Repair v3 supervision, native tool context, and evaluation isolation

Add explicit v3 choice and native auxiliary configurations. Regenerate semantic pairs, restore Hermes requests and call sets, recover Nemotron schemas, restore legacy validation, mask unsupported NLU++ labels, and preserve frozen default/v2 data. All new configs passed structural, lineage, encoder and datasets loading checks.

LICENSE.md

@@ -13,3 +13,9 @@ FastBrain-Instruct is a collection of upstream material and adaptations. Each co
 Changes in v2 include filtering, grouping and split isolation; assistant-reviewed intent descriptions and alias mapping; candidate sampling and paired target removal; restored available policy/context; and schema/metadata reformatting. Source utterances are retained. The MTOP/TOPv2 derived semantic data, including the adapted catalog, is distributed under CC BY-SA 4.0. Attribution does not imply upstream endorsement of this adaptation.
 
 The frozen `default` and retained legacy replay include Google Research SGD (CC BY-SA 4.0), Amazon MASSIVE (CC BY 4.0), Glaive v2 and NousResearch Hermes (Apache 2.0 as recorded in the original card), attributed Nemotron records, NL2Bash (consult its source terms), Git natural-language commands (consult its source terms), and maintainer DevOps templates (Apache 2.0). Their original links and notices remain in the [archived legacy data card](docs/legacy-data-card.md). The v2 release does not claim to resolve unspecified legacy-source terms. The four original data files are unchanged from `a84ea0337a424e287dc6f36f6b9fa320fb920fb0`.
+
+## v3 additions and modifications
+
+The repaired semantic descriptions, conflict catalog and MTOP/TOPv2 examples retain CC BY-SA 4.0. Recovered Hermes requests, call sets and schemas retain the original NousResearch source terms; native Nemotron tool schemas retain the NVIDIA component terms listed above. Repairs include filtering, descriptor clarification, candidate regeneration, source recovery, additional group reservations and explicit supervision masks. These changes do not certify upstream outputs or imply author endorsement.
+
+HINT3 v2 (Curekart and SOFMattress), by Haptik and Gaurav Arora, Chirag Jain, Manas Chaturvedi and Krupal Modi, is provided as a separate native auxiliary database. The database is under ODbL 1.0 and individual contents under DbCL 1.0. The adapted HINT3 view retains those terms. See the [verbatim original notice](licenses/HINT3-ODbL-DbCL.txt), [attribution and source links](licenses/HINT3-attribution.md), and [pinned file hashes](curation/v3/hint3_source_manifest.json). HINT3 labels are not silently relicensed as MIT or merged into the choice catalog.

README.md

@@ -18,6 +18,8 @@ tags:
 - multi-label
 - synthetic
 - aggregated
+- dataset-repair
+- fastbrain-v3
 source_datasets:
 - glaiveai/glaive-function-calling-v2
 - NousResearch/hermes-function-calling-v1
@@ -80,66 +82,130 @@ configs:
     path: data/semantic_original_labels/validation.parquet
   - split: test
     path: data/semantic_original_labels/test.parquet
+- config_name: choice_v3
+  data_files:
+  - split: train
+    path: data/choice_v3/train.parquet
+  - split: validation
+    path: data/choice_v3/validation.parquet
+  - split: test
+    path: data/choice_v3/test.parquet
+- config_name: legacy_replay_v3
+  data_files:
+  - split: train
+    path: data/legacy_replay_v3/train.parquet
+  - split: validation
+    path: data/legacy_replay_v3/validation.parquet
+  - split: test
+    path: data/legacy_replay_v3/test.parquet
+- config_name: semantic_v3
+  data_files:
+  - split: train
+    path: data/semantic_v3/train.parquet
+  - split: validation
+    path: data/semantic_v3/validation.parquet
+  - split: test
+    path: data/semantic_v3/test.parquet
+- config_name: tool_context_v3
+  data_files:
+  - split: train
+    path: data/tool_context_v3/train.parquet
+  - split: validation
+    path: data/tool_context_v3/validation.parquet
+  - split: test
+    path: data/tool_context_v3/test.parquet
+- config_name: tool_environment_holdout_v3
+  data_files:
+  - split: validation
+    path: data/tool_environment_holdout_v3/validation.parquet
+  - split: test
+    path: data/tool_environment_holdout_v3/test.parquet
+- config_name: hermes_actions_v3
+  data_files:
+  - split: train
+    path: data/hermes_actions_v3/train.parquet
+  - split: validation
+    path: data/hermes_actions_v3/validation.parquet
+  - split: test
+    path: data/hermes_actions_v3/test.parquet
+- config_name: nlupp_hotels_v3
+  data_files:
+  - split: train
+    path: data/nlupp_hotels_v3/train.parquet
+  - split: validation
+    path: data/nlupp_hotels_v3/validation.parquet
+  - split: test
+    path: data/nlupp_hotels_v3/test.parquet
+- config_name: hint3_native_v3
+  data_files:
+  - split: train
+    path: data/hint3_native_v3/train.parquet
+  - split: validation
+    path: data/hint3_native_v3/validation.parquet
+  - split: test
+    path: data/hint3_native_v3/test.parquet
 ---
 
 # FastBrain-Instruct
 
-FastBrain-Instruct aggregates public intent and tool-selection data for decision heads. The **2026-10-14 v2** release adds semantic boundary pairs, richer tool context and a separate multi-label task while preserving the original files.
-
-Load the integrated release explicitly:
+The **2026-10-14 v3** release repairs documented supervision and evaluation defects in v2. It regenerates semantic candidate pairs, restores full Hermes requests and native call sets, recovers Nemotron schemas, restores legacy validation, and disables unsupported NLU++ Boolean labels. Model training and capability-retention experiments have not been run for this release.
 
 ```python
 from datasets import load_dataset
 
-ds = load_dataset("TypeScript/FastBrain-Instruct", "choice_v2")
-print(ds["train"][0])
+ds = load_dataset("TypeScript/FastBrain-Instruct", "choice_v3")
 ```
 
 ## Configurations
 
-| Config | Train | Validation | Test | Use |
+| Config | Train | Validation | Test | Purpose |
 |---|---:|---:|---:|---|
+| `choice_v3` | 113,163 | 9,505 | 21,502 | Integrated repaired choice corpus; recommended for the choice head |
+| `legacy_replay_v3` | 61,131 | 825 | 7,181 | Legacy coarse routing and restored Hermes single-call choices |
+| `semantic_v3` | 26,922 | 7,230 | 12,220 | Regenerated MTOP/TOPv2 candidate-relative match/no-match pairs |
+| `tool_context_v3` | 25,110 | 1,450 | 2,101 | Observed Funcdex/Nemotron tool choices with available full schemas |
+| `tool_environment_holdout_v3` | — | 534 | 561 | Additional held-out Nemotron policy/catalog groups; evaluation only |
+| `hermes_actions_v3` | 868 | 24 | 123 | Native observed call sets; separate auxiliary task |
+| `nlupp_hotels_v3` | 765 | 97 | 98 | Native multi-label intents and masked slot values; separate auxiliary task |
+| `hint3_native_v3` | 801 | 68 | 1,348 | Native brand-relative intent/OOS labels; separate auxiliary task |
 | `default` | 97,450 | — | 10,828 | Frozen original release |
-| `choice_v2` | 116,440 | 8,206 | 24,639 | Integrated choice corpus: legacy replay + semantic pairs + tool context |
-| `semantic_v2` | 26,976 | 7,230 | 12,228 | MTOP/TOPv2 intent boundaries and paired candidate-relative no-match |
-| `tool_context_v2` | 26,241 | 976 | 1,583 | Funcdex and repaired Nemotron observed next-tool steps |
-| `nlupp_hotels_v2` | 765 | 97 | 98 | Original multi-label intent modules and slots; separate task format |
-| `semantic_original_labels` | 21,966 | 3,151 | 9,131 | Original labels for complex/ambiguous/unsupported/low-support cases; auxiliary only |
+| `choice_v2` | 116,440 | 8,206 | 24,639 | Frozen previous integrated release |
+
+`legacy_replay_v3`, `semantic_v3`, and `tool_context_v3` are exact subsets of `choice_v3`. **Do not concatenate these subsets with the integrated config.** `tool_environment_holdout_v3` overlaps the tool validation/test splits. All three native auxiliary configs have their own schema and are not automatic choice-head training data. Some native Hermes single-call observations also appear in the choice view; `included_in_choice_v3` records this overlap.
+
+Calling `load_dataset("TypeScript/FastBrain-Instruct")` still loads `default`. All original and v2 data files remain unchanged from [the pre-repair revision](https://huggingface.co/datasets/TypeScript/FastBrain-Instruct/tree/11e6f657f6d8830da230d4e49c65d0f67f106585). The [v2 data card](docs/v2-data-card.md) is archived; its known defects are addressed by the explicit v3 configs, not by silently replacing old splits.
 
-`semantic_v2` and `tool_context_v2` are **overlapping views of `choice_v2`**, not additional examples to concatenate with it. `nlupp_hotels_v2` and `semantic_original_labels` use their own schemas. Calling `load_dataset("TypeScript/FastBrain-Instruct")` still loads `default`.
+## Repairs and evidence
 
-The original `train.parquet`, `test.parquet`, `train.jsonl` and `test.jsonl` remain unchanged from [the frozen baseline](https://huggingface.co/datasets/TypeScript/FastBrain-Instruct/tree/a84ea0337a424e287dc6f36f6b9fa320fb920fb0). All 10,828 original test rows also appear in `choice_v2/test` with their original seven core fields intact. The original data card is [archived here](docs/legacy-data-card.md); its historical claims should be read together with the new audit below.
+- **Semantic supervision:** all retained pairs are regenerated with three repaired operation descriptions and a conservative conflict graph. All 44 previously confirmed false-no-match cases are repaired or explicitly quarantined. Known context-dependent/ambiguous requests and unsupported persistent call rules are excluded from the choice view. See [case-level outcomes](audit/v3/known_issue_resolution.json) and [the versioned catalog](catalogs/semantic_v3.json). The catalog still contains 68 operations; this is not a 151-operation natural intent dataset.
+- **Legacy replay:** restore 825 protected validation rows across all seven legacy sources. Repair documented Git/shell/MASSIVE descriptions, replace a known equivalent pager distractor, and reserve lexical groups against existing holdouts. MASSIVE cooking-label boundaries and other unreviewed labels remain limitations. Missing context is not guessed.
+- **Hermes:** all 1,147 original rows are uniquely mapped to the pinned native file, with full requests, system messages, tool schemas and original call sets recovered. Contract failures, documented semantic contradictions and specifically reviewed financial operations are quarantined. Multi-call responses are retained as observed sets in `hermes_actions_v3`; their first call is not made unique gold. Only eligible single-call records with at least two native tools enter the choice view. Source order does not certify parallel independence or a unique execution order.
+- **Tool context:** quarantine 31 reviewed inconsistent/ambiguous parents, including contaminated preceding histories. Recover the primary catalog for every published Nemotron source step before filtering: 10,810 native conversations / 23,275 targets. In the final tool view, 23,259 Nemotron rows have full native schemas. Only 10,792 pass both the strict source-prefix check and native argument contract checks needed to expose argument supervision. Other recovered arguments remain explicitly unverified. No execution-correctness or unique-valid-action claim is made.
+- **NLU++:** preserve all 960 native texts, intent lists and original slot dictionaries. Disable the generic Boolean export (`head_questions_json={}`, `head_expected_json={}`, `use_for_head_training=false`), because an absent native annotation does not certify a broadly worded predicate false. Mask the 22 reversed stay-date pairs from `slots_for_value_training_json`; retain the original values in `slots_json`.
+- **HINT3:** add a separate native auxiliary view of Curekart/SOFMattress with pinned source files and ODbL/DbCL attribution. Source OOS refers to its original brand catalog, not FastBrain's global no-match option. Native test/OOS is never promoted into training. Financing, lexical overlap, empty text and duplicates are handled explicitly; this filtered view is not the canonical HINT3 benchmark.
 
-## What changed
+## Choice adapter and task contract
 
-| Main training component | Rows |
-|---|---:|
-| Retained legacy replay | 63,223 |
-| MTOP/TOPv2 semantic pairs | 26,976 |
-| Funcdex | 4,010 |
-| Repaired Nemotron | 22,231 |
+Use [head_adapter_v3.py](scripts/v3/head_adapter_v3.py) for the current Clef joint-schema format. It assigns deterministic random option IDs per record, preserves the ID-to-name answer mapping, and accounts for the encoder sorting IDs. Merely shuffling the `choices` list does not randomize the encoder's final order. `encode_without_truncation` raises on oversized input instead of letting the upstream encoder crop state.
 
-- **Semantic coverage:** 68 retained intent classes after alias consolidation, split/template isolation, label screening and a minimum of 20 independent training requests per class. Each selected request has one positive and one answer-removed no-match example. Candidate budgets target 16/32/64 ordinary operations and shrink when conflict exclusions require it; `__no_match__` is counted separately. This release does not provide a 151- or 256-intent natural-language catalog.
-- **Tool context:** preserve the full available pre-answer prefix for Funcdex; restore Nemotron system policy and exclude steps with known missing earlier history or non-system truncation. Tools remain **observed expert-action labels**; schema validity and context recovery do not establish that every other tool is invalid. No text-only/no-call record is relabeled as no-match.
-- **Holdout isolation:** the audit identified 2,687 raw Nemotron conversations shared by the original train and test sets. The v2 rebuild blocks 3,286 historical held-out parents before extracting training steps. The frozen `default` retains its historical split limitations.
-- **Multi-label preservation:** NLU++ hotels keeps all annotated intent modules and slots. Only 576 of its 765 train rows are eligible for intent training; empty-intent context fragments have `use_for_intent_training=false` and no fabricated negative labels.
+The documented adapter includes the complete published `query` plus candidate names/descriptions; it does **not** append `tools_json` to state. All 144,170 choice records were encoded with the actual Clef tokenizer/encoder: maximum 11,482 tokens, below 16,384. See [the measured token budget](audit/v3/token_budget.json). Adding full schemas, more head questions or media requires a new budget check.
 
-## Record format
+For observed tool records, `target` is the source-observed action and `acceptable_targets` is a declared positive label list, **not an exhaustive set of valid actions**. Do not turn unobserved alternatives into independently certified `noul=false` or score-zero labels. The supplied adapter exports only the observed choice task. Auxiliary call-set and multi-label records are rejected by that adapter.
 
-Choice configs retain `id`, `domain`, `category`, `query`, `target`, `choices` (a list of name/description objects), and `source`. Added fields record `parent_id`, `source_family`, `source_repo`, `source_revision`, `source_split`, `source_row_id`, `task_type`, `label_status`, `is_rejection`, `candidate_count`, `catalog_id`, `provenance_json`, `context_json`, and `tools_json`.
+## Evaluation and retention
 
-`provenance_json` contains the source file and row, construction method, group, and applicable license for new rows. For tool records, `query` renders the structured `context_json`. Funcdex retains full original schemas in `tools_json`; Nemotron retains only names, descriptions and required-parameter names because the intermediate source discarded the rest.
+The v3 test has changed requests, descriptions, candidates and filtering; its accuracy is not directly comparable to v2 accuracy. Preserve old regression runs on frozen files, and re-evaluate both the baseline head and candidate head on the same v3 protocol.
 
-NLU++ uses `text`, `intents`, `slots_json`, source/fold identifiers, `use_for_intent_training`, `head_questions_json` and `head_expected_json`. The optional head fields contain 40 independent boolean module questions. Filter on `use_for_intent_training` before intent training. Auxiliary semantic rows retain source labels and parse trees and explicitly set `use_for_single_choice=false`.
+The additional Nemotron environment holdout reserves 23 validation and 23 test policy/catalog groups absent from v3 tool training. Existing in-environment validation/test remains in `tool_context_v3` and is identified by provenance. This is exact policy/catalog grouping within this dataset, not proof that a previous model has never seen the public source or semantically similar workflows.
 
-## Curation and use
+The historical legacy validation is a restored reserved set, not newly unseen data. Historical MASSIVE native-split mixing and unavailable original conversation identities in several legacy sources remain disclosed. Conservative lexical grouping does not establish original dialogue or template ancestry.
 
-[Curation recipe and split policy](docs/curation-v2.md) · [Source revisions and file hashes](audit/source_manifest.json) · [Dataset counts and hashes](audit/release_summary.json) · [Validation report](audit/validation.json) · [Semantic catalog](catalogs/semantic_v2.json) · [Source exclusions](curation/excluded_source_families.json)
+Use [fixed evaluation membership](curation/v3/evaluation_membership.jsonl.gz), [evaluation contract](curation/v3/evaluation_protocol.json), and [ablation definitions](curation/v3/ablation_configs.json). Report source/operation-family slices, candidate-count buckets, present/absent results and in-/out-of-environment tool results separately. Paired semantic rows share a parent; uncertainty estimates should group by parent/template/environment as applicable. Predeclare retention tolerances before training; a lower aggregate loss is not a retention gate.
 
-Use `source`, `source_family`, `task_type` and `provenance_json.group` to make explicit ablations. The paired semantic data has an artificial 1:1 present/absent ratio within that view; this is not a deployment class prior. The release does not establish model improvement or preservation of an existing head's capabilities; those require training and evaluation.
+## Provenance, exclusions and limitations
 
-The source-family exclusions from the original corpus are retained. Additional lexical checks use the exact cached files recorded in the exclusion manifest. They do not certify absence of paraphrases or unobserved source ancestry. Curation uses assistant ontology review, deterministic rules and targeted training-example inspection, **not exhaustive human annotation**. Ambiguous single-choice cases are retained only in the explicitly auxiliary view when possible.
+[Release counts and hashes](audit/v3/release_summary.json) · [Validation](audit/v3/validation.json) · [Repair recipe](docs/curation-v3.md) · [Input evidence](audit/v3/input_manifest.json) · [Quarantine ledger](curation/v3/quarantine.jsonl.gz) · [Historical membership](curation/v3/historical_holdout_membership.json) · [Licenses](LICENSE.md)
 
-## Attribution and licenses
+The quarantine ledger distinguishes quarantined rows, split reservations, deduplication and records retained in a separate task. It is audit data, not a training config; its row count is not a label-error rate. Native source text and original labels remain available for traceability.
 
