Repository navigation
fix(evaluation): let scenario_meta.json route scoring, and make the F… - #646
Merged
Merged
Conversation
…MEA threshold inclusive The loader backfilled scoring_method to static_json at load time, so evaluator.py's 'scenario.scoring_method or self.default_scorer' could never fall through and --scorer-default was inert. fmea.py has therefore never scored a trial. Dropping the backfill makes the default live; the default itself moves from llm_judge to static_json so a scenario with no metadata lands on a deterministic scorer rather than a billed judge. fmea.py's pass test was strictly greater, which made f1_pass_threshold=1.0 reject every answer instead of meaning exact match. It is now inclusive. No trial in the 191-trial asset_modes cohort sits exactly at 0.70, so recorded numbers are unchanged. Signed-off-by: Dhaval Patel <pateldha@us.ibm.com>
2PR/(P+R) can land one ulp under an exact boundary: coverage 6/7 with precision 6/8 is F1 0.8 but computes as 0.7999999999999999, so the inclusive 0.80 threshold still failed it while reporting score=0.8 and "F1=0.800 (< 0.80)". The pass test now uses the score rounded to the six places it is reported at, so the verdict always agrees with the score. Across the 304 recorded FMEA trials this changes 5 verdicts from fail to pass, all at F1 = 0.8: three on 9001 (this fix), and one each on 9003 and 9015 (from making the threshold inclusive). Signed-off-by: Shuxin Lin <linshuhsin@gmail.com>
ShuxinLin
approved these changes
Oct 19, 2026
This was referenced Oct 19, 2026
Merged
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
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
FMEA threshold inclusive
The loader backfilled scoring_method to static_json at load time, so evaluator.py's 'scenario.scoring_method or self.default_scorer' could never fall through and --scorer-default was inert. fmea.py has therefore never scored a trial. Dropping the backfill makes the default live; the default itself moves from llm_judge to static_json so a scenario with no metadata lands on a deterministic scorer rather than a billed judge.
fmea.py's pass test was strictly greater, so f1_pass_threshold=1.0 rejected every answer instead of requiring an exact match. It is now inclusive. No trial in the 191-trial asset_modes cohort sits exactly at 0.80, so recorded numbers are unchanged.