---
pretty_name: XV-ML full-ledger paper classification
tags:
- papers
- machine-learning
- research-metadata
- provenance
configs:
- config_name: default
  data_files:
  - split: train
    path: data/*.parquet
- config_name: curated_references
  data_files:
  - split: train
    path: coverage/references.parquet
- config_name: review_candidates
  data_files:
  - split: train
    path: review/data/*.parquet
---

# XV-ML full-ledger paper classification

This is the `modelomics/xv-ml` release, generated from a captured full-ledger pass.
Release `modelome-xv-ml-2026-10-08` is `complete and ready for release`. The manifest is the source of runtime
totals; this card does not invent counts or describe an incomplete manifest as
published. The direct dataset is [modelomics/xv-ml](https://huggingface.co/datasets/modelomics/xv-ml).

The capture is bounded by the recorded live-ledger high-water mark. It is a
full pass over that current ledger with curated supplements, not a global
coverage claim; unresolved source gaps remain possible.

## Storage and counts

Parquet files use the canonical `train` storage split. This is a storage name,
not a training/evaluation benchmark claim.

| Measure | Rows |
| --- | ---: |
| `classifier_included` | 1530803 |
| `curated_only` | 2047 |
| `examined_rows` | 33323692 |
| `included_rows` | 1532850 |
| `review_rows` | 8429993 |
| `runner_included_rows` | 9960930 |
| `scored_rows` | 33313758 |

| Source | Rows |
| --- | ---: |
| `acl-test-of-time` | 13 |
| `arxiv-complete-snapshot` | 700358 |
| `biorxiv` | 2832 |
| `crossref` | 34183 |
| `crossref-public-data` | 56 |
| `cvpr-longuet-higgins` | 5 |
| `daturkel-landmark-ml-papers` | 74 |
| `dblp-xml-snapshot` | 378818 |
| `epoch-all-ai-models` | 1726 |
| `europe-pmc` | 49201 |
| `gh-ml-hf-daily-papers` | 4 |
| `gh-ml-repository-doi-refs` | 14 |
| `hf-ml-doi-snapshot` | 1 |
| `hf-ml-snapshot` | 9 |
| `icml-test-of-time` | 9 |
| `medrxiv` | 390 |
| `neurips-test-of-time` | 3 |
| `openalex-snapshot-crossfield-ml` | 19696 |
| `openalex-snapshot-cs` | 344805 |
| `openalex-works` | 593 |
| `pubmed-bulk` | 5 |
| `spdin-deep-learning-chronology` | 55 |

| Record type | Rows |
| --- | ---: |
| `blog` | 152 |
| `cited_document` | 1168 |
| `model_card` | 248 |
| `paper` | 1531068 |
| `technical_report` | 214 |

| Metadata status | Rows |
| --- | ---: |
| (not reported) | — |



Chronology and coverage are recorded by the run manifest:

| (not reported) | — |

## Row schema

Rows preserve title, abstract, DOI or arXiv URL, source, source revision,
source license, article license, group identity, exact text hash, model score,
model inclusion, final inclusion, inclusion reason, classifier or curated
reference provenance, text basis, language, and eligibility status. Each row
also has `record_type` (`paper`, `technical_report`, `model_card`,
`announcement/blog`, or `cited_document`) and `metadata_status`
(`corpus_metadata`, `corpus_matched`, `hydrated`, `citation_only`, or
`unavailable`). Text basis is one of `title_abstract`, `title_only`,
`abstract_only`, or `no_text`.

This is a broad ML-research dataset, not a claim that every primary record is
a conventional paper. Curated direct-document references remain typed even
when they have no DOI, using a stable source URL key and metadata-availability
flags. A missing document URL remains an audit configuration such as
`no_document_reference`; it does not create a guessed paper.
An explicit citation without a URL can still retain its `reference_label`,
title, authors, year, stable source key, and metadata-availability flag.

