Chapter 6 of 6
Confirm the result#
The module is complete when:
- the diagnostic setting exists on the Foundry account and points to the approved Log Analytics workspace;
- the action group is enabled and owned by the operating team;
- the throttling, latency, and token-volume metric alert rules are enabled, tagged with
implementationSession=optional-module-continuous-evaluation-platform-alerts, and scoped to the approved Foundry account; - the continuous evaluation rule is enabled for the governed agent; and
- recent evaluation runs show status after sampled traffic arrives.
Stop if diagnostic rows never arrive, any alert is disabled, the action group has the wrong owner, or evaluation runs remain absent after traffic and the sampling cap should allow them.
After implementation#
| What remains | Owner |
|---|---|
| Diagnostic setting, alert rules, action group, and thresholds | Observability owner |
| Continuous evaluation rule, evaluator set, thresholds, and sampling cap | AI quality owner |
| Foundry account, project, managed identity, and role assignment | Platform owner |
| Incident route and restore decision | Service owner |
| KQL checks and alert tuning notes | Azure Monitor operations owner |
Restore through the owning change paths:
- Disable only the noisy module-owned alert rule while the owner corrects the threshold.
- Disable the continuous evaluation rule before deleting it if evaluator cost or preview behavior becomes a problem.
- Remove the diagnostic setting only after the observability owner confirms another route keeps the required platform logs.
- Delete the action group only when no alert rule still uses it.
- Keep Application Insights, Log Analytics, incident records, and release evaluation gates in place.