# Datasets

> Arieta Data sells three datasets cut from one monthly snapshot (corpus, history, graph), plus eval sets built to your spec on request. Custom cuts per jurisdiction, language, date range or document type are prepared within 5⁠–⁠6 business days. Each ships as Parquet (zstd) and gzip JSONL with a manifest, datasheet and licence file.

- [Arieta Data Legal Corpus](https://data.arietalabs.com/datasets/corpus/): Article-level EU and national legal text with full metadata.
- [Arieta Data Version History](https://data.arietalabs.com/datasets/history/): Every content-hash version we have observed, for point-in-time training and evals.
- [Arieta Data Citation Graph](https://data.arietalabs.com/datasets/graph/): Typed citation edges across EU and national law.
- [Arieta Data Legal Eval](https://data.arietalabs.com/datasets/eval/): Retrieval QA and citation QA sets, built to your spec from the graph and the history. (on request)

---

# Arieta Data Legal Corpus

> Arieta Data Legal Corpus is article-level text and metadata for EU legislation, CJEU case law and national law from European jurisdictions, one record per document and language at its latest version, with licence and attribution on every row.

URL: https://data.arietalabs.com/datasets/corpus/
Croissant: https://data.arietalabs.com/datasets/corpus/croissant.json
Datasheet: https://data.arietalabs.com/datasets/corpus/datasheet/
Free sample: https://data.arietalabs.com/samples/corpus.jsonl.gz

- Records: 419,355 documents as of the November 2026 snapshot
- Unit: one record per document and language (latest version)
- Formats: Apache Parquet (zstd-compressed, sharded at about 512 MB, one directory per SKU and lane); JSON Lines (gzip) (jsonl.gz, same records and field names as Parquet)
- Updates: monthly delta snapshots with upserts and tombstones
- Use cases: Retrieval-augmented generation over EU and national law; Supervised fine-tuning and continued pre-training; Article-level chunking without re-parsing HTML

## Schema

| Field | Type | Nullable | Description |
| --- | --- | --- | --- |
| id | string | no | Stable record id: the CELEX number for EU documents, the lane id (e.g. CZ-SB-2012-89) for national ones. |
| celex | string | no | CELEX number (EU) or national id; same as id for national documents. |
| language | string | no | ISO 639-1 language code of this rendering. |
| kind | string | no | legislation, caseLaw or other. |
| title | string | no | Full official title. |
| documentType | string | yes | Upstream document type (Regulation, Directive, Judgment, Zákon, …). |
| author | string | yes | Issuing institution or court. |
| dateOfDocument | date | yes | Date of the document (adoption or decision). |
| dateOfEffect | date | yes | Date of entry into force or effect, where known. |
| dateOfPublication | date | yes | Date of publication in the official journal or gazette. |
| ecli | string | yes | European Case Law Identifier, where assigned. |
| eli | string | yes | European Legislation Identifier URI, where assigned. |
| subjects | list<string> | no | Subject descriptors (EuroVoc for EU documents where available). |
| keywords | list<string> | no | Keywords from the upstream record. |
| recitals | list<string> | no | Recitals, one string each, in order. |
| articles | list<struct> | no | Articles (or sections, paragraphs, grounds) in document order. |
| articles.id | string | no | Article id within the document (e.g. art_5). |
| articles.number | string | no | Article number as printed. |
| articles.title | string | yes | Article heading. |
| articles.text | string | no | Article text, normalised whitespace, no HTML. |
| articles.repealed | boolean | no | True when the source marks the article as repealed (schema 1.2.0). Filter on it to keep in-force articles only. |
| jurisdiction | string | no | Lane code: EU or an ISO 3166-1 alpha-2 country code (UK for the United Kingdom). |
| version | int64 | no | Content version of the document as observed by the pipeline; increments when the parsed text changes. |
| contentHash | string | no | SHA-256 of the normalised parsed text of this version. |
| source | struct | no | Upstream provenance of the record. |
| source.name | string | no | Upstream publisher or portal. |
| source.url | string | no | Canonical upstream URL. |
| source.licence | string | no | Upstream licence name. |
| source.licenceUrl | string | no | Upstream licence URL. |
| source.attribution | string | no | Attribution text to pass through. |
| snapshotId | string | no | Snapshot the record belongs to, YYYY-MM. |

---

# Arieta Data Version History

> Arieta Data Version History is every content-hash version of every document the pipeline has observed, with fetch timestamps, so models and evals can be built point-in-time without leakage from later amendments.

URL: https://data.arietalabs.com/datasets/history/
Croissant: https://data.arietalabs.com/datasets/history/croissant.json
Datasheet: https://data.arietalabs.com/datasets/history/datasheet/
Free sample: https://data.arietalabs.com/samples/history.jsonl.gz

- Records: 15,098 versions as of the November 2026 snapshot
- Unit: one record per document, language and version
- Formats: Apache Parquet (zstd-compressed, sharded at about 512 MB, one directory per SKU and lane); JSON Lines (gzip) (jsonl.gz, same records and field names as Parquet)
- Updates: monthly delta snapshots with upserts and tombstones
- Use cases: Point-in-time evals without future leakage; Temporal and amendment-aware training; Diff supervision: what changed between versions

## Schema

| Field | Type | Nullable | Description |
| --- | --- | --- | --- |
| celex | string | no | CELEX number (EU) or national id. |
| language | string | no | ISO 639-1 language code. |
| jurisdiction | string | no | Lane code. |
| version | int64 | no | Version number, 1 = first observation. |
| contentHash | string | no | SHA-256 of the normalised parsed text of this version. |
| fetchedAt | timestamp | no | When this version was observed upstream. Version N is current from its fetchedAt until version N+1’s. |
| title | string | no | Title at this version. |
| articleCount | int64 | no | Number of articles in this version. |
| diffSummary | string | yes | Short human-readable summary of what changed from the previous version. |
| articles | list<struct> | no | Full article text of this version (same structure as the corpus). |
| articles.id | string | no | Article id within the document (e.g. art_5). |
| articles.number | string | no | Article number as printed. |
| articles.title | string | yes | Article heading. |
| articles.text | string | no | Article text, normalised whitespace, no HTML. |
| articles.repealed | boolean | no | True when the source marks the article as repealed (schema 1.2.0). Filter on it to keep in-force articles only. |
| snapshotId | string | no | Snapshot the record belongs to. |

---

# Arieta Data Citation Graph

> Arieta Data Citation Graph is a typed, directed edge list linking EU legislation, CJEU case law and national law: references, amendments, repeals, corrigenda, implementation, legal bases, proposals and national transposition.

URL: https://data.arietalabs.com/datasets/graph/
Croissant: https://data.arietalabs.com/datasets/graph/croissant.json
Datasheet: https://data.arietalabs.com/datasets/graph/datasheet/
Free sample: https://data.arietalabs.com/samples/graph.jsonl.gz

- Records: 801,149 edges as of the November 2026 snapshot
- Unit: one record per directed, typed edge
- Formats: Apache Parquet (zstd-compressed, sharded at about 512 MB, one directory per SKU and lane); JSON Lines (gzip) (jsonl.gz, same records and field names as Parquet)
- Updates: monthly delta snapshots with upserts and tombstones
- Use cases: Hard-negative mining and positives for legal retrieval; Graph-aware RAG and re-ranking; Amendment and transposition tracking

## Schema

| Field | Type | Nullable | Description |
| --- | --- | --- | --- |
| source | string | no | Citing document id. |
| target | string | no | Cited document id. |
| sourceJurisdiction | string | no | Lane code of the citing document. |
| targetJurisdiction | string | no | Lane code of the cited document. |
| type | string | no | reference, amendment, repeal, corrigendum, implementation, legal-basis, proposal or transposition. |
| origin | string | no | Where the edge comes from: notice, cellar, notice+cellar, cellar-nim, source or resolver. |
| context | string | yes | Text around the citation in the citing document, where captured. |
| snapshotId | string | no | Snapshot the record belongs to. |

## Edge types

| Type | Meaning |
| --- | --- |
| reference | The citing document refers to the target. |
| amendment | The citing act amends the target (including partial replacement). |
| repeal | The citing act repeals the whole target act. |
| corrigendum | The citing document corrects a language version of the target. |
| implementation | A delegated or implementing act that completes the target. |
| legal-basis | The target is the legal basis of the citing act. |
| proposal | A Commission proposal and the act it would amend or became. |
| transposition | A national measure notified for, or recorded as transposing, an EU directive. |