Curated references are post-hoc additions retained only after an explicit
curator source audit. A curated override is provenance, not evidence that the
classifier has 100% recall. Rows with `inclusion_reason=curated_reference`
make that override explicit. Curated scholarly references from Epoch and picked
chronologies are retained irrespective of the classifier decision.

The high-recall policy is `high_recall_with_review_v1`: scorer threshold `0.35`, primary/default inclusion threshold `0.5`, and every scorer-positive row is retained in the audit collection. Curated references override the score. Non-curated rows in the `[0.35, 0.5)` band are retained in the optional `review_candidates` config with `included=false`, `inclusion_reason=review_candidate`, and review flags. Language and title-only uncertainty are flags, not vetoes. This review is not a population statistic or an accuracy estimate.

## Classifier provenance

| Field | Value |
| --- | --- |
| `rubric_version` | ml-methods-neural-apps-v1 |
| `model_version` | paper-ml-tfidf-logreg-v1 |
| `model_sha256` | df506fb1480a729f827f9825714b330a251499a07cdb1d223c5405095ac5e8a0 |
| `threshold` | 0.35 |
| `language_policy` | English normlangid confidence >= 0.90 is a review flag after model scoring; it does not veto model-positive inclusion |
| `provenance.evaluation_counts` | 79 |
| `provenance.language_detector` | langid-1.1.6-normalized |
| `provenance.training_counts` | 320 |
| `provenance.training_fingerprint` | 0d050304377d4727ccf627947c7bc33b76f66710ae6b6852521000864599554c |

| Release metadata | Value |
| --- | --- |
| `dataset_scope` | broad_ml_research_v1 |
| `inclusion_policy` | high_recall_with_review_v1 |

The current classifier was trained with 320 assistant-origin training
annotations and evaluated on 79 assistant-origin evaluation annotations under
the earlier scope. It remains a coverage anchor for blind spots; it does not
define the recall-first inclusion policy. Those metrics are agreement checks and
are not validated accuracy, human gold evaluation, or a population estimate.
Title-only and language-uncertain rows are retained with flags.

## Licensing and reproducibility

Source-specific licenses are preserved per row. This card assigns no global
redistribution license, and the release contains paper metadata/abstract
fields rather than paper full text. Keep the immutable manifest, captured
high-water mark, source revisions, model checksum, chunk receipts, and any
curation audit together when reproducing the run.

Files declared by the manifest:

- `README.md`
- `coverage/references.parquet`
- `coverage/summary.json`
- `data/part-00000.parquet`
- `data/part-00001.parquet`
- `data/part-00002.parquet`
- `data/part-00003.parquet`
- `data/part-00004.parquet`
- `data/part-00005.parquet`
- `data/part-00006.parquet`
- `data/part-00007.parquet`
- `data/part-00008.parquet`
- `data/part-00009.parquet`
- `data/part-00010.parquet`
- `data/part-00011.parquet`
- `data/part-00012.parquet`
- `data/part-00013.parquet`
- `data/part-00014.parquet`
- `data/part-00015.parquet`
- `data/part-00016.parquet`
- `data/part-00017.parquet`
- `data/part-00018.parquet`
- `data/part-00019.parquet`
- `data/part-00020.parquet`
- `data/part-00021.parquet`
- `data/part-00022.parquet`
- `data/part-00023.parquet`
- `data/part-00024.parquet`
- `data/part-00025.parquet`
- `data/part-00026.parquet`
- `data/part-00027.parquet`
- `data/part-00028.parquet`
- `data/part-00029.parquet`
- `data/part-00030.parquet`
- `data/part-00031.parquet`
- `data/part-00032.parquet`
- `data/part-00033.parquet`
- `data/part-00034.parquet`
- `data/part-00035.parquet`
- `data/part-00036.parquet`
- `data/part-00037.parquet`
- `data/part-00038.parquet`
- `data/part-00039.parquet`
- `data/part-00040.parquet`
- `data/part-00041.parquet`
- `data/part-00042.parquet`
- `data/part-00043.parquet`
- `data/part-00044.parquet`
- `data/part-00045.parquet`
- `data/part-00046.parquet`
- `provenance/model-manifest.json`
- `provenance/source-manifest.json`
- `review/data/part-00000.parquet`
- `review/data/part-00001.parquet`
- `review/data/part-00002.parquet`
- `review/data/part-00003.parquet`
- `review/data/part-00004.parquet`
- `review/data/part-00005.parquet`
- `review/data/part-00006.parquet`
- `review/data/part-00007.parquet`
- `review/data/part-00008.parquet`
- `review/data/part-00009.parquet`
- `review/data/part-00010.parquet`
- `review/data/part-00011.parquet`
- `review/data/part-00012.parquet`
- `review/data/part-00013.parquet`
- `review/data/part-00014.parquet`
- `review/data/part-00015.parquet`
- `review/data/part-00016.parquet`
- `review/data/part-00017.parquet`
- `review/data/part-00018.parquet`
- `review/data/part-00019.parquet`
- `review/data/part-00020.parquet`
- `review/data/part-00021.parquet`
- `review/data/part-00022.parquet`
- `review/data/part-00023.parquet`
- `review/data/part-00024.parquet`
- `review/data/part-00025.parquet`
- `review/data/part-00026.parquet`
- `review/data/part-00027.parquet`
- `review/data/part-00028.parquet`
- `review/data/part-00029.parquet`
- `review/data/part-00030.parquet`
- `review/data/part-00031.parquet`
- `review/data/part-00032.parquet`
- `review/data/part-00033.parquet`
- `review/data/part-00034.parquet`
- `review/data/part-00035.parquet`
- `review/data/part-00036.parquet`
- `review/data/part-00037.parquet`
- `review/data/part-00038.parquet`
- `review/data/part-00039.parquet`
- `review/data/part-00040.parquet`
- `review/data/part-00041.parquet`
- `review/data/part-00042.parquet`
- `review/data/part-00043.parquet`
- `review/data/part-00044.parquet`
- `review/data/part-00045.parquet`
- `review/data/part-00046.parquet`
- `review/data/part-00047.parquet`
- `review/data/part-00048.parquet`
- `review/data/part-00049.parquet`
- `review/data/part-00050.parquet`
- `review/data/part-00051.parquet`
- `review/data/part-00052.parquet`
- `review/data/part-00053.parquet`
- `review/data/part-00054.parquet`
- `review/data/part-00055.parquet`
- `review/data/part-00056.parquet`
- `review/data/part-00057.parquet`
- `review/data/part-00058.parquet`
- `review/data/part-00059.parquet`
- `review/data/part-00060.parquet`
- `review/data/part-00061.parquet`
- `review/data/part-00062.parquet`
- `review/data/part-00063.parquet`
- `review/data/part-00064.parquet`
- `review/data/part-00065.parquet`
- `review/data/part-00066.parquet`
- `review/data/part-00067.parquet`
- `review/data/part-00068.parquet`
- `review/data/part-00069.parquet`
- `review/data/part-00070.parquet`
- `review/data/part-00071.parquet`
- `review/data/part-00072.parquet`
- `review/data/part-00073.parquet`
- `review/data/part-00074.parquet`
