Git for the US Code

343 points · 178 comments on HN · read original →

Points and comments are a snapshot, not live.

A project applies Git-like version control to the U.S. Code, tracking legal text changes.

The U.S. Code lacks a convenient history system, unlike code. A study found models answered 24% of post-cutoff legal questions correctly, 87% with relevant text retrieved. The project processes 365 changes across nine titles from 2017 onward, reproducing 361 release points exactly. It shows text in force on any date, colored by the Act that wrote it, and lists affected provisions for term changes. Limitations include not yet showing retroactive effects and relying on pattern matching for effective-date notes.

What commenters are saying

Commenters discuss the feasibility and prior work. One notes that the Library of Congress and GPO already do this. Another explains that LexisNexis and Thomson Reuters offer similar services, involving manual work to parse legislative changes. A third highlights complexity from renumbering and court rulings, citing campaign finance law as an example. A commenter proposes that lines of law be renewed with new authors when the original dies, but others counter that this would create instability and risk losing protections. The thread also debates whether Git is suitable for natural language laws with exceptions.