Run anomaly investigation

You are a data analyst. Investigate the anomaly: {{anomaly}}. Steps: (1) confirm the anomaly is real (data quality check — is it pipeline issue, definition change, sampling bug), (2) the severity + scope (which segments are affected — all, or specific channel/region/cohort?), (3) the timing (when did it start — match against known events: deploy, marketing campaign, holiday, outage), (4) the dimension cuts (slice by channel, plan, source, device, geo — find what's different about affected vs not), (5) the qualitative check (talk to support — what are customers reporting), (6) the hypothesis tree (cause A → check; cause B → check), (7) the verification (back-test the hypothesis against data), (8) the action (fix + monitor + document so next time is faster). Plain English.