- `review/data/part-00075.parquet`
- `review/data/part-00076.parquet`
- `review/data/part-00077.parquet`
- `review/data/part-00078.parquet`
- `review/data/part-00079.parquet`
- `review/data/part-00080.parquet`
- `review/data/part-00081.parquet`
- `review/data/part-00082.parquet`
- `review/data/part-00083.parquet`
- `review/data/part-00084.parquet`
- `review/data/part-00085.parquet`
- `review/data/part-00086.parquet`
- `review/data/part-00087.parquet`
- `review/data/part-00088.parquet`
- `review/data/part-00089.parquet`
- `review/data/part-00090.parquet`
- `review/data/part-00091.parquet`
- `review/data/part-00092.parquet`
- `review/data/part-00093.parquet`
- `review/data/part-00094.parquet`
- `review/data/part-00095.parquet`
- `review/data/part-00096.parquet`
- `review/data/part-00097.parquet`
- `review/data/part-00098.parquet`
- `review/data/part-00099.parquet`
- `review/data/part-00100.parquet`
- `review/data/part-00101.parquet`
- `review/data/part-00102.parquet`
- `review/data/part-00103.parquet`
- `review/data/part-00104.parquet`
- `review/data/part-00105.parquet`
- `review/data/part-00106.parquet`
- `review/data/part-00107.parquet`
- `review/data/part-00108.parquet`
- `review/data/part-00109.parquet`
- `review/data/part-00110.parquet`
- `review/data/part-00111.parquet`
- `review/data/part-00112.parquet`
- `review/data/part-00113.parquet`
- `review/data/part-00114.parquet`
- `review/data/part-00115.parquet`
- `review/data/part-00116.parquet`
- `review/data/part-00117.parquet`
- `review/data/part-00118.parquet`
- `review/data/part-00119.parquet`
- `review/data/part-00120.parquet`
- `review/data/part-00121.parquet`
- `review/data/part-00122.parquet`
- `review/data/part-00123.parquet`
- `review/data/part-00124.parquet`
- `review/data/part-00125.parquet`
- `review/data/part-00126.parquet`
- `review/data/part-00127.parquet`
- `review/data/part-00128.parquet`
- `review/data/part-00129.parquet`
- `review/data/part-00130.parquet`
- `review/data/part-00131.parquet`
- `review/data/part-00132.parquet`
- `review/data/part-00133.parquet`
- `review/data/part-00134.parquet`
- `review/data/part-00135.parquet`
- `review/data/part-00136.parquet`
- `review/data/part-00137.parquet`
- `review/data/part-00138.parquet`
- `review/data/part-00139.parquet`
- `review/data/part-00140.parquet`
- `review/data/part-00141.parquet`
- `review/data/part-00142.parquet`
- `review/data/part-00143.parquet`
- `review/data/part-00144.parquet`
- `review/data/part-00145.parquet`
- `review/data/part-00146.parquet`
- `review/data/part-00147.parquet`
- `review/data/part-00148.parquet`
- `review/data/part-00149.parquet`
- `review/data/part-00150.parquet`
- `review/data/part-00151.parquet`
- `review/data/part-00152.parquet`
- `review/data/part-00153.parquet`
- `review/data/part-00154.parquet`
- `review/data/part-00155.parquet`
- `review/data/part-00156.parquet`
- `review/data/part-00157.parquet`
- `review/data/part-00158.parquet`
- `review/data/part-00159.parquet`
- `review/data/part-00160.parquet`
- `review/data/part-00161.parquet`
- `review/data/part-00162.parquet`
- `review/data/part-00163.parquet`
- `review/data/part-00164.parquet`
- `review/data/part-00165.parquet`
- `review/data/part-00166.parquet`
- `review/data/part-00167.parquet`
- `review/data/part-00168.parquet`
- `review/data/part-00169.parquet`
- `review/data/part-00170.parquet`
- `review/data/part-00171.parquet`
- `review/data/part-00172.parquet`
- `review/data/part-00173.parquet`
- `review/data/part-00174.parquet`
- `review/data/part-00175.parquet`
- `review/data/part-00176.parquet`
- `review/data/part-00177.parquet`
- `review/data/part-00178.parquet`
- `review/data/part-00179.parquet`
- `review/data/part-00180.parquet`
- `review/data/part-00181.parquet`
- `review/data/part-00182.parquet`
- `review/data/part-00183.parquet`
- `review/data/part-00184.parquet`
- `review/data/part-00185.parquet`
- `review/data/part-00186.parquet`
- `review/data/part-00187.parquet`
- `review/data/part-00188.parquet`
- `review/data/part-00189.parquet`
- `review/data/part-00190.parquet`