LOCAL WORKBENCH · 03

Run Compare

Compare agent runs on the same cases and rubric. Required failures remain visible; missing measurements stay unknown.

Results JSON Template CSV Baseline Candidate

Human judgments and measured values remain distinct.