Production-Readiness Checklist

Mark each point: ✅ done · ⚠️ partly · ❌ missing. Your answers stay in this browser.

Access / Auth

1Every admin UI sