# COLD Cases: 8.3 Million US Court Decisions with Opinions and Metadata (Harvard Library Innovation Lab, CourtListener)

## Facts

- Provider: Harvard Library Innovation Lab; Free Law Project (CourtListener)
- License: Public Domain (https://www.courtlistener.com/help/api/bulk-data/)
- Size: 35 files
- Format: Parquet (in sampled files)
- Stored at: Hugging Face, https://huggingface.co/datasets/harvard-lil/cold-cases
- Access: public, readable without an account
- Category: Natural Language Processing
- Source: https://github.com/harvard-lil/cold-cases-export
- Page: https://scigantic.com/archive/fe972a4d-d556-5d91-987d-6cbd32d2aa2f
- Keywords: Natural Language Processing, Parquet, Hugging Face

## About

Collaborative Open Legal Data (COLD) Cases: 8.3 million published United States court decisions (caselaw) with full opinion text and metadata, one record per decision. The Harvard Library Innovation Lab, working with the Free Law Project, reformats CourtListener's bulk data so that each decision carries its majority, concurring and dissenting opinions with their authors, the head matter, the judges, the filing date (with an approximate-date flag), and the case name in short and full forms. This is the main open corpus for empirical legal studies and judicial politics, for example how judges write and cite precedent, how courts disagree, and how legal language changes over time. It is also used for legal NLP and retrieval. The data is in 32 gzip-compressed Parquet files (about 32 GB); stream it with Hugging Face datasets, or read it with pyarrow or DuckDB. CourtListener states that its bulk data files are free of known copyright restrictions (Public Domain Mark). Credit the Harvard Library Innovation Lab and the Free Law Project (CourtListener).

## Load the data

```python
from datasets import load_dataset
ds = load_dataset("harvard-lil/cold-cases")
print(ds)
```

## Data structure

File types in sampled files: Parquet

Example file: `part-00000-bd7b5d85-e1bd-440b-8817-95ec234884f3-c000.gz.parquet`

## Open it

- In a hosted notebook with the data mounted: https://scigantic.com/archive/fe972a4d-d556-5d91-987d-6cbd32d2aa2f
- Full record as JSON (schema card included): https://api.scigantic.com/api/archives/fe972a4d-d556-5d91-987d-6cbd32d2aa2f
- Over MCP: `get_schema_card` with id `fe972a4d-d556-5d91-987d-6cbd32d2aa2f` at https://api.scigantic.com/mcp

## Related topics

- [Parquet datasets](https://scigantic.com/topics/parquet)
- [Natural Language Processing datasets](https://scigantic.com/topics/natural-language-processing)