-This is an aggregation and adaptation of upstream data. MTOP and TOPv2 are **CC BY-SA 4.0**; NLU++ and the attributed Nemotron component are **CC BY 4.0**; Funcdex declares **MIT**. Legacy records retain their original source terms, including SGD's CC BY-SA 4.0. The collection is not licensed uniformly as MIT. See [license and attribution notices](LICENSE.md), the pinned source manifest and the archived legacy card.
+Existing excluded source families/benchmarks remain excluded. Finance filtering is scoped and recorded: it does not establish that every legacy row is finance-free. Ordinary retail price, refund and cash-on-delivery questions are not automatically equated with banking. The semantic 1:1 present/absent construction is an artificial sampling ratio, not a deployment prior. This release does not establish real no-call/clarification policy competence, exhaustive semantic correctness, human gold annotation, or a model-quality gain.

audit/v3/hint3_repair.json

@@ -0,0 +1,105 @@
+{
+  "counters": {
+    "excluded_train": 50,
+    "excluded_test": 40,
+    "train_duplicates": 8
+  },
+  "counts": {
+    "train": {
+      "rows": 801,
+      "bytes": 40869,
+      "sha256": "d789a0db0c2a5e70c3275569a05fa5993f60f4f1bf72dddb134ad219c0932291",
+      "native_oos": 0,
+      "by_brand": {
+        "curekart": 535,
+        "sofmattress": 266
+      }
+    },
+    "validation": {
+      "rows": 68,
+      "bytes": 12648,
+      "sha256": "2da5a239515f0a8ac85454c19607a7184cb0e805476113d713ee06a2071b27c6",
+      "native_oos": 0,
+      "by_brand": {
+        "curekart": 49,
+        "sofmattress": 19
+      }
+    },
+    "test": {
+      "rows": 1348,
+      "bytes": 73960,
+      "sha256": "fa4ea54984aff79bb1aa4b793051b0e9d554881f2c24028528c53b40881e8472",
+      "native_oos": 656,
+      "by_brand": {
+        "curekart": 973,
+        "sofmattress": 375
+      }
+    }
+  },
+  "original_counts": {
+    "train": 927,
+    "test": 1388,
+    "test_native_oos": 676
+  },
+  "native_catalogs": {
+    "curekart": [
+      "CALL_CENTER",
+      "CANCEL_ORDER",
+      "CHAT_WITH_AGENT",
+      "CHECK_PINCODE",
+      "CONSULT_START",
+      "DELAY_IN_PARCEL",
+      "EXPIRY_DATE",
+      "FRANCHISE",
+      "IMMUNITY",
+      "INTERNATIONAL_SHIPPING",
+      "MODES_OF_PAYMENTS",
+      "MODIFY_ADDRESS",
+      "ORDER_QUERY",
+      "ORDER_STATUS",
+      "ORDER_TAKING",
+      "ORIGINAL_PRODUCT",
+      "PAYMENT_AND_BILL",
+      "PORTAL_ISSUE",
+      "RECOMMEND_PRODUCT",
+      "REFER_EARN",
+      "REFUNDS_RETURNS_REPLACEMENTS",
+      "RESUME_DELIVERY",
+      "SIDE_EFFECT",
+      "SIGN_UP",
+      "START_OVER",
+      "STORE_INFORMATION",
+      "USER_GOAL_FORM",
+      "WORK_FROM_HOME"
+    ],
+    "sofmattress": [
+      "100_NIGHT_TRIAL_OFFER",
+      "ABOUT_SOF_MATTRESS",
+      "CANCEL_ORDER",
+      "CHECK_PINCODE",
+      "COD",
+      "COMPARISON",
+      "DELAY_IN_DELIVERY",
+      "DISTRIBUTORS",
+      "EMI",
+      "ERGO_FEATURES",
+      "LEAD_GEN",
+      "MATTRESS_COST",
+      "OFFERS",
+      "ORDER_STATUS",
+      "ORTHO_FEATURES",
+      "PILLOWS",
+      "PRODUCT_VARIANTS",
+      "RETURN_EXCHANGE",
+      "SIZE_CUSTOMIZATION",
+      "WARRANTY",
+      "WHAT_SIZE_TO_ORDER"
+    ]
+  },
+  "seed": 2026101403,
+  "no_native_oos_training_data": true,
+  "global_head_no_match_conversion": false,
+  "main_choice_normalized_queries_screened": 120975,
+  "test_note": "Filtered native test is auxiliary evaluation, not the canonical HINT3 paper benchmark.",
+  "true_151_candidate_semantic_coverage_established": false
+}

audit/v3/input_manifest.json

@@ -0,0 +1,83 @@
+{
+  "previous_dataset": {
+    "repo": "TypeScript/FastBrain-Instruct",
+    "revision": "11e6f657f6d8830da230d4e49c65d0f67f106585"
+  },
+  "original_dataset_revision": "a84ea0337a424e287dc6f36f6b9fa320fb920fb0",
+  "evidence_files": [
+    {
+      "local_artifact": "fastbrain_dataset_repair_20261014/source_research/legacy/repair_decisions.json",
+      "bytes": 853673,
+      "sha256": "a3153dc4d7aa85e8dad520ce06154e0680f5495fa1e4ad985d203debde68a866"
+    },
+    {
+      "local_artifact": "fastbrain_dataset_repair_20261014/source_research/legacy/hermes_restored_records.jsonl",
+      "bytes": 19267908,
+      "sha256": "cc9fa296aeb84750d4a84a913d4d3543dc6c2a92d301fc2a71e5a940ff94d785"
+    },
+    {
+      "local_artifact": "fastbrain_dataset_repair_20261014/source_research/legacy/hermes_source_mapping.json",
+      "bytes": 2437721,
+      "sha256": "f7db59bd07a351e6874959b1eee368c6232cfb4d8bfd660e3ab87d6f916cf557"
+    },
+    {
+      "local_artifact": "fastbrain_dataset_repair_20261014/source_research/legacy/historical_holdouts.json",
+      "bytes": 2103898,
+      "sha256": "bed96087547c827f2961d6675cd8eef1de9c3466af5270c6696b059f95b561b8"
+    },
+    {
+      "local_artifact": "fastbrain_dataset_repair_20261014/source_research/tools/nemotron_raw_recovered.jsonl",
+      "bytes": 122112609,
+      "sha256": "28bd5ddbb74bf9bcb8d03bf4ab22d52d5b59b45b4b6d1edd97548c285e270943"
+    },
+    {
+      "local_artifact": "fastbrain_dataset_repair_20261014/source_research/tools/nemotron_raw_recovery_summary.json",
+      "bytes": 2372,
+      "sha256": "b825f47b24e59a318f9d53bb40194a85f167d507f09788d71fa2c8d0f2c7c467"
+    },
+    {
+      "local_artifact": "fastbrain_dataset_repair_20261014/source_research/tools/repair_decisions.json",
+      "bytes": 276230,
+      "sha256": "df8123dc3c876b727602881cb1823414af8cefa19215fd443979843a87c12255"
+    },
+    {
+      "local_artifact": "fastbrain_dataset_repair_20261014/source_research/semantic/hint3_selected_manifest.json",
+      "bytes": 8969,
+      "sha256": "cf07feab6b8d841ed5d0bef7d3c97568d1bda7174bab0f74d8e8bfdc4e6bf71d"
+    },
+    {
+      "local_artifact": "fastbrain_dataset_repair_20261014/source_research/semantic/hint3_row_audit.jsonl",
+      "bytes": 1756661,
+      "sha256": "b15ce05c475ace9cbfa1ea044b6a22bbc984a68b9e37987b7cc85e5444a9ca99"
+    },
+    {
+      "local_artifact": "fastbrain_dataset_reaudit_20261014/semantic/findings.json",
+      "bytes": 2985851,
+      "sha256": "4798e18aa300a4d7d093939256b4e6dcf4e479f22487ae49dc6d33db043f6f6f"
+    },
+    {
+      "local_artifact": "fastbrain_dataset_reaudit_20261014/semantic/semantic_assistant_judgments.jsonl",
+      "bytes": 469741,
+      "sha256": "178b4c4651e11cd329b0243b18acfd22a0027e41de3d8e1482f9e7ceecde76d6"
+    },
+    {
+      "local_artifact": "fastbrain_dataset_reaudit_20261014/tools/case_judgments.json",
+      "bytes": 357043,
+      "sha256": "b20e22ee4c183abcde6b33719057585934436bc4a6db7d0cacff763f44273230"
+    }
+  ],
+  "source_links": {
+    "hermes": "curation/v3/hermes_source_mapping.json",
+    "hint3": "curation/v3/hint3_source_manifest.json",
+    "other_upstream_sources": "audit/source_manifest.json"
+  },
+  "primary_nemotron": {
+    "repo": "nvidia/Nemotron-SFT-Agentic-v2",
+    "revision": "7c804833427f633ccd53b582dbf02525fd680f78",
+    "file": "data/interactive_agent.jsonl",
+    "bounded_prefix_last_row": 15512,
+    "recovered_native_parents": 10810,
+    "recovered_source_steps": 23275
+  },
+  "reproducibility_note": "Evidence hashes identify the local pinned-source research artifacts. Per-row published provenance carries native locators; original source files remain at the linked pinned revisions."
+}

audit/v3/known_issue_resolution.json

@@ -0,0 +1,20 @@
+{
+  "S01_route_retrieval_alternative_remains": {
+    "regenerated_with_repair": 35
+  },
+  "S02_online_contact_filter_alternative_remains": {
+    "regenerated_with_repair": 9
+  },
+  "S03_nutritional_query_description_gap": {
+    "regenerated_with_repair": 97
+  },
+  "S04_alarm_rename_description_gap": {
+    "regenerated_with_repair": 12
+  },
+  "S05_timer_history_description_gap": {
+    "regenerated_with_repair": 2
+  },
+  "S06_persistent_call_policy_description_gap": {
+    "explicitly_quarantined": 7
+  }
+}

audit/v3/legacy_repair.json

@@ -0,0 +1,162 @@
+{
+  "counters": {
+    "records_with_description_repairs": 17314,
+    "records_with_candidate_repairs": 1,
+    "The request explicitly names mypy and flake8 but the target is an undefined pylint_typecheck wrapper. No implementation/native schema establishes that it invokes the requested tools.": 1,
+    "The request combines filesystem search and FTP upload; matching native commands invoke curl inside find -exec. The exposed find capability alone does not certify the complete operation, and multiple native rows remain matched.": 1,
+    "The native command is git rm -r --cached build/ followed by appending build/ to .gitignore. git_rm alone cannot perform the requested ignore-file edit.": 1,
+    "The entire request is a seat reservation confirmation; both bus-ticket purchase and one-way-flight reservation are offered. The service/dialogue context that distinguishes them was not recovered.": 1,
+    "Let us call it Trip to Hawaii identifies a name but not the object or action; the earlier conversation needed to justify create_todo_list is absent.": 1,
+    "The request covers all pods in a namespace, while k8s_rollout_restart describes a single deployment. The missing cluster/controller state does not establish equivalence.": 1,
+    "The same native query maps both to a full comm/sort/grep process-substitution command and to corrupt comm -23 < <. Preserve both matches; neither first-match selection nor description rewriting recovers a unique executable source command.": 1,
+    "Weird things people want to know is labeled cooking_recipe without any visible cooking request. Do not expand a recipe description to cover it.": 1,
+    "Get me a one-way flight does not uniquely distinguish searching from reserving in the exposed SearchOnewayFlight/ReserveOnewayFlight choice set.": 1,
+    "The only input is olly, labeled recommendation_events; no event-discovery request is visible.": 1,
+    "The request asks for things to do and dates of events. FindEvents and GetEventDates are both candidates; the missing service state does not establish a unique action.": 1,
+    "The only input is fast, labeled audio_volume_up; no volume-increase intent is established.": 1,
+    "scope_finance: Calculate the payment on a loan from principal, interest rate, and loan term.": 1,
+    "Go back is labeled play_podcasts but no podcast/media context is present.": 1,
+    "The user requests a reminder tomorrow, while create_todo only promises creating a todo item. No recovered native schema establishes timed-reminder support.": 1,
+    "How many hours ahead is NYC is natively datetime_query; retained training examples also place offset questions in datetime_convert. Do not invent an exclusive boundary or silently relabel.": 1,
+    "What are jalapenos is natively labeled cooking_recipe, but it asks for an ingredient definition. No automatic relabeling or expansion of recipe to all food questions is justified.": 1,
+    "Change my alarm is labeled alarm_remove, but the request does not distinguish deletion, snoozing, or editing. Do not redefine remove as all alarm changes.": 1,
+    "Send a text to Alice is labeled email_querycontact; contact lookup may be a prerequisite, but no native dialogue/tool state establishes it as the unique next action.": 1,
+    "The nature of artificial intelligence is labeled general_quirky. Its boundary with factual/definition questions is not established by an authoritative ontology contract; retain the native label as evidence, not a repaired unique action.": 1,
+    "Lights out is labeled iot_hue_lighton. Its ordinary reading does not justify turning lights on; isolate without silently flipping the label.": 1,
+    "indistinguishable_target_candidate_descriptions": 1,
+    "hermes_native_aux_only_rows": 752,
+    "Native first-round call contract failure: argument_schema. Preserve raw calls; do not silently repair nested names or arguments.": 49,
+    "scope_finance: Initiate a USD transfer to a PayPal recipient.": 1,
+    "scope_finance: Create an actual Venmo peer-to-peer payment.": 1,
+    "Native first-round call contract failure: missing_top_level_name. Preserve raw calls; do not silently repair nested names or arguments.": 60,
+    "scope_finance: Approve a consumer loan and compute its payments and amortization schedule.": 1,
+    "scope_finance: Transfer Bitcoin between blockchain wallets.": 1,
+    "scope_finance: Initiate a credit-card payment through a payment gateway; this is an explicit payment API operation, not merely a shopping-price mention.": 1,
+    "scope_finance: Retrieve cryptocurrency-wallet financial transaction history.": 1,
+    "scope_finance: Retrieve PayPal financial transaction history.": 1,
+    "scope_finance: Transfer USD funds between e-wallet users.": 1,
+    "scope_finance: Send a Venmo peer-to-peer payment.": 1,
+    "scope_finance: Link a credit card to Coinbase as the first action of a cryptocurrency-purchase workflow.": 1,
+    "extract_files_by_extension creates and populates Compliance_Audit_2023, then generate_file_report reads that folder. The user also says parallel, but the visible dependency and contradictory timing prevent certifying the co-emitted set as ready to run in parallel.": 1,
+    "scope_finance: Initiate a Revolut international money transfer.": 1,
+    "scope_finance: Purchase Bitcoin through Coinbase.": 1,
+    "scope_finance: Create a cryptocurrency wallet for asset management and payments.": 1,
+    "scope_finance: Create a Klarna buy-now-pay-later installment-payment session; the requested operation is credit/payment processing.": 1,
+    "scope_finance: Transfer funds between checking and savings bank accounts.": 1,
+    "scope_finance: Create a PayPal payment request; distinguish this explicit payment-network action from ordinary invoice document generation.": 1,
+    "list_container_services receives cluster=\"web-app-service\", although the user provided a service name and task definition, not its cluster. Passing the native schema does not ground this argument.": 1,
+    "scope_finance: Create a brokerage account and update an investment profile for securities trading.": 1,
+    "scope_finance: Link a checking bank account to PayPal as a funding source.": 1,
+    "scope_finance: Retrieve PayPal transaction history and account balance.": 1,
+    "scope_finance: Send a USD peer-to-peer payment through Cash App.": 1,
+    "scope_finance: Purchase Bitcoin through Cash App and query the USD/BTC exchange rate.": 1,
+    "train_lexical_group_reserved": 1307,
+    "validation_lexical_group_reserved": 12
+  },
+  "choice_counts": {
+    "train": {
+      "rows": 61131,
+      "parents": 61131,
+      "sources": {
+        "sgd": 33724,
+        "nl2bash": 2634,
+        "glaive_v2": 9992,
+        "amazon_massive": 13763,
+        "git_nl_commands": 777,
+        "curated_devops": 18,
+        "hermes_v1": 223
+      },
+      "task_types": {
+        "single_choice": 60908,
+        "observed_next_tool": 223
+      },
+      "candidate_counts": {
+        "2": 60,
+        "3": 81,
+        "4": 61,
+        "5": 60928,
+        "6": 1
+      },
+      "rejection_rows": 0
+    },
+    "validation": {
+      "rows": 825,
+      "parents": 825,
+      "sources": {
+        "amazon_massive": 407,
+        "glaive_v2": 97,
+        "sgd": 231,
+        "git_nl_commands": 31,
+        "nl2bash": 51,
+        "curated_devops": 2,
+        "hermes_v1": 6
+      },
+      "task_types": {
+        "single_choice": 819,
+        "observed_next_tool": 6
+      },
+      "candidate_counts": {
+        "2": 2,
+        "3": 2,
+        "4": 2,
+        "5": 819
+      },
+      "rejection_rows": 0
+    },
+    "test": {
+      "rows": 7181,
+      "parents": 7172,
+      "sources": {
+        "sgd": 3889,
+        "git_nl_commands": 75,
+        "glaive_v2": 1304,
+        "amazon_massive": 1531,
+        "nl2bash": 345,
+        "curated_devops": 3,
+        "hermes_v1": 34
+      },
+      "task_types": {
+        "single_choice": 7147,
+        "observed_next_tool": 34
+      },
+      "candidate_counts": {
+        "2": 9,
+        "3": 16,
+        "4": 9,
+        "5": 7147
+      },
+      "rejection_rows": 0
+    }
+  },
+  "hermes_action_counts": {
+    "train": {
+      "rows": 868,
+      "bytes": 1623002,
+      "sha256": "bf4e001ed82700ec73d48bfad64304517784c00a3c4ff4bd7533a1bf58d2e6f9",
+      "single_call": 372,
+      "multiple_calls": 496
+    },
+    "validation": {
+      "rows": 24,
+      "bytes": 83582,
+      "sha256": "429cedc1b030a4b16bb818c41ffd80c8f5cfde7fb76fed61e474923379591dfe",
+      "single_call": 12,
+      "multiple_calls": 12
+    },
+    "test": {
+      "rows": 123,
+      "bytes": 267399,
+      "sha256": "3227e75f589673a0a0269008d8b2761f379cd7a41159a931cd2ea05a0f49bcbe",
+      "single_call": 55,
+      "multiple_calls": 68
+    }
+  },
+  "historical_validation_restored": true,
+  "old_cropped_nemotron_in_choice_v3": false,
+  "historical_validation_is_new_unseen_data": false,
+  "native_legacy_split_isolation_proven": false,
+  "split_rule": "Existing test > historical validation > train over connected exact/greeting/quoted-number lexical groups. Lower-priority records are reserved, never promoted.",
+  "quarantine_is_not_an_error_rate": true,
+  "retained_legacy_not_exhaustively_reannotated": true,
+  "lexical_template_grouping_proves_source_parent_isolation": false
+}

audit/v3/loading_validation.json

@@ -0,0 +1,62 @@
+{
+  "status": "pass",
+  "available_configurations": [
+    "default",
+    "choice_v2",
+    "semantic_v2",
+    "tool_context_v2",
+    "nlupp_hotels_v2",
+    "semantic_original_labels",
+    "choice_v3",
+    "legacy_replay_v3",
+    "semantic_v3",
+    "tool_context_v3",
+    "tool_environment_holdout_v3",
+    "hermes_actions_v3",
+    "nlupp_hotels_v3",
+    "hint3_native_v3"
+  ],
+  "full_local_load": {
+    "choice_v3": {
+      "train": 113163,
+      "validation": 9505,
+      "test": 21502
+    },
+    "legacy_replay_v3": {
+      "train": 61131,
+      "validation": 825,
+      "test": 7181
+    },
+    "semantic_v3": {
+      "train": 26922,
+      "validation": 7230,
+      "test": 12220
+    },
+    "tool_context_v3": {
+      "train": 25110,
+      "validation": 1450,
+      "test": 2101
+    },
+    "tool_environment_holdout_v3": {
+      "validation": 534,
+      "test": 561
+    },
+    "hermes_actions_v3": {
+      "train": 868,
+      "validation": 24,
+      "test": 123
+    },
+    "nlupp_hotels_v3": {
+      "train": 765,
+      "validation": 97,
+      "test": 98
+    },
+    "hint3_native_v3": {
+      "train": 801,
+      "validation": 68,
+      "test": 1348
+    }
+  },
+  "method": "datasets.load_dataset on merged immutable old/new local snapshot",
+  "input_revision": "11e6f657f6d8830da230d4e49c65d0f67f106585"
+}

