Repository navigation
fix(config): load evaluator.re_evaluations from YAML (#12 was unenable-able) - #15
Merged
Merged
Conversation
The evaluator section was filtered through a hand-maintained allowlist that omitted re_evaluations (added for #12), so it was silently dropped on load: setting evaluator.re_evaluations in YAML had no effect, the noise-aware re-evaluation never ran, and #12's A/B was invalid (both arms ran re_eval=1). Derive the allowlist from EvaluatorConfig's dataclass fields instead, so any field loads from YAML automatically and can't silently drop again. Add test_config_loading.py to cover the YAML-load path that was untested. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Bug
load_configfiltered theevaluator:YAML section through a hand-maintained allowlist that didn't includere_evaluations(added for #12). Soevaluator.re_evaluationswas silently dropped on load — setting it had no effect, the noise-aware re-evaluation never ran, and #12's A/B was invalid (both arms ranre_eval=1). The #12 unit tests passed only because they builtEvaluatorConfig(re_evaluations=3)directly, bypassing the YAML path.Fix
Derive the allowed keys from
EvaluatorConfig's dataclass fields instead of a hardcoded set, so any field loads from YAML automatically and can't silently drop again. The allowlist's real purpose (rejecting keys that aren't config fields) is preserved.Adds
tests/test_config_loading.pycovering the YAML-load path that was untested (the gap that hid this).Note on #12
With this fix,
re_evaluations=3actually runs (verified:combined_score_stdnow persists). I re-ran the #12 A/B properly — it's inconclusive/negative, not the +0.145 originally reported (that run hadre_evalstuck at 1). #12 stays merged but default-off and explicitly unproven; the record is corrected on issue #12. This PR is the bug fix that makes the feature enable-able at all.🤖 Generated with Claude Code