Case law data

Opinion text and case facts (parties, court, date, docket, citation strings) are public-domain government material. LAW.co's summaries, holdings, topics and citation graph are LAW.co's own work.

Data.co keeps the two apart. Every record carries provenance for the original content, and any AI enrichment comes in a separate block with its own provenance and an is-derived flag. AI summaries are labeled as machine-generated and are not citable as authority.

A court opinion record carries a connector-scoped id (lawco:case:<id>), the case name, reporter citation strings, the court, the jurisdiction, the decision date and the docket number. Judges, parties and the opinion author sit in a fields object. When you fetch a single record, the opinion text comes back in a body block; search results return a snippet instead.

Two lists connect a record to the rest of the corpus: authorities (the cases this opinion cites) and cited-by (later cases that cite it). Both come from LAW.co's citation graph, and LAW.co's public case endpoint caps them at 200 authorities and 50 citing cases per record, so a long list may be truncated.

The record shape is designed to support grounded answers: retrieve opinions, quote the opinion text, and cite the case name, reporter citation and source URL from the record rather than from model memory. Keep the enrichment block out of anything you present as authority.

Expect search to be noisy until LAW.co ships a paginated search endpoint. Today's upstream search samples up to 50 matches at random, so the same query can return different cases on different calls. The API passes that caveat through in meta.warnings.