Do not use performance-anomaly email silence as a real-time health check

Source facts

Performance-anomaly analysis runs daily against the preceding UTC day and needs at least eight days of sufficient telemetry to establish normal behavior. Microsoft does not promise to detect every abnormal condition or replace metric alerts. An issue produces one message, not repeated reminders; a persistent issue is updated in the smart-detection feed. Performance-anomaly emails are limited to one per resource per day when a new issue exists. Microsoft Learn.

Applicability

This interpretation concerns performance-anomaly notifications, not an application’s real-time availability contract. Separate a historical detection, its email delivery, and the evidence that the application is currently responding acceptably.

DSE recommendation

Use the notification to open an investigation, and assign separate evidence for present health and recovery. Give the investigator the affected operation, observed period and relevant deployment history. Do not close the investigation because tomorrow’s mailbox contains no reminder. Identify the checks that will demonstrate recovery and retain an owner until those checks have been evaluated.

Verification

Compare the notification’s detection period with current application measurements rather than treating them as simultaneous observations. Inspect the feed for the issue’s updated status and record the independent service checks used to reach a recovery conclusion. When assessing a newly instrumented application, document whether enough representative telemetry exists for the stated learning period. Keep missing notification, missing telemetry and successful recovery as different findings.

Official references

Microsoft Learn: Smart detection performance anomalies. Source reviewed September 9, 2026.