Reading Eval Reports

Reports are viewed by batch — select a batch number on the Eval Reports page, or jump in via "View Report" from the execution records.

View	What to Look At
Overview	The batch's composite score, per-metric scores, and pass rates
Case Details	Each case's answer, score, and verdict — click in to see the full execution trace of a single case
Batch Comparison	Side-by-side comparison of multiple batches on the same case set, showing changes before and after configuration adjustments
Fact Verification	Details of how facts decomposed from answers compare against the fact base

✓

Batch comparison requires the same case set — which is exactly why we recommend fixing a case set as your "regression set": run the same set after every configuration change so the trend stays comparable.