so staffing accounts for part of what the opening comparison puts down to the missing reports.\n\nUnits the register lists as merged during the phase are dropped rather than counted at their predecessor's reports, and they are 5 per cent of units. The annex note carries the grouping rule, the pull date plus the base of each share.\n\nMention: \"Rolling Operations Monitoring Registry (ROMR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Rolling Operations Monitoring Registry relates missing reports to cases closed within the unit's own limit.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:512e8080e26c3fe974b2b2d2", "occurrence_id": "5599699d52fc08ea8e9eec60", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Tertiary study and health were put together for this review from the Municipal Tertiary Education Collection, which PROCOMER compiled for Eastern Europe in 2019, and the readings below are the review's own.\n\nMunicipalities were grouped by the share of its young people the collection records enrolled in tertiary study and the share of its households reporting an unmet medical need computed: 12 per cent of municipalities above a quarter enrolled against 26 per cent of those below a tenth. The collection carries an enrolment count for each municipality and a health return for the same municipality without ever joining the two.\n\nMaking the groups alike on the median income the collection records for the municipality leaves 6 points of that distance, which leaves income as part of the gap the opening comparison puts down to enrolment.\n\nMunicipalities the collection lists with a campus but no enrolled resident are dropped rather than counted at their resident share, and they are 4 per cent of municipalities. The annex note carries the grouping rule, the pull date, and the base of each share.\n\nMention: \"Municipal Tertiary Education Collection\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Municipal Tertiary Education Collection relates tertiary enrolment to households reporting an unmet medical need.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:2d2146a457df555a5df27215", "occurrence_id": "85a3c24faf14583bb43dc7a6", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Output area conditions were derived for this chapter from the Nationwide Output Area Report, the report UNDESA issued for Poland in 2017, and the figures below are the chapter's own, not the report's.\n\nAreas were grouped by the share of its residents the report places in the lowest income quintile and the mean years the report records for a dwelling's age computed: 46 years at areas above a third in the lowest quintile against 21 years at those below a tenth. The report carries a quintile share for each area and a dwelling age for the same area without ever joining the two.\n\nThat comparison supports one claim, that the refurbishment allotment should follow the quintile share rather than the dwelling count. The contractor lines the proposal quotes come from the council's own works schedule, from which the report supplied nothing.\n\nAreas the report lists without a dwelling count are dropped rather than counted at the national mean, and they are 6 per cent of areas. The bands and the counts behind each mean are recorded beneath the table.\n\nMention: \"Nationwide Output Area Report\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Nationwide Output Area Report supports a quintile-based refurbishment allotment built from lowest-quintile share and dwelling age.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:626d6d072a180b6c1675fa9e", "occurrence_id": "96264bd22d4b66c6cccbe21c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Service catchments were derived for this chapter from the Panel Service Catchments Census, the census the General Statistics Office conducted in Nairobi in 2021, and the figures below belong to the chapter rather than to the census's.\n\nWards were grouped by the share of its households the census places outside every facility catchment and the mean minutes the council records for a request to be answered computed: 38 minutes at wards above a fifth outside a catchment against 14 minutes at those below a tenth. The census holds a catchment for each facility and the council records a request with an answer time, and neither reports the two together.\n\nThat comparison supports one claim, that the mobile unit should be scheduled by catchment gap rather than by ward boundary. The fuel lines the proposal quotes come from the council's own vehicle log, from which the census supplied nothing.\n\nWards the census maps without a facility count are dropped rather than counted at the city mean, and they are 5 per cent of wards. The bands and the counts behind each mean are recorded beneath the table.\n\nMention: \"Panel Service Catchments Census\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Panel Service Catchments Census supports a catchment-based mobile unit schedule built from households outside catchments and request answer times.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:244a750f805b738522ff427a", "occurrence_id": "d2a8950a254df457f0125c90", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Poverty and shock were reconstructed for this review from the Midline Poverty and Vulnerability Survey (MPVS), Seventh Version, the survey Maddison follows, and the readings here are the review's, not the survey's.\n\nHouseholds were grouped by the number of shocks the survey records for them in the year and the share standing below the survey's adequacy line computed: 58 per cent of households with three shocks or more against 21 per cent of those with none. The survey carries a shock list for each household and an adequacy measure for the same household without ever joining the two.\n\nHolding the household's land holding even leaves 19 points of that distance, so the land accounts for part of what the opening comparison puts down to the shocks.\n\nHouseholds the survey records a shock for but no date were dropped rather than counted as shocked, and they are 6 per cent of households. The annex note carries the grouping rule, the pull date, with the base of each share alongside.\n\nMention: \"Midline Poverty and Vulnerability Survey (MPVS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Midline Poverty and Vulnerability Survey relates shocks in the year to households below the adequacy line.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:396aed2e283c68eb0d51b226", "occurrence_id": "05f1086055e0763774f04416", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Border crossing figures for each part of the city are drawn from the Urban Border Crossing Report (UBCR), the report CNTS issued for 2023, and the four values reach this report unchanged.\n\nThe report puts the number of crossings recorded at the city's posts in the year at 214 000, the share of those crossings made by households the report classes as in transit at 31 per cent, the mean hours a household waits at a post before passing at 6, and the share of transit households the report places below the city's adequacy line at 44 per cent. None of them was recalculated here, and the report's own reading of a crossing, which requires the post's own stamp in the record, is attached to them.\n\nA crossing refused at the post was counted, and a crossing made on a pass the report could not verify was not. Each figure keeps the post it was drawn from kept with it, instead of being rolled up to the city.\n\nBeneath the table sit the report page behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Urban Border Crossing Report (UBCR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Urban Border Crossing Report supplies crossings recorded, transit share, waiting hours and adequacy incidence among transit households.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:5aed0d34dd4ccfc068c4a706", "occurrence_id": "225ba1a5818def898023736b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Referral completion and cost were put together for this review from the Linked Referral Pathways Survey, which the Department of Defense conducted at Zatari camp in 2017, and the readings below are the review's own.\n\nHouseholds were grouped by the number of referrals the survey records for them in the year and the share the survey finds standing below the camp's adequacy line computed: 49 per cent of households with two referrals or more against 18 per cent of those with none. The survey carries a referral for each household and an adequacy measure for the same household without ever joining the two.\n\nMaking the groups alike on the months the household had been at the camp leaves 12 points of that distance, which leaves the months at the camp as part of the gap the opening comparison puts down to the referrals.\n\nHouseholds the survey finds paying for part of a referral are dropped rather than counted as referred free, and they are 7 per cent of households. The annex note carries the grouping rule, the pull date and the base of each share.\n\nMention: \"Linked Referral Pathways Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Linked Referral Pathways Survey relates referrals in the year to households below the camp's adequacy line.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:c6b60a1a84d0248c440393dc", "occurrence_id": "8437efd0a90fd59499afba90", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Transfers and sowing were reconstructed for this review from the Multi-Topic Cash Transfer Registry Survey (MCTRS), Wave 7, the survey SABESP follows, and the readings here are the review's, not the survey's.\n\nHouseholds were grouped by the months the survey records them receiving a transfer and the share sowing improved seed at the visit computed: 46 per cent of households receiving two years or more against 22 per cent of those receiving under a year. The survey carries a first-receipt date for each household and a seed item for the same household without ever joining the two.\n\nHolding the area a household sows even leaves 9 points of that distance, so the sown area accounts for part of what the opening comparison puts down to the months received.\n\nHouseholds the survey lists as registered but never paid are dropped rather than counted as recipients, and they are 5 per cent of households. The annex note carries the grouping rule, the pull date and, alongside it, the base of each share.\n\nMention: \"Multi-Topic Cash Transfer Registry Survey (MCTRS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Multi-Topic Cash Transfer Registry Survey relates months of transfer receipt to households sowing improved seed.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:c3b56529ddb445161f1de17c", "occurrence_id": "ca51195ea8257a0e811727b2", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Working conditions were derived for this chapter from the Quality-Assured Working Conditions Census, the census the International Trade Centre conducted in Eritrea in 2024, and the figures below are counted by the chapter rather than by the census's.\n\nWorkplaces were grouped by the share of its workers the census records without any schooling completed and the share reporting no drinking water at the workplace computed: 52 per cent of workplaces above a third unschooled against 17 per cent of those below a tenth. The census carries a schooling item for each worker and a workplace item for the same workplace without ever joining the two.\n\nThat comparison supports one claim, that the inspection round should begin with workplaces whose schooling share is highest rather than with the largest workplaces. The training lines the proposal quotes come from the ministry's own course register, from which the census supplied nothing.\n\nWorkplaces the census visits during a work stoppage are dropped rather than counted at their usual response, and they are 6 per cent of workplaces. The bands and the counts behind each share are recorded beneath the table.\n\nMention: \"Quality-Assured Working Conditions Census\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Quality-Assured Working Conditions Census supports a schooling-led inspection round built from unschooled workers and workplace water access.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:662fa674b60464c899b2089d", "occurrence_id": "41e4de57a53acc8e3f25e97a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Operations practice was derived for this chapter from the Linked Operations Monitoring Index (LOMI), the indicator kept for Bolivia in 2018, and the figures below are the chapter's own figures and not the index's.\n\nOffices were grouped by the months the index records since their last operations return and the share of their returns carrying a figure the index could trace computed: 69 per cent of offices returning within the quarter against 32 per cent of those returning later. The index carries a return date for each office and a traceable flag for the same return without ever joining the two.\n\nThat comparison supports one claim, that a return should be filed monthly rather than at the close of the half-year. The staffing lines the proposal quotes come from the ministry's own establishment return, from which the index supplied nothing.\n\nOffices the index lists as relocated during the reference year are dropped rather than counted at their predecessor's return, and they are 4 per cent of offices. The bands and the counts behind each share are recorded beneath the table.\n\nMention: \"Linked Operations Monitoring Index (LOMI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Linked Operations Monitoring Index supports a monthly-filing claim built from return age and traceable figures.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:647b078e1c9733f74952f180", "occurrence_id": "55ccc86fe410acb06971bc34", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Skills and adequacy were put together for this review from the Comparative Skills and Employment Survey, which the Growth Lab at Harvard conducted in Delhi in 2021, and the readings below are the review's own.\n\nWorkers were grouped by the skill certificate the survey records for them and the share whose household stands below the city's adequacy line computed: 33 per cent of workers with no certificate against 9 per cent of those holding an advanced one. The survey carries a certificate item for each worker and a consumption aggregate for the same household without ever joining the two.\n\nMaking the groups alike on the years of schooling the worker reports leaves 7 points of that distance, which leaves schooling as part of the gap the opening comparison puts down to the certificate.\n\nWorkers the survey lists as studying rather than working at the visit are dropped rather than counted as unskilled, and they are 5 per cent of workers. The annex note carries the grouping rule, the pull date as well as the base of each share.\n\nMention: \"Comparative Skills and Employment Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Comparative Skills and Employment Survey relates skill certificates to workers whose household stands below adequacy.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:8940f0bf602c0b6af83b29fd", "occurrence_id": "ec5d48c24f1357e55346ef5a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Three land cover lines in the annex come from the Joint Land Cover Registry (JLCR), Version 6, the register kept for Nigeria in 2017, and all three are copied across unchanged.\n\nThe register puts the share of the frame the register classes as woodland at 28 per cent, the mean hectares a class boundary moved between versions at 640, and the share of the frame whose class the register could confirm on the ground at 52 per cent. Nothing was recomputed for this report, and the register's own reading of woodland, which requires the canopy to close over a third of the cell, travels with the figures.\n\nThe annex's other lines are compiled from the service's own field return, so the register carries three lines while the return accounts for the remaining lines. A figure reported for one state keeps that state attached rather than being averaged across the country, and where register and return disagree about a boundary the register's figure is kept for these three lines and the return's for the rest, the difference being noted beneath the table rather than smoothed away.\n\nCells the register marks as burned within the version are excluded from all three lines, since the register re-classes them in the next one. The annex log carries the register layer behind each value, the day of the pull and the base each share is taken on.\n\nMention: \"Joint Land Cover Registry (JLCR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Joint Land Cover Registry supplies woodland share, boundary movement and ground-confirmed classes.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:b94649ff249e6eb5181ade5d", "occurrence_id": "bde19a443c21c6e02e47c9a0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Three hiring lines in the labour annex come from the Triennial Hiring and Vacancies Index (THVI), Third Release, the indicator kept for Addis Ababa in 2017, and all three are copied across unchanged.\n\nThe index puts the share of establishments reporting a vacancy open at the release at 18 per cent, the mean days a vacancy stands open before it is filled at 47, and the share of filled vacancies taken from a school leaver at 26 per cent. Nothing was recomputed for this report, and the index's own reading of a vacancy, which requires the establishment to have advertised it, travels with the figures.\n\nThe annex's other lines are compiled from the office's own permit file, so the index carries three lines while the file carries the rest. A figure reported for one sub-city keeps that sub-city attached rather than being averaged across the city, and where index and file disagree about a vacancy the index's figure is kept for these three lines and the file's for the rest, the difference being noted beneath the table rather than smoothed away.\n\nVacancies the index records as withdrawn by the establishment are excluded from all three lines, since the file still carries them. The annex log carries the index row behind each value, the extraction day together with the counts the shares rest on.\n\nMention: \"Triennial Hiring and Vacancies Index (THVI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Triennial Hiring and Vacancies Index supplies open vacancies, days to fill and vacancies taken by school leavers.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:8da46829d9c5c3be0d3d5d6a", "occurrence_id": "ecf58868f1549c9df162293d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Poverty and vulnerability figures measured across the country are read from the Enumerative Poverty and Vulnerability Survey, the survey Coface conducted in Bolivia in 2022, and the four values reach this report unchanged.\n\nThe survey puts the share of households below the national adequacy line at 31 per cent, the share of those households holding no land of their own at 38 per cent, the mean hectares a household in the lowest fifth sows at 1.2, and the share of households reporting a crop loss in the last season at 22 per cent. None of them was recalculated here, and the survey's own reading of adequacy, which is measured on consumption including home production, is stated with them.\n\nA plot held on a sharecropping term was counted as no land of the household's own, and a loss the household put at under a tenth of the crop was not counted. Each figure keeps the department it was drawn from named with it, instead of averaged over the country.\n\nBeneath the table sit the survey row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Enumerative Poverty and Vulnerability Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Enumerative Poverty and Vulnerability Survey supplies adequacy incidence, landless households, sown hectares of the lowest fifth and crop loss.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:4bef679c6c085b400f6b95a7", "occurrence_id": "824996e4ec6cbe93e11eea6f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Consumption and hardship were derived for this chapter from the Decennial Household Consumption Survey (DHCS), Version 4, the survey the Supreme Court follows, and the figures below come from the chapter and not from the survey's.\n\nDistricts were grouped by the mean monthly consumption the survey records for a household and the share of households the survey finds going without a meal in the reference week computed: 27 per cent of districts below the national median against 6 per cent of those above twice it. The survey carries a consumption aggregate for each household and a meal item for the same household, and it publishes both only at the household level.\n\nThat comparison supports one claim, that the supplement should be raised by district rather than held at one national rate. The budget lines the proposal quotes come from the treasury's own vote, from which the survey supplied nothing.\n\nDistricts the survey lists with too few households for a mean are dropped rather than counted at the national figure, and they are 7 per cent of districts. The bands and the counts behind each share are recorded beneath the table.\n\nMention: \"Decennial Household Consumption Survey (DHCS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Decennial Household Consumption Survey supports a district-level supplement built from mean consumption and meals skipped.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:5d5455ef406de188dd98fbea", "occurrence_id": "26c300f0b453a8901a2bbe18", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Adequacy and work were reconstructed for this review from the Quarterly Poverty and Vulnerability Microdata (QPVM), Version 8, the microdata World Inequality Lab maintains for Pakistan, and the readings here are the review's, not the microdata's.\n\nHouseholds were grouped by the number of quarters the microdata records them in a workless household and the share standing below the national adequacy line computed: 62 per cent of households in four quarters or more against 18 per cent of those in none. The microdata carries a work flag for each adult and an adequacy measure for the same household without ever joining the two.\n\nHolding the size of the household even leaves 14 points of that distance, so household size accounts for part of what the opening comparison puts down to the quarters workless.\n\nHouseholds the microdata lists with an adult in unpaid family work are dropped rather than counted as workless, and they are 6 per cent of households. The annex note carries the grouping rule, the pull date plus the base of each share.\n\nMention: \"Quarterly Poverty and Vulnerability Microdata (QPVM)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Quarterly Poverty and Vulnerability Microdata relates quarters in a workless household to households below adequacy.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:356deb33053bdeec9e9ae6b6", "occurrence_id": "02bb4adea6260d90b1cfdbca", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Bulletin practice was derived for this chapter from the Standard Statistical Bulletin System, the register the National Statistical Office maintains for Mexico City in 2023, and the figures below come from the chapter, not from the register's.\n\nOffices were grouped by the months the register records since their last bulletin and the share of their tables carrying a method note computed: 71 per cent of offices issuing within the quarter against 29 per cent of those issuing later. The register carries an issue date for each office and a method-note flag for the same bulletin without ever joining the two.\n\nThat comparison supports one claim, that a bulletin should carry its method note at first issue rather than at the next revision. The staffing lines the proposal quotes come from the office's own establishment return, from which the register supplied nothing.\n\nOffices the register lists as renamed during the year are dropped rather than counted at their predecessor's bulletin, and they are 4 per cent of offices. The bands and the counts behind each share are recorded beneath the table.\n\nMention: \"Standard Statistical Bulletin System\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Standard Statistical Bulletin System supports a first-issue method note claim built from bulletin age and method notes.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:bbb25a042f35f34d2f43538b", "occurrence_id": "ffa7c7bd707861afd4005613", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Working conditions figures for the country are reported from the Rolling Working Conditions Survey (RWCS), Fourth Edition, the survey World Bank-IFC follows for Egypt, and the four values reach this report unchanged.\n\nThe survey puts the share of workers reporting no written contract at 57 per cent, the share working more than forty-eight hours in the longest week at 34 per cent, the mean days of leave a worker reports at 6, and the share of workplaces with a first-aider present at 41 per cent. None of them was recalculated here, and the survey's own reading of a written contract, which requires the worker to have seen the signed copy, comes with them.\n\nA contract held only by the employer was counted as no contract, and leave bought from a colleague was not counted. Each figure keeps the governorate it was drawn from listed with it, rather than smoothed across the country.\n\nBeneath the table sit the survey row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Rolling Working Conditions Survey (RWCS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Rolling Working Conditions Survey supplies missing contracts, long weeks, leave days and workplaces with a first-aider.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:5880bda602d3de616de4d708", "occurrence_id": "6ccb6b5fadc004a86023cabf", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Assets and adequacy were put together for this review from the Periodic Asset Ownership Microdata, which CPCB maintains for Colombia in 2018, and the readings below are the review's own.\n\nHouseholds were grouped by the number of asset classes the microdata records for them and the share standing below the national adequacy line computed: 68 per cent of households holding one class or none against 12 per cent of those holding four or more. The microdata carries an asset list for each household and an adequacy measure for the same household without ever joining the two.\n\nMaking the groups alike on the education of the household head leaves 15 points of that distance, which leaves education as part of the gap the opening comparison puts down to the assets.\n\nHouseholds the microdata records owing a mortgage larger than their annual income are dropped rather than counted as owning their dwelling, and they are 3 per cent of households. The annex note carries the grouping rule, the pull date, and the base of each share.\n\nMention: \"Periodic Asset Ownership Microdata\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Periodic Asset Ownership Microdata relates asset classes held to households below the adequacy line.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:99ec4169d834b08e401a20f5", "occurrence_id": "85781e1e583713facf77a52d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Adequacy and transfer receipt were reconstructed for this review from the Nationwide Socioeconomic Survey, which the Tax Justice Network conducted in the CHT and Sylhet divisions in 2022, and the readings here are the review's, not the survey's.\n\nHouseholds were grouped by whether the survey records any transfer received in the year and the share standing below the national adequacy line computed: 41 per cent of households receiving no transfer against 22 per cent of those receiving one. The survey carries a transfer item for each household and a consumption aggregate for the same household without ever joining the two.\n\nHolding the land a household sows even leaves 8 points of that distance, so the land accounts for part of what the opening comparison puts down to the transfer.\n\nHouseholds the survey finds receiving a transfer only once in the year are dropped rather than counted as recipients, and they are 5 per cent of households. The annex note carries the grouping rule, the pull date, with the base of each share alongside.\n\nMention: \"Nationwide Socioeconomic Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Nationwide Socioeconomic Survey relates transfer receipt in the year to households below the adequacy line.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:9afefb2f70c0d7656829742c", "occurrence_id": "6b313b66e7b51843cbd3585c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Informal trade the country figures are compiled from the Clustered Informal Trade Survey, the survey S&P conducted in Mozambique in 2021, and the four values reach this report unchanged.\n\nThe survey puts the share of traders working without a stall licence at 64 per cent, the share of those traders who arrived in the town within five years at 37 per cent, the mean daily takings a trader reports at 210, and the share of traders moved on by the council in the last year at 18 per cent. None of them was recalculated here, and the survey's own reading of a trader, which requires the person to have traded at the place on the day of visit, is kept with them.\n\nA trader selling from a bicycle was counted, and a trader trading only on market days was counted at the market days. Each figure keeps the cluster it was drawn from carried with it, rather than mixed across the country.\n\nBeneath the table sit the survey row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Clustered Informal Trade Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Clustered Informal Trade Survey supplies unlicensed trading, recent arrival, daily takings and traders moved on.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:3922c15a2c5e2d9653066e8c", "occurrence_id": "2b14d88277996bbf55ce99d7", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Three apprenticeship lines in the labour annex come from the Rapid Apprenticeship Registry (RAR), Wave 8, the register kept for the CHT and Sylhet divisions in 2017, and all three are copied across unchanged.\n\nThe registry puts the share of establishments holding a registered apprentice at 13 per cent, the mean months a spell runs before it closes at 11, and the share of leavers the registry traces into a job in the same trade at 39 per cent. Nothing was recomputed for this report, and the registry's own reading of a spell, which requires the agreement to be filed with the divisional office, travels with the figures.\n\nThe annex's other lines are compiled from the institute's own course roll, so the registry carries three lines while the roll holds the other lines. A figure reported for one division keeps that division attached rather than being averaged across both, and where registry and roll disagree about a spell the registry's figure is kept for these three lines and the roll's for the rest, the difference being noted beneath the table rather than smoothed away.\n\nSpells the registry records as transferred to another establishment are excluded from all three lines, since the roll counts them once at the receiving institute. The annex log carries the registry row behind each value, the extraction date and the base of each share.\n\nMention: \"Rapid Apprenticeship Registry (RAR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Rapid Apprenticeship Registry supplies establishments holding apprentices, spell length and leavers placed in the same trade.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:c62ff1016a064a5f64acf2c9", "occurrence_id": "6aaf248370a09bef3809d27d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The apprenticeship figures in this chapter are the ministry's own, taken from its course register, and the Recurrent Apprenticeship Report (RAR), Fifth Collection, the report issued for Jordan in 2016, is quoted only as background to them.\n\nThe report puts the share of establishments taking an apprentice in the year at 15 per cent, the mean months a spell runs at 12, and the share of leavers placed in the trade they trained for at 41 per cent. All three are given as the report issues them, with its own establishment list stated, and no chapter figure rests on them.\n\nTheir use here is to show how apprenticeship stands across the governorates beside the eleven institutes whose register the chapter works from. The chapter's argument rests on that register, which names each enrolment rather than sampling establishments, and the two are kept apart rather than merged.\n\nThe annex lists the report pages consulted and the day the table was pulled, and the report is used for nothing beyond that background.\n\nMention: \"Recurrent Apprenticeship Report (RAR)\""}, {"role": "assistant", "content": "{\"role\": \"background\", \"usage_action\": \"source\", \"summary\": \"The Recurrent Apprenticeship Report is quoted as background beside the ministry's course register.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:bffddbce4bfbcca0aedffdec", "occurrence_id": "8e0c0ea1c5ced2143ec7ad8e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Record linkage and coverage were put together for this review from the Inter-Agency Administrative Records Index, which ECLAC compiled for Florida in 2016, and the readings below are the review's own.\n\nAgencies were grouped by the share of its records the index finds matched to another agency's file and the share of its caseload the index finds closed within the stated limit computed: 77 per cent of agencies above half matched against 45 per cent of those below a fifth. The index carries a match flag for each record and a closing date for the same case without ever joining the two.\n\nMaking the groups alike on the number of cases an agency carries leaves 9 points of that distance, which leaves the caseload as part of the gap the opening comparison puts down to the matching.\n\nAgencies the index lists with records but no closing dates are dropped rather than counted as unclosed, and they are 6 per cent of agencies. The annex note carries the grouping rule, the pull date and, alongside it, the base of each share.\n\nMention: \"Inter-Agency Administrative Records Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Inter-Agency Administrative Records Index relates matched records to caseload closed within the stated limit.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:c086895170e4fc59f0f25442", "occurrence_id": "3a9ed61fb031a0e831ae2b17", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Public records figures for the whole frame are sourced from the Quarterly Public Records Registry (QPRR), Second Release, the register MODIS follows, and the four values reach this report unchanged.\n\nThe register puts the share of offices holding a retention schedule the register accepts at 58 per cent, the mean quarters a series waits before appraisal at 14, the share of series the register finds held in a format it can open at 44 per cent, and the share of offices whose register was reviewed within the release at 31 per cent. None of them was recalculated here, and the register's own reading of an accepted schedule, which requires the signature of the office head, accompanies the figures.\n\nA schedule signed by a predecessor was still counted, and a series held only on a personal machine was counted as unlisted. Each figure keeps the office it was drawn from with it, rather than pooled across the frame.\n\nBeneath the table sit the register row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Quarterly Public Records Registry (QPRR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Quarterly Public Records Registry supplies accepted schedules, appraisal waits, openable formats and reviewed registers.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:079621824af0d7207c284512", "occurrence_id": "da88a6416cae5a376fcb8a4a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Temperature records for the country are reported from the Government Temperature Records Report, the report Oxford issued for Romania in 2020, and the four values reach this report unchanged.\n\nThe report puts the share of the country covered by a station with a complete year at 66 per cent, the mean hot days a covered station records in the season at 17, the share of the country's readings taken at a station the report rates as urban at 29 per cent, and the share of stations whose reading the report revised after issue at 6 per cent. None of them was recalculated here, and the report's own reading of a hot day, which requires the threshold to be exceeded at the station's own hour, comes down with them.\n\nA day interpolated from a neighbouring station was not counted, and a station moved within the year was counted at both sites. Each figure keeps the county it was drawn from attached, instead of being blended across the country.\n\nBeneath the table sit the report row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Government Temperature Records Report\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Government Temperature Records Report supplies station cover, hot days, urban share and post-issue revisions.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:a71b0236e33b3d402efafbfa", "occurrence_id": "5689d66982b280a67b88d31d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Shock coping and work were reconstructed for this review from the Endline Shock Coping Survey (ESCS), the survey conducted in Peru in 2019, and the readings here are the review's, not the survey's.\n\nHouseholds were grouped by the number of coping strategies the survey records them using and the share of adults in the household seeking work outside it computed: 51 per cent of households using three strategies or more against 23 per cent of those using none. The survey carries a coping list for each household and a work-search item for the same adult without ever joining the two.\n\nHolding the household's land holding even leaves 11 points of that distance, so the land accounts for part of what the opening comparison puts down to the strategies.\n\nHouseholds the survey records a strategy for but no shock were dropped rather than counted as coping, and they are 5 per cent of households. The annex note carries the grouping rule, the pull date as well as the base of each share.\n\nMention: \"Endline Shock Coping Survey (ESCS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Endline Shock Coping Survey relates coping strategies used to adults seeking work outside the household.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:9ad538328305a81124edf5a9", "occurrence_id": "85d8fa8bc11503b1c98aa412", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The delivery figure in this chapter is the department's own, taken from its own service log, and the Sentinel Service Delivery System (SSDS), Eighth Round, the register kept for Moscow in 2016, is cited to test it.\n\nThe department's log shows 90 per cent of requests answered within its stated limit. The register shows 82 per cent for the same districts, because it dates an answer from the day the applicant receives it, while the department's log dates it from the day the officer closes the entry. Neither basis reworks the other's log.\n\nThe 8-point distance is smaller than the movement the log shows between two rounds, so the department's figure is printed as logged and the register supplies no revision to it and no requests its round did not sample.\n\nExcluded are requests the register finds reopened in the same month, 150 of them, since the log carries them once and the register twice. Which requests each source counts is recorded in the annex's method note.\n\nMention: \"Sentinel Service Delivery System (SSDS)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Sentinel Service Delivery System checks the department's share of requests answered within its stated limit.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:dc5eb717c595cd868a902c21", "occurrence_id": "91fe0c8af614efb2d4022808", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The facility figure in this chapter is the ministry's own, taken from its establishment list, and the Joint Facility Assessment Census (JFAC), the census the IMPACT Initiative follows for China, is cited to test it.\n\nThe ministry's list shows 95 per cent of its facilities offering the full service package. The census shows 87 per cent for the same counties, because it scores a facility on what the assessment team saw in use, while the ministry's list carries the package the facility is established to offer. Neither basis was applied to the other's list.\n\nThe 8-point distance is smaller than the movement the list shows between two years, so the ministry's figure is printed as listed and the census supplies no revision to it and no facilities its teams did not visit.\n\nExcluded are facilities the census finds under construction at the visit, 240 of them, since the list counts them as open. Which facilities each source counts is stored in the annex's method sheet.\n\nMention: \"Joint Facility Assessment Census (JFAC)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Joint Facility Assessment Census checks the ministry's share of facilities offering the full service package.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:ab8fe72bdfaf085503532806", "occurrence_id": "09af0d0a4217e975a1548ae6", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Informal work and licensing were put together for this review from the Rural Informal Sector System (RISS), Version 2, the register kept for Sulaymaniya in 2023, and the readings below are the review's own.\n\nUnits were grouped by the months the register records a unit trading without a licence and the share the register finds registering within the following year computed: 34 per cent of units under a year unlicensed against 11 per cent of those above three years. The register carries a licence date for each unit and a trading start for the same unit without ever joining the two.\n\nMaking the groups alike on the size of the unit's holding leaves 7 points of that distance, which leaves the holding as part of the gap the opening comparison puts down to the months unlicensed.\n\nUnits the register lists as closed during the year are dropped rather than counted as unregistered, and they are 6 per cent of units. The annex note carries the grouping rule, the pull date together with the base of each share.\n\nMention: \"Rural Informal Sector System (RISS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Rural Informal Sector System relates months trading unlicensed to units registering within the following year.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:100c8fc7bf686c7b5e32f9bb", "occurrence_id": "03e4ac337142020c6850f34e", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The credit figures in this chapter are drawn from the bank's own loan book, and the Regional Rural Credit Census, the census OCHA conducted in Fiji in 2020, is quoted only to place those figures in context.\n\nThe census puts the share of holdings holding a loan from any formal lender at 28 per cent, the mean outstanding balance a borrowing holding carries at 900, and the share of holdings turned away when last they asked at 17 per cent. All three are given as the census issues them, with its own holding list stated, and no chapter figure was built from them.\n\nTheir use here is to show how rural borrowing stands across the region beside the six branches whose loan book the chapter works from. The chapter's argument rests on that book, which names each loan rather than sampling holdings at a visit, and the two are kept apart rather than merged.\n\nThe annex lists the census tables consulted and the day of the pull, which is the full extent of the census's use in this chapter.\n\nMention: \"Regional Rural Credit Census\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Regional Rural Credit Census is quoted as context beside the bank's rural loan book.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:892a0525479cc52a945745d2", "occurrence_id": "a2420b8b8040159efc3a435f", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Tertiary education the continent figures are tabulated by the Civil Tertiary Education Survey (CTES), Release 4, the survey followed across Africa for 2016, and the four values reach this report unchanged.\n\nThe survey puts the share of young people enrolled in tertiary study at 9 per cent, the share of enrolled students at a private institution at 24 per cent, the mean annual fee a family reports at 1 800, and the share of graduates employed within a year of leaving at 46 per cent. None of them was recalculated here, and the survey's own reading of enrolment, which requires the institution to hold the student on its register at the visit, is carried alongside them.\n\nA student on a year of placement was counted as enrolled, and a student who had paid but not registered was not. Each figure keeps the country it was drawn from shown with it, instead of combined into one continent figure.\n\nBeneath the table sit the survey row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Civil Tertiary Education Survey (CTES)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Civil Tertiary Education Survey supplies tertiary enrolment, private share, annual fees and employment within a year.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:b908623893d60375020fdb78", "occurrence_id": "25985e4db0a8531f3ece2019", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Whether the office's own count of filed returns holds was tested against the Local Public Sector Monitoring Registry, the register MODIS maintains for Kyrgyzstan in 2015, and the register is the principal source for the test.\n\nThe office's control sheet gives 1 040 returns filed by the close of the quarter. The register gives 1 190 for the same departments, because it counts a return filed when the register's own log receives it, which the office's sheet records only when the department's officer signs it off. No basis was used on the other's count.\n\nThe register therefore carries the chapter's count of filed returns, and the office's smaller figure is printed beside it so the reader can see that the distance is the sign-off rule rather than a disagreement about the quarter. Where a return appears in both, the office's sign-off date is kept and the two counts are not added.\n\nReturns the register holds a date for but no department were dropped from both rather than counted by the register alone, 6 per cent of its listings. The sign-off rule each source works to is listed in the annex notes.\n\nMention: \"Local Public Sector Monitoring Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Local Public Sector Monitoring Registry tests the office's count of returns filed by the close of the quarter.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:a6edca982096c4518f3d1a0f", "occurrence_id": "12529c63eb8a11f5285a2754", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Whether the ministry's own count of non-state schools holds was tested against the Sentinel Non-State Education System (SNES), Collection 4, the register kept for Ecuador in 2017, and the register is the principal source for the test.\n\nThe ministry's permit file gives 830 schools operating under a permit. The register gives 965 for the same provinces, because it counts a school operating where it enrolled pupils at the collection, which the ministry's file records only where the permit had not lapsed. The bases stay separate: neither is applied to the other's count.\n\nThe register therefore carries the section's count of non-state schools, and the ministry's smaller figure is printed beside it so the reader can see that the distance is the permit rule rather than a disagreement about the year. Where a school appears in both, the ministry's permit date is kept and the two counts are not added.\n\nSchools the register lists with an enrolment but no province were dropped from both rather than counted by the register alone, 5 per cent of its listings. The permit rule each source works to is set down in the annex file.\n\nMention: \"Sentinel Non-State Education System (SNES)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Sentinel Non-State Education System tests the ministry's count of schools operating under a permit.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:c4bc409e9c2e1a07ab920cf3", "occurrence_id": "1b8675b3aa1bdc3c77f30421", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Three operations lines in the control annex come from the Representative Operations Monitoring Registry, the register SPAM maintains for Eritrea in 2020, and all three are copied across unchanged.\n\nThe register puts the share of units whose monthly return reached the centre at 71 per cent, the mean days a return waits at a district office before moving at 8, and the share of returns the centre could reconcile with the unit's own ledger at 49 per cent. Nothing was recomputed for this report, and the register's own reading of a reached return, which requires the centre's own receipt, travels with the figures.\n\nThe annex's other lines are compiled from the unit's own diary, so the register carries three lines while the diary carries the lines the table does not. A figure reported for one region keeps that region attached rather than being averaged across the country, and where register and diary disagree about a movement the register's figure is kept for these three lines and the diary's for the rest, the difference being noted beneath the table rather than smoothed away.\n\nUnits the register lists as suspended within the year are excluded from all three lines, since the diary still counts them. The annex log carries the register row behind each value, the day the rows were taken and the counts each share divides.\n\nMention: \"Representative Operations Monitoring Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Representative Operations Monitoring Registry supplies returns reaching the centre, district delay and reconciled returns.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:61815b7ad8d6d30ed8f23ad3", "occurrence_id": "9cf0dfa995f9e43c0a3d3bc6", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Patient movement and bed use were put together for this review from the Rolling Patient Flow Survey, which FISE conducted in Egypt in 2015, and the readings below are the review's own.\n\nWards were grouped by the mean days the survey records a patient waiting for a bed and the share of beds the survey finds occupied at the visit computed: 88 per cent of wards waiting under a day against 96 per cent of those waiting three days or more. The survey carries a wait for each admission and a bed count for the same ward without ever joining the two.\n\nMaking the groups alike on the number of nurses the survey records on duty leaves 4 points of that distance, which leaves nursing as part of the gap the opening comparison puts down to the wait.\n\nWards the survey finds closed for works on the day of visit are dropped rather than counted at their usual occupancy, and they are 5 per cent of wards. The annex note carries the grouping rule, the pull date plus the base of each share.\n\nMention: \"Rolling Patient Flow Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Rolling Patient Flow Survey relates the wait for a bed to bed occupancy at the ward at the visit.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:4012d0343a62d44791cea1d1", "occurrence_id": "bf7196d4baf232b6e2081b9a", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The water stress figure in this chapter is the authority's own, taken from its abstraction ledger, and the Joint Water Stress Dataset (JWSD), Round 8, the geospatial dataset compiled for Tunisia in 2022, is cited to test it.\n\nThe authority's ledger shows 58 per cent of its abstraction licences standing over the basin's allotment. The dataset shows 66 per cent for the same basins, because it measures the draw at the point of abstraction, while the ledger carries the volume the licence grants rather than the volume taken. Each basis keeps its own ledger untouched.\n\nThe 8-point distance is smaller than the movement the ledger shows between two rounds, so the authority's figure is printed as ledged and the dataset supplies no revision to it and no licences for the basins its round did not observe.\n\nExcluded are licences the dataset places on two basins, 130 of them, since the allotment cannot be settled against either. Which licences each source counts is laid out in the papers behind the annex.\n\nMention: \"Joint Water Stress Dataset (JWSD)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Joint Water Stress Dataset checks the authority's share of abstraction licences standing over the basin allotment.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:fad7a28594b8990617fdbf0d", "occurrence_id": "af890243c2c1ef90f104c057", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Rural livelihood the figures for the frame are reported by the Pilot Rural Livelihoods Survey (PRLS), Collection 9, the survey Jacobs et al follows for 2022, and the four values reach this report unchanged.\n\nThe survey puts the share of working adults whose main work is on their own holding at 54 per cent, the share taking paid work off the holding in the season at 29 per cent, the mean months a household goes without any off-holding pay at 5, and the share of households holding a written contract for that work at 17 per cent. None of them was recalculated here, and the survey's own reading of main work, which requires the adult to name it as the greater part of the year, is printed beside them.\n\nWork done for a relative in exchange for crop was counted as off-holding work, and a season's work at a mill without pay was not. Each figure keeps the ward it was drawn from reported with it, rather than blended into a single frame share.\n\nBeneath the table sit the survey row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Pilot Rural Livelihoods Survey (PRLS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Pilot Rural Livelihoods Survey supplies own-holding work, off-holding pay, unpaid months and written contracts.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:844f23ea271806f2043a821c", "occurrence_id": "876448e3f1daa20733449087", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Enrolment and retention figures covering the city are drawn from the Baseline Enrolment and Retention System (BERS), Second Phase, the register kept for Riga in 2017, and the four values reach this report unchanged.\n\nThe register puts the number of pupils held on a school's roll at the phase at 84 000, the share still on a roll at the same school a year later at 91 per cent, the mean moves a pupil makes between schools within the phase at 1.3, and the share of pupils the register holds a previous school for at 63 per cent. None of them was recalculated here, and the register's own reading of retention, which requires the roll to carry the pupil at both phases, is given next to them.\n\nA pupil on a year at a sister school was counted as retained, and a pupil whose family moved out of the city was counted as lost. Each figure keeps the district it was drawn from kept with it, instead of being rolled up to the city.\n\nBeneath the table sit the register row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Baseline Enrolment and Retention System (BERS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Baseline Enrolment and Retention System supplies roll size, year-on-year retention, school moves and traceable previous schools.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:aa3ea01b9b7a96bd13cf7dc9", "occurrence_id": "56910fa44de54d95c1be41bd", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Flood extent figures covering the whole city derive from the Periodic Flood Extent Registry, the register the Chinese Wind Energy Association maintains for Bhopal in 2018, and the four values reach this report unchanged.\n\nThe register puts the share of the frame the register held under water at some point in the season at 19 per cent, the mean days a flooded area the register maps stays under water at 6, the share of the flooded area falling inside a ward the register lists as drained at 44 per cent, and the share of the frame whose extent changed between the registry's own readings at 8 per cent. None of them was recalculated here, and the register's own reading of flood, which requires water standing over ground it holds dry in the previous reading, is bound to the figures.\n\nPonding behind a road embankment was counted, and water standing only on a road the register classes as unsealed was not. Each figure keeps the ward it was drawn from named with it, instead of averaged over the city.\n\nBeneath the table sit the register row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Periodic Flood Extent Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Periodic Flood Extent Registry supplies flooded share, days under water, drained-ward overlap and reading-to-reading change.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:486215f2166c2300cce443b0", "occurrence_id": "cd01651178bd10eac95e10c3", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The skills figure in this chapter is the office's own, taken from its certificate file, and the Multi-Topic Skills and Employment Survey (MSES), the survey MOH follows for Timor-Leste, is cited to test it.\n\nThe office's file shows 62 per cent of its certificate-holders recorded in work three months after issue. The survey shows 53 per cent for the same districts, because it asks the holder whether paid work followed, while the office's file carries the employer's own placement note. Neither basis was carried across to the other's file.\n\nThe 9-point distance is smaller than the movement the file shows between two collections, so the office's figure is printed as filed and the survey supplies no revision to it and no holders its teams could not reach.\n\nExcluded are holders the survey finds enrolled in a second course, 210 of them, since the file counts them as waiting and the survey as studying. Which holders each source counts is set out in the working papers behind the annex.\n\nMention: \"Multi-Topic Skills and Employment Survey (MSES)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Multi-Topic Skills and Employment Survey checks the office's share of certificate-holders in work three months on.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:3ff6a5dcd3dd4810d3f585fa", "occurrence_id": "81cde2184009fe7e6838afdd", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Youth transition the numbers held for the region come from the Coordinated Youth Transition Index, the indicator Statistics Indonesia compiled for Sub-Saharan Africa in 2017, and the four values reach this report unchanged.\n\nThe index puts the share of young people neither in study nor in work at 24 per cent, the share in study beyond lower secondary at 18 per cent, the mean years between leaving school and first steady work at 2.6, and the share of young people in work holding a written contract at 31 per cent. None of them was recalculated here, and the index's own reading of steady work, which requires four months of the same employer, goes with the figures.\n\nA season's work on a relative's holding was not counted as work, and unpaid internship with a stipend was. Each figure keeps the country it was drawn from listed with it, rather than smoothed across the region.\n\nBeneath the table sit the index row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Coordinated Youth Transition Index\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Coordinated Youth Transition Index supplies inactivity, study beyond lower secondary, school-to-work years and contract cover.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:c2072747e13f2cbd1e2fca98", "occurrence_id": "5cf417b46e2cee2a060b8432", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The emission figure in this chapter is the inventory office's own, taken from its own sector returns, and the Quarterly Emission Inventory Dataset (QEID), Collection 5, the geospatial dataset compiled for Turkey in 2020, is cited to test it.