audit/v3/nlupp_repair.json

@@ -0,0 +1,286 @@
+{
+  "counters": {
+    "train_rows": 765,
+    "train_removed_boolean_fields": 23040,
+    "train_reversed_stay_dates": 19,
+    "validation_rows": 97,
+    "validation_removed_boolean_fields": 2840,
+    "validation_reversed_stay_dates": 2,
+    "test_rows": 98,
+    "test_removed_boolean_fields": 2800,
+    "test_reversed_stay_dates": 1
+  },
+  "invalid_slot_values": [
+    {
+      "id": "fbv3:nlupp:hotels:0:39",
+      "original_id": "fbv2:nlupp:hotels:0:39",
+      "split": "train",
+      "source_file": "nlupp/data/hotels/fold0.json",
+      "source_row_id": "39",
+      "reason": "inherited_checkin_after_checkout",
+      "masked_value_fields": [
+        "date_from",
+        "date_to"
+      ]
+    },
+    {
+      "id": "fbv3:nlupp:hotels:1:41",
+      "original_id": "fbv2:nlupp:hotels:1:41",
+      "split": "train",
+      "source_file": "nlupp/data/hotels/fold1.json",
+      "source_row_id": "41",
+      "reason": "inherited_checkin_after_checkout",
+      "masked_value_fields": [
+        "date_from",
+        "date_to"
+      ]
+    },
+    {
+      "id": "fbv3:nlupp:hotels:1:45",
+      "original_id": "fbv2:nlupp:hotels:1:45",
+      "split": "train",
+      "source_file": "nlupp/data/hotels/fold1.json",
+      "source_row_id": "45",
+      "reason": "inherited_checkin_after_checkout",
+      "masked_value_fields": [
+        "date_from",
+        "date_to"
+      ]
+    },
+    {
+      "id": "fbv3:nlupp:hotels:2:0",
+      "original_id": "fbv2:nlupp:hotels:2:0",
+      "split": "train",
+      "source_file": "nlupp/data/hotels/fold2.json",
+      "source_row_id": "0",
+      "reason": "inherited_checkin_after_checkout",
+      "masked_value_fields": [
+        "date_from",
+        "date_to"
+      ]
+    },
+    {
+      "id": "fbv3:nlupp:hotels:2:38",
+      "original_id": "fbv2:nlupp:hotels:2:38",
+      "split": "train",
+      "source_file": "nlupp/data/hotels/fold2.json",
+      "source_row_id": "38",
+      "reason": "inherited_checkin_after_checkout",
+      "masked_value_fields": [
+        "date_from",
+        "date_to"
+      ]
+    },
+    {
+      "id": "fbv3:nlupp:hotels:3:47",
+      "original_id": "fbv2:nlupp:hotels:3:47",
+      "split": "train",
+      "source_file": "nlupp/data/hotels/fold3.json",
+      "source_row_id": "47",
+      "reason": "inherited_checkin_after_checkout",
+      "masked_value_fields": [
+        "date_from",
+        "date_to"
+      ]
+    },
+    {
+      "id": "fbv3:nlupp:hotels:4:24",
+      "original_id": "fbv2:nlupp:hotels:4:24",
+      "split": "train",
+      "source_file": "nlupp/data/hotels/fold4.json",
+      "source_row_id": "24",
+      "reason": "inherited_checkin_after_checkout",
+      "masked_value_fields": [
+        "date_from",
+        "date_to"
+      ]
+    },
+    {
+      "id": "fbv3:nlupp:hotels:4:38",
+      "original_id": "fbv2:nlupp:hotels:4:38",
+      "split": "train",
+      "source_file": "nlupp/data/hotels/fold4.json",
+      "source_row_id": "38",
+      "reason": "inherited_checkin_after_checkout",
+      "masked_value_fields": [
+        "date_from",
+        "date_to"
+      ]
+    },
+    {
+      "id": "fbv3:nlupp:hotels:6:34",
+      "original_id": "fbv2:nlupp:hotels:6:34",
+      "split": "train",
+      "source_file": "nlupp/data/hotels/fold6.json",
+      "source_row_id": "34",
+      "reason": "inherited_checkin_after_checkout",
+      "masked_value_fields": [
+        "date_from",
+        "date_to"
+      ]
+    },
+    {
+      "id": "fbv3:nlupp:hotels:7:25",
+      "original_id": "fbv2:nlupp:hotels:7:25",
+      "split": "train",
+      "source_file": "nlupp/data/hotels/fold7.json",
+      "source_row_id": "25",
+      "reason": "inherited_checkin_after_checkout",
+      "masked_value_fields": [
+        "date_from",
+        "date_to"
+      ]
+    },
+    {
+      "id": "fbv3:nlupp:hotels:7:49",
+      "original_id": "fbv2:nlupp:hotels:7:49",
+      "split": "train",
+      "source_file": "nlupp/data/hotels/fold7.json",
+      "source_row_id": "49",
+      "reason": "inherited_checkin_after_checkout",
+      "masked_value_fields": [
+        "date_from",
+        "date_to"
+      ]
+    },
+    {
+      "id": "fbv3:nlupp:hotels:8:44",
+      "original_id": "fbv2:nlupp:hotels:8:44",
+      "split": "train",
+      "source_file": "nlupp/data/hotels/fold8.json",
+      "source_row_id": "44",
+      "reason": "inherited_checkin_after_checkout",
+      "masked_value_fields": [
+        "date_from",
+        "date_to"
+      ]
+    },
+    {
+      "id": "fbv3:nlupp:hotels:9:48",
+      "original_id": "fbv2:nlupp:hotels:9:48",
+      "split": "train",
+      "source_file": "nlupp/data/hotels/fold9.json",
+      "source_row_id": "48",
+      "reason": "inherited_checkin_after_checkout",
+      "masked_value_fields": [
+        "date_from",
+        "date_to"
+      ]
+    },
+    {
+      "id": "fbv3:nlupp:hotels:10:47",
+      "original_id": "fbv2:nlupp:hotels:10:47",
+      "split": "train",
+      "source_file": "nlupp/data/hotels/fold10.json",
+      "source_row_id": "47",
+      "reason": "inherited_checkin_after_checkout",
+      "masked_value_fields": [
+        "date_from",
+        "date_to"
+      ]
+    },
+    {
+      "id": "fbv3:nlupp:hotels:12:42",
+      "original_id": "fbv2:nlupp:hotels:12:42",
+      "split": "train",
+      "source_file": "nlupp/data/hotels/fold12.json",
+      "source_row_id": "42",
+      "reason": "inherited_checkin_after_checkout",
+      "masked_value_fields": [
+        "date_from",
+        "date_to"
+      ]
+    },
+    {
+      "id": "fbv3:nlupp:hotels:13:38",
+      "original_id": "fbv2:nlupp:hotels:13:38",
+      "split": "train",
+      "source_file": "nlupp/data/hotels/fold13.json",
+      "source_row_id": "38",
+      "reason": "inherited_checkin_after_checkout",
+      "masked_value_fields": [
+        "date_from",
+        "date_to"
+      ]
+    },
+    {
+      "id": "fbv3:nlupp:hotels:14:38",
+      "original_id": "fbv2:nlupp:hotels:14:38",
+      "split": "train",
+      "source_file": "nlupp/data/hotels/fold14.json",
+      "source_row_id": "38",
+      "reason": "inherited_checkin_after_checkout",
+      "masked_value_fields": [
+        "date_from",
+        "date_to"
+      ]
+    },
+    {
+      "id": "fbv3:nlupp:hotels:14:47",
+      "original_id": "fbv2:nlupp:hotels:14:47",
+      "split": "train",
+      "source_file": "nlupp/data/hotels/fold14.json",
+      "source_row_id": "47",
+      "reason": "inherited_checkin_after_checkout",
+      "masked_value_fields": [
+        "date_from",
+        "date_to"
+      ]
+    },
+    {
+      "id": "fbv3:nlupp:hotels:15:40",
+      "original_id": "fbv2:nlupp:hotels:15:40",
+      "split": "train",
+      "source_file": "nlupp/data/hotels/fold15.json",
+      "source_row_id": "40",
+      "reason": "inherited_checkin_after_checkout",
+      "masked_value_fields": [
+        "date_from",
+        "date_to"
+      ]
+    },
+    {
+      "id": "fbv3:nlupp:hotels:16:39",
+      "original_id": "fbv2:nlupp:hotels:16:39",
+      "split": "validation",
+      "source_file": "nlupp/data/hotels/fold16.json",
+      "source_row_id": "39",
+      "reason": "inherited_checkin_after_checkout",
+      "masked_value_fields": [
+        "date_from",
+        "date_to"
+      ]
+    },
+    {
+      "id": "fbv3:nlupp:hotels:17:40",
+      "original_id": "fbv2:nlupp:hotels:17:40",
+      "split": "validation",
+      "source_file": "nlupp/data/hotels/fold17.json",
+      "source_row_id": "40",
+      "reason": "inherited_checkin_after_checkout",
+      "masked_value_fields": [
+        "date_from",
+        "date_to"
+      ]
+    },
+    {
+      "id": "fbv3:nlupp:hotels:18:40",
+      "original_id": "fbv2:nlupp:hotels:18:40",
+      "split": "test",
+      "source_file": "nlupp/data/hotels/fold18.json",
+      "source_row_id": "40",
+      "reason": "inherited_checkin_after_checkout",
+      "masked_value_fields": [
+        "date_from",
+        "date_to"
+      ]
+    }
+  ],
+  "native_text_intents_slots_preserved": true,
+  "generic_boolean_supervision_enabled": false,
+  "contract": "Original positive modules remain available for the native annotation task. Unannotated modules are not certified false natural-language predicates. Head adapter refuses these rows until an independently verified boolean task exists.",
+  "original_slots_kept_for_provenance": true,
+  "value_training_field": "slots_for_value_training_json",
+  "value_mask_field": "slot_value_mask_json",
+  "automatic_date_corrections_guessed": 0
+}

audit/v3/quarantine_summary.json

@@ -0,0 +1,95 @@
+{
+  "records": 2524,
+  "by_reason": {
+    "The request explicitly names mypy and flake8 but the target is an undefined pylint_typecheck wrapper. No implementation/native schema establishes that it invokes the requested tools.": 1,
+    "The request combines filesystem search and FTP upload; matching native commands invoke curl inside find -exec. The exposed find capability alone does not certify the complete operation, and multiple native rows remain matched.": 1,
+    "The native command is git rm -r --cached build/ followed by appending build/ to .gitignore. git_rm alone cannot perform the requested ignore-file edit.": 1,
+    "The entire request is a seat reservation confirmation; both bus-ticket purchase and one-way-flight reservation are offered. The service/dialogue context that distinguishes them was not recovered.": 1,
+    "Let us call it Trip to Hawaii identifies a name but not the object or action; the earlier conversation needed to justify create_todo_list is absent.": 1,
+    "The request covers all pods in a namespace, while k8s_rollout_restart describes a single deployment. The missing cluster/controller state does not establish equivalence.": 1,
+    "The same native query maps both to a full comm/sort/grep process-substitution command and to corrupt comm -23 < <. Preserve both matches; neither first-match selection nor description rewriting recovers a unique executable source command.": 1,
+    "Weird things people want to know is labeled cooking_recipe without any visible cooking request. Do not expand a recipe description to cover it.": 1,
+    "Get me a one-way flight does not uniquely distinguish searching from reserving in the exposed SearchOnewayFlight/ReserveOnewayFlight choice set.": 1,
+    "The only input is olly, labeled recommendation_events; no event-discovery request is visible.": 1,
+    "The request asks for things to do and dates of events. FindEvents and GetEventDates are both candidates; the missing service state does not establish a unique action.": 1,
+    "The only input is fast, labeled audio_volume_up; no volume-increase intent is established.": 1,
+    "scope_finance: Calculate the payment on a loan from principal, interest rate, and loan term.": 1,
+    "Go back is labeled play_podcasts but no podcast/media context is present.": 1,
+    "The user requests a reminder tomorrow, while create_todo only promises creating a todo item. No recovered native schema establishes timed-reminder support.": 1,
+    "How many hours ahead is NYC is natively datetime_query; retained training examples also place offset questions in datetime_convert. Do not invent an exclusive boundary or silently relabel.": 1,
+    "What are jalapenos is natively labeled cooking_recipe, but it asks for an ingredient definition. No automatic relabeling or expansion of recipe to all food questions is justified.": 1,
+    "Change my alarm is labeled alarm_remove, but the request does not distinguish deletion, snoozing, or editing. Do not redefine remove as all alarm changes.": 1,
+    "Send a text to Alice is labeled email_querycontact; contact lookup may be a prerequisite, but no native dialogue/tool state establishes it as the unique next action.": 1,
+    "The nature of artificial intelligence is labeled general_quirky. Its boundary with factual/definition questions is not established by an authoritative ontology contract; retain the native label as evidence, not a repaired unique action.": 1,
+    "Lights out is labeled iot_hue_lighton. Its ordinary reading does not justify turning lights on; isolate without silently flipping the label.": 1,
+    "indistinguishable_target_candidate_descriptions": 1,
+    "native_call_set_kept_outside_single_choice": 752,
+    "Native first-round call contract failure: argument_schema. Preserve raw calls; do not silently repair nested names or arguments.": 49,
+    "scope_finance: Initiate a USD transfer to a PayPal recipient.": 1,
+    "scope_finance: Create an actual Venmo peer-to-peer payment.": 1,
+    "Native first-round call contract failure: missing_top_level_name. Preserve raw calls; do not silently repair nested names or arguments.": 60,
+    "scope_finance: Approve a consumer loan and compute its payments and amortization schedule.": 1,
+    "scope_finance: Transfer Bitcoin between blockchain wallets.": 1,
+    "scope_finance: Initiate a credit-card payment through a payment gateway; this is an explicit payment API operation, not merely a shopping-price mention.": 1,
+    "scope_finance: Retrieve cryptocurrency-wallet financial transaction history.": 1,
+    "scope_finance: Retrieve PayPal financial transaction history.": 1,
+    "scope_finance: Transfer USD funds between e-wallet users.": 1,
+    "scope_finance: Send a Venmo peer-to-peer payment.": 1,
+    "scope_finance: Link a credit card to Coinbase as the first action of a cryptocurrency-purchase workflow.": 1,
+    "extract_files_by_extension creates and populates Compliance_Audit_2023, then generate_file_report reads that folder. The user also says parallel, but the visible dependency and contradictory timing prevent certifying the co-emitted set as ready to run in parallel.": 1,
+    "scope_finance: Initiate a Revolut international money transfer.": 1,
+    "scope_finance: Purchase Bitcoin through Coinbase.": 1,
+    "scope_finance: Create a cryptocurrency wallet for asset management and payments.": 1,
+    "scope_finance: Create a Klarna buy-now-pay-later installment-payment session; the requested operation is credit/payment processing.": 1,
+    "scope_finance: Transfer funds between checking and savings bank accounts.": 1,
+    "scope_finance: Create a PayPal payment request; distinguish this explicit payment-network action from ordinary invoice document generation.": 1,
+    "list_container_services receives cluster=\"web-app-service\", although the user provided a service name and task definition, not its cluster. Passing the native schema does not ground this argument.": 1,
+    "scope_finance: Create a brokerage account and update an investment profile for securities trading.": 1,
+    "scope_finance: Link a checking bank account to PayPal as a funding source.": 1,
+    "scope_finance: Retrieve PayPal transaction history and account balance.": 1,
+    "scope_finance: Send a USD peer-to-peer payment through Cash App.": 1,
+    "scope_finance: Purchase Bitcoin through Cash App and query the USD/BTC exchange rate.": 1,
+    "lexical_group_reserved_for_existing_holdout": 1319,
+    "reviewed_semantic_ambiguous_multiple_valid": 16,
+    "reviewed_semantic_needs_context": 24,
+    "persistent_call_rule_requires_distinct_operation": 14,
+    "persistent_call_rule_screen": 8,
+    "reviewed_tool_parent_inconsistent_or_ambiguous": 139,
+    "normalized_overlap_with_selected_native_test": 15,
+    "benchmark_exact|normalized_overlap_with_selected_native_test": 1,
+    "benchmark_exact": 14,
+    "query_already_in_main_choice_config": 4,
+    "financial_scope_requires_review": 1,
+    "explicit_financing_lexical_match": 7,
+    "empty_or_nonlexical_query_requires_review": 10,
+    "explicit_financing_lexical_match|explicit_financing_quarantine|native_label_installment_financing": 22,
+    "explicit_financing_lexical_match|explicit_financing_quarantine|native_label_installment_financing|normalized_overlap_with_selected_native_test": 1,
+    "explicit_financing_quarantine|financial_scope_requires_review|native_label_installment_financing": 1,
+    "explicit_financing_quarantine|native_label_installment_financing": 1,
+    "native_brand_query_duplicate": 8,
+    "explicit_financing_lexical_match|native_label_installment_financing": 10,
+    "explicit_financing_lexical_match|financial_scope_requires_review|native_label_installment_financing": 1,
+    "native_label_installment_financing": 2
+  },
+  "by_disposition": {
+    "quarantine": 426,
+    "separate_task": 752,
+    "split_reserved": 1338,
+    "deduplicated": 8
+  },
+  "by_group": {
+    "legacy": 22,
+    "hermes": 884,
+    "legacy_split": 1319,
+    "semantic": 62,
+    "tools": 139,
+    "hint3": 98
+  },
+  "by_original_split": {
+    "train": 2240,
+    "test": 190,
+    "validation": 94
+  },
+  "note": "Quarantined, split-reserved and separate-task rows are distinguished. These counts are not an error-rate estimate.",
+  "sha256": "73214987fbaec508b01d478dac26e863fabc7daa6c24bcd98a5bef7c0ce42b22"
+}

