planning-monitoring-checks

Define the signals that tell whether "everything is OK" or "we must stop/roll back" Change goal + critical user scenarios SLA/SLO, existing metrics/alerts Risks (especially high-impact) Define 3-7 key signals: errors, latency, business metrics, resource saturation For each signal: source, aggregation, threshold, window, alerting Define a post-rollout burn-in observation window (e.g., 1-24h) and owners Define stop/rollback criteria as measurable conditions

Key signals

Alerts + thresholds

Dashboards / views

Burn-in window

Stop / rollback criteria

</output_format> Signals are tied to target behavior and risks, not "metrics in general" Thresholds and windows are defined (otherwise it is not controllable) </quality_rules>