\n\nThe office's returns show 410 megatonnes released in the year. The dataset shows 448 for the same provinces, because it counts a release where its own model places one at the plant, while the office's returns carry only what the operator filed. One basis was never used to adjust the other's returns.\n\nThe dataset therefore carries the section's total, and the office's smaller figure is printed beside it so the reader can see that the distance is the filing rule rather than a disagreement about the year. Where a plant appears in both, the office's filing date is kept and the two totals are not added.\n\nPlants the dataset places on a boundary between two provinces were dropped from both rather than counted by the dataset alone, 6 per cent of its listings. The filing rule each source works to is written up in the annex working papers.\n\nMention: \"Quarterly Emission Inventory Dataset (QEID)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Quarterly Emission Inventory Dataset tests the inventory office's national release total against operator filings.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:6583559419357ff2cb69d9c5", "occurrence_id": "e0ade15951f45bbe6ef91585", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Household income and spending records were catalogued into the study's working file from the Verified Household Income and Expenditure Registry (VHIER), Collection 7, the register kept for Spain in 2017, and the work done on them was cataloguing rather than measurement.\n\nEach household was given the file's own spending band and its reference dates checked against the registry's collection window; where the registry carries a household the file has no band for, it was parked for the analyst rather than entered under the nearest one. Households the registry lists without a spending total were left unbanded rather than assumed to have spent nothing, and 830 of them entered the file in that condition. The registry's own collection stamp was kept on every household so a later collection can be matched without re-deriving it.\n\nThe study's regional weights come from the statistical office's own population register, from which the registry supplied nothing; it is the household coding that lets those weights be read against spending for the distribution table the report draws up.\n\nThe coding sheet, the count of parked households and the day of deposit are held in the annex log.\n\nMention: \"Verified Household Income and Expenditure Registry (VHIER)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The Verified Household Income and Expenditure Registry supplies the coded household records behind the study's working file.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:6207b4254cb08766d2d7ba4b", "occurrence_id": "e4c948307fa93a2256bb0f54", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Durable solution frame figures are assembled from the Ministerial Durable Solutions Registry (MDSR), Seventh Phase, the register OSIPTEL follows, and the four values reach this report unchanged.\n\nThe register puts the share of displaced households the register holds a solution for at 37 per cent, the mean months a household waits before its solution is recorded at 26, the share of recorded solutions the register classes as local integration at 44 per cent, and the share of households whose solution the register later reopened at 7 per cent. None of them was recalculated here, and the register's own reading of a solution, which requires the household's own consent entered in the file, travels with the values.\n\nA household resettled twice was counted at its latest entry, and a household whose return the register could not confirm was counted as open. Each figure keeps the area it was drawn from carried with it, rather than mixed across the frame.\n\nBeneath the table sit the register row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Ministerial Durable Solutions Registry (MDSR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Ministerial Durable Solutions Registry supplies solutions recorded, waiting months, integration share and reopened cases.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:264527ae73cc2ffb41460adf", "occurrence_id": "b542059ab08615c19e43e36c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Three delivery lines in the service annex come from the Baseline Service Delivery Index (BSDI), Release 7, the indicator Lane Financial compiled for 2021, and all three are copied across unchanged.\n\nThe index puts the share of facilities meeting its own staffing norm at 58 per cent, the mean days a request waits before the office answers it at 12, and the share of facilities publishing their service standard where the public can read it at 39 per cent. Nothing was recomputed for this report, and the index's own reading of an answered request, which requires the answer to reach the applicant, travels with the figures.\n\nThe annex's other lines are compiled from the authority's own log, so the index carries three lines while the log carries the balance. A figure reported for one district keeps that district attached rather than being averaged across the frame, and where index and log disagree about an answer the index's figure is kept for these three lines and the log's for the rest, the difference being noted beneath the table rather than smoothed away.\n\nFacilities the index lists as newly opened within the release are excluded from all three lines, since the log carries no period for them. The annex log carries the index row behind each value, the day the extract ran, and the counts under each share.\n\nMention: \"Baseline Service Delivery Index (BSDI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"source\", \"summary\": \"The Baseline Service Delivery Index supplies facilities meeting the staffing norm, request delay and published standards.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:e8ab6829a1704531bbdf2fef", "occurrence_id": "330a12ab1aa7438d3dd37b10", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Multidimensional poverty figures for the urban frame are reported from the Urban Multidimensional Poverty Survey (UMPS), the survey conducted in Samoa in 2020, and the four values reach this report unchanged.\n\nThe survey puts the share of households poor on its own indicator set at 17 per cent, the share holding an unmet medical need at 22 per cent, the mean number of indicators a poor household fails at 4, and the share of poor households failing a health indicator at 61 per cent. None of them was recalculated here, and the survey's own reading of poverty, which requires a household to fail a third of its indicators, follows the figures.\n\nA household failing an indicator only because its member was away at the visit was still counted, and a need met by a relative was counted as met. Each figure keeps the ward it was drawn from with it, rather than pooled across the frame.\n\nBeneath the table sit the survey row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Urban Multidimensional Poverty Survey (UMPS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Urban Multidimensional Poverty Survey supplies indicator poverty, unmet medical need, indicators failed and health failure among poor households.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:64d457b4b1de2233aed41ea9", "occurrence_id": "8b78fb8e5f26b93254d2275b", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Working hours and adequacy were put together for this review from the Comparative Working Conditions Survey, which TRMM conducted in Senegal in 2020, and the readings below are the review's own.\n\nWorkers were grouped by the hours the survey records them working in the longest week and the share whose household stands below the national adequacy line computed: 29 per cent of workers above sixty hours against 11 per cent of those under forty. The survey carries an hours item for each worker and a consumption aggregate for the same household without ever joining the two.\n\nMaking the groups alike on the wage the worker reports leaves 6 points of that distance, which leaves the wage as part of the gap the opening comparison puts down to the hours.\n\nWorkers the survey lists as working a second job without pay were dropped rather than counted at their paid hours, and they are 5 per cent of workers. The annex note carries the grouping rule, the pull date, with the base of each share alongside.\n\nMention: \"Comparative Working Conditions Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Comparative Working Conditions Survey relates hours in the longest week to households below the adequacy line.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:569f05b478742affbefab73d", "occurrence_id": "99176c2a8f8af4ef9eb06414", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Elevation and household conditions for the frame are reported from the Record-Based Elevation Model Census (REMC), First Version, the census Stanford University follows, and the four values reach this report unchanged.\n\nThe census puts the share of settlements the census places above one thousand metres at 23 per cent, the mean slope per kilometre the census records between a settlement and its nearest road at 46 metres, the share of households in those settlements the census holds without a written tenure document at 58 per cent, and the share of high settlements whose elevation the census revised between versions at 6 per cent. None of them was recalculated here, and the census's own reading of a settlement, which requires a structure standing at the enumeration visit, is reported with them.\n\nA hamlet occupied only in the dry season was counted, and a structure roofed with palm leaf was counted as a structure. Each figure keeps the district it was drawn from attached, instead of being blended across the frame.\n\nBeneath the table sit the census row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Record-Based Elevation Model Census (REMC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Record-Based Elevation Model Census supplies high-settlement share, slope to road, tenure documents and elevation revisions.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:5b5bf0fa65a1ec85ec6ff15a", "occurrence_id": "cf209a720f39d55392e71008", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Integrated records figures spanning the city are taken from the Comparative Integrated Records System, the register CONARE maintains for Addis Ababa in 2016, and the four values reach this report unchanged.\n\nThe register puts the share of offices holding their series on a list the register accepts at 54 per cent, the mean years a series waits before appraisal at 9, the share of series held in a format the reading room can open at 41 per cent, and the share of offices whose register was reviewed within the year at 27 per cent. None of them was recalculated here, and the register's own reading of an accepted list, which requires the signature of the office head, is attached to them.\n\nA list signed by a predecessor was still counted, and a series held only on a personal machine was counted as unlisted. Each figure keeps the sub-city it was drawn from shown with it, instead of combined into one city figure.\n\nBeneath the table sit the register row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Comparative Integrated Records System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Comparative Integrated Records System supplies accepted series lists, appraisal waits, readable formats and reviewed registers.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:2f3785180063d258659a4004", "occurrence_id": "35224e57e7006c6b4b212e69", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Records series were catalogued into the review's inventory from the Decennial Public Records Registry, the database ITESM compiled for Moscow in 2018, and the work done on them was cataloguing rather than measurement.\n\nEach series was given the inventory's own retention band and its review dates checked against the registry's reference window; where the registry carries a series the inventory has no band for, it was parked for the analyst rather than entered under the nearest one. Series the registry lists without a holding office were left unbanded rather than assigned to the nearest archive, and 310 of them entered the inventory in that condition. The registry's own deposit stamp was kept on every series so a later deposit can be matched without re-deriving it.\n\nThe inventory's volume counts come from the archive's own accession book, from which the registry supplied nothing; it is the series coding that lets those counts be read against retention for the appraisal table the report draws up.\n\nThe coding sheet, the count of parked series and the day of deposit are detailed in the annex log.\n\nMention: \"Decennial Public Records Registry\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The Decennial Public Records Registry supplies the coded series behind the review's inventory.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:ac5626c5f4beb54ec6f862c0", "occurrence_id": "a64ab9946c6dc04c3b401c56", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Building footprints and tenure were reconstructed for this review from the Linked Building Footprint Dataset (LBFD), Release 3, the geospatial dataset TRMM maintains, and the readings here are the review's, not the dataset's.\n\nStructures were matched to the footprint the dataset traces for them and grouped by whether that footprint changed between releases, and the share the office holds a tenure document for computed: 67 per cent of structures whose footprint stands unchanged against 49 per cent of those whose footprint moved. The dataset holds a footprint per structure and the office records a document per holding, and neither reports the two together.\n\nHolding the size of the structure even leaves 6 points of that distance, so the size accounts for part of what the opening comparison puts down to the change.\n\nStructures the dataset traces twice within the release are dropped rather than counted as changed, and they are 4 per cent of structures. The annex note carries the matching rule, the pull date and the base of each share.\n\nMention: \"Linked Building Footprint Dataset (LBFD)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Linked Building Footprint Dataset relates footprint change between releases to structures holding a tenure document.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:ccbe52bf9200cfff13d07ac8", "occurrence_id": "f57b7ccd16c23061642de1f1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Adequacy and hardship were put together for this review from the Record-Based Poverty and Vulnerability Survey (RPVS), Wave 2, the survey Amadeus follows, and the readings below are the review's own.