COLD Cases: 8.3 Million U.S. Legal Decisions with Text and Metadata

COLD Cases is a dataset of 8.3 million United States legal decisions with text and metadata, formatted as compressed parquet files. The dataset is maintained by harvard-lil and was last updated on March 26, 2024. It exists to support the open legal movement and projects like Pile of Law and LegalBench.

Use Cases

• Train legal language models based on the corpus of 8.3 million case texts.
• Analyze legal precedent and citation networks based on case metadata.
• Benchmark legal reasoning tasks as part of projects like LegalBench.
• Extract structured information from unstructured legal decision text.
• Study the evolution of legal language and concepts across the dataset.

Strengths

• Contains 8.3 million records, providing a large-scale corpus for analysis.
• Includes both full text and metadata for each legal decision.
• Formatted as compressed parquet files for efficient storage and access.
• Explicitly supports established open legal projects, indicating community alignment.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is known, but specific data quality, completeness, and potential biases require manual inspection.
• Freshness should be verified as the last update timestamp is 2024-03-26 15:50:21.

Provenance

Source

harvard-lil (Harvard Law School Library Innovation Lab?) via Hugging Face.

Collection Method

Likely aggregated from published, precedential decisions of United States courts; part of the Collaborative Open Legal Data (COLD) project.

Freshness

Last updated 2024-03-26 15:50:21.

Geography

United States