audit/v3/release_summary.json

@@ -0,0 +1,544 @@
+{
+  "version": "2026-10-14-v3",
+  "base_revision": "11e6f657f6d8830da230d4e49c65d0f67f106585",
+  "configurations": {
+    "choice_v3": {
+      "train": {
+        "rows": 113163,
+        "bytes": 165993665,
+        "sha256": "1a6f5d3a1dead75be71ac23ce61f7812eecc5fa65289770adfbf38b470e2fb9c",
+        "parents": 84904,
+        "sources": {
+          "topv2": 20040,
+          "sgd": 33724,
+          "nemotron_agentic": 21131,
+          "glaive_v2": 9992,
+          "nl2bash": 2634,
+          "hermes_v1": 223,
+          "amazon_massive": 13763,
+          "git_nl_commands": 777,
+          "mtop": 6882,
+          "funcdex": 3979,
+          "curated_devops": 18
+        },
+        "task_types": {
+          "intent_match": 13461,
+          "single_choice": 60908,
+          "observed_next_tool": 25333,
+          "intent_no_match": 13461
+        },
+        "candidate_counts": {
+          "2": 60,
+          "3": 81,
+          "4": 61,
+          "5": 60928,
+          "6": 1,
+          "9": 68,
+          "10": 70,
+          "11": 178,
+          "12": 552,
+          "13": 1319,
+          "14": 1340,
+          "15": 1652,
+          "16": 2404,
+          "17": 11963,
+          "18": 2689,
+          "19": 2582,
+          "20": 5563,
+          "21": 1270,
+          "22": 1092,
+          "23": 814,
+          "24": 383,
+          "25": 64,
+          "26": 30,
+          "27": 80,
+          "28": 126,
+          "29": 11,
+          "30": 54,
+          "33": 8898,
+          "62": 266,
+          "64": 132,
+          "65": 8432
+        },
+        "rejection_rows": 13461
+      },
+      "validation": {
+        "rows": 9505,
+        "bytes": 9193104,
+        "sha256": "ff824bfffcc1a3f5bfcf3ac15d14df7ddb429f1d74ca5e35f2291827c20dfc61",
+        "parents": 5023,
+        "sources": {
+          "topv2": 5676,
+          "funcdex": 389,
+          "nemotron_agentic": 1061,
+          "amazon_massive": 407,
+          "mtop": 1554,
+          "sgd": 231,
+          "nl2bash": 51,
+          "glaive_v2": 97,
+          "git_nl_commands": 31,
+          "hermes_v1": 6,
+          "curated_devops": 2
+        },
+        "task_types": {
+          "intent_no_match": 3615,
+          "observed_next_tool": 1456,
+          "intent_match": 3615,
+          "single_choice": 819
+        },
+        "candidate_counts": {
+          "2": 2,
+          "3": 2,
+          "4": 2,
+          "5": 819,
+          "10": 7,
+          "11": 10,
+          "12": 19,
+          "13": 65,
+          "14": 50,
+          "15": 114,
+          "16": 164,
+          "17": 2679,
+          "18": 78,
+          "19": 64,
+          "20": 418,
+          "21": 36,
+          "22": 103,
+          "23": 13,
+          "24": 46,
+          "25": 3,
+          "26": 32,
+          "27": 5,
+          "33": 2404,
+          "62": 48,
+          "64": 26,
+          "65": 2296
+        },
+        "rejection_rows": 3615
+      },
+      "test": {
+        "rows": 21502,
+        "bytes": 15544004,
+        "sha256": "a2645fa4468779772fc7569c5b70b23222e400f72c2b1d05e4b903d5ed81f9ee",
+        "parents": 13916,
+        "sources": {
+          "topv2": 9478,
+          "nl2bash": 345,
+          "sgd": 3889,
+          "mtop": 2742,
+          "nemotron_agentic": 1067,
+          "glaive_v2": 1304,
+          "amazon_massive": 1531,
+          "funcdex": 1034,
+          "git_nl_commands": 75,
+          "hermes_v1": 34,
+          "curated_devops": 3
+        },
+        "task_types": {
+          "intent_no_match": 6110,
+          "single_choice": 7147,
+          "intent_match": 6110,
+          "observed_next_tool": 2135
+        },
+        "candidate_counts": {
+          "2": 9,
+          "3": 16,
+          "4": 9,
+          "5": 7147,
+          "9": 3,
+          "11": 8,
+          "12": 12,
+          "13": 58,
+          "14": 84,
+          "15": 49,
+          "16": 63,
+          "17": 4194,
+          "18": 160,
+          "19": 205,
+          "20": 1078,
+          "21": 78,
+          "22": 33,
+          "23": 54,
+          "24": 5,
+          "27": 5,
+          "28": 12,
+          "30": 4,
+          "33": 4078,
+          "62": 102,
+          "64": 40,
+          "65": 3996
+        },
+        "rejection_rows": 6110
+      }
+    },
+    "legacy_replay_v3": {
+      "train": {
+        "rows": 61131,
+        "bytes": 10028293,
+        "sha256": "1a1d562eaf528a11a6c2964da09340d247b595265d2a80279faa48e7929d7455",
+        "parents": 61131,
+        "sources": {
+          "sgd": 33724,
+          "glaive_v2": 9992,
+          "nl2bash": 2634,
+          "hermes_v1": 223,
+          "amazon_massive": 13763,
+          "git_nl_commands": 777,
+          "curated_devops": 18
+        },
+        "task_types": {
+          "single_choice": 60908,
+          "observed_next_tool": 223
+        },
+        "candidate_counts": {
+          "2": 60,
+          "3": 81,
+          "4": 61,
+          "5": 60928,
+          "6": 1
+        },
+        "rejection_rows": 0
+      },
+      "validation": {
+        "rows": 825,
+        "bytes": 156799,
+        "sha256": "f1f24745eff42a5989e810272b8a5bf42d31f616f96ea77cb1fc109ad69c6695",
+        "parents": 825,
+        "sources": {
+          "amazon_massive": 407,
+          "sgd": 231,
+          "nl2bash": 51,
+          "glaive_v2": 97,
+          "git_nl_commands": 31,
+          "hermes_v1": 6,
+          "curated_devops": 2
+        },
+        "task_types": {
+          "single_choice": 819,
+          "observed_next_tool": 6
+        },
+        "candidate_counts": {
+          "2": 2,
+          "3": 2,
+          "4": 2,
+          "5": 819
+        },
+        "rejection_rows": 0
+      },
+      "test": {
+        "rows": 7181,
+        "bytes": 1215350,
+        "sha256": "a8fa27f305f8d50bc564b84652578a664bb82fb501930b7db081d9388da9edb6",
+        "parents": 7172,
+        "sources": {
+          "nl2bash": 345,
+          "sgd": 3889,
+          "glaive_v2": 1304,
+          "amazon_massive": 1531,
+          "git_nl_commands": 75,
+          "hermes_v1": 34,
+          "curated_devops": 3
+        },
+        "task_types": {
+          "single_choice": 7147,
+          "observed_next_tool": 34
+        },
+        "candidate_counts": {
+          "2": 9,
+          "3": 16,
+          "4": 9,
+          "5": 7147
+        },
+        "rejection_rows": 0
+      }
+    },
+    "semantic_v3": {
+      "train": {
+        "rows": 26922,
+        "bytes": 7556898,
+        "sha256": "16ec519f3f9690cfe7eb3f489bac6feb30a9be4a3f7b93b55d0a30a362f0a3f2",
+        "parents": 13461,
+        "sources": {
+          "topv2": 20040,
+          "mtop": 6882
+        },
+        "task_types": {
+          "intent_match": 13461,
+          "intent_no_match": 13461
+        },
+        "candidate_counts": {
+          "17": 9194,
+          "33": 8898,
+          "62": 266,
+          "64": 132,
+          "65": 8432
+        },
+        "rejection_rows": 13461
+      },
+      "validation": {
+        "rows": 7230,
+        "bytes": 1992132,
+        "sha256": "9160a88e709e74b6bf385894dba280ed3fca1102879448e23b52d65b41a7839b",
+        "parents": 3615,
+        "sources": {
+          "topv2": 5676,
+          "mtop": 1554
+        },
+        "task_types": {
+          "intent_no_match": 3615,
+          "intent_match": 3615
+        },
+        "candidate_counts": {
+          "17": 2456,
+          "33": 2404,
+          "62": 48,
+          "64": 26,
+          "65": 2296
+        },
+        "rejection_rows": 3615
+      },
+      "test": {
+        "rows": 12220,
+        "bytes": 3324079,
+        "sha256": "2a106193f9a32ce905dca30e2b3763ac967c34c81887f83b96e660652d390543",
+        "parents": 6110,
+        "sources": {
+          "topv2": 9478,
+          "mtop": 2742
+        },
+        "task_types": {
+          "intent_no_match": 6110,
+          "intent_match": 6110
+        },
+        "candidate_counts": {
+          "17": 4004,
+          "33": 4078,
+          "62": 102,
+          "64": 40,
+          "65": 3996
+        },
+        "rejection_rows": 6110
+      }
+    },
+    "tool_context_v3": {
+      "train": {
+        "rows": 25110,
+        "bytes": 126211737,
+        "sha256": "7ec090fb52e83b5d31638093b72fe5d17c3b48170f69be5ce024912ec49956ec",
+        "parents": 10312,
+        "sources": {
+          "nemotron_agentic": 21131,
+          "funcdex": 3979
+        },
+        "task_types": {
+          "observed_next_tool": 25110
+        },
+        "candidate_counts": {
+          "9": 68,
+          "10": 70,
+          "11": 178,
+          "12": 552,
+          "13": 1319,
+          "14": 1340,
+          "15": 1652,
+          "16": 2404,
+          "17": 2769,
+          "18": 2689,
+          "19": 2582,
+          "20": 5563,
+          "21": 1270,
+          "22": 1092,
+          "23": 814,
+          "24": 383,
+          "25": 64,
+          "26": 30,
+          "27": 80,
+          "28": 126,
+          "29": 11,
+          "30": 54
+        },
+        "rejection_rows": 0
+      },
+      "validation": {
+        "rows": 1450,
+        "bytes": 5340066,
+        "sha256": "5c037c09d48adaeb7c34c1838049c8232dbc4c40fcc28d3195e21bccec1c7e4d",
+        "parents": 583,
+        "sources": {
+          "funcdex": 389,
+          "nemotron_agentic": 1061
+        },
+        "task_types": {
+          "observed_next_tool": 1450
+        },
+        "candidate_counts": {
+          "10": 7,
+          "11": 10,
+          "12": 19,
+          "13": 65,
+          "14": 50,
+          "15": 114,
+          "16": 164,
+          "17": 223,
+          "18": 78,
+          "19": 64,
+          "20": 418,
+          "21": 36,
+          "22": 103,
+          "23": 13,
+          "24": 46,
+          "25": 3,
+          "26": 32,
+          "27": 5
+        },
+        "rejection_rows": 0
+      },
+      "test": {
+        "rows": 2101,
+        "bytes": 7724286,
+        "sha256": "4d38b21265fafcfd0e6b32685ce501f0f7141545c0d6da73016a627be15ab1c1",
+        "parents": 634,
+        "sources": {
+          "nemotron_agentic": 1067,
+          "funcdex": 1034
+        },
+        "task_types": {
+          "observed_next_tool": 2101
+        },
+        "candidate_counts": {
+          "9": 3,
+          "11": 8,
+          "12": 12,
+          "13": 58,
+          "14": 84,
+          "15": 49,
+          "16": 63,
+          "17": 190,
+          "18": 160,
+          "19": 205,
+          "20": 1078,
+          "21": 78,
+          "22": 33,
+          "23": 54,
+          "24": 5,
+          "27": 5,
+          "28": 12,
+          "30": 4
+        },
+        "rejection_rows": 0
+      }
+    },
+    "tool_environment_holdout_v3": {
+      "validation": {
+        "rows": 534,
+        "bytes": 979530,
+        "sha256": "916608566e8821f2a4fa1ae84566f031e28429041eeba7da5bd1733410841403",
+        "parents": 237,
+        "sources": {
+          "nemotron_agentic": 534
+        },
+        "task_types": {
+          "observed_next_tool": 534
+        },
+        "candidate_counts": {
+          "13": 19,
+          "14": 33,
+          "15": 63,
+          "16": 81,
+          "17": 142,
+          "18": 23,
+          "19": 14,
+          "21": 10,
+          "22": 79,
+          "24": 38,
+          "26": 32
+        },
+        "rejection_rows": 0
+      },
+      "test": {
+        "rows": 561,
+        "bytes": 1008820,
+        "sha256": "4e36c812a6b54a321a45fce805bef13c784b196a177db91c6fcfc87f6c251e1b",
+        "parents": 260,
+        "sources": {
+          "nemotron_agentic": 561
+        },
+        "task_types": {
+          "observed_next_tool": 561
+        },
+        "candidate_counts": {
+          "13": 39,
+          "14": 29,
+          "15": 20,
+          "16": 7,
+          "17": 115,
+          "18": 97,
+          "19": 154,
+          "20": 8,
+          "21": 50,
+          "23": 35,
+          "28": 7
+        },
+        "rejection_rows": 0
+      }
+    },
+    "hermes_actions_v3": {
+      "train": {
+        "rows": 868,
+        "bytes": 1623002,
+        "sha256": "bf4e001ed82700ec73d48bfad64304517784c00a3c4ff4bd7533a1bf58d2e6f9"
+      },
+      "validation": {
+        "rows": 24,
+        "bytes": 83582,
+        "sha256": "429cedc1b030a4b16bb818c41ffd80c8f5cfde7fb76fed61e474923379591dfe"
+      },
+      "test": {
+        "rows": 123,
+        "bytes": 267399,
+        "sha256": "3227e75f589673a0a0269008d8b2761f379cd7a41159a931cd2ea05a0f49bcbe"
+      }
+    },
+    "nlupp_hotels_v3": {
+      "train": {
+        "rows": 765,
+        "bytes": 96788,
+        "sha256": "d7b41fabbc83925de08f2e930392126e15d065bb4796f606880a02c6ab8e272d"
+      },
+      "validation": {
+        "rows": 97,
+        "bytes": 27733,
+        "sha256": "77c1886044a790b6af777e6a09a92dcf8f49d83b5e3ea37e0cfee2d10456e5b7"
+      },
+      "test": {
+        "rows": 98,
+        "bytes": 27170,
+        "sha256": "f95736ffdbc0353673a6e323bc39cad702494e7ee82e609f5fdb9bd3b98912f3"
+      }
+    },
+    "hint3_native_v3": {
+      "train": {
+        "rows": 801,
+        "bytes": 40869,
+        "sha256": "d789a0db0c2a5e70c3275569a05fa5993f60f4f1bf72dddb134ad219c0932291"
+      },
+      "validation": {
+        "rows": 68,
+        "bytes": 12648,
+        "sha256": "2da5a239515f0a8ac85454c19607a7184cb0e805476113d713ee06a2071b27c6"
+      },
+      "test": {
+        "rows": 1348,
+        "bytes": 73960,
+        "sha256": "fa4ea54984aff79bb1aa4b793051b0e9d554881f2c24028528c53b40881e8472"
+      }
+    }
+  },
+  "counts_note": "choice_v3 contains exactly the legacy_replay_v3, semantic_v3 and tool_context_v3 views; environment holdouts overlap the tools view. Auxiliary configs are not concatenated automatically.",
+  "old_data_files_modified": [],
+  "no_training_experiments_run": true,
+  "no_human_gold_annotation_claim": true,
+  "fixed_evaluation_membership_sha256": "b2ee4187546f5326bfb20de788290f445459f903e14098001405e1b7b5c6a0ff",
+  "legacy_native_split_isolation": false,
+  "new_source_test_promoted_to_train": false
+}

audit/v3/semantic_repair.json

@@ -0,0 +1,54 @@
+{
+  "counters": {
+    "retained_train_parents": 13461,
+    "reviewed_semantic_ambiguous_multiple_valid_parents": 8,
+    "reviewed_semantic_needs_context_parents": 12,
+    "persistent_call_rule_requires_distinct_operation_parents": 7,
+    "persistent_call_rule_screen_parents": 4,
+    "retained_validation_parents": 3615,
+    "retained_test_parents": 6110
+  },
+  "inspected_ambiguous_parents": {
+    "mtop_topv2:2de023213157b15aef9741d08d46785b": "reviewed_semantic_needs_context",
+    "mtop_topv2:9e233727fb9d4949484938463f8faec7": "reviewed_semantic_ambiguous_multiple_valid",
+    "mtop_topv2:fe9dfb74857cd2bcd0b20f88f02d8326": "reviewed_semantic_ambiguous_multiple_valid",
+    "mtop_topv2:7e19b483fd7bca8ec68cc2ca968e6f3b": "reviewed_semantic_ambiguous_multiple_valid",
+    "mtop_topv2:da7dd1aea1c962631db73f3a57742310": "reviewed_semantic_needs_context",
+    "mtop_topv2:74bf27f378640289e37cc27e4fe2eb7d": "reviewed_semantic_ambiguous_multiple_valid",
+    "mtop_topv2:fd13eb57d7d0137f2c9c3f388d5a693a": "reviewed_semantic_ambiguous_multiple_valid",
+    "mtop_topv2:dce7556488d134cbbe994fba8b0bb794": "reviewed_semantic_needs_context",
+    "mtop_topv2:80277df55135f398707b395ae25b7781": "persistent_call_rule_requires_distinct_operation",
+    "mtop_topv2:cc3b254031999953358e63124b1dbdbe": "reviewed_semantic_needs_context",
+    "mtop_topv2:c1c3ad3eec6872f57d4ebfb5f215c4ff": "reviewed_semantic_needs_context",
+    "mtop_topv2:029d455b634546d7ab37651e4306c985": "reviewed_semantic_ambiguous_multiple_valid",
+    "mtop_topv2:5528bd22f96505e1708205248913e697": "reviewed_semantic_needs_context",
+    "mtop_topv2:59973bf29842db821ea21b53f8506ba5": "reviewed_semantic_ambiguous_multiple_valid",
+    "mtop_topv2:e03756b41538e1776fb6d9a3f62b021e": "reviewed_semantic_needs_context",
+    "mtop_topv2:63a2bf13fa5788faf17ab4f1a818e7c8": "reviewed_semantic_needs_context",
+    "mtop_topv2:7b60dca163f0689c10fa73a2e6b5a81c": "reviewed_semantic_needs_context",
+    "mtop_topv2:547fc0741b515385d9e3fb4961f12301": "reviewed_semantic_needs_context",
+    "mtop_topv2:be5fa5a51c541a0d78aff216e2521616": "reviewed_semantic_needs_context",
+    "mtop_topv2:33a2f3e6bce0182b644f4fc4301a2605": "reviewed_semantic_ambiguous_multiple_valid",
+    "mtop_topv2:7bd32b7787562657929b248792f7b413": "reviewed_semantic_needs_context",
+    "mtop_topv2:6a6f693f421f94529b667ec18e472e44": "persistent_call_rule_requires_distinct_operation",
+    "mtop_topv2:328a48fe493175d3437f3a36eb0147e2": "persistent_call_rule_requires_distinct_operation",
+    "mtop_topv2:82696824b493ed4aa9b6e4eb59d7c672": "persistent_call_rule_requires_distinct_operation",
+    "mtop_topv2:aad52c8398f5e9ce06d2ec6c0a4a27c9": "persistent_call_rule_requires_distinct_operation",
+    "mtop_topv2:a9970e0a625aa451b3dc125dc9ca4280": "persistent_call_rule_requires_distinct_operation",
+    "mtop_topv2:0e5c8708d146af1842d71d5a2c998d86": "persistent_call_rule_requires_distinct_operation"
+  },
+  "expected_resolved_findings": [
+    "S01",
+    "S02",
+    "S03",
+    "S04",
+    "S05"
+  ],
+  "separate_operation_quarantined": [
+    "S06"
+  ],
+  "all_pairs_regenerated": true,
+  "source_splits_unchanged": true,
+  "natural_catalog_limit": 68,
+  "synthetic_oos_not_added": true
+}

audit/v3/token_budget.json

@@ -0,0 +1,430 @@
+{
+  "limit": 16384,
+  "encoder_sha256": "0e304cf7c6500e8bb59bef7e2afd2c6373f82596dfb3b57d1aa93c175e2dc3a3",
+  "tokenizer_manifest": {
+    "repo": "Cloudflare/clef-flash",
+    "revision": "8b2e5fd17c09fd49bc2880b805d5323ec7ff4ff3",
+    "files": [
+      {
+        "file": "tokenizer.json",
+        "bytes": 19989325,
+        "sha256": "06b9509352d2af50381ab2247e083b80d32d5c0aba91c272ca9ff729b6a0e523"
+      },
+      {
+        "file": "tokenizer_config.json",
+        "bytes": 1075,
+        "sha256": "91a08f825d370d085d692e04cf117cdd7faad7bf18e996f1e6031b6dab03db72"
+      }
+    ]
+  },
+  "adapter_sha256": "619ad2cd0afeeb74db8ae099707c6963dd1268009734602c79026930eb172e6c",
+  "splits": {
+    "train": {
+      "rows": 113163,
+      "max": 8051,
+      "quantiles": {
+        "0.5": 371.0,
+        "0.9": 3183.0,
+        "0.99": 3721.3800000000047,
+        "1": 8051.0
+      },
+      "by_source": {
+        "topv2": {
+          "rows": 20040,
+          "max": 3438
+        },
+        "sgd": {
+          "rows": 33724,
+          "max": 406
+        },
+        "nemotron_agentic": {
+          "rows": 21131,
+          "max": 5113
+        },
+        "glaive_v2": {
+          "rows": 9992,
+          "max": 510
+        },
+        "nl2bash": {
+          "rows": 2634,
+          "max": 440
+        },
+        "hermes_v1": {
+          "rows": 223,
+          "max": 1876
+        },
+        "amazon_massive": {
+          "rows": 13763,
+          "max": 404
+        },
+        "git_nl_commands": {
+          "rows": 777,
+          "max": 380
+        },
+        "mtop": {
+          "rows": 6882,
+          "max": 3420
+        },
+        "funcdex": {
+          "rows": 3979,
+          "max": 8051
+        },
+        "curated_devops": {
+          "rows": 18,
+          "max": 403
+        }
+      },
+      "over_limit": 0
+    },
+    "validation": {
+      "rows": 9505,
+      "max": 5756,
+      "quantiles": {
+        "0.5": 1783.0,
+        "0.9": 3380.0,
+        "0.99": 3713.959999999999,
+        "1": 5756.0
+      },
+      "by_source": {
+        "topv2": {
+          "rows": 5676,
+          "max": 3428
+        },
+        "funcdex": {
+          "rows": 389,
+          "max": 5756
+        },
+        "nemotron_agentic": {
+          "rows": 1061,
+          "max": 4407
+        },
+        "amazon_massive": {
+          "rows": 407,
+          "max": 389
+        },
+        "mtop": {
+          "rows": 1554,
+          "max": 3417
+        },
+        "sgd": {
+          "rows": 231,
+          "max": 381
+        },
+        "nl2bash": {
+          "rows": 51,
+          "max": 373
+        },
+        "glaive_v2": {
+          "rows": 97,
+          "max": 391
+        },
+        "git_nl_commands": {
+          "rows": 31,
+          "max": 378
+        },
+        "hermes_v1": {
+          "rows": 6,
+          "max": 1150
+        },
+        "curated_devops": {
+          "rows": 2,
+          "max": 376
+        }
+      },
+      "over_limit": 0
+    },
+    "test": {
+      "rows": 21502,
+      "max": 11482,
+      "quantiles": {
+        "0.5": 1001.0,
+        "0.9": 3377.0,
+        "0.99": 4293.880000000019,
+        "1": 11482.0
+      },
+      "by_source": {
+        "topv2": {
+          "rows": 9478,
+          "max": 3428
+        },
+        "nl2bash": {
+          "rows": 345,
+          "max": 381
+        },
+        "sgd": {
+          "rows": 3889,
+          "max": 401
+        },
+        "mtop": {
+          "rows": 2742,
+          "max": 3419
+        },
+        "nemotron_agentic": {
+          "rows": 1067,
+          "max": 4275
+        },
+        "glaive_v2": {
+          "rows": 1304,
+          "max": 505
+        },
+        "amazon_massive": {
+          "rows": 1531,
+          "max": 390
+        },
+        "funcdex": {
+          "rows": 1034,
+          "max": 11482
+        },
+        "git_nl_commands": {
+          "rows": 75,
+          "max": 369
+        },
+        "hermes_v1": {
+          "rows": 34,
+          "max": 1455
+        },
+        "curated_devops": {
+          "rows": 3,
+          "max": 369
+        }
+      },
+      "over_limit": 0
+    }
+  },
+  "no_match_position_histograms": {
+    "33": {
+      "8": 229,
+      "15": 236,
+      "26": 242,
+      "29": 222,
+      "14": 238,
+      "9": 228,
+      "10": 235,
+      "20": 244,
+      "0": 219,
+      "31": 230,
+      "23": 241,
+      "24": 230,
+      "6": 229,
+      "5": 222,
+      "25": 210,
+      "2": 244,
+      "16": 227,
+      "27": 233,
+      "7": 255,
+      "13": 213,
+      "18": 256,
+      "21": 257,
+      "12": 239,
+      "32": 234,
+      "22": 241,
+      "19": 224,
+      "28": 212,
+      "11": 211,
+      "4": 244,
+      "17": 218,
+      "30": 245,
+      "1": 249,
+      "3": 233
+    },
+    "65": {
+      "14": 113,
+      "5": 116,
+      "39": 110,
+      "46": 106,
+      "55": 105,
+      "0": 103,
+      "56": 118,
+      "11": 103,
+      "16": 124,
+      "40": 116,
+      "37": 113,
+      "15": 109,
+      "1": 121,
+      "51": 109,
+      "48": 125,
+      "44": 115,
+      "28": 103,
+      "49": 121,
+      "52": 107,
+      "13": 114,
+      "3": 95,
+      "19": 117,
+      "54": 102,
+      "60": 119,
+      "57": 109,
+      "61": 128,
+      "27": 145,
+      "47": 110,
+      "50": 138,
+      "6": 113,
+      "9": 115,
+      "35": 108,
+      "17": 138,
+      "45": 105,
+      "32": 116,
+      "43": 123,
+      "58": 104,
+      "29": 128,
+      "23": 100,
+      "42": 100,
+      "38": 120,
+      "4": 109,
+      "21": 134,
+      "30": 112,
+      "12": 125,
+      "8": 114,
+      "62": 97,
+      "25": 114,
+      "36": 100,
+      "63": 112,
+      "22": 115,
+      "10": 105,
+      "34": 134,
+      "2": 113,
+      "53": 112,
+      "20": 94,
+      "64": 113,
+      "26": 106,
+      "24": 105,
+      "59": 110,
+      "33": 115,
+      "7": 115,
+      "41": 118,
+      "31": 89,
+      "18": 117
+    },
+    "17": {
+      "14": 457,
+      "11": 457,
+      "4": 448,
+      "6": 467,
+      "16": 443,
+      "1": 487,
+      "7": 465,
+      "5": 452,
+      "15": 456,
+      "2": 458,
+      "3": 460,
+      "12": 491,
+      "9": 457,
+      "8": 450,
+      "0": 447,
+      "10": 473,
+      "13": 459
+    },
+    "62": {
+      "15": 3,
+      "22": 4,
+      "31": 7,
+      "13": 3,
+      "25": 4,
+      "29": 6,
+      "54": 6,
+      "39": 6,
+      "59": 4,
+      "33": 2,
+      "56": 3,
+      "9": 4,
+      "18": 3,
+      "14": 3,
+      "23": 3,
+      "6": 4,
+      "30": 10,
+      "11": 2,
+      "5": 2,
+      "49": 4,
+      "55": 4,
+      "57": 1,
+      "24": 2,
+      "19": 2,
+      "52": 5,
+      "16": 1,
+      "0": 5,
+      "35": 6,
+      "38": 3,
+      "58": 2,
+      "50": 2,
+      "1": 4,
+      "61": 4,
+      "7": 7,
+      "17": 5,
+      "20": 3,
+      "41": 3,
+      "10": 2,
+      "26": 5,
+      "44": 5,
+      "51": 2,
+      "8": 2,
+      "60": 3,
+      "48": 3,
+      "32": 2,
+      "37": 2,
+      "3": 5,
+      "36": 2,
+      "28": 4,
+      "12": 7,
+      "4": 5,
+      "45": 3,
+      "43": 2,
+      "47": 2,
+      "53": 1,
+      "40": 2,
+      "27": 2,
+      "34": 3,
+      "21": 1,
+      "42": 1
+    },
+    "64": {
+      "51": 2,
+      "41": 4,
+      "29": 3,
+      "36": 1,
+      "31": 4,
+      "54": 3,
+      "30": 2,
+      "34": 1,
+      "24": 1,
+      "57": 2,
+      "26": 1,
+      "37": 3,
+      "22": 2,
+      "52": 2,
+      "39": 1,
+      "42": 1,
+      "53": 3,
+      "49": 2,
+      "2": 2,
+      "59": 1,
+      "13": 2,
+      "43": 5,
+      "33": 4,
+      "5": 3,
+      "27": 1,
+      "32": 1,
+      "28": 2,
+      "35": 2,
+      "45": 1,
+      "55": 1,
+      "63": 4,
+      "1": 3,
+      "50": 2,
+      "0": 1,
+      "23": 1,
+      "40": 1,
+      "17": 4,
+      "19": 2,
+      "47": 1,
+      "61": 2,
+      "8": 4,
+      "3": 1,
+      "10": 2,
+      "48": 1,
+      "16": 1,
+      "14": 2,
+      "21": 1,
+      "11": 2,
+      "4": 1
+    }
+  }
+}

