We just published a fresh snapshot of the Open Legal Data corpus to the Hugging Face Hub. The 2026-05-20 dump covers every available record in the platform: 423,941 German court decisions, 113,537 laws across 7,058 law books, and the 7,438,459 citation edges between them. All three datasets are versioned per dump date, public, and load directly with the 🤗 datasets library.

This is the first release after long time of inactivity. Expect more updates soon. Join us on Discord to contribute to the project.

Datasets

Each dataset is published as three configs:

• dump-20260520 — the full snapshot
• dump-20260520-10k — random sample of 10,000 rows (seed=42)
• dump-20260520-1k — random sample of 1,000 rows (seed=42; strict subset of the 10k)

The 1k and 10k subsets are intended for quick exploration; both preserve the full schema and processing pipeline so prototypes scale to the full corpus without code changes.

Loading the data

from datasets import load_dataset

# Full court decisions corpus
cases = load_dataset(
    "openlegaldata/court-decisions-germany",
    name="dump-20260520",
    split="train",
)
print(cases[0]["markdown_content"][:200])

# Small subsample for exploration
sample = load_dataset(
    "openlegaldata/court-decisions-germany",
    name="dump-20260520-1k",
    split="train",
)

Each court decision carries both the original HTML in content and a clean Markdown rendering in markdown_content. Reference markers extracted with legal-reference-extraction are attached as a JSON string in reference_markers. Laws follow the same content + markdown_content convention; references are a flat denormalised table of slug-based edges (no internal database IDs).

What’s in the data

Court decisions (423,941)

By year:

By state:

By level of appeal:

Laws (113,537 sections across 7,058 books)

Largest law books by section count:

Citation graph (7,438,459)

By source side:

By target side:

Most-cited law books:

Usage ideas

• Citation analysis — the openlegaldata/legal-citation-graph-germany dataset is a slug-keyed edge list you can join straight back to the court-decision and laws datasets with pandas / DuckDB / Polars.
• Legal text generation / retrieval — markdown_content is clean enough to feed directly to tokenisers; the 1k subset is ideal for pipeline prototyping.
• Court-by-court analyses — every decision carries court metadata (court.name, court.jurisdiction, court.level_of_appeal, etc.) so you can filter or group without joins.

The older notebooks under openlegaldata/oldp-notebooks still work and show end-to-end examples of citation extraction, litigation-value extraction, and n-gram modelling on the corpus.

Licensing

In Germany, acts, statutory instruments, official decrees and official notices, as well as decisions (including court decisions) and official head notes of decisions do not enjoy copyright protection (UrhG § 5). The collection and metadata are released under the Open Database License (ODbL) v1.0.

Citation

If you use these datasets, please cite our research paper:

@inproceedings{ostendorff2020oldp,
  author    = {Ostendorff, Malte and Blume, Till and Ostendorff, Saskia},
  title     = {Towards an Open Platform for Legal Information},
  year      = {2020},
  isbn      = {9781450375856},
  publisher = {Association for Computing Machinery},
  doi       = {10.1145/3383583.3398616},
  booktitle = {Proceedings of the ACM/IEEE Joint Conference on Digital Libraries in 2020},
  pages     = {385--388},
  series    = {JCDL '20},
}

Questions, ideas or bugs? Join us on Discord or open an issue on the oldp repo.