Backups
Backups run through an authenticated cron route — the endpoint requires the backup cron secret and refuses unauthenticated requests. A backup program needs a schedule, an encryption boundary, retention, ownership, a restore test, and an incident record. The operations pulse in the platform console surfaces backup freshness alongside deployment health and queue lag, so stale backups are visible before they matter.
Restore
A backup is only as good as its last verified restore. Run restore tests on a schedule, record the result, and keep the incident record that a real recovery generates. Recovery documentation includes a degraded mode, a communication route for stakeholders, and a post-incident review — "we have backups" is not a runbook.
Secret rotation
Rotation is a staged procedure: inventory the credential, stage the replacement, validate the new value works, roll back capability if validation fails, and revoke the old value. Service accounts stay separate from people throughout. Cron, metrics, webhook, and provider secrets all rotate through this procedure — none are rotated by editing running configuration ad hoc.
Test the failure, not the happy path
Schedule a quarterly exercise that simulates a dependency failure: kill Redis, watch rate limiting deny; cut the database, watch booking reads error cleanly; unset the alert secret, watch the sink refuse. The failure model only earns its keep when it has been observed working.