Operations Runbook

Failures are visible, explainable, and recoverable.

Failure flow

flowchart TD
    A[Job triggered] --> B{Gate pass?}
    B -- Yes --> C[Publish pack + manifest]
    B -- No --> D[Record incident]
    D --> E[Notify channels]
    E --> F[Open recovery item]
    F --> G[Retry with idempotency key]

Incident sequence

1. Identify failing gate and impacted locality scope.
2. Pause publish for affected scope if certainty drops.
3. Capture audit evidence and open tracked incident.
4. Apply fix with explicit rollback step.
5. Re-run targeted checks and close only on proof.

Rollback checklist

Whisper-audit wording

[Intention: Amanah] Publish remains paused for this scope until gate confidence recovers and rollback evidence is complete.