Repository navigation
fix(evaluator): retain valid feedback from partial judge ensembles - #511
Merged
codelion merged 1 commit intoOct 11, 2026
Conversation
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
One malformed LLM judge response currently makes
_llm_evaluatediscard every judge's feedback, including valid responses already parsed. A missing metric is also implicitly scored as zero: with weights 0.25/0.75, a clarity score of 0.6 reported only by the first judge becomes 0.15.Parse each judge independently and compute each metric's weighted mean over the judges that supplied it. Invalid JSON and non-object responses are skipped with a warning. Zero-weight judges do not introduce zero-valued metrics. If all responses are invalid, the measured program fitness remains unchanged.
The regression tests cover invalid responses in either position, missing metrics, zero-weight judges, and the resulting combined fitness through
evaluate_program. They use mocked judge responses and a temporary evaluator, with no model calls.Validation: 22 targeted tests passed. The official
OPENAI_API_KEY=test-key-for-unit-tests python -m unittest discover testssuite passed 640 tests (2 skipped). The repository's isort/Black pre-commit hooks andgit diff --checkpassed.