Lexifina builds a "git" for the US Code

A programmer can point at any line of code and see who wrote it, when, and what the file looked like on any past day. A lawyer reading a federal statute usually cannot, short of working through the notes at the foot of the section. A post on the Lexifina blog proposes giving lawyers the same system of record. The need is sharper now that so much first-pass legal research runs through language models, which typically lack access to a law's history.

Why it's hard

The US Code is published by the Office of the Law Revision Counsel as a series of release points. Each one is current through a named public law. In titles Congress has not enacted as positive law, the Code is only prima facie evidence of the law, and the Statutes at Large govern where the two differ. Which public law wrote a sentence is recorded in prose amendment notes. When a change took effect is often in a separate effective-date note. Where Congress said nothing, a law takes effect on the day it is enacted.

The authors illustrate this with the definition of a small business debtor in 11 USC § 1182(1). It was created in 2019. In March 2020 the CARES Act temporarily raised the debt limit to $7.5 million, and the increase was later extended to March 2022. It lapsed on 27 March 2022. On 21 June Congress restored it for two more years, and did so retroactively: for cases commenced on or after 27 March 2020 and still pending. The higher limit expired again in June 2024. After the April 2025 adjustment it stands at $3,424,000. For a case filed in May 2022, the release point printed at the time gives the lower limit, and only a later Act changes the answer.

Models and time

A Technical University of Munich study identified two failures in models answering questions on German statutes. They apply rules superseded after their training cutoff, and they prefer the newer provision even when an older one governs the facts. One model answered 24% of post-cutoff questions correctly on its own and 87% when the relevant statutory text was retrieved for it. Asking a model to reconstruct the old text is no fix either. Its output looked 93–99% similar to the official text, yet only half of single amendments and a fifth of amendment chains matched exactly, and experts found legally significant errors in the near misses. The authors' conclusion: the history has to be checkable.

What was built

For each pair of release points from 2017 onward, the team took the public laws enacted in between and applied their amending instructions to the older text in date order. The result was then compared with the newer version. Of 365 changes across nine titles, 361 reproduced the next release point exactly. The other four are labelled with how far they could be confirmed. Each change is tied to its law and effective date: 261 took effect on enactment by default, 72 by an effective-date note, and 29 by a date in the amendment itself. One more depends on an agency determination that may never come, so it was left out of the text.

As a result, a reader can pick a date and get the text in force on it, with the wording that has since changed marked. Each phrase is coloured by the Act that wrote it. A change to a defined term lists every provision that uses it. A section can be followed to see what has changed since the last visit.

Limits

The system does not yet reflect a later law's retroactive reach. For a case filed in the 2022 gap, it shows the lower limit without a warning. Effective-date notes are parsed by pattern, and unusual wording falls back to the enactment date. The authors advise reading the original law for anything that matters, and admit they are still working out the right form for the product.

Source: Hacker News