Repository navigation
Conversation
CI grades with SKIP_JUDGE=1, so the line that caps a judged reward after a critical failure never ran in tests. Stub the judge, simulator and checks and assert the cap for an IR task and an RNA task: capped at 0.3 with a critical failure, unchanged below the cap or without one. Co-Authored-By: Claude <noreply@anthropic.com>
This branch has not been deployed
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
CI grades with
SKIP_JUDGE=1, so the line that caps a judged reward after a critical failure (reward = min(score, CRITICAL_CAP) if critical else score) never runs in tests; the existingtest_sonnet_rna_protocol_that_overdraws_ethanol_is_cappedchecks the trigger, not the cap. This adds two tests intests/test_benchmark_regressions.py, one for an IR task'sgrade()and one for the RNAmain(), with the judge, simulator and checks stubbed (no API, no simulator): capped at 0.3 with a critical failure, unchanged below the cap or without one.Checked: both tests fail when the cap is removed from the two graders and pass on
main; fullpytest -q -rswith harbor 0.23.0: 220 passed, 6 skipped. Test-only change, no task files touched.Generated with Claude noreply@anthropic.com