遇见数据集

FL-Stale

收藏
OpenML2026-09-02 更新2026-09-08 收录
官方服务:

资源简介:

FL-Stale is a benchmark for STATUTORY CURRENCY: whether a system knows that a statute has changed since a court construed it. Language models routinely recite holdings whose governing statute has since been amended, because the reports they were trained on say nothing about the amendment. 1,500 items over three tasks. amended_since asks whether a Florida statute section was amended after a given year. last_amended asks for the most recent amendment year. text_changed asks whether the operative text differs between two editions of the code. Ground truth is MECHANICAL rather than interpretive. Every answer is checkable against the section's history trail or a diff of two held editions, so there is no contestable answer key. The binary tasks are balanced so a constant answer scores 0.50; the last_amended majority-class baseline is 0.064. Those baselines ship with the data so any reported score has something honest to beat. Built from a local corpus of United States judicial opinions joined to 30 annual editions of the Florida Statutes (1997-2026). Code and method: https://github.com/stepuplaw/legal-empirics DOI: https://doi.org/10.5281/zenodo.22247377 Author: Kevin D. Klagge, ORCID 0009-0002-1385-8498 Licence: CC BY 4.0. Not legal advice.

创建时间:
2026-09-02
二维码
社区交流群
二维码
科研交流群
商业服务