Git for US Legislation

Federal statutes are difficult to read historically: for many titles of the US Code, the code is prima facie evidence, while the Statutes at Large control if they conflict. Amendment notes identify which law changed a provision, but effective dates and retroactive reach may be recorded elsewhere. The changing small-business debt limit in bankruptcy law illustrates the problem: temporary increases expired and were restored with provisions reaching back to earlier cases, so the text printed at a given time did not necessarily reflect the rule ultimately applicable to a case. This history matters for lawyers and for language models, which can apply outdated rules or choose a newer rule even when an earlier version governs.

A proposed Git-like record lets readers select a date, view the text then in force, highlight words that differ from today, and trace each phrase to the Act that introduced it. The project reconstructed changes between release points from 2017 onward across nine titles: 361 of 365 changes matched the next official release exactly; the remainder carry confidence labels. It also records how changes took effect, distinguishing enactment by default, specified effective-date notes, and dates stated in amendments. Research cited on statutory question-answering found retrieval raised one model’s accuracy from 24% to 87%, while automated amendment reconstruction often produced near-matches with legally significant errors. The system’s key limitation is that it does not yet apply later laws retroactively to earlier dates, and unusual effective-date language may be missed.