\n\nHouseholds were grouped by the number of hardship items the survey records against them and the share standing below the survey's adequacy line computed: 64 per cent of households with four items or more against 14 per cent of those with none. The survey carries a hardship list for each household and an adequacy measure for the same household without ever joining the two.\n\nMaking the groups alike on the size of the household leaves 16 points of that distance, which leaves household size as part of the gap the opening comparison puts down to the items.\n\nHouseholds the survey records an item for but no response to the adequacy module were dropped rather than counted as poor, and they are 5 per cent of households. The annex note carries the grouping rule, the pull date and, alongside it, the base of each share.\n\nMention: \"Record-Based Poverty and Vulnerability Survey (RPVS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Record-Based Poverty and Vulnerability Survey relates hardship items to households below the survey's adequacy line.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:64870ea4409c9fa77c9945b3", "occurrence_id": "b95f7f9d1c8c53c5bbd11fe1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Records coverage and appraisal were put together for this review from the Representative Integrated Records System (RIRS), Third Release, the register kept for Italy in 2019, and the readings below are the review's own.\n\nOffices were grouped by the share of its series the release finds listed on an accepted schedule and the mean years the release records a series waiting for appraisal computed: 11 years at offices above three-quarters listed against 23 years at those below half. The release carries a schedule flag for each series and an appraisal date for the same series without ever joining the two.\n\nMaking the groups alike on the number of series an office holds leaves 4 years of that distance, which leaves the number of series as part of the gap the opening comparison puts down to the listing.\n\nOffices the release lists with series but no appraisal dates are dropped rather than counted as never appraised, and they are 6 per cent of offices. The annex note carries the grouping rule, the pull date as well as the base of each figure.\n\nMention: \"Representative Integrated Records System (RIRS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Representative Integrated Records System relates series listed on accepted schedules to appraisal waiting years.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:612ba8d21fb6279648c3a99c", "occurrence_id": "6474458b0d872a6499778858", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Climate and flood risk figures at the frame level are taken from the Quality-Assured Climate and Flood Risk Registry (QCFRR), Version 2, the register WTO follows, and the four values reach this report unchanged.\n\nThe registry puts the share of the frame the registry rates at high flood risk at 14 per cent, the mean years the registry holds a risk reading back for a rated unit at 9, the share of rated units whose reading the registry drew from a gauge rather than a model at 46 per cent, and the share of units whose rating changed between versions at 8 per cent. None of them was recalculated here, and the registry's own reading of high risk, which requires a one-in-a-hundred-year event to reach the unit, is stated with them.\n\nA unit protected by a embankment the registry holds unbuilt was still rated, and a unit flooded only by surface water was rated separately. Each figure keeps the region it was drawn from reported with it, rather than blended into a single frame share.\n\nBeneath the table sit the registry row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Quality-Assured Climate and Flood Risk Registry (QCFRR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Quality-Assured Climate and Flood Risk Registry supplies high-risk share, reading depth, gauge-based readings and rating change.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:231f0dcd512952de996dfdeb", "occurrence_id": "f29a57dfc78613574a7840c1", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Spending and heat were derived for this chapter from the Harmonised Household Income and Expenditure Survey (HHIES), Version 4, the survey SPAM follows for France, and the figures below are the chapter's own rather than the survey's.\n\nHouseholds were grouped by the share of its spending the survey records on cooling and the share standing below the national adequacy line computed: 33 per cent of households spending nothing on cooling against 12 per cent of those spending above a tenth. The survey carries a spending line for each household and an adequacy measure for the same household without ever joining the two.\n\nThat comparison supports one claim, that the heat warning should reach households without cooling first rather than the oldest households alone. The equipment lines the proposal quotes come from the utility's own load register, from which the survey supplied nothing.\n\nHouseholds the survey records in a dwelling with no fixed cooling provision are dropped rather than counted as spending nothing, and they are 7 per cent of households. The bands and the counts behind each share are recorded beneath the table.\n\nMention: \"Harmonised Household Income and Expenditure Survey (HHIES)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"analyze\", \"summary\": \"The Harmonised Household Income and Expenditure Survey supports a cooling-led heat warning built from cooling spending and adequacy incidence.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:1af92d9f3c47f0463855f87b", "occurrence_id": "b37d103a753742aeecf96b2d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Service register entries were catalogued into the review's inventory from the District Service Register Registry, the database Penn World table maintains for Lebanon in 2020, and the work done on them was cataloguing rather than measurement.\n\nEach entry was given the inventory's own service band and its review dates checked against the registry's reference window; where the registry carries an entry the inventory has no band for, it was parked for the analyst rather than entered under the nearest one. Entries the registry lists without a holding office were left unbanded rather than assigned to the nearest district, and 270 of them entered the inventory in that condition. The registry's own round stamp was kept on every entry so a later round can be matched without re-deriving it.\n\nThe inventory's coverage counts come from the ministry's own establishment return, from which the registry supplied nothing; it is the entry coding that lets those counts be read against service band for the coverage table the report draws up.\n\nThe coding sheet, the count of parked entries and the day of deposit are carried in the annex log.\n\nMention: \"District Service Register Registry\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"curate\", \"summary\": \"The District Service Register Registry supplies the coded service entries behind the review's inventory.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:ccebea8a28871336b4287b50", "occurrence_id": "b4bc575fa1e6e1f27be9e299", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Records coverage and revision were put together for this review from the Quality-Assured Public Records Registry (QPRR), Round 4, the register kept for South Asia in 2018, and the readings below are the review's own.\n\nInstitutions were grouped by the share of its series the round finds held under a retention band and the share of its published tables the round finds revised within a year computed: 28 per cent of institutions above three-quarters banded against 9 per cent of those below half. The round carries a band flag for each series and a revision date for the same table without ever joining the two.\n\nMaking the groups alike on the number of series an institution holds leaves 6 points of that distance, which leaves the number of series as part of the gap the opening comparison puts down to the banding.\n\nInstitutions the round lists with series but no published tables are dropped rather than counted as never revising, and they are 5 per cent of institutions. The annex note carries the grouping rule, the pull date together with the base of each share.\n\nMention: \"Quality-Assured Public Records Registry (QPRR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Quality-Assured Public Records Registry relates series held under retention bands to published tables revised within a year.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:532a6570645aa9155913f349", "occurrence_id": "d950d076ab31da43f26e0183", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Minimum needs and medical need were reconstructed for this review from the High-Frequency Minimum Needs Survey (HMNS), Wave 7, the survey New York Fed follows for Russia, and the readings here are the review's, not the survey's.\n\nHouseholds were grouped by the number of basket items the survey records them going without and the share reporting a medical need they did not act on computed: 52 per cent of households without three items or more against 15 per cent of those without none. The survey carries a basket list for each household and a medical item for the same household without ever joining the two.\n\nHolding the age of the household head even leaves 14 points of that distance, so age accounts for part of what the opening comparison puts down to the basket.\n\nHouseholds the survey records a basket item for but no answer to the medical module were dropped rather than counted as unmet, and they are 5 per cent of households. The annex note carries the grouping rule, the pull date plus the base of each share.\n\nMention: \"High-Frequency Minimum Needs Survey (HMNS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The High-Frequency Minimum Needs Survey relates basket items gone without to medical needs left unacted on.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:d0b285d38446bf468ff8abee", "occurrence_id": "fcda32880ea545b7456886cb", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: School spending and health were put together for this review from the Nationwide Education Finance Survey (NEFS), Release 8, the survey CONDUSEF follows, and the readings below are the review's own.\n\nSchools were grouped by the annual spending per pupil the survey records and the share of its pupils the survey finds screened for vision in the year computed: 79 per cent of schools above the national median against 38 per cent of those below a third of it. The survey carries a spending line for each school and a screening item for the same school without ever joining the two.\n\nMaking the groups alike on the number of pupils a school holds leaves 12 points of that distance, which leaves the roll as part of the gap the opening comparison puts down to the spending.\n\nSchools the survey lists with spending but no pupils at the visit are dropped rather than counted at the district mean, and they are 4 per cent of schools. The annex note carries the grouping rule, the pull date, and the base of each share.\n\nMention: \"Nationwide Education Finance Survey (NEFS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Nationwide Education Finance Survey relates spending per pupil to pupils screened for vision in the year.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:feb3e452fa256fbfd75e8af8", "occurrence_id": "8951568a2498a098513e1710", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Transfer coverage and distance were put together for this review from the Coordinated Cash Transfer Registry Census (CCTRC), Sixth Release, the census UNDESA compiled for 2018, and the readings below are the review's own.\n\nAreas were grouped by the share of its eligible households the census finds on the register and the mean kilometres the census records to the nearest payment point computed: 9 kilometres at areas below half registered against 3 kilometres at those above nine-tenths. The census carries a registration flag for each household and a point location for the same area without ever joining the two.\n\nMaking the groups alike on the number of eligible households an area holds leaves 2 kilometres of that distance, which leaves the number of households as part of the gap the opening comparison puts down to the distance.\n\nAreas the census maps without a payment point are dropped rather than counted at the national mean, and they are 6 per cent of areas. The annex note carries the grouping rule, the pull date, with the base of each figure alongside.\n\nMention: \"Coordinated Cash Transfer Registry Census (CCTRC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Coordinated Cash Transfer Registry Census relates registration of eligible households to distance from a payment point.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:4ec9127c82bf387a111c7bb9", "occurrence_id": "1e1ab465cd264539d240ed8d", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Living standards island-level figures come from the Ministerial Living Standards Survey, the survey the Brazilian Ministry of Labor conducted in Cyprus in 2017, and the four values reach this report unchanged.\n\nThe survey puts the share of households below the national adequacy line at 16 per cent, the share of adults in a household with no paid work at 21 per cent, the mean monthly consumption of a household in the lowest fifth at 480, and the share of households unable to cover an unexpected cost of 500 at 27 per cent. None of them was recalculated here, and the survey's own reading of adequacy, which is measured on consumption after housing costs, comes with them.\n\nA household with a member on unpaid leave was counted as having that member out of paid work, and a cost met by borrowing was counted as not covered. Each figure keeps the district it was drawn from kept with it, instead of being rolled up to the island.\n\nBeneath the table sit the survey row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Ministerial Living Standards Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Ministerial Living Standards Survey supplies adequacy incidence, workless households, lowest-fifth consumption and shock coverage.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:130314c71cd7ba5ec8804664", "occurrence_id": "62317142def64304840f2aba", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The targeting figure in this chapter is the ministry's own, taken from its payment file, and the Multi-Topic Targeting and Payments Microdata (MTPM), the microdata kept for Europe and Central Asia in 2023, is cited to test it.\n\nThe ministry's file shows 81 per cent of its payments reaching a household its own score ranks in the bottom two fifths. The microdata shows 72 per cent for the same countries, because it ranks the household on what the survey interview records, while the ministry's file ranks it on the register entry. The two bases were not applied to each other's file.\n\nThe 9-point distance is smaller than the movement the file shows between two rounds, so the ministry's figure is printed as filed and the microdata supplies no revision to it and no households whose interview was never completed.