Changelog
2026-10-09 · release
OpenDisclosure system rows re-measured
The OpenDisclosure Redaction and Rules only rows were re-measured on the same 135-document tuning pilot with the current pipeline (labels 1.0.0-r3+labels.20261006.1). The bare model row is unchanged from 2026-10-07. These diagnostics do not establish held-out performance.
2026-10-07 · release
R3 tuning pilot preview
Model, OpenDisclosure system and rules-only measurements on 135 documents and 892 pages from the reviewed R3 tuning revision. These diagnostics do not establish held-out performance; breakdowns are shown only when all reporting floors are met.