Locked rubric
A set of document-review criteria approved by attorneys and frozen by a cryptographic hash before any classification begins, so that the standard applied to documents cannot change during or after the review.
First used in We Ran a Blind eDiscovery Review with No Humans in the Loop
In the isResponsive methodology for technology-assisted review, rubric error is treated as upstream of model error: if the rubric marks the wrong issue as responsive, a perfect classifier will be perfectly wrong. The case team therefore reviews, edits and approves the per-request rubrics, then locks them, and the matter's workflow will not begin classification until the lock is in place.
In the blind review Legawrite.AI reported, the rubric held twenty responsiveness criteria frozen by a SHA-256 hash, and thresholds were fixed before execution.
The hash follows every downstream artifact. A classification result records it, and a defensibility certificate can be issued only if the rubric hash it records matches the locked rubric.