\n\nExcluded are households the microdata finds listed under two ministries, 340 of them, since the two records cannot be settled. Which households each source counts is detailed in the method note in the annex.\n\nMention: \"Multi-Topic Targeting and Payments Microdata (MTPM)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Multi-Topic Targeting and Payments Microdata checks the ministry's share of payments reaching the bottom two fifths.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:372753f192ac43c226693ab8", "occurrence_id": "a1f82f1eb83413b3a9f89e3c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Time use and adequacy figures for the small areas are reported from the Small-Area Time Use Survey (STUS), Second Wave, the survey DANE follows, and the four values reach this report unchanged.\n\nThe survey puts the mean hours a week an adult reports in paid work at 38, the mean hours a week women report for unpaid care at 26, the share of adults reporting no paid work in the reference week at 29 per cent, and the share of households with no adult in paid work at 21 per cent. None of them was recalculated here, and the survey's own reading of paid work, which requires the hour to be reported against a named employer or customer, is kept with them.\n\nWork on the family holding sold for cash was counted as paid work, and help for a neighbour in exchange for food was not. Each figure keeps the small area it was drawn from named with it, instead of averaged over the frame.\n\nBeneath the table sit the survey row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Small-Area Time Use Survey (STUS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Small-Area Time Use Survey supplies paid hours, women's care hours, adults without paid work and workless households.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:7c574df1e183d1d6b23257c1", "occurrence_id": "394343cdbf26645d6cf198c6", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The road figure in this chapter is the authority's own, taken from its own maintenance schedule, and the Rural Road Segment Collection (RRSC), Third Edition, the dataset DMSP/OLS follows, is cited to test it.\n\nThe authority's schedule shows 68 per cent of its segments standing at their classed surface. The collection shows 59 per cent for the same districts, because it reads a segment from the surface it observes at the pass, while the schedule carries the class recorded when the segment was last graded. Neither basis reworks the other's schedule.\n\nThe 9-point distance is smaller than the movement the schedule shows between two editions, so the authority's figure is printed as scheduled and the collection supplies no revision to it and no segments for the districts its round did not pass.\n\nExcluded are segments the collection places on two districts at once, 120 of them, since the class cannot be settled against either. Which segments each source counts is kept in the annex's method note.\n\nMention: \"Rural Road Segment Collection (RRSC)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Rural Road Segment Collection checks the authority's share of segments standing at their classed surface.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:a35666cde5192c287aaa7499", "occurrence_id": "d888efa25d467c9c14210bd8", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Administrative record figures for each part of the frame are drawn from the Follow-Up Administrative Records Registry (FARR), Fifth Round, the register Fitch follows, and the four values reach this report unchanged.\n\nThe register puts the share of records the register follows up within its stated interval at 63 per cent, the mean months between a record's opening and its follow-up at 7, the share of followed-up records the register holds closed at 74 per cent, and the share of records the register reopened after closing at 9 per cent. None of them was recalculated here, and the register's own reading of a follow-up, which requires the officer's own entry against the record, accompanies the figures.\n\nA follow-up made by telephone was counted, and a reminder sent without a reply was not. Each figure keeps the office it was drawn from listed with it, rather than smoothed across the frame.\n\nBeneath the table sit the register row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Follow-Up Administrative Records Registry (FARR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Follow-Up Administrative Records Registry supplies timely follow-ups, opening-to-follow-up months, closures and reopenings.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:dd2bcfbe67b80831916766ec", "occurrence_id": "7829e62f9556b3dd030f4c75", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Informal trade figures measured across the frame are read from the Recurrent Informal Trade Registry (RITR), the register HLP AoR maintains for 2022, and the four values reach this report unchanged.\n\nThe register puts the share of trading places the register holds without a permit at 57 per cent, the mean days a place trades between two register visits at 46, the share of places the register finds moved on between visits at 12 per cent, and the share of places whose trader the register holds a second place for at 8 per cent. None of them was recalculated here, and the register's own reading of a trading place, which requires goods to change hands at the spot on the visit, comes down with them.\n\nA place trading only after dark was counted, and a place used for storage only was not. Each figure keeps the ward it was drawn from carried with it, rather than mixed across the frame.\n\nBeneath the table sit the register row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Recurrent Informal Trade Registry (RITR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Recurrent Informal Trade Registry supplies unlicensed places, days between visits, places moved on and traders holding a second place.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:6f05e19124a902408506934a", "occurrence_id": "e76f1bde5bc6554dd46998e6", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The operations figures in this chapter are drawn from the council's own control sheet, and the Urban Operations Monitoring Index (UOMI), Round 4, the indicator kept for Dar-es-Salaam in 2018, is referred to only as background to them.\n\nThe index puts the share of depots filing their weekly sheet at 72 per cent, the mean hours a job the index records stays open at 21, and the share of sheets the index could reconcile with the depot's own book at 51 per cent. All three are given as the index issues them, with its own depot list stated, and no chapter figure rests on them.\n\nNo chapter table is built from its sheets. The figures the chapter argues from come from the council's control sheet, which names each job rather than sampling depots at a visit, and where the two accounts of an open job differ the difference is noted rather than reconciled.\n\nThe annex lists the index tables consulted and the pull date, and the index is used for nothing beyond that background.\n\nMention: \"Urban Operations Monitoring Index (UOMI)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"inform\", \"summary\": \"The Urban Operations Monitoring Index is background to the council's own operations control sheet.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:6cdc807f2bd0e92ee10bde87", "occurrence_id": "7ee113bc2473d87ae3dc8e41", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Targeting and adequacy were put together for this review from the Civil Targeting and Payments Survey (CTPS), Version 6, the survey DOLISA follows, and the readings below are the review's own.\n\nHouseholds were grouped by whether the survey finds them reached by a payment and the share standing below the survey's adequacy line computed: 45 per cent of households reached by no payment against 24 per cent of those reached by one. The survey carries a payment item for each household and an adequacy measure for the same household without ever joining the two.\n\nMaking the groups alike on the score the survey computes for the household leaves 7 points of that distance, which leaves the score as part of the gap the opening comparison puts down to the payment.\n\nHouseholds the survey finds registered but never paid are dropped rather than counted as reached, and they are 6 per cent of households. The annex note carries the grouping rule, the pull date and the base of each share.\n\nMention: \"Civil Targeting and Payments Survey (CTPS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Civil Targeting and Payments Survey relates being reached by a payment to households below the adequacy line.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:76ad38317affe315b6ef96f5", "occurrence_id": "5acd1beeb1705dccefef80af", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Living standards and work were reconstructed for this review from the Combined Living Standards Collection (CLSC), the collection DAC compiled for 2018, and the readings here are the review's, not the collection's.\n\nHouseholds were grouped by the number of adults the collection records in paid work and the share standing below the national adequacy line computed: 57 per cent of households with no adult in paid work against 9 per cent of those with two or more. The collection carries a work item for each adult and a consumption aggregate for the same household without ever joining the two.\n\nHolding the number of children in the household even leaves 12 points of that distance, so the children account for part of what the opening comparison puts down to the paid work.\n\nHouseholds the collection records an adult in unpaid family work for are dropped rather than counted as workless, and they are 5 per cent of households. The annex note carries the grouping rule, the pull date and, alongside it, the base of each share.\n\nMention: \"Combined Living Standards Collection (CLSC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Combined Living Standards Collection relates adults in paid work to households below the national adequacy line.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:27a8759cf7a176b033e10975", "occurrence_id": "ec0bdbe144c820d54d1b65ea", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Operations index figures for the region are reported from the Decennial Operations Index Registry (DOIR), Wave 8, the register kept for Amhara region in 2019, and the four values reach this report unchanged.\n\nThe register puts the share of woredas filing their operations return within the wave at 66 per cent, the mean weeks a filed return waits before the register logs it at 5, the share of returns carrying a figure the register could check at 48 per cent, and the share of woredas whose return the register queried before acceptance at 19 per cent. None of them was recalculated here, and the register's own reading of a logged return, which requires the register's own date stamp, is carried alongside them.\n\nA return handed to a field officer was counted once logged rather than on handover, and a return withdrawn by its woreda was not counted. Each figure keeps the zone it was drawn from with it, rather than pooled across the region.\n\nBeneath the table sit the register row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Decennial Operations Index Registry (DOIR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Decennial Operations Index Registry supplies timely filings, logging delay, checkable figures and queried returns.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:375a99ae1277f23b32eabf1c", "occurrence_id": "ba09feedcf56a9e1c5b08a95", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Boundary records were catalogued into the review's inventory from the Triennial Administrative Boundary Dataset (TABD), Third Phase, the geospatial dataset compiled for Mexico in 2023, and the work done on them was cataloguing rather than measurement.\n\nEach boundary was given the inventory's own revision band and its dates checked against the dataset's reference window; where the dataset carries a boundary the inventory has no band for, it was parked for the analyst rather than entered under the nearest one. Boundaries the dataset traces without a governing decree were left unbanded rather than assigned to the nearest instrument, and 240 of them entered the inventory in that condition. The dataset's own phase stamp was kept on every boundary so a later phase can be matched without re-deriving it.\n\nThe inventory's unit counts come from the statistical office's own register of divisions, from which the dataset supplied nothing; it is the boundary coding that lets those counts be read against revision for the harmonisation table the report draws up.\n\nThe coding sheet, the count of parked boundaries and the day of deposit are held in the annex log.\n\nMention: \"Triennial Administrative Boundary Dataset (TABD)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"curate\", \"summary\": \"The Triennial Administrative Boundary Dataset supplies the coded boundary records behind the review's inventory.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:3b086bc6e6600af02b229d2b", "occurrence_id": "6a9655928b2f5cf17efdc222", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Assets and adequacy were put together for this review from the Coordinated Asset Ownership System (CAOS), Wave 7, the register KNBS compiled for 2016, and the readings below are the review's own.\n\nHouseholds were grouped by the number of asset classes the register records for them and the share standing below the national adequacy line computed: 61 per cent of households holding one class or none against 13 per cent of those holding four or more. The register carries an asset list for each household and an adequacy measure for the same household without ever joining the two.\n\nMaking the groups alike on the education of the household head leaves 14 points of that distance, which leaves education as part of the gap the opening comparison puts down to the assets.\n\nHouseholds the register records holding a dwelling jointly with a non-member are dropped rather than counted as owning it, and they are 4 per cent of households. The annex note carries the grouping rule, the pull date as well as the base of each share.\n\nMention: \"Coordinated Asset Ownership System (CAOS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Coordinated Asset Ownership System relates asset classes held to households below the national adequacy line.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:a591913b4308ebfb88333dc5", "occurrence_id": "c266a94a3dc71b94f44f5d9c", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Whether the ministry's own count of workplaces meeting the heat rule holds was tested against the Continuous Working Conditions Survey, the survey UNHCR conducted in Ecuador in 2023, and the survey is the principal source for the test.