Chapter 12 / 14

Operations

Observe

Readiness distinguishes process, database, Redis, signing, and storage dependencies. Structured logs use request/correlation IDs and bounded dimensions. Metrics cover latency, errors, concurrency, provider outcomes, queue depth, retry state, content freshness, authorization decisions, and deletion lifecycle without recording personal content.

Recover

Retries use bounded attempts, deadlines, backoff, and idempotency. Circuit breakers prevent cascading vendor failure. Durable tasks retain integrity metadata and explicit terminal states. Long-running work can be resumed by a worker; user-facing requests do not wait indefinitely.

Routine ownership

Operators review stale source evidence, failed publication scans, expired sessions/tokens, retention candidates, voice deletion retries, provider circuit state, task dead letters, OTA health, and deployment drift. Every intervention records an outcome without copying sensitive payloads into tickets or logs.

Private procedures for credentials, emergency access, signing, incidents, and provider consoles remain access-controlled. This handbook explains when a runbook is needed, not the secret-bearing steps inside it.