Open Builder > Evaluations to review conversation quality. The Overview, Run history, and Compare runs tabs help you inspect current results and changes between runs.
Evaluation criteria, filters, and pending results

Evaluation overview before the first completed result.

Choose a run, channel, evidence filter, and date range. The overview tracks evaluated, pending, and failed conversations and scores six criteria: retrieval quality, answer accuracy, conversation continuity, lead capture execution, redundancy, and conciseness. Higher scores are better, including redundancy; only applicable criteria contribute to the overall score. Select a criterion and open a conversation under Conversations to review to inspect its evidence. No score or Pending means a completed result is not yet available. Use Run evaluation to start a quality check and Export for the available results. When comparing runs, use matched conversations where possible: different conversation populations can change aggregate scores.
Evaluations are Beta. Review failed transcripts and evidence before deciding that a change is ready for release.