\n\nThe ministry's return gives 74 per cent of inspected workplaces meeting the rule. The survey gives 63 per cent for the same provinces, because it asks the worker whether shade and drinking water stood available through the hottest shift, which the ministry's return records only where the inspector signed the sheet. Neither basis was applied to the other's count.\n\nThe survey therefore carries the section's count of workplaces meeting the rule, and the ministry's larger figure is printed beside it so the reader can see that the distance is the inspection rule rather than a disagreement about the year. Where a workplace appears in both, the ministry's inspection date is kept and the two counts are not averaged.\n\nWorkplaces the survey dates to a season rather than a shift were dropped from both rather than counted by the survey alone, 7 per cent of its listings. The inspection rule each source works to is laid out in the annex's method sheet.\n\nMention: \"Continuous Working Conditions Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"validate\", \"summary\": \"The Continuous Working Conditions Survey tests the ministry's count of workplaces meeting the heat rule.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:84d55e5aeb78bce50edc30ea", "occurrence_id": "64a3493a2aac09b3992fb490", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Asset ownership and adequacy were reconstructed for this review from the Ministerial Asset Ownership Survey, which the Tax Justice Network conducted in Dominican Rep. in 2021, and the readings here are the review's, not the survey's.\n\nHouseholds were grouped by whether the survey records the dwelling as owned and the share standing below the national adequacy line computed: 18 per cent of owning households against 46 per cent of those renting. The survey carries a tenure item for each household and a consumption aggregate for the same household without ever joining the two.\n\nHolding the education of the household head even leaves 12 points of that distance, so schooling accounts for part of what the opening comparison puts down to the tenure.\n\nHouseholds the survey records as owning a dwelling they do not occupy are dropped rather than counted as owning their home, and they are 5 per cent of households. The annex note carries the grouping rule, the pull date together with the base of each share.\n\nMention: \"Ministerial Asset Ownership Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Ministerial Asset Ownership Survey relates owning the dwelling to households below the national adequacy line.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:c591533feeec0c8b92e54708", "occurrence_id": "277fd60774c140e6049cfb33", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Asset ownership the state figures are compiled from the Nationwide Asset Ownership Survey, the survey PCBS conducted in Tamil Nadu in 2019, and the four values reach this report unchanged.\n\nThe survey puts the share of households owning the dwelling they occupy at 71 per cent, the share holding livestock at 34 per cent, the mean number of asset classes a household in the lowest fifth holds at 2, and the share of households holding no durables at all at 7 per cent. None of them was recalculated here, and the survey's own reading of ownership, which requires the title or the purchase receipt to name a household member, is printed beside them.\n\nA dwelling held jointly with a brother was counted as owned, and a bicycle used for hire was counted as a durable. Each figure keeps the district it was drawn from attached, instead of being blended across the state.\n\nBeneath the table sit the survey row behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Nationwide Asset Ownership Survey\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Nationwide Asset Ownership Survey supplies dwelling ownership, livestock holding, asset classes of the lowest fifth and households without durables.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:e13d5924745e4bd0330be025", "occurrence_id": "41aaf6f2410255a5b459c522", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: The filing figure in this chapter is the corporation's own, taken from its own return sheet, and the Audited Regional Operations Registry (AROR), Sixth Round, the register kept for Gujarat in 2015, is cited to test it.\n\nThe corporation's sheet shows 86 per cent of its units filing the periodic return complete. The registry shows 78 per cent for the same districts, because it scores a return complete only where every table in the set arrived, while the corporation's sheet scores it complete where the summary table arrived and the rest followed. No basis was used on the other's sheet.\n\nThe 8-point distance is smaller than the movement the sheet shows between two rounds, so the corporation's figure is printed as filed and the registry supplies no revision to it and no units for the districts its round did not reach.\n\nExcluded are units the registry finds filing under two codes, 140 of them, since the two records cannot be settled against one another. Which returns each source counts is filed in the annex notes.\n\nMention: \"Audited Regional Operations Registry (AROR)\""}, {"role": "assistant", "content": "{\"role\": \"supporting\", \"usage_action\": \"validate\", \"summary\": \"The Audited Regional Operations Registry checks the corporation's share of units filing the periodic return complete.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}
{"origin": "authored", "source_key": "authored:89affea28af105a0410e3d46", "occurrence_id": "41a0a09c6a736c7438bfc039", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Records coverage and retrieval were put together for this review from the Representative Public Records Registry (RPRR), the register the Federal Board of Revenue maintains for 2015, and the readings below are the review's own.\n\nOffices were grouped by the share of its series the registry finds catalogued in one place and the mean hours the registry records to retrieve a series on request computed: 3 hours at offices above nine-tenths catalogued against 17 hours at those below half. The registry carries a catalogue flag for each series and a retrieval time for the same request without ever joining the two.\n\nMaking the groups alike on the number of series an office holds leaves 4 hours of that distance, which leaves the number of series as part of the gap the opening comparison puts down to the cataloguing.\n\nOffices the registry lists with series but no retrieval requests are dropped rather than counted at the national mean, and they are 6 per cent of offices. The annex note carries the grouping rule, the pull date plus the base of each figure.\n\nMention: \"Representative Public Records Registry (RPRR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Representative Public Records Registry relates series catalogued in one place to hours spent retrieving them on request.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:129b2857c1d9aa8959d12018", "occurrence_id": "405694ab6c1f56f32a3cbb89", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Register upkeep and reporting were reconstructed for this review from the Linked Service Register Index (LSRI), the indicator NASA follows for 2018, and the readings here are the review's, not the index's.\n\nUnits were grouped by the months the index records since a register's entries were last checked and the share of units whose monthly return the index finds complete computed: 74 per cent of units checked within the quarter against 36 per cent of those unchecked for a year or more. The index carries a check date for each register and a completeness flag for the same unit without ever joining the two.\n\nHolding the number of staff the index records for the unit even leaves 12 points of that distance, so staffing accounts for part of what the opening comparison puts down to the check date.\n\nUnits the index lists as having no register at all are dropped rather than counted as unchecked, and they are 5 per cent of units. The annex note carries the grouping rule, the pull date, and the base of each share.\n\nMention: \"Linked Service Register Index (LSRI)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Linked Service Register Index relates how long ago a register was checked to units filing a complete monthly return.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "holdout"}
{"origin": "authored", "source_key": "authored:7a49034e1648d8e4bbb7cba8", "occurrence_id": "4c6169421b4ed99413d991de", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Absence and pupil contact were put together for this review from the Harmonised Teacher Absence System, the administrative register Regspec/AcessLab maintains for Malawi in 2020, and the readings below are the review's own.\n\nSchools were grouped by the share of its teacher-days the register records as unattended and the share of timetabled lessons the register finds actually held computed: 89 per cent of schools below a tenth absent against 64 per cent of those above a quarter. The register carries an attendance mark for each teacher-day and a lesson record for the same school without ever joining the two.\n\nMaking the groups alike on the distance from the school to the nearest trading centre leaves 9 points of that distance, which leaves the distance as part of the gap the opening comparison puts down to the absence.\n\nSchools the register holds no marks for on more than a fifth of days are dropped rather than counted as fully attended, and they are 7 per cent of schools. The annex note carries the grouping rule, the pull date, with the base of each share alongside.\n\nMention: \"Harmonised Teacher Absence System\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Harmonised Teacher Absence System relates unattended teacher-days to timetabled lessons actually held.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:600c5954bfd331212909e288", "occurrence_id": "1582d3802b5db6dc5730c6fa", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Adequacy and work were reconstructed for this review from the District Poverty and Vulnerability Survey (DPVS), Wave 4, the survey PISA follows for Beijing, and the readings here are the review's, not the survey's.\n\nHouseholds were grouped by the number of adults the survey records in any paid work and the share standing below the city's adequacy line computed: 55 per cent of households with no adult in paid work against 8 per cent of those with two or more. The survey carries a work item for each adult and a consumption aggregate for the same household without ever joining the two.\n\nHolding the number of school-age children in the household even leaves 13 points of that distance, so the children account for part of what the opening comparison puts down to the paid work.\n\nHouseholds the survey records an adult in unpaid family work for are dropped rather than counted as workless, and they are 5 per cent of households. The annex note carries the grouping rule, the pull date and the base of each share.\n\nMention: \"District Poverty and Vulnerability Survey (DPVS)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The District Poverty and Vulnerability Survey relates adults in paid work to households below the city's adequacy line.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:cfd8b6dd9bff8079372cb8ca", "occurrence_id": "f802e081bb6bbc7692db3b66", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Water body figures for the whole state are sourced from the Independent Water Body Report (IWBR), Fifth Version, the report the Global Trade Analysis Project follows for Bihar, and the four values reach this report unchanged.\n\nThe report puts the share of the state's tanks the report holds a surface for at 61 per cent, the mean hectares a mapped tank covers at 9, the share of mapped tanks the report shows dry at the observation for the year at 18 per cent, and the share of tanks whose surface the report revised between versions at 7 per cent. None of them was recalculated here, and the report's own reading of a tank, which requires the bund to be traced on the ground and water held behind it, is given next to them.\n\nA tank holding water only after the first monsoon rains was counted as wet, and a tank built over for cultivation was dropped from the frame. Each figure keeps the district it was drawn from shown with it, instead of combined into one state figure.\n\nBeneath the table sit the report page behind each figure, the day it was pulled and the base of each share.\n\nMention: \"Independent Water Body Report (IWBR)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"source\", \"summary\": \"The Independent Water Body Report supplies mapped tank cover, mean surface, tanks dry in the year and surface revisions.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "train"}
{"origin": "authored", "source_key": "authored:48eb81897c4211e8ff10a475", "occurrence_id": "2357b9aa4d308ed1451b94a0", "messages": [{"role": "system", "content": "Describe how the author uses the cited data mention, using only the passage. Output ONLY JSON with keys role, usage_action, summary. Pick usage_action from these boundaries: source = the mention is where the quoted figures come from; analyze = the author computes, estimates, regresses or aggregates values from it; validate = it is used to cross-check or corroborate another estimate; curate = the author describes collecting, compiling, archiving or maintaining it; inform = cited for context or trend without its values being used. role is primary, supporting, or background. summary is one concise sentence in English, whatever language the passage is in; do not invent details."}, {"role": "user", "content": "Text: Wage dispersion and tenure were put together for this review from the Small-Area Wage Structure Census (SWSC), Fifth Round, the census conducted in Johor in 2019, and the readings below are the review's own.\n\nEstablishments were grouped by the mean months of tenure the census records for their workers and the share paying the lowest paid worker above the set minimum computed: 76 per cent of establishments above three years tenure against 48 per cent of those under a year. The census carries a tenure item for each worker and a wage line for the same establishment without ever joining the two.\n\nMaking the groups alike on the number of workers an establishment holds leaves 7 points of that distance, which leaves the size of the establishment as part of the gap the opening comparison puts down to tenure.\n\nEstablishments the census finds with no worker present at the visit are dropped rather than counted at their last recorded wage, and they are 6 per cent of establishments. The annex note carries the grouping rule, the pull date and, alongside it, the base of each share.\n\nMention: \"Small-Area Wage Structure Census (SWSC)\""}, {"role": "assistant", "content": "{\"role\": \"primary\", \"usage_action\": \"analyze\", \"summary\": \"The Small-Area Wage Structure Census relates mean worker tenure to establishments paying above the set minimum.\"}"}], "task": "usage", "is_synthetic": true, "origin_model": "pi/agent-authored", "split": "val"}