Verdicts
Every replay ends in a verdict. Verdicts are evidence, not commands — they summarize what the replay showed so you can decide, and they’re recorded in the evidence ledger.
The three verdicts
Section titled “The three verdicts”| Verdict | What it means | Suggested action |
|---|---|---|
| Fix Verified | The candidate fix resolved the failure and didn’t break sibling traces in the cluster | Safe to ship; promote to the regression library |
| Fix Regresses | The fix either didn’t resolve the failure, or resolved it while breaking another trace | Don’t ship; revisit the fix |
| Uncertain | The evidence isn’t strong enough to call — often the original failure didn’t cleanly reproduce | Gather more traces or make the repro deterministic before deciding |
Why “Uncertain” is a feature
Section titled “Why “Uncertain” is a feature”Lumni would rather say “I can’t prove this” than give you false confidence. An Uncertain verdict usually means one of:
- The failure was flaky or environment-dependent and didn’t reproduce.
- The replay bundle was missing inputs needed for a faithful re-run.
- The fix changed behavior in a way the comparison can’t cleanly attribute.
Treat it as a signal to improve the repro, not as a soft pass.