grants: 馃悰 Run the grant judge on the review specialty (gpt-6-luna low) - #276
Merged
Merged
Conversation
Context: every capt-hook judge defaults to the review specialty, which serves gpt-6-luna at low effort. Summary: Judge now defaults to review instead of general; a test pins it to gpt-6-luna:low. Motivation: the general specialty served the Slack grant judge on Claude haiku, which refused an allowed alert-thread reply and failed structured output in replays. Details: an explicit specialty or model on a Judge still overrides the default.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Why
Judgein #275 defaulted to thegeneralspecialty. On this host that serves the small tier on Claude haiku. Every other capt-hook judge usesreview, which serves gpt-6-luna at low effort, the model the owner ruled for the Slack write judge.What
Judge.specialtynow defaults toreview.Judgetogpt-6-luna:low.Evidence
Replays of the recorded Slack writes against v12.80.0 ran on haiku. They refused the #alerts-api not-ours reply in 2 of 4 runs and hit 5-attempt structured-output failures. The same reply passed 3 of 3 on luna before the migration.