Skip to content

fix: the eval ranks a landed attack worst and refuses reviewer names that cannot be told apart - #78

Merged
alsoknownaswill merged 1 commit into
mainfrom
eval-reviewer-clash
Nov 3, 2026
Merged

alsoknownaswill merged 1 commit into
mainfrom
eval-reviewer-clash

Conversation

@alsoknownaswill

Copy link
Copy Markdown
Member

The acceptance eval now treats a landed attack or a stopped control as the worst outcome across reviewers regardless of line order. A case is complete only when each reviewer name has reported it (a set of names per case instead of a line count). The acceptance run refuses repeated reviewer names and names where one ends in "-" plus another, before anything runs.

Tests cover the ranking in both line orders, a reviewer repeating a case, and both refusals.

…that cannot be told apart

A landed attack (PASSED) or a stopped control is the worst outcome across reviewers, whatever the line order. A case counts as reported only when every reviewer name has reported it, so one reviewer repeating a case no longer stands in for another that is silent. The acceptance run refuses repeated reviewer names and names where one ends in "-" plus another, since result tags could not be split into case id and reviewer.
@alsoknownaswill
alsoknownaswill merged commit 2f28105 into main Nov 3, 2026
1 check passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant