Baseline
Canary
Ramp stage
Decision
ADVANCE
NEXT STEP
25% traffic
Rampwatch turns a few operational signals into an explicit release decision. It uses fixed guardrails and Wilson intervals so a "looks fine" dashboard does not quietly become the release policy.
Policy: rollback at ≥30% p95 regression, or a ≥0.5 pp error increase with separated 95% intervals; hold small cohorts and moderate regressions. These thresholds are demo policy, not universal SRE guidance. Source and tests ↗