Skip to content
Docs

Replaying saved evidence

Replay lets teams re-score a saved agent run after changing assertions, rubrics or report rendering.

Terminal window
uv run dialogassert replay \
  -c dialogassert.openai.business-acceptance.yaml \
  --from-report reports/openai-business-acceptance.json \
  --output reports/openai-business-acceptance-replayed.json \
  --html reports/openai-business-acceptance-replayed.html \
  --no-fail-on-findings

Replay does not call the agent again. It evaluates saved evidence.

Use replay when you:

  • changed acceptance rubric wording;
  • improved deterministic assertions;
  • want faster report iteration;
  • want stable debugging without paying for another model run.