All research
paperPreprint
Point-in-time retrieval over amended statutes
Legal text is versioned, and most retrieval systems flatten it. We describe the indexing scheme behind NexLex's as-of queries and the failure cases it removes.
- retrieval
- legal
- systems
Statutes are amended, repealed and revived. A retrieval index that stores only the current text will confidently answer a question about 2019 with the law of 2026 — a failure that is invisible to the reader because the answer is still fluent and still cited.
Interval-indexed provisions
We index at provision level with validity intervals, so each embedded passage carries the range of dates over which it was in force. Retrieval filters on the interval before ranking, which means the model never sees text that was not law on the requested date.
- Provision-level rather than document-level granularity
- Half-open validity intervals per provision version
- Date filter applied pre-ranking, never post-hoc
- Amendment chains preserved for explanation
More research
- Spacing intervals at topic level, not card levelAn in-progress evaluation of scheduling review by topic mastery rather than per-item, using StudyFlow's early cohort data.
- Attribution is not accuracyGrounded generation systems are routinely evaluated as if faithfulness to retrieved text were the same thing as being right. It is not. We separate the two and show where the confusion costs you.