lm-as-a-judge
Acceptance testing for LLM features, with verdicts corrected for the judge's own error rate.