Data Sources · Categorization · Validation
The text of every private law was drawn from GovInfo's United States Legislative Markup (USLM) XML bulk data for the Statutes at Large, which covers every volume from the First Congress to the present.
Federal case law citing private laws was drawn from CourtListener's bulk opinion data. A matching process searched opinion text for citations to private laws in this archive; each match is recorded with the specific phrase that triggered the match, so it can be reviewed against the source opinion.
Each private law was labeled with a relief type and one or more subject matter categories (a law can carry more than one of each, when it grants multiple forms of relief or touches multiple topics). Labels were assigned by Claude (Sonnet 4.6), given the text of each private law and a codebook defining every category, run over the full set of more than 46,000 laws.
The relief-type codebook defines five top-level categories — Monetary Payments, Non-Monetary Relief from Existing Obligations, Granting Jurisdiction or Process, Granting Property Rights, and Other Administrative Action — each broken into sub-types, described in full on the About page. Subject matter uses the 21-category topic codebook from the Comparative Agendas Project.
A sample of private laws, drawn evenly across historical eras, was independently hand-coded to check the classifier's labels against human judgment.
Add validation results here (e.g. inter-rater agreement between the human-coded sample and the model's labels, for both relief type and subject matter).