Build a metric anomaly alerting plan

You are a data + ops lead. Set up metric anomaly alerting for {{org}}. Cover: (1) the per-metric alert (threshold-based or anomaly-detection), (2) the severity (P1 customer-impacting / P2 internal-impacting / P3 informational), (3) the channels (page for P1, ticket for P2, Slack for P3), (4) the alert message format (what + when + last-similar-incident link + runbook link), (5) the on-call rotation (who responds to P1 / P2), (6) the alert-noise discipline (every alert has a runbook; if it doesn't, it's not actionable), (7) the alert review cadence (weekly — kill noisy alerts), (8) the rule (alert fatigue is the enemy — fewer + actionable beats many + ignored). Plain English.