AMP28
AI Observability & Incident Response
Monitor assistant health, evidence coverage, feedback drift, action backlog, release gates, and automation safety. Incidents draft degraded mode, rollback, and postmortem plans for human approval.
Step 1
Observe health
Audit events, evidence coverage, feedback, queues, release gates, and shadow runs.
Step 2
Create incident
Detect failures, drift, automation risk, degraded mode, and rollback steps.
Step 3
Review recovery
Queue incident review before automation or release controls are changed.
Audit events
4
Actions
0
Automations
0
Signals
1
Health score
90
Live Health Preview
Score 90/100 · severity HIGH · evidence 100% · negative feedback 0% · release gate BLOCKED
DEGRADED_REVIEW_ONLY: Assistant should operate in review-only mode for risky surfaces until evidence, feedback, and automation risks are reviewed.
HIGH RELEASE_GATE_BLOCKED
Release gate score 60/75.
Observability Incidents
No observability incidents yet.