audit/v3/tool_repair.json

@@ -0,0 +1,94 @@
+{
+  "counters": {
+    "nemotron_steps_native_schema_restored": 23259,
+    "nemotron_steps_arguments_available": 10792,
+    "funcdex_train_quarantined_steps": 31,
+    "funcdex_steps_with_undocumented_argument_keys": 51,
+    "nemotron_agentic_train_quarantined_steps": 5,
+    "funcdex_validation_quarantined_steps": 56,
+    "nemotron_agentic_validation_quarantined_steps": 4,
+    "funcdex_test_quarantined_steps": 36,
+    "nemotron_agentic_test_quarantined_steps": 7
+  },
+  "blocked_parent_count": 31,
+  "retained_counts": {
+    "train": 25110,
+    "validation": 1450,
+    "test": 2101
+  },
+  "original_reviewed_blocked_parent_count": 31,
+  "new_native_contract_invalid_parent_count": 0,
+  "primary_recovery_sha256": "28bd5ddbb74bf9bcb8d03bf4ab22d52d5b59b45b4b6d1edd97548c285e270943",
+  "recovered_steps": 23275,
+  "native_contract_status_counts": {
+    "native_contract_valid_not_executed": 23246,
+    "native_parameter_schema_unspecified": 29
+  },
+  "reserved_environment_ids": {
+    "validation": [
+      "nemotron_env:0868967b50316f10869cee3adceb2434",
+      "nemotron_env:0d42cef464dac5995c3a0d2f16d39415",
+      "nemotron_env:10ee2c010f79c683fe1e7e84bd127e5f",
+      "nemotron_env:119cbabe108db9849e91a528fffe0d23",
+      "nemotron_env:16e6edd32fa3109ae06d1ef5676a7152",
+      "nemotron_env:1b3ae54558b5b06ffc6fd19986aa06c7",
+      "nemotron_env:25359fd156332413e4bbf2c0c52df386",
+      "nemotron_env:3129418a2a804b8f9e2448a5cfa47b5c",
+      "nemotron_env:3603801b75e7295c535cf44c1021e9ef",
+      "nemotron_env:38f1b53e0458235dde486fb356a2e5f3",
+      "nemotron_env:609b27d16c32518870e2664a5072cd32",
+      "nemotron_env:6c4abc6fc433a2482e482909b90dfc5c",
+      "nemotron_env:7ff63b9426fc15f6e5d04eea7b25604d",
+      "nemotron_env:aeb7718f391cdcadb0478fd12be2c076",
+      "nemotron_env:af243ed2c5eb9541f8eae35d9211893e",
+      "nemotron_env:d3f373a7cb0a0d532f496a9cf1c6b8f1",
+      "nemotron_env:d579c257750e1b0b716990c189c5567d",
+      "nemotron_env:d6593091e2614549d785a034fde22e50",
+      "nemotron_env:d7b96ced61cb54b2d9e95dbda47a49fc",
+      "nemotron_env:ded377ab5346d80cfd9ed6b3f1705b54",
+      "nemotron_env:e709a844af8881d076c2f7ef13c04fbf",
+      "nemotron_env:ef03cc103d0800fa3c93dfd60b789dfe",
+      "nemotron_env:fb247ef93ce78d760b11da1d36c95381"
+    ],
+    "test": [
+      "nemotron_env:00553a72407b45907545ab5c83e9986f",
+      "nemotron_env:066bdb257fbebc6b6cf49c54df8ece97",
+      "nemotron_env:07551020808df088e5b060e5bb0db551",
+      "nemotron_env:13b88c59e5fede932e47a0ac34634dd0",
+      "nemotron_env:231b2e50ccca4c137eff5cd61b5ace70",
+      "nemotron_env:239f885e3e367adcd583918b6c8ca81a",
+      "nemotron_env:3bd536e81280810eb5a839432c0af353",
+      "nemotron_env:4929b42c5eaec39cfdd5c866ea107e8a",
+      "nemotron_env:5e796b6c983f414e45e67380578cf23a",
+      "nemotron_env:6ce9c3a0fe34d8a969a52271771f13f3",
+      "nemotron_env:7ac6875bc6135dba4aa49f815b30ad42",
+      "nemotron_env:7b684b23aa60fd4aeb1221f70a7cb4fc",
+      "nemotron_env:9c7bca8266b2dc3bd8e5238032741f10",
+      "nemotron_env:af9d5b20ecb6834f223f6fb390922e37",
+      "nemotron_env:b4952198936ac3b1cf6ca1659d1b83f7",
+      "nemotron_env:bbba8913ce5ddf01d78722edd2698a56",
+      "nemotron_env:c491c9bbba8551128275b3198ebf4dd3",
+      "nemotron_env:cc8bc4f4f18bbcf42b7333d018c55c07",
+      "nemotron_env:cff3638dcc62fa85ea52b5d8e874ce71",
+      "nemotron_env:d619b461e5fa0859f918840145bb03a1",
+      "nemotron_env:de352bdfdfeda33f65730e35991269b9",
+      "nemotron_env:e884e2022e07a53c264f1e6a28448c6f",
+      "nemotron_env:f96cfdd12a80a84da7a985c284dceaff"
+    ]
+  },
+  "reserved_environment_counts": {
+    "validation": 23,
+    "test": 23
+  },
+  "reserved_holdout_rows": {
+    "validation": 534,
+    "test": 561
+  },
+  "selection_seed": 2026101403,
+  "prior_train_only_environment_count": 466,
+  "source_eval_promoted_to_train": false,
+  "original_query_context_and_target_preserved": true,
+  "original_brief_tools_replaced_with_native_schemas": true,
+  "no_call_supervision_claimed": false,
+  "scope": "The environment holdout is disjoint from v3 training inside the Nemotron source. Existing in-environment eval remains separate via provenance. Public source exposure of previously trained models is not ruled out."
+}

audit/v3/validation.json

@@ -0,0 +1,79 @@
+{
+  "status": "pass",
+  "base_revision": "11e6f657f6d8830da230d4e49c65d0f67f106585",
+  "configs": {
+    "choice_v3": {
+      "train": 113163,
+      "validation": 9505,
+      "test": 21502
+    },
+    "legacy_replay_v3": {
+      "train": 61131,
+      "validation": 825,
+      "test": 7181
+    },
+    "semantic_v3": {
+      "train": 26922,
+      "validation": 7230,
+      "test": 12220
+    },
+    "tool_context_v3": {
+      "train": 25110,
+      "validation": 1450,
+      "test": 2101
+    },
+    "tool_environment_holdout_v3": {
+      "validation": 534,
+      "test": 561
+    },
+    "hermes_actions_v3": {
+      "train": 868,
+      "validation": 24,
+      "test": 123
+    },
+    "nlupp_hotels_v3": {
+      "train": 765,
+      "validation": 97,
+      "test": 98
+    },
+    "hint3_native_v3": {
+      "train": 801,
+      "validation": 68,
+      "test": 1348
+    }
+  },
+  "checks": {
+    "choice_subset_identity_and_structure": "pass",
+    "quarantined_main_parents_absent": "pass",
+    "all_semantic_pairs_and_44_known_false_none": "pass",
+    "historical_holdouts_and_legacy_validation_coverage": "pass",
+    "hermes_full_input_native_calls_and_single_choice_eligibility": "pass",
+    "nlu_native_annotation_boolean_mask_and_22_date_pairs": "pass",
+    "native_tool_schemas_and_environment_holdouts": "pass",
+    "hint3_no_test_oos_promoted_to_training": "pass",
+    "actual_encoder_all_choice_tokens_no_truncation": "pass",
+    "random_ids_expected_mapping_and_auxiliary_rejection": "pass"
+  },
+  "known_issues": {
+    "S01_route_retrieval_alternative_remains": {
+      "regenerated_with_repair": 35
+    },
+    "S02_online_contact_filter_alternative_remains": {
+      "regenerated_with_repair": 9
+    },
+    "S03_nutritional_query_description_gap": {
+      "regenerated_with_repair": 97
+    },
+    "S04_alarm_rename_description_gap": {
+      "regenerated_with_repair": 12
+    },
+    "S05_timer_history_description_gap": {
+      "regenerated_with_repair": 2
+    },
+    "S06_persistent_call_policy_description_gap": {
+      "explicitly_quarantined": 7
+    }
+  },
+  "native_schema_rows_verified": 23259,
+  "known_issue_count_scope": "Previously documented cases only; no dataset-wide human error-rate claim."
+}

catalogs/hint3_native_labels_v3.json

@@ -0,0 +1,563 @@
+[
+  {
+    "brand": "curekart",
+    "native_label": "CALL_CENTER",
+    "description_zh": "呼叫中心营业与疫情期间可用性",
+    "finance_category": "nonfinancial_retail_health_or_service_label",
+    "recommendation": "apply_row_level_filters",
+    "native_train_rows": 21,
+    "native_test_rows": 5,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "curekart",
+    "native_label": "CANCEL_ORDER",
+    "description_zh": "取消零售订单",
+    "finance_category": "nonfinancial_retail_health_or_service_label",
+    "recommendation": "apply_row_level_filters",
+    "native_train_rows": 12,
+    "native_test_rows": 32,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "curekart",
+    "native_label": "CHAT_WITH_AGENT",
+    "description_zh": "联系人工客服",
+    "finance_category": "nonfinancial_retail_health_or_service_label",
+    "recommendation": "apply_row_level_filters",
+    "native_train_rows": 40,
+    "native_test_rows": 14,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "curekart",
+    "native_label": "CHECK_PINCODE",
+    "description_zh": "邮编区域是否配送；不是金融 PIN",
+    "finance_category": "nonfinancial_retail_health_or_service_label",
+    "recommendation": "apply_row_level_filters",
+    "native_train_rows": 14,
+    "native_test_rows": 22,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "curekart",
+    "native_label": "CONSULT_START",
+    "description_zh": "开始健康、营养或体重管理咨询",
+    "finance_category": "nonfinancial_retail_health_or_service_label",
+    "recommendation": "apply_row_level_filters",
+    "native_train_rows": 26,
+    "native_test_rows": 16,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "curekart",
+    "native_label": "DELAY_IN_PARCEL",
+    "description_zh": "包裹配送延迟，含疫情影响",
+    "finance_category": "nonfinancial_retail_health_or_service_label",
+    "recommendation": "apply_row_level_filters",
+    "native_train_rows": 23,
+    "native_test_rows": 10,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "curekart",
+    "native_label": "EXPIRY_DATE",
+    "description_zh": "商品保质期或收到过期商品",
+    "finance_category": "nonfinancial_retail_health_or_service_label",
+    "recommendation": "apply_row_level_filters",
+    "native_train_rows": 8,
+    "native_test_rows": 13,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "curekart",
+    "native_label": "FRANCHISE",
+    "description_zh": "加盟、卖家或零售合作咨询",
+    "finance_category": "nonfinancial_retail_health_or_service_label",
+    "recommendation": "apply_row_level_filters",
+    "native_train_rows": 12,
+    "native_test_rows": 12,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "curekart",
+    "native_label": "IMMUNITY",
+    "description_zh": "免疫力相关健康信息",
+    "finance_category": "nonfinancial_retail_health_or_service_label",
+    "recommendation": "apply_row_level_filters",
+    "native_train_rows": 6,
+    "native_test_rows": 0,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "curekart",
+    "native_label": "INTERNATIONAL_SHIPPING",
+    "description_zh": "国际配送可用性",
+    "finance_category": "nonfinancial_retail_health_or_service_label",
+    "recommendation": "apply_row_level_filters",
+    "native_train_rows": 3,
+    "native_test_rows": 0,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "curekart",
+    "native_label": "MODES_OF_PAYMENTS",
+    "description_zh": "零售结账支付方式",
+    "finance_category": "retail_transaction_or_monetary_offer",
+    "recommendation": "apply_shared_retail_policy_and_text_financing_screen",
+    "native_train_rows": 7,
+    "native_test_rows": 0,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "curekart",
+    "native_label": "MODIFY_ADDRESS",
+    "description_zh": "更改配送地址",
+    "finance_category": "nonfinancial_retail_health_or_service_label",
+    "recommendation": "apply_row_level_filters",
+    "native_train_rows": 16,
+    "native_test_rows": 3,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "curekart",
+    "native_label": "NO_NODES_DETECTED",
+    "description_zh": "相对于该品牌完整原生目录的 OOS；不是全局无匹配",
+    "finance_category": "source_relative_oos",
+    "recommendation": "native_test_only_and_text_finance_screen",
+    "native_train_rows": 0,
+    "native_test_rows": 532,
+    "explicit_financing_quarantine_rows": 4,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "curekart",
+    "native_label": "ORDER_QUERY",
+    "description_zh": "一般订单支持请求",
+    "finance_category": "nonfinancial_retail_health_or_service_label",
+    "recommendation": "apply_row_level_filters",
+    "native_train_rows": 7,
+    "native_test_rows": 0,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "curekart",
+    "native_label": "ORDER_STATUS",
+    "description_zh": "零售订单与物流状态",
+    "finance_category": "nonfinancial_retail_health_or_service_label",
+    "recommendation": "apply_row_level_filters",
+    "native_train_rows": 47,
+    "native_test_rows": 12,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "curekart",
+    "native_label": "ORDER_TAKING",
+    "description_zh": "是否接受订单，含疫情期间运营",
+    "finance_category": "nonfinancial_retail_health_or_service_label",
+    "recommendation": "apply_row_level_filters",
+    "native_train_rows": 39,
+    "native_test_rows": 19,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "curekart",
+    "native_label": "ORIGINAL_PRODUCT",
+    "description_zh": "商品真伪或正品保证",
+    "finance_category": "nonfinancial_retail_health_or_service_label",
+    "recommendation": "apply_row_level_filters",
+    "native_train_rows": 23,
+    "native_test_rows": 43,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "curekart",
+    "native_label": "PAYMENT_AND_BILL",
+    "description_zh": "零售支付、扣款、账单或订单未生成问题",
+    "finance_category": "retail_transaction_or_monetary_offer",
+    "recommendation": "apply_shared_retail_policy_and_text_financing_screen",
+    "native_train_rows": 26,
+    "native_test_rows": 3,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 1
+  },
+  {
+    "brand": "curekart",
+    "native_label": "PORTAL_ISSUE",
+    "description_zh": "购物网站或界面故障",
+    "finance_category": "nonfinancial_retail_health_or_service_label",
+    "recommendation": "apply_row_level_filters",
+    "native_train_rows": 4,
+    "native_test_rows": 0,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "curekart",
+    "native_label": "RECOMMEND_PRODUCT",
+    "description_zh": "营养或健康商品推荐与商品咨询",
+    "finance_category": "nonfinancial_retail_health_or_service_label",
+    "recommendation": "apply_row_level_filters",
+    "native_train_rows": 95,
+    "native_test_rows": 205,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "curekart",
+    "native_label": "REFER_EARN",
+    "description_zh": "零售推荐奖励及兑换",
+    "finance_category": "retail_transaction_or_monetary_offer",
+    "recommendation": "apply_shared_retail_policy_and_text_financing_screen",
+    "native_train_rows": 13,
+    "native_test_rows": 0,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "curekart",
+    "native_label": "REFUNDS_RETURNS_REPLACEMENTS",
+    "description_zh": "零售退款、退货或更换",
+    "finance_category": "retail_transaction_or_monetary_offer",
+    "recommendation": "apply_shared_retail_policy_and_text_financing_screen",
+    "native_train_rows": 54,
+    "native_test_rows": 20,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "curekart",
+    "native_label": "RESUME_DELIVERY",
+    "description_zh": "何时恢复配送，含疫情影响",
+    "finance_category": "nonfinancial_retail_health_or_service_label",
+    "recommendation": "apply_row_level_filters",
+    "native_train_rows": 51,
+    "native_test_rows": 9,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "curekart",
+    "native_label": "SIDE_EFFECT",
+    "description_zh": "商品副作用咨询",
+    "finance_category": "nonfinancial_retail_health_or_service_label",
+    "recommendation": "apply_row_level_filters",
+    "native_train_rows": 4,
+    "native_test_rows": 5,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "curekart",
+    "native_label": "SIGN_UP",
+    "description_zh": "注册购物服务账号",
+    "finance_category": "nonfinancial_retail_health_or_service_label",
+    "recommendation": "apply_row_level_filters",
+    "native_train_rows": 7,
+    "native_test_rows": 3,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "curekart",
+    "native_label": "START_OVER",
+    "description_zh": "重新开始或回到主菜单的对话控制",
+    "finance_category": "nonfinancial_retail_health_or_service_label",
+    "recommendation": "apply_row_level_filters",
+    "native_train_rows": 5,
+    "native_test_rows": 0,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "curekart",
+    "native_label": "STORE_INFORMATION",
+    "description_zh": "线下店铺信息与营业状态",
+    "finance_category": "nonfinancial_retail_health_or_service_label",
+    "recommendation": "apply_row_level_filters",
+    "native_train_rows": 14,
+    "native_test_rows": 11,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "curekart",
+    "native_label": "USER_GOAL_FORM",
+    "description_zh": "用户健康目标或资料重新评估",
+    "finance_category": "nonfinancial_retail_health_or_service_label",
+    "recommendation": "apply_row_level_filters",
+    "native_train_rows": 12,
+    "native_test_rows": 2,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "curekart",
+    "native_label": "WORK_FROM_HOME",
+    "description_zh": "公司办公室疫情期间运营状态",
+    "finance_category": "nonfinancial_retail_health_or_service_label",
+    "recommendation": "apply_row_level_filters",
+    "native_train_rows": 10,
+    "native_test_rows": 0,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "sofmattress",
+    "native_label": "100_NIGHT_TRIAL_OFFER",
+    "description_zh": "床垫 100 晚试睡政策",
+    "finance_category": "retail_transaction_or_monetary_offer",
+    "recommendation": "apply_shared_retail_policy_and_text_financing_screen",
+    "native_train_rows": 18,
+    "native_test_rows": 5,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "sofmattress",
+    "native_label": "ABOUT_SOF_MATTRESS",
+    "description_zh": "品牌介绍及区别",
+    "finance_category": "nonfinancial_retail_health_or_service_label",
+    "recommendation": "apply_row_level_filters",
+    "native_train_rows": 11,
+    "native_test_rows": 3,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "sofmattress",
+    "native_label": "CANCEL_ORDER",
+    "description_zh": "取消零售订单",
+    "finance_category": "nonfinancial_retail_health_or_service_label",
+    "recommendation": "apply_row_level_filters",
+    "native_train_rows": 10,
+    "native_test_rows": 6,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "sofmattress",
+    "native_label": "CHECK_PINCODE",
+    "description_zh": "邮编区域是否配送；不是金融 PIN",
+    "finance_category": "nonfinancial_retail_health_or_service_label",
+    "recommendation": "apply_row_level_filters",
+    "native_train_rows": 10,
+    "native_test_rows": 25,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "sofmattress",
+    "native_label": "COD",
+    "description_zh": "零售货到付款可用性",
+    "finance_category": "retail_transaction_or_monetary_offer",
+    "recommendation": "apply_shared_retail_policy_and_text_financing_screen",
+    "native_train_rows": 12,
+    "native_test_rows": 12,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "sofmattress",
+    "native_label": "COMPARISON",
+    "description_zh": "Ergo 与 Ortho 产品比较",
+    "finance_category": "nonfinancial_retail_health_or_service_label",
+    "recommendation": "apply_row_level_filters",
+    "native_train_rows": 11,
+    "native_test_rows": 18,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "sofmattress",
+    "native_label": "DELAY_IN_DELIVERY",
+    "description_zh": "配送延迟；一些原生表达依赖对话上下文",
+    "finance_category": "nonfinancial_retail_health_or_service_label",
+    "recommendation": "apply_row_level_filters",
+    "native_train_rows": 11,
+    "native_test_rows": 15,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "sofmattress",
+    "native_label": "DISTRIBUTORS",
+    "description_zh": "线下展厅、经销商或体验地点",
+    "finance_category": "nonfinancial_retail_health_or_service_label",
+    "recommendation": "apply_row_level_filters",
+    "native_train_rows": 34,
+    "native_test_rows": 13,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "sofmattress",
+    "native_label": "EMI",
+    "description_zh": "分期付款、零息融资及首付",
+    "finance_category": "explicit_financing",
+    "recommendation": "quarantine_all_rows_preserve_native_label",
+    "native_train_rows": 25,
+    "native_test_rows": 13,
+    "explicit_financing_quarantine_rows": 38,
+    "other_finance_lexical_review_rows": 2
+  },
+  {
+    "brand": "sofmattress",
+    "native_label": "ERGO_FEATURES",
+    "description_zh": "Ergo 床垫功能；产品变体不代表新通用意图",
+    "finance_category": "nonfinancial_retail_health_or_service_label",
+    "recommendation": "apply_row_level_filters",
+    "native_train_rows": 11,
+    "native_test_rows": 3,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "sofmattress",
+    "native_label": "LEAD_GEN",
+    "description_zh": "联系销售或人工服务",
+    "finance_category": "nonfinancial_retail_health_or_service_label",
+    "recommendation": "apply_row_level_filters",
+    "native_train_rows": 21,
+    "native_test_rows": 17,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "sofmattress",
+    "native_label": "MATTRESS_COST",
+    "description_zh": "床垫零售价格",
+    "finance_category": "retail_transaction_or_monetary_offer",
+    "recommendation": "apply_shared_retail_policy_and_text_financing_screen",
+    "native_train_rows": 22,
+    "native_test_rows": 23,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "sofmattress",
+    "native_label": "NO_NODES_DETECTED",
+    "description_zh": "相对于该品牌完整原生目录的 OOS；不是全局无匹配",
+    "finance_category": "source_relative_oos",
+    "recommendation": "native_test_only_and_text_finance_screen",
+    "native_train_rows": 0,
+    "native_test_rows": 144,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "sofmattress",
+    "native_label": "OFFERS",
+    "description_zh": "零售优惠；个别测试查询同时涉及 EMI",
+    "finance_category": "retail_transaction_or_monetary_offer",
+    "recommendation": "apply_shared_retail_policy_and_text_financing_screen",
+    "native_train_rows": 10,
+    "native_test_rows": 11,
+    "explicit_financing_quarantine_rows": 1,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "sofmattress",
+    "native_label": "ORDER_STATUS",
+    "description_zh": "零售订单或配送状态",
+    "finance_category": "nonfinancial_retail_health_or_service_label",
+    "recommendation": "apply_row_level_filters",
+    "native_train_rows": 21,
+    "native_test_rows": 8,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "sofmattress",
+    "native_label": "ORTHO_FEATURES",
+    "description_zh": "Ortho 床垫功能；产品变体不代表新通用意图",
+    "finance_category": "nonfinancial_retail_health_or_service_label",
+    "recommendation": "apply_row_level_filters",
+    "native_train_rows": 17,
+    "native_test_rows": 11,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "sofmattress",
+    "native_label": "PILLOWS",
+    "description_zh": "枕头商品是否销售或相关咨询",
+    "finance_category": "nonfinancial_retail_health_or_service_label",
+    "recommendation": "apply_row_level_filters",
+    "native_train_rows": 10,
+    "native_test_rows": 13,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "sofmattress",
+    "native_label": "PRODUCT_VARIANTS",
+    "description_zh": "床垫产品变体信息",
+    "finance_category": "nonfinancial_retail_health_or_service_label",
+    "recommendation": "apply_row_level_filters",
+    "native_train_rows": 21,
+    "native_test_rows": 8,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "sofmattress",
+    "native_label": "RETURN_EXCHANGE",
+    "description_zh": "零售退货、换货或退款；个别测试查询同时涉及 EMI",
+    "finance_category": "retail_transaction_or_monetary_offer",
+    "recommendation": "apply_shared_retail_policy_and_text_financing_screen",
+    "native_train_rows": 14,
+    "native_test_rows": 12,
+    "explicit_financing_quarantine_rows": 2,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "sofmattress",
+    "native_label": "SIZE_CUSTOMIZATION",
+    "description_zh": "床垫尺寸修改或定制",
+    "finance_category": "nonfinancial_retail_health_or_service_label",
+    "recommendation": "apply_row_level_filters",
+    "native_train_rows": 9,
+    "native_test_rows": 15,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "sofmattress",
+    "native_label": "WARRANTY",
+    "description_zh": "商品保修政策",
+    "finance_category": "nonfinancial_retail_health_or_service_label",
+    "recommendation": "apply_row_level_filters",
+    "native_train_rows": 10,
+    "native_test_rows": 0,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  },
+  {
+    "brand": "sofmattress",
+    "native_label": "WHAT_SIZE_TO_ORDER",
+    "description_zh": "选择购买尺寸",
+    "finance_category": "nonfinancial_retail_health_or_service_label",
+    "recommendation": "apply_row_level_filters",
+    "native_train_rows": 20,
+    "native_test_rows": 22,
+    "explicit_financing_quarantine_rows": 0,
+    "other_finance_lexical_review_rows": 0
+  }
+]

catalogs/nlupp_hotels_modules_v3.json

@@ -0,0 +1,243 @@
+{
+  "accesibility": {
+    "description": "is the intent to ask something related to accessibility?",
+    "domain": [
+      "hotels"
+    ]
+  },
+  "acknowledge": {
+    "description": "is the intent to acknowledge what was said?",
+    "domain": [
+      "general"
+    ]
+  },
+  "affirm": {
+    "description": "is the intent to affirm something?",
+    "domain": [
+      "general"
+    ]
+  },
+  "booking": {
+    "description": "is the intent to talk about some booking?",
+    "domain": [
+      "hotels"
+    ]
+  },
+  "cancel_close_leave_freeze": {
+    "description": "is the intent asking about canceling or closing something?",
+    "domain": [
+      "general"
+    ]
+  },
+  "change": {
+    "description": "is the intent to change or modify something?",
+    "domain": [
+      "general"
+    ]
+  },
+  "check_in": {
+    "description": "is the intent to ask about check in?",
+    "domain": [
+      "hotels"
+    ]
+  },
+  "check_out": {
+    "description": "is the intent to ask about check out?",
+    "domain": [
+      "hotels"
+    ]
+  },
+  "deny": {
+    "description": "is the intent to deny something?",
+    "domain": [
+      "general"
+    ]
+  },
+  "dont_know": {
+    "description": "is the intent to say I don’t know?",
+    "domain": [
+      "general"
+    ]
+  },
+  "end_call": {
+    "description": "is the intent to end call or say goodbye?",
+    "domain": [
+      "general"
+    ]
+  },
+  "existing": {
+    "description": "is the intent asking about something that already exists?",
+    "domain": [
+      "general"
+    ]
+  },
+  "greet": {
+    "description": "is the intent to greet someone?",
+    "domain": [
+      "general"
+    ]
+  },
+  "gym": {
+    "description": "is the intent to ask something related to gym?",
+    "domain": [
+      "hotels"
+    ]
+  },
+  "handoff": {
+    "description": "is the intent to speak to a human or hand off?",
+    "domain": [
+      "general"
+    ]
+  },
+  "housekeeping": {
+    "description": "is the intent to talk about housekeeping issues?",
+    "domain": [
+      "hotels"
+    ]
+  },
+  "how": {
+    "description": "is the intent asking how to do something?",
+    "domain": [
+      "general"
+    ]
+  },
+  "how_long": {
+    "description": "is the intent asking about how long something takes?",
+    "domain": [
+      "general"
+    ]
+  },
+  "how_much": {
+    "description": "is the intent asking about some quantity or how much?",
+    "domain": [
+      "general"
+    ]
+  },
+  "less_lower_before": {
+    "description": "is the intent to indicate something less, lower, before or decreasing?",
+    "domain": [
+      "general"
+    ]
+  },
+  "limits": {
+    "description": "is the intent asking about some sort of limit?",
+    "domain": [
+      "general"
+    ]
+  },
+  "lost_stolen": {
+    "description": "is the intent asking about something being lost or stolen?",
+    "domain": [
+      "general"
+    ]
+  },
+  "make_open_apply_setup_get_activate": {
+    "description": "is the intent to make, open, apply, set up or activate something?",
+    "domain": [
+      "general"
+    ]
+  },
+  "more_higher_after": {
+    "description": "is the intent to indicate something more, higher, after or increasing?",
+    "domain": [
+      "general"
+    ]
+  },
+  "new": {
+    "description": "is the intent asking about something new?",
+    "domain": [
+      "general"
+    ]
+  },
+  "parking": {
+    "description": "is the intent to ask something related to parking?",
+    "domain": [
+      "hotels"
+    ]
+  },
+  "pets": {
+    "description": "is the intent to ask something related to pets?",
+    "domain": [
+      "hotels"
+    ]
+  },
+  "refund": {
+    "description": "is the intent to ask about some refund?",
+    "domain": [
+      "banking",
+      "hotels"
+    ]
+  },
+  "repeat": {
+    "description": "is the intent asking to repeat the previous sentence?",
+    "domain": [
+      "general"
+    ]
+  },
+  "request_info": {
+    "description": "is the intent to ask or request some information?",
+    "domain": [
+      "general"
+    ]
+  },
+  "restaurant": {
+    "description": "is the intent to ask something related to restaurant?",
+    "domain": [
+      "hotels"
+    ]
+  },
+  "room_ammenities": {
+    "description": "is the intent to ask something related to some room amenities?",
+    "domain": [
+      "hotels"
+    ]
+  },
+  "room_service": {
+    "description": "is the intent to talk about room service?",
+    "domain": [
+      "hotels"
+    ]
+  },
+  "spa": {
+    "description": "is the intent to ask something related to spa or beauty services?",
+    "domain": [
+      "hotels"
+    ]
+  },
+  "swimming_pool": {
+    "description": "is the intent to ask something related to the swimming pool?",
+    "domain": [
+      "hotels"
+    ]
+  },
+  "thank": {
+    "description": "is the intent to thank someone?",
+    "domain": [
+      "general"
+    ]
+  },
+  "when": {
+    "description": "is the intent to ask about when or what time something happens?",
+    "domain": [
+      "general"
+    ]
+  },
+  "why": {
+    "description": "is the intent to ask why something happened or needs to be done?",
+    "domain": [
+      "general"
+    ]
+  },
+  "wifi": {
+    "description": "is the intent to ask something related to wifi or wireless?",
+    "domain": [
+      "hotels"
+    ]
+  },
+  "wrong_notworking_notshowing": {
+    "description": "is the intent asking about something wrong, missing or not working?",
+    "domain": [
+      "general"
+    ]
+  }
+}

curation/v3/ablation_configs.json

@@ -0,0 +1,47 @@
+{
+  "version": "2026-10-14-v3",
+  "execution_status": "data_definitions_only_no_training_run",
+  "experiments": [
+    {
+      "name": "legacy_baseline",
+      "choice_configs": [
+        "legacy_replay_v3"
+      ]
+    },
+    {
+      "name": "legacy_plus_semantic",
+      "choice_configs": [
+        "legacy_replay_v3",
+        "semantic_v3"
+      ]
+    },
+    {
+      "name": "legacy_plus_tools",
+      "choice_configs": [
+        "legacy_replay_v3",
+        "tool_context_v3"
+      ]
+    },
+    {
+      "name": "legacy_plus_semantic_plus_tools",
+      "choice_configs": [
+        "choice_v3"
+      ]
+    }
+  ],
+  "separate_objective_experiments": [
+    "hermes_actions_v3",
+    "nlupp_hotels_v3",
+    "hint3_native_v3"
+  ],
+  "required_controls": [
+    "same initial frozen backbone/head checkpoint revision",
+    "same optimizer and scheduler",
+    "matched training token/update budget",
+    "predeclared source and parent sampling weights",
+    "same fixed validation and test membership",
+    "same retention/distillation configuration",
+    "checkpoint selection on validation only; test evaluated after selection"
+  ],
+  "disallowed_concatenation": "Do not combine choice_v3 with any of its subsets, or auto-convert native auxiliary tasks to choice."
+}

curation/v3/evaluation_membership.jsonl.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b2ee4187546f5326bfb20de788290f445459f903e14098001405e1b7b5c6a0ff
+size 936365

curation/v3/evaluation_protocol.json

@@ -0,0 +1,21 @@
+{
+  "version": "2026-10-14-v3",
+  "membership_sha256": "b2ee4187546f5326bfb20de788290f445459f903e14098001405e1b7b5c6a0ff",
+  "selection_metrics": [
+    "legacy accuracy by source and operation family",
+    "semantic present and absent accuracy separately",
+    "tool observed-action accuracy by source and environment exposure",
+    "candidate-count buckets"
+  ],
+  "do_not_select_on_loss_alone": true,
+  "no_test_driven_checkpoint_selection": true,
+  "baseline_must_be_recomputed_on_same_v3_inputs": true,
+  "retention_tolerances": "Predeclare per-slice noninferiority tolerances before training; no fabricated universal tolerance is implied by these data.",
+  "uncertainty_units": {
+    "semantic": "source query parent",
+    "legacy": "available query/template group; native dialogues not all recovered",
+    "tools": "trajectory parent, with environment-cluster sensitivity analysis"
+  },
+  "small_sample_policy": "Report counts and uncertainty. Tiny DevOps/per-tool slices do not prove broad capability retention.",
+  "score_comparability": "Frozen default/v2 regression and cleaned v3 evaluation are different protocols."
+}

curation/v3/hint3_source_manifest.json

@@ -0,0 +1,204 @@
+{
+  "schema_version": 1,
+  "source": "HINT3",
+  "source_version": "v2",
+  "repository": "https://github.com/hellohaptik/HINT3",
+  "revision": "99965c892ea9a6801a87083d3861b0b04c91a4e7",
+  "source_attribution": "Haptik; Gaurav Arora, Chirag Jain, Manas Chaturvedi, Krupal Modi",
+  "paper_url": "https://aclanthology.org/2020.insights-1.16/",
+  "selected_brands": [
+    "curekart",
+    "sofmattress"
+  ],
+  "excluded_brand": "powerplay11",
+  "native_oos_label": "NO_NODES_DETECTED",
+  "native_oos_scope": "original complete intent catalog for the same brand",
+  "no_match_conversion_permitted": false,
+  "native_split_policy": "train remains train; all original test rows remain held out; derived validation is grouped inside native train only",
+  "row_locator": "row_index is 0-based among CSV records after the header; csv_record_number_including_header is row_index + 2, not a physical line number",
+  "normalization": "Unicode NFKC, casefold, re.findall(r'\\w+', text), join tokens with ASCII space",
+  "license": {
+    "database": "ODbL-1.0",
+    "individual_contents": "DbCL-1.0",
+    "verbatim_notice": "LICENSE\n==============\nHINT3 dataset is made available under the Open Database License: http://opendatacommons.org/licenses/odbl/1.0/.\nAny rights in individual contents of the database are licensed under the Database Contents License: http://opendatacommons.org/licenses/dbcl/1.0/\n",
+    "sha256": "a5df081ff80f725f7bcada010abb9fc76ccb72afe7416c7a2665af022b300940",
+    "source_url": "https://github.com/hellohaptik/HINT3/blob/99965c892ea9a6801a87083d3861b0b04c91a4e7/LICENSE.md",
+    "license_urls": [
+      "http://opendatacommons.org/licenses/odbl/1.0/",
+      "http://opendatacommons.org/licenses/dbcl/1.0/"
+    ]
+  },
+  "files": [
+    {
+      "brand": "curekart",
+      "native_split": "train",
+      "source_relative_path": "dataset/v2/train/curekart_train.csv",
+      "bytes": 29653,
+      "sha256": "99620b357d20801cbbebe2823d606d1afecd9414074d97a7386aaf2493f8911f",
+      "rows": 599,
+      "schema": [
+        "sentence",
+        "label"
+      ],
+      "label_counts": {
+        "CALL_CENTER": 21,
+        "CANCEL_ORDER": 12,
+        "CHAT_WITH_AGENT": 40,
+        "CHECK_PINCODE": 14,
+        "CONSULT_START": 26,
+        "DELAY_IN_PARCEL": 23,
+        "EXPIRY_DATE": 8,
+        "FRANCHISE": 12,
+        "IMMUNITY": 6,
+        "INTERNATIONAL_SHIPPING": 3,
+        "MODES_OF_PAYMENTS": 7,
+        "MODIFY_ADDRESS": 16,
+        "ORDER_QUERY": 7,
+        "ORDER_STATUS": 47,
+        "ORDER_TAKING": 39,
+        "ORIGINAL_PRODUCT": 23,
+        "PAYMENT_AND_BILL": 26,
+        "PORTAL_ISSUE": 4,
+        "RECOMMEND_PRODUCT": 95,
+        "REFER_EARN": 13,
+        "REFUNDS_RETURNS_REPLACEMENTS": 54,
+        "RESUME_DELIVERY": 51,
+        "SIDE_EFFECT": 4,
+        "SIGN_UP": 7,
+        "START_OVER": 5,
+        "STORE_INFORMATION": 14,
+        "USER_GOAL_FORM": 12,
+        "WORK_FROM_HOME": 10
+      },
+      "native_oos_rows": 0,
+      "browse_url": "https://github.com/hellohaptik/HINT3/blob/99965c892ea9a6801a87083d3861b0b04c91a4e7/dataset/v2/train/curekart_train.csv",
+      "raw_url": "https://raw.githubusercontent.com/hellohaptik/HINT3/99965c892ea9a6801a87083d3861b0b04c91a4e7/dataset/v2/train/curekart_train.csv"
+    },
+    {
+      "brand": "curekart",
+      "native_split": "test",
+      "source_relative_path": "dataset/v2/test/curekart_test.csv",
+      "bytes": 49506,
+      "sha256": "47bf651cafa8b8f476ca4bea3052ee9a29c5bc6a4d841f2162f12ad16064f5a0",
+      "rows": 991,
+      "schema": [
+        "sentence",
+        "label"
+      ],
+      "label_counts": {
+        "CALL_CENTER": 5,
+        "CANCEL_ORDER": 32,
+        "CHAT_WITH_AGENT": 14,
+        "CHECK_PINCODE": 22,
+        "CONSULT_START": 16,
+        "DELAY_IN_PARCEL": 10,
+        "EXPIRY_DATE": 13,
+        "FRANCHISE": 12,
+        "MODIFY_ADDRESS": 3,
+        "NO_NODES_DETECTED": 532,
+        "ORDER_STATUS": 12,
+        "ORDER_TAKING": 19,
+        "ORIGINAL_PRODUCT": 43,
+        "PAYMENT_AND_BILL": 3,
+        "RECOMMEND_PRODUCT": 205,
+        "REFUNDS_RETURNS_REPLACEMENTS": 20,
+        "RESUME_DELIVERY": 9,
+        "SIDE_EFFECT": 5,
+        "SIGN_UP": 3,
+        "STORE_INFORMATION": 11,
+        "USER_GOAL_FORM": 2
+      },
+      "native_oos_rows": 532,
+      "browse_url": "https://github.com/hellohaptik/HINT3/blob/99965c892ea9a6801a87083d3861b0b04c91a4e7/dataset/v2/test/curekart_test.csv",
+      "raw_url": "https://raw.githubusercontent.com/hellohaptik/HINT3/99965c892ea9a6801a87083d3861b0b04c91a4e7/dataset/v2/test/curekart_test.csv"
+    },
+    {
+      "brand": "sofmattress",
+      "native_split": "train",
+      "source_relative_path": "dataset/v2/train/sofmattress_train.csv",
+      "bytes": 11866,
+      "sha256": "0addf4dc4b9df1e71509da5318d5a9f6e9a5d6eb98b9584d55cab3da7badf4b1",
+      "rows": 328,
+      "schema": [
+        "sentence",
+        "label"
+      ],
+      "label_counts": {
+        "100_NIGHT_TRIAL_OFFER": 18,
+        "ABOUT_SOF_MATTRESS": 11,
+        "CANCEL_ORDER": 10,
+        "CHECK_PINCODE": 10,
+        "COD": 12,
+        "COMPARISON": 11,
+        "DELAY_IN_DELIVERY": 11,
+        "DISTRIBUTORS": 34,
+        "EMI": 25,
+        "ERGO_FEATURES": 11,
+        "LEAD_GEN": 21,
+        "MATTRESS_COST": 22,
+        "OFFERS": 10,
+        "ORDER_STATUS": 21,
+        "ORTHO_FEATURES": 17,
+        "PILLOWS": 10,
+        "PRODUCT_VARIANTS": 21,
+        "RETURN_EXCHANGE": 14,
+        "SIZE_CUSTOMIZATION": 9,
+        "WARRANTY": 10,
+        "WHAT_SIZE_TO_ORDER": 20
+      },
+      "native_oos_rows": 0,
+      "browse_url": "https://github.com/hellohaptik/HINT3/blob/99965c892ea9a6801a87083d3861b0b04c91a4e7/dataset/v2/train/sofmattress_train.csv",
+      "raw_url": "https://raw.githubusercontent.com/hellohaptik/HINT3/99965c892ea9a6801a87083d3861b0b04c91a4e7/dataset/v2/train/sofmattress_train.csv"
+    },
+    {
+      "brand": "sofmattress",
+      "native_split": "test",
+      "source_relative_path": "dataset/v2/test/sofmattress_test.csv",
+      "bytes": 19035,
+      "sha256": "9c4746f7cf445b5db276260e1c05b9e7b102fa2bca0ce1ace20f51df0ba53aae",
+      "rows": 397,
+      "schema": [
+        "sentence",
+        "label"
+      ],
+      "label_counts": {
+        "100_NIGHT_TRIAL_OFFER": 5,
+        "ABOUT_SOF_MATTRESS": 3,
+        "CANCEL_ORDER": 6,
+        "CHECK_PINCODE": 25,
+        "COD": 12,
+        "COMPARISON": 18,
+        "DELAY_IN_DELIVERY": 15,
+        "DISTRIBUTORS": 13,
+        "EMI": 13,
+        "ERGO_FEATURES": 3,
+        "LEAD_GEN": 17,
+        "MATTRESS_COST": 23,
+        "NO_NODES_DETECTED": 144,
+        "OFFERS": 11,
+        "ORDER_STATUS": 8,
+        "ORTHO_FEATURES": 11,
+        "PILLOWS": 13,
+        "PRODUCT_VARIANTS": 8,
+        "RETURN_EXCHANGE": 12,
+        "SIZE_CUSTOMIZATION": 15,
+        "WHAT_SIZE_TO_ORDER": 22
+      },
+      "native_oos_rows": 144,
+      "browse_url": "https://github.com/hellohaptik/HINT3/blob/99965c892ea9a6801a87083d3861b0b04c91a4e7/dataset/v2/test/sofmattress_test.csv",
+      "raw_url": "https://raw.githubusercontent.com/hellohaptik/HINT3/99965c892ea9a6801a87083d3861b0b04c91a4e7/dataset/v2/test/sofmattress_test.csv"
+    }
+  ],
+  "benchmark_screen_index": {
+    "local_directory": "scratch/fastbrain_instruct_integration_20261014/exclusion",
+    "exact_bin_sha256": "af41710b35e4481f003eacceb9ea33936c65c6dcbd2dc89f8a83213165c3da81",
+    "grams_bin_sha256": "f8f761744fee732134b41465afce05b7d644dfa32db629b589002c3c4ccdb71f",
+    "rule": "normalized full query BLAKE2b-128 OR contiguous 10 normalized tokens BLAKE2b-64"
+  },
+  "evidence_artifacts": [
+    "hint3_native_label_finance_map.json",
+    "hint3_split_finance_audit.json",
+    "hint3_row_audit.jsonl",
+    "HINT3_ATTRIBUTION.md"
+  ]
+}

curation/v3/quarantine.jsonl.gz

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:73214987fbaec508b01d478dac26e863fabc7daa6c24bcd98a5bef7c0ce42b22
+size 1089399

curation/v3/tool_parent_decisions.json

@@ -0,0 +1,699 @@
+{
+  "blocked_parents": [
+    "funcdex:0364bfec3755531ed937890077a1ff99",
+    "funcdex:0fd8892a0b771f338e74ceb9b5bb1155",
+    "funcdex:20c1538d3140baa081417575cc3e4629",
+    "funcdex:3e2f6df490c982d76e8c86676feadff5",
+    "funcdex:579dffc45352291585fbbef351ad8402",
+    "funcdex:5e58c002adf82f92eb871a6e282b7efc",
+    "funcdex:61f3ba4a3a48ec671cd40e2fccc467d0",
+    "funcdex:6976507c216d698bfcfa8ade0dd7664c",
+    "funcdex:6b01cfbe55ad13f33b4d491caa63ac42",
+    "funcdex:776cf8163c1a6444d347c7faf1f9c8de",
+    "funcdex:7833a9588df8a8c31fa2a24a31826c44",
+    "funcdex:80cb95a02eb23fc96a10ed598235e341",
+    "funcdex:8876827e1fb21db748aa4289fea153ea",
+    "funcdex:8bf115f0d106414de347e0a82de851ca",
+    "funcdex:9cb4f097b8ea6d50eaf92620f968ab66",
+    "funcdex:a16efbee10f9de5cf97c6830912c3848",
+    "funcdex:b0070760204890aa7fe9f214bd0df951",
+    "funcdex:b4d7ebd0025899856b5343d927f7f52a",
+    "funcdex:c3bcc573819feb8ab22267607a159803",
+    "funcdex:cf1ba1d6be84abf7c864f72bc6c8545f",
+    "funcdex:db382ec2856dfb346fbe7c14155dbd10",
+    "funcdex:f42cdf67d9e8486666de2042d2a46393",
+    "nemotron:237e0f2c0a137bb563778649adbb1361",
+    "nemotron:2696a3a63b4513ff56fb9713e356c253",
+    "nemotron:3ec31689dcd30946f97fb4b33437d7d7",
+    "nemotron:41b1622c1877653364c378e2b86acb97",
+    "nemotron:b55a089c8dcf0f3804553a82b788b493",
+    "nemotron:bbe1ebbec2d84e4140c01a9fb3f2e386",
+    "nemotron:d4ff1b76ed59c9d7bd3dc2ac3709a5ac",
+    "nemotron:f214408657f160e8c683d69359b26634",
+    "nemotron:f9a3d258f9fbbebe4d2352fae872e20f"
+  ],
+  "reviewed_parent_decisions": {
+    "funcdex:f42cdf67d9e8486666de2042d2a46393": [
+      {
+        "sample_id": "F001",
+        "step_id": "fbv2:funcdex:train:1316:3",
+        "status": "clear_contradiction_or_mismatch",
+        "block_parent": true,
+        "reason": "完整policy要求首次调用RETRIEVE_TASKS_FOR_PROJECT；用户已给project_gid，标签却GET_A_PROJECT（描述明确不返回任务）。原始row1316/index3同一调用，属源轨迹与policy冲突。",
+        "source_row_id": "1316",
+        "source_split": "train"
+      }
+    ],
+    "funcdex:776cf8163c1a6444d347c7faf1f9c8de": [
+      {
+        "sample_id": "F002",
+        "step_id": "fbv2:funcdex:train:280:6",
+        "status": "clear_contradiction_or_mismatch",
+        "block_parent": true,
+        "reason": "GET_ISSUE_TYPE_SCHEME要求方案ID；前缀仅说default scheme，没有default-scheme这一ID来源，且可用GET_ALL_ISSUE_TYPE_SCHEMES先查ID。原始调用把占位式ID送入接口并得到合成成功响应；不能因schema接受string就当执行前提成立。",
+        "source_row_id": "280",
+        "source_split": "train"
+      }
+    ],
+    "funcdex:d2d5b7ab099793e502a27629b58df829": [
+      {
+        "sample_id": "F003",
+        "step_id": "fbv2:funcdex:train:677:52",
+        "status": "acceptable",
+        "block_parent": false,
+        "reason": "当前请求是给刚创建的Launch task加评论；task_5004来自工具返回，CREATE_COMMENT合理。评论内容属于用户授权起草；未证明唯一合法下一步。",
+        "source_row_id": "677",
+        "source_split": "train"
+      }
+    ],
+    "funcdex:b0070760204890aa7fe9f214bd0df951": [
+      {
+        "sample_id": "F004",
+        "step_id": "fbv2:funcdex:train:1142:18",
+        "status": "acceptable",
+        "block_parent": true,
+        "reason": "用户给finance_david并请求添加watcher；EVT-101来自CREATE_ISSUE，目标和当前参数均支持。",
+        "source_row_id": "1142",
+        "source_split": "train"
+      }
+    ],
+    "funcdex:6976507c216d698bfcfa8ade0dd7664c": [
+      {
+        "sample_id": "F005",
+        "step_id": "fbv2:funcdex:train:1173:33",
+        "status": "acceptable",
+        "block_parent": true,
+        "reason": "最终邮件草稿已展示且用户明确批准发送，SEND_EMAIL与body/recipient一致。",
+        "source_row_id": "1173",
+        "source_split": "train"
+      }
+    ],
+    "funcdex:9cb4f097b8ea6d50eaf92620f968ab66": [
+      {
+        "sample_id": "F006",
+        "step_id": "fbv2:funcdex:train:848:23",
+        "status": "acceptable",
+        "block_parent": true,
+        "reason": "用户明确说季度结束要求移除webhook，目标删除调用使用前序工具返回的WH ID，满足本步前提。",
+        "source_row_id": "848",
+        "source_split": "train"
+      }
+    ],
+    "funcdex:3e2f6df490c982d76e8c86676feadff5": [
+      {
+        "sample_id": "F007",
+        "step_id": "fbv2:funcdex:train:1171:5",
+        "status": "clear_contradiction_or_mismatch",
+        "block_parent": true,
+        "reason": "用户要合同PDF→Legal Review task→回复邮件，未请求release/milestone；policy仅在必须建立版本时允许CREATE_VERSION，原始助手却自行创建合同版本。该副作用并非完成请求的必要步骤。",
+        "source_row_id": "1171",
+        "source_split": "train"
+      }
+    ],
+    "funcdex:abf9e17c3a8652857a6dcb24434873a7": [
+      {
+        "sample_id": "F008",
+        "step_id": "fbv2:funcdex:train:1351:60",
+        "status": "acceptable",
+        "block_parent": false,
+        "reason": "为新建Healthcare task添加用户指定的VP/CTO followers，ID皆来自前缀；同样可先添加已获授权附件，先后次序不唯一。",
+        "source_row_id": "1351",
+        "source_split": "train"
+      }
+    ],
+    "funcdex:5e58c002adf82f92eb871a6e282b7efc": [
+      {
+        "sample_id": "F009",
+        "step_id": "fbv2:funcdex:train:1005:3",
+        "status": "context_ambiguous",
+        "block_parent": true,
+        "reason": "policy既要求检查message templates，又要求对超过30天未聊天人群立即拒绝发送；GET_MESSAGE_TEMPLATES是可行的预检，但可以直接文字拒绝且无需查询。没有发送违规操作，不将预检本身当明确错标。",
+        "source_row_id": "1005",
+        "source_split": "train"
+      }
+    ],
+    "funcdex:7833a9588df8a8c31fa2a24a31826c44": [
+      {
+        "sample_id": "F010",
+        "step_id": "fbv2:funcdex:train:624:1",
+        "status": "context_ambiguous",
+        "block_parent": true,
+        "reason": "policy允许获取当前日期，目标工具可接受但用户已经给定活动日期；timezone=-5没有地点、时区或默认值根据。工具名不一定错，参数不宜作为gold复用。",
+        "source_row_id": "624",
+        "source_split": "train"
+      }
+    ],
+    "funcdex:010dbaa0d71eeab99faf256abb3c0a39": [
+      {
+        "sample_id": "F061",
+        "step_id": "fbv2:funcdex:train:890:6",
+        "status": "acceptable",
+        "block_parent": false,
+        "reason": "三位manager获授权逐一邀请；第一位成功后邀请第二位符合policy和必要ID。原始user内容(intermediate)是人工拼接的继续信号，现实交互自然度有限。",
+        "source_row_id": "890",
+        "source_split": "train"
+      }
+    ],
+    "funcdex:8876827e1fb21db748aa4289fea153ea": [
+      {
+        "sample_id": "F062",
+        "step_id": "fbv2:funcdex:train:554:3",
+        "status": "clear_contradiction_or_mismatch",
+        "block_parent": true,
+        "reason": "policy只允许确认标签不存在后CREATE_LABEL，前缀无LIST_GMAIL_LABELS或用户确认不存在；该查询工具可用，标签却直接创建。",
+        "source_row_id": "554",
+        "source_split": "train"
+      }
+    ],
+    "funcdex:6b01cfbe55ad13f33b4d491caa63ac42": [
+      {
+        "sample_id": "F063",
+        "step_id": "fbv2:funcdex:train:1313:11",
+        "status": "clear_contradiction_or_mismatch",
+        "block_parent": true,
+        "reason": "GET_STORIES_FOR_TASK的task_101不在前缀出现；此前只是列项目setup计划，无创建任务工具，status正文却声称所有任务/附件/自动化已经完成。GET_MULTIPLE_TASKS可用，应先获得真实任务ID。",
+        "source_row_id": "1313",
+        "source_split": "train"
+      }
+    ],
+    "funcdex:0364bfec3755531ed937890077a1ff99": [
+      {
+        "sample_id": "F064",
+        "step_id": "fbv2:funcdex:train:885:4",
+        "status": "clear_contradiction_or_mismatch",
+        "block_parent": true,
+        "reason": "用户为刚创建的event type生成链接；目标args却把owner设为助手此前臆造的host用户URI、owner_type=user，没有绑定已返回的event-type URI。schema只有string，无枚举或引用校验。",
+        "source_row_id": "885",
+        "source_split": "train"
+      }
+    ],
+    "funcdex:db382ec2856dfb346fbe7c14155dbd10": [
+      {
+        "sample_id": "F065",
+        "step_id": "fbv2:funcdex:train:907:24",
+        "status": "acceptable",
+        "block_parent": true,
+        "reason": "用户明确报告Candidate02缺席，invitee URI来自LIST_EVENTS，重复事件已取消；当前no-show调用有支持。",
+        "source_row_id": "907",
+        "source_split": "train"
+      }
+    ],
+    "funcdex:80cb95a02eb23fc96a10ed598235e341": [
+      {
+        "sample_id": "F066",
+        "step_id": "fbv2:funcdex:train:101:40",
+        "status": "clear_contradiction_or_mismatch",
+        "block_parent": true,
+        "reason": "用户撤销exec2权限，target UPDATE_PERMISSION以permission.role=none实现，候选已有DELETE_PERMISSION。官方Drive permissions文档不支持none角色；嵌套permission只定义object使静态验证漏掉。",
+        "source_row_id": "101",
+        "source_split": "train"
+      }
+    ],
+    "funcdex:0fd8892a0b771f338e74ceb9b5bb1155": [
+      {
+        "sample_id": "F067",
+        "step_id": "fbv2:funcdex:train:834:12",
+        "status": "acceptable",
+        "block_parent": true,
+        "reason": "当前single-use link使用已返回event-type URI，max_event_count=1对应请求；CREATE_SCHEDULING_LINK描述也能完成，gold非唯一。",
+        "source_row_id": "834",
+        "source_split": "train"
+      }
+    ],
+    "funcdex:579dffc45352291585fbbef351ad8402": [
+      {
+        "sample_id": "F068",
+        "step_id": "fbv2:funcdex:train:936:10",
+        "status": "acceptable",
+        "block_parent": true,
+        "reason": "两封PDF已取回，用户明确要求发给legal，当前FORWARD_EMAIL_MESSAGE参数直接引用msg_001和指定收件人。",
+        "source_row_id": "936",
+        "source_split": "train"
+      }
+    ],
+    "funcdex:2098c3b0485f35e20b2f24cd5e797a3a": [
+      {
+        "sample_id": "F069",
+        "step_id": "fbv2:funcdex:train:64:17",
+        "status": "acceptable",
+        "block_parent": false,
+        "reason": "用户给旧报告Drive ID并要求复制到审计包；COPY_FILE以该ID复制是合理下一步，新copy完成后仍需MOVE_FILE。",
+        "source_row_id": "64",
+        "source_split": "train"
+      }
+    ],
+    "funcdex:61f3ba4a3a48ec671cd40e2fccc467d0": [
+      {
+        "sample_id": "F070",
+        "step_id": "fbv2:funcdex:train:1417:7",
+        "status": "clear_contradiction_or_mismatch",
+        "block_parent": true,
+        "reason": "PATCH_CALENDAR实际修改日历metadata，源policy却把它描述成修改event；当前args包含schema未定义attendees/start_datetime并缺event ID和Drive URL。宽松additionalProperties使错误对象层次通过schema验证，执行不能生成所需会议且可能重命名primary calendar。",
+        "source_row_id": "1417",
+        "source_split": "train"
+      }
+    ],
+    "funcdex:20c1538d3140baa081417575cc3e4629": [
+      {
+        "sample_id": "F081",
+        "step_id": "fbv2:funcdex:test:26:4",
+        "status": "context_ambiguous",
+        "block_parent": true,
+        "reason": "用户需找Sales Data Q3 spreadsheet，SEARCH_DOCUMENTS的描述仅称Google documents，返回folder的前史使实现是否支持通用Drive MIME筛选不明。读查询可能合理，但没有执行实现可确认跨类型检索能力。",
+        "source_row_id": "26",
+        "source_split": "test"
+      }
+    ],
+    "funcdex:1527d3f77a0dbf965c1a49d2bada4c9a": [
+      {
+        "sample_id": "F082",
+        "step_id": "fbv2:funcdex:test:246:3",
+        "status": "acceptable",
+        "block_parent": false,
+        "reason": "用户给template document ID、vendor和amendment序号；copy目标与title均满足policy，余下replace依赖新ID。",
+        "source_row_id": "246",
+        "source_split": "test"
+      }
+    ],
+    "funcdex:b4d7ebd0025899856b5343d927f7f52a": [
+      {
+        "sample_id": "F083",
+        "step_id": "fbv2:funcdex:test:32:19",
+        "status": "clear_contradiction_or_mismatch",
+        "block_parent": true,
+        "reason": "用户要求在文档末尾插入的表格之后加图；原args location.index=5没有任何文档结构返回支持，已知markdown先有长标题和摘要，index5不可能位于末尾表格后。工具名合理但位置参数错误。",
+        "source_row_id": "32",
+        "source_split": "test"
+      }
+    ],
+    "funcdex:93911677922f73486eea29879cb93a3d": [
+      {
+        "sample_id": "F084",
+        "step_id": "fbv2:funcdex:test:230:3",
+        "status": "acceptable",
+        "block_parent": false,
+        "reason": "用户提供原Gmail message ID，policy明确先读取原邮件再取附件；当前FETCH_MESSAGE_BY_MESSAGE_ID的ID和me均有根据。",
+        "source_row_id": "230",
+        "source_split": "test"
+      }
+    ],
+    "funcdex:cf1ba1d6be84abf7c864f72bc6c8545f": [
+      {
+        "sample_id": "F085",
+        "step_id": "fbv2:funcdex:test:173:18",
+        "status": "clear_contradiction_or_mismatch",
+        "block_parent": true,
+        "reason": "源对话中途user/assistant角色反转：user说Here is the link / I will cancel，assistant却自称Maya并给booking UUID、请求取消。本步CANCEL_EVENT接在user“I will cancel”后，原始角色不能正确表示谁授权谁执行。",
+        "source_row_id": "173",
+        "source_split": "test"
+      }
+    ],
+    "funcdex:c3bcc573819feb8ab22267607a159803": [
+      {
+        "sample_id": "F086",
+        "step_id": "fbv2:funcdex:test:245:21",
+        "status": "acceptable",
+        "block_parent": true,
+        "reason": "用户明确两项global替换，先替换HOST_NAME是合理单调用且新document_id已返回。",
+        "source_row_id": "245",
+        "source_split": "test"
+      }
+    ],
+    "funcdex:a16efbee10f9de5cf97c6830912c3848": [
+      {
+        "sample_id": "F087",
+        "step_id": "fbv2:funcdex:test:37:13",
+        "status": "acceptable",
+        "block_parent": true,
+        "reason": "复制文件已返回ID，目标新文件夹已创建，MOVE_FILE用于搬运copy符合任务。",
+        "source_row_id": "37",
+        "source_split": "test"
+      }
+    ],
+    "funcdex:8bf115f0d106414de347e0a82de851ca": [
+      {
+        "sample_id": "F088",
+        "step_id": "fbv2:funcdex:test:21:10",
+        "status": "context_ambiguous",
+        "block_parent": true,
+        "reason": "用户要求after sharing再hide，前缀只有共享操作的手工说明，未确认用户执行了共享；直接hide是��能下一步但先决状态不确定。",
+        "source_row_id": "21",
+        "source_split": "test"
+      }
+    ],
+    "funcdex:8ba1d3ba07890c625b689151c6b781e4": [
+      {
+        "sample_id": "F089",
+        "step_id": "fbv2:funcdex:test:137:82",
+        "status": "acceptable",
+        "block_parent": false,
+        "reason": "HR任务已创建且最新已更新due date；用户指定Signed clause comment原文，CREATE_COMMENT引用task_5004正确。用户此前允许仅在comment文本提及PDF，不能误判为必须真实上传。",
+        "source_row_id": "137",
+        "source_split": "test"
+      }
+    ],
+    "funcdex:3e8ab49aeb82b9ab97195e8f34169953": [
+      {
+        "sample_id": "F090",
+        "step_id": "fbv2:funcdex:test:235:9",
+        "status": "acceptable",
+        "block_parent": false,
+        "reason": "用户明确查Insider Threat主题草稿，LIST_DRAFTS verbose=true可取内容后本地筛选；schema无subject查询字段，不应误判缺query。后续分页仍需根据响应处理。",
+        "source_row_id": "235",
+        "source_split": "test"
+      }
+    ],
+    "nemotron:bbe1ebbec2d84e4140c01a9fb3f2e386": [
+      {
+        "sample_id": "N001",
+        "step_id": "fbv2:nemotron:ad2ceab085166eeb054acf3cc76cd282",
+        "status": "context_ambiguous",
+        "block_parent": true,
+        "reason": "用户要求HIPAA/医疗隐私合规书面保证，转人工是合理限制边界；policy列出的transfer触发只含insurance/custom contract/security，未写此场景。不能证明transfer唯一gold，也不宜直接判错。",
+        "source_row_id": "4230",
+        "source_split": "interactive_agent"
+      }
+    ],
+    "nemotron:1698ba4ac79f74377220e886cb0a30c3": [
+      {
+        "sample_id": "N002",
+        "step_id": "fbv2:nemotron:4d82b68cf253105f32fc3c4c219f1942",
+        "status": "acceptable",
+        "block_parent": false,
+        "reason": "authenticate_user工具已确认Tier 1及resident_id；用户明确optometrist/routine annual eye exam，协调专科咨询的前提和required names齐备。",
+        "source_row_id": "11580",
+        "source_split": "interactive_agent"
+      }
+    ],
+    "nemotron:74737c18d67239b8e1cec1a5be781d7d": [
+      {
+        "sample_id": "N003",
+        "step_id": "fbv2:nemotron:c25c379fd97df212684b3c80eac072e3",
+        "status": "acceptable",
+        "block_parent": false,
+        "reason": "用户选择full name+ZIP核验并给出两项，verify_user_identity是合理下一步；用户新ZIP是否与档案一致应由工具决定，不能预判验证结果。",
+        "source_row_id": "2734",
+        "source_split": "interactive_agent"
+      }
+    ],
+    "nemotron:41b1622c1877653364c378e2b86acb97": [
+      {
+        "sample_id": "N004",
+        "step_id": "fbv2:nemotron:b0694ea49102952d7066346fd187b45e",
+        "status": "context_ambiguous",
+        "block_parent": true,
+        "reason": "未认证用户要求经理；policy一方面要求before any action验证，一方面要求client requests manager则transfer。认证失败是否可直接转人工未定义，属于政策冲突而非可确定错标。",
+        "source_row_id": "3845",
+        "source_split": "interactive_agent"
+      }
+    ],
+    "nemotron:237e0f2c0a137bb563778649adbb1361": [
+      {
+        "sample_id": "N005",
+        "step_id": "fbv2:nemotron:888135f69fb2706f5d7f07c57c0ee2f6",
+        "status": "clear_contradiction_or_mismatch",
+        "block_parent": true,
+        "reason": "航空policy要求访问reservation之前先发送并验证验证码；完整原始prefix只有system+首条user，目标get_reservation_details直接访问订票信息。已核对NVIDIA原始row312/index2和完整schema，policy逐字相同，排除本次截断修复造成。",
+        "source_row_id": "312",
+        "source_split": "interactive_agent"
+      }
+    ],
+    "nemotron:3ec31689dcd30946f97fb4b33437d7d7": [
+      {
+        "sample_id": "N006",
+        "step_id": "fbv2:nemotron:5a89a1d38a0545d1dc29bf5c7d23b9ce",
+        "status": "clear_contradiction_or_mismatch",
+        "block_parent": true,
+        "reason": "policy要求account access前验证User ID+注册email/phone；用户仅给User ID，历史get_user_profile直接返回VIP资料，再选择get_subscription_details，仍无认证。恢复后的前缀暴露认证前提未满足。",
+        "source_row_id": "12955",
+        "source_split": "interactive_agent"
+      }
+    ],
+    "nemotron:64ab5841bbeafc5585a503c1947d8bba": [
+      {
+        "sample_id": "N007",
+        "step_id": "fbv2:nemotron:ea83f2ae0b03b36d3a345545b2936dae",
+        "status": "acceptable",
+        "block_parent": false,
+        "reason": "用户给Client ID和规定的电话尾号验证信息，authenticate_client合理；退赔时限和专家判断尚未执行，不把后续可能资格不符算到当前认证调用。",
+        "source_row_id": "9965",
+        "source_split": "interactive_agent"
+      }
+    ],
+    "nemotron:f27d491ed368e3cb744aab64bd21aced": [
+      {
+        "sample_id": "N008",
+        "step_id": "fbv2:nemotron:f370f37982c4b2840ded81fc1d7a2e86",
+        "status": "acceptable",
+        "block_parent": false,
+        "reason": "发送、提交、验证验证码已有成功工具响应并返回guest_id，随后get_guest_profile用于检查会员资格合理。",
+        "source_row_id": "3029",
+        "source_split": "interactive_agent"
+      }
+    ],
+    "nemotron:233177b1f7d713d274f3b4c2e8a44a30": [
+      {
+        "sample_id": "N009",
+        "step_id": "fbv2:nemotron:2f4bcf68cf992763ed37bbd411092d2d",
+        "status": "acceptable",
+        "block_parent": false,
+        "reason": "注册用户给出email，请求查迟送费；policy先send authentication code，当前工具符合认证步骤，不因用户声称延迟便直接退款。",
+        "source_row_id": "8863",
+        "source_split": "interactive_agent"
+      }
+    ],
+    "nemotron:265cb1677862fd7931c1a568400e2b03": [
+      {
+        "sample_id": "N010",
+        "step_id": "fbv2:nemotron:f84aea370dd777966f22edafe19cd9a4",
+        "status": "acceptable",
+        "block_parent": false,
+        "reason": "用户明确要billing agent，现有工具无法添加副卡，transfer_to_human_agent符合能力边界；不臆答扣款顺序和完成结果。",
+        "source_row_id": "6606",
+        "source_split": "interactive_agent"
+      }
+    ],
+    "nemotron:a7b76c9ddca26f8da17b76239ec2419c": [
+      {
+        "sample_id": "N061",
+        "step_id": "fbv2:nemotron:4779b2d734a37a8042d2bb44cd754da3",
+        "status": "acceptable",
+        "block_parent": false,
+        "reason": "用户保险direct billing咨询触发明确insurance escalation，transfer合理；不执行账户或付款动作。",
+        "source_row_id": "11383",
+        "source_split": "interactive_agent"
+      }
+    ],
+    "nemotron:76a7853afd0a65322ffa2ecca623e65e": [
+      {
+        "sample_id": "N062",
+        "step_id": "fbv2:nemotron:fb7b458e871334545722a45bbad110b4",
+        "status": "acceptable",
+        "block_parent": false,
+        "reason": "booking ID+last name符合guest验证方法，先authenticate而非直接cancel，日期/退款比例留待核验。",
+        "source_row_id": "4552",
+        "source_split": "interactive_agent"
+      }
+    ],
+    "nemotron:5a11d5539003171d6b0399b29066d85a": [
+      {
+        "sample_id": "N063",
+        "step_id": "fbv2:nemotron:2296cbbd2d8824fffeaddf1025278740",
+        "status": "acceptable",
+        "block_parent": false,
+        "reason": "用户无法找到ID并明确要求supervisor，transfer工具说明覆盖complex issues；policy列举触发未写only，不据此判错。",
+        "source_row_id": "2744",
+        "source_split": "interactive_agent"
+      }
+    ],
+    "nemotron:4aeb08fdb50d2d4ab5ca6a0f0639f95a": [
+      {
+        "sample_id": "N064",
+        "step_id": "fbv2:nemotron:9ca7a48d1250497e60b7f4425c57c80e",
+        "status": "acceptable",
+        "block_parent": false,
+        "reason": "身份已由工具确认；用户要联系管理方问staffing，无staffing数据工具而有human transfer，下一步转人工合理。",
+        "source_row_id": "6677",
+        "source_split": "interactive_agent"
+      }
+    ],
+    "nemotron:4d58740b841903c23abd8835b9f3e823": [
+      {
+        "sample_id": "N065",
+        "step_id": "fbv2:nemotron:163b28d0c0bf05af8d913033a4b92f0f",
+        "status": "acceptable",
+        "block_parent": false,
+        "reason": "船舶事故保险理赔为policy明确转人工场景；该步不直接修改账户或发起支付，转人工合理。",
+        "source_row_id": "2916",
+        "source_split": "interactive_agent"
+      }
+    ],
+    "nemotron:0dc84454ed457035d454d9081e7cc32e": [
+      {
+        "sample_id": "N066",
+        "step_id": "fbv2:nemotron:c992041f87ee49101d7c72f32cf2a9d3",
+        "status": "acceptable",
+        "block_parent": false,
+        "reason": "用户最终明确请求pastoral staff谈宗教教育事宜；工具转pastoral staff对应policy，未违反无酒精预订规则。",
+        "source_row_id": "3323",
+        "source_split": "interactive_agent"
+      }
+    ],
+    "nemotron:2696a3a63b4513ff56fb9713e356c253": [
+      {
+        "sample_id": "N067",
+        "step_id": "fbv2:nemotron:5ba17f0b6796a01c47518321a450aa42",
+        "status": "context_ambiguous",
+        "block_parent": true,
+        "reason": "authenticate_user要求certification_level；用户称Guide，policy列Novice/Certified/Professional，缺完整enum和原始args，不知是否允许Guide或需映射/澄清。工具名合理，不能认证参数已准备好。",
+        "source_row_id": "12756",
+        "source_split": "interactive_agent"
+      }
+    ],
+    "nemotron:d4ff1b76ed59c9d7bd3dc2ac3709a5ac": [
+      {
+        "sample_id": "N068",
+        "step_id": "fbv2:nemotron:5d51461d5e94dc257c16277bfdaef7c0",
+        "status": "context_ambiguous",
+        "block_parent": true,
+        "reason": "完成证书不在工具能力内；transfer工具限定per policy triggers，而列举触发未覆盖该普通证书请求。转人工或文字解释限制都合理，无法证明单一label必选。",
+        "source_row_id": "2138",
+        "source_split": "interactive_agent"
+      }
+    ],
+    "nemotron:809a0749e0104413e055ef81800101ba": [
+      {
+        "sample_id": "N069",
+        "step_id": "fbv2:nemotron:c1b7d7d5d0b6cb7c25c4dfa454e28067",
+        "status": "acceptable",
+        "block_parent": false,
+        "reason": "用户报告data update latency，check_service_status可先核实是否故障；不能据它承诺SEO刷新频率，但当前诊断查询合理。",
+        "source_row_id": "6364",
+        "source_split": "interactive_agent"
+      }
+    ],
+    "nemotron:bb5d498dc838ec45fba642f948d60f6d": [
+      {
+        "sample_id": "N070",
+        "step_id": "fbv2:nemotron:4ab93fe049238eae5b1a6ee00a2f04cd",
+        "status": "acceptable",
+        "block_parent": false,
+        "reason": "完整验证码认证成功并返回customer_id，get_customer_details查当前服务计划是升级前合理准备。",
+        "source_row_id": "11235",
+        "source_split": "interactive_agent"
+      }
+    ],
+    "nemotron:e564ee8531922f45d211a6311bbea706": [
+      {
+        "sample_id": "N081",
+        "step_id": "fbv2:nemotron:6d273024535e5eaf0f376e96fb35119f",
+        "status": "acceptable",
+        "block_parent": false,
+        "reason": "profile和matches已返回，get_subscription_details可核实每日额度/会员tier；policy未明确要求此次查询必须先调用独立认证。",
+        "source_row_id": "924",
+        "source_split": "interactive_agent"
+      }
+    ],
+    "nemotron:9dbbb7f09d6263f706251448e5da4c16": [
+      {
+        "sample_id": "N082",
+        "step_id": "fbv2:nemotron:b761675a69bbfd65342af0ddd003dcc5",
+        "status": "acceptable",
+        "block_parent": false,
+        "reason": "authenticate_customer已成功，get_customer_profile可核验合约tier和设备信息，再决定远程诊断。",
+        "source_row_id": "9794",
+        "source_split": "interactive_agent"
+      }
+    ],
+    "nemotron:1f00689599faf44052857c7667fcb4e7": [
+      {
+        "sample_id": "N083",
+        "step_id": "fbv2:nemotron:119ff0656fe5df1118ef19014852e1c4",
+        "status": "acceptable",
+        "block_parent": false,
+        "reason": "brand ID与业务税号尾部验证信息均由用户给出，authenticate_brand符合政策身份方法。后续COPPA文件是否可取不影响本步。",
+        "source_row_id": "6907",
+        "source_split": "interactive_agent"
+      }
+    ],
+    "nemotron:f214408657f160e8c683d69359b26634": [
+      {
+        "sample_id": "N084",
+        "step_id": "fbv2:nemotron:ffaa07515484209b2c2b709ef1e7fe64",
+        "status": "context_ambiguous",
+        "block_parent": true,
+        "reason": "当前verify_user_identity方向合理，用户给验证材料；但历史声称已发送phone/email验证码，没有发送工具或发送记录，challenge来源未证实，且现有工具集中无code发送接口。不能认定参数可执行性。",
+        "source_row_id": "7069",
+        "source_split": "interactive_agent"
+      }
+    ],
+    "nemotron:a8a10dd52487852b579fad194f1cf80e": [
+      {
+        "sample_id": "N085",
+        "step_id": "fbv2:nemotron:503bd10c286f30b0fdcd70754e5284d5",
+        "status": "acceptable",
+        "block_parent": false,
+        "reason": "farm_id/device_pin/equipment_id均由用户给出；诊断工具自身接受认证凭据，工具集中无独立authenticate，因此不能机械地把未先call authenticate判错。用户同意诊断但不可据此执行被禁止的重校准。",
+        "source_row_id": "10697",
+        "source_split": "interactive_agent"
+      }
+    ],
+    "nemotron:f9a3d258f9fbbebe4d2352fae872e20f": [
+      {
+        "sample_id": "N086",
+        "step_id": "fbv2:nemotron:faa7e0c5c618b7a16cc572292e28ac50",
+        "status": "context_ambiguous",
+        "block_parent": true,
+        "reason": "用户给验证材料后authenticate_user合理，但前缀无验证码发送记录，系统与toolset也未提供发送接口；真实challenge存在与否未知。",
+        "source_row_id": "7613",
+        "source_split": "interactive_agent"
+      }
+    ],
+    "nemotron:4105cf069a4a9497479730ab2703a5b7": [
+      {
+        "sample_id": "N087",
+        "step_id": "fbv2:nemotron:309ebf168c1f996e31bfbc4f6ededfa9",
+        "status": "acceptable",
+        "block_parent": false,
+        "reason": "用户给User ID，查询积分/会员档案可验证余额；最新规则问题也可用policy文字答复，当前read是合理非唯一动作。",
+        "source_row_id": "13837",
+        "source_split": "interactive_agent"
+      }
+    ],
+    "nemotron:b55a089c8dcf0f3804553a82b788b493": [
+      {
+        "sample_id": "N088",
+        "step_id": "fbv2:nemotron:04c54c74ff648ec99fc8d0ef7a210984",
+        "status": "context_ambiguous",
+        "block_parent": true,
+        "reason": "已补查NVIDIA原始row425/index4：policy要求发码并校验；用户只给email要求send，观察调用verify_identity(method=parent_email)。完整schema的method枚举仅student_id/parent_email/guest_code，没有send模式，也无接收验证code字段。源policy与认证接口冲突已证实；当前调用符合接口说明但无法证明遵守policy，也不存在已验证唯一替代动作，因此保留动作gold歧义。",
+        "source_row_id": "425",
+        "source_split": "interactive_agent"
+      }
+    ],
+    "nemotron:fb9df802ae975b484dceacbb07e8983b": [
+      {
+        "sample_id": "N089",
+        "step_id": "fbv2:nemotron:f779d72da5251d2ded1ccb6fe0e51840",
+        "status": "acceptable",
+        "block_parent": false,
+        "reason": "用户问情感支持动物在比赛中的便利安排，明确属于policy accessibility accommodations，转人工无需臆断法律/服务动物资格。",
+        "source_row_id": "14373",
+        "source_split": "interactive_agent"
+      }
+    ],
+    "nemotron:70b766eacac85d080e8470ce5c411501": [
+      {
+        "sample_id": "N090",
+        "step_id": "fbv2:nemotron:2ae4f48d9fc620550632b8c06d3b0e57",
+        "status": "acceptable",
+        "block_parent": false,
+        "reason": "用户VIN和registered phone已齐，authenticate_customer匹配policy；不可把其后无法回答的plan terms当作当前认证错标。",
+        "source_row_id": "7752",
+        "source_split": "interactive_agent"
+      }
+    ]
+  },
+  "selection_rule": "Block full reviewed parent for explicit contradiction, current-action ambiguity, or substantive contaminated history. Never invent replacement labels.",
+  "human_annotation": false
+}

data/choice_v3/test.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a2645fa4468779772fc7569c5b70b23222e400f72c2b1d05e4b903d5ed81f9ee
+size 15544004

data/choice_v3/train.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1a6f5d3a1dead75be71ac23ce61f7812eecc5fa65289770adfbf38b470e2fb9c
+size 165993665

data/choice_v3/validation.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ff824bfffcc1a3f5bfcf3ac15d14df7ddb429f1d74ca5e35f2291827c20dfc61
+size 9193104

data/hermes_actions_v3/test.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3227e75f589673a0a0269008d8b2761f379cd7a41159a931cd2ea05a0f49bcbe
+size 267399

data/hermes_actions_v3/train.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bf4e001ed82700ec73d48bfad64304517784c00a3c4ff4bd7533a1bf58d2e6f9
+size 1623002

data/hermes_actions_v3/validation.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:429cedc1b030a4b16bb818c41ffd80c8f5cfde7fb76fed61e474923379591dfe
+size 83582

data/hint3_native_v3/test.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fa4ea54984aff79bb1aa4b793051b0e9d554881f2c24028528c53b40881e8472
+size 73960

data/hint3_native_v3/train.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d789a0db0c2a5e70c3275569a05fa5993f60f4f1bf72dddb134ad219c0932291
+size 40869

data/hint3_native_v3/validation.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2da5a239515f0a8ac85454c19607a7184cb0e805476113d713ee06a2071b27c6
+size 12648

data/legacy_replay_v3/test.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a8fa27f305f8d50bc564b84652578a664bb82fb501930b7db081d9388da9edb6
+size 1215350

data/legacy_replay_v3/train.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1a1d562eaf528a11a6c2964da09340d247b595265d2a80279faa48e7929d7455
+size 10028293

data/legacy_replay_v3/validation.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f1f24745eff42a5989e810272b8a5bf42d31f616f96ea77cb1fc109ad69c6695
+size 156799

data/nlupp_hotels_v3/test.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f95736ffdbc0353673a6e323bc39cad702494e7ee82e609f5fdb9bd3b98912f3
+size 27170

data/nlupp_hotels_v3/train.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d7b41fabbc83925de08f2e930392126e15d065bb4796f606880a02c6ab8e272d
+size 96788

data/nlupp_hotels_v3/validation.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:77c1886044a790b6af777e6a09a92dcf8f49d83b5e3ea37e0cfee2d10456e5b7
+size 27733

data/semantic_v3/test.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2a106193f9a32ce905dca30e2b3763ac967c34c81887f83b96e660652d390543
+size 3324079

data/semantic_v3/train.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:16ec519f3f9690cfe7eb3f489bac6feb30a9be4a3f7b93b55d0a30a362f0a3f2
+size 7556898

data/semantic_v3/validation.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9160a88e709e74b6bf385894dba280ed3fca1102879448e23b52d65b41a7839b
+size 1992132

data/tool_context_v3/test.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4d38b21265fafcfd0e6b32685ce501f0f7141545c0d6da73016a627be15ab1c1
+size 7724286

data/tool_context_v3/train.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7ec090fb52e83b5d31638093b72fe5d17c3b48170f69be5ce024912ec49956ec
+size 126211737

data/tool_context_v3/validation.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5c037c09d48adaeb7c34c1838049c8232dbc4c40fcc28d3195e21bccec1c7e4d
+size 5340066

data/tool_environment_holdout_v3/test.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4e36c812a6b54a321a45fce805bef13c784b196a177db91c6fcfc87f6c251e1b
+size 1008820

data/tool_environment_holdout_v3/validation.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:916608566e8821f2a4fa1ae84566f031e28429041eeba7da5bd1733410841403
+size 979530