QA Team
The QA team (benchmark agent) runs full test suites and benchmarks. Distinct from the code-reviewer's small regression test with early stop.
Regression Thresholds
Configurable in project-config.md. Defaults (shared with
code-reviewer for consistency):
| Level | Default threshold |
|---|---|
| Watch | > +2% |
| Regressed | > +5% |
| Critical | > +10% |
Verdicts
| Verdict | Condition |
|---|---|
| PASS | No metric in Regressed or Critical range |
| WATCH | Some Watch, none Regressed |
| REGRESSED | Any metric in Regressed range |
| CRITICAL | Any metric in Critical range or new failure |
CRITICAL triggers HALT (see Rollback Protocol). Baselines are never silently updated---always require user approval.