Regression detection

Compare two runs; find what got worse.

Every meaningful release changes prompts, models, tools or retrieval. Compare a baseline run (e.g. last week's scheduled run) against the candidate run from your release branch.

GET /api/v1/test-runs/{baseline_id}/compare/{candidate_id}
POST /api/v1/test-runs/{id}/detect-regressions

The comparison surfaces tests that flipped from pass to fail, score deltas per dimension and the specific failure reasons — labeled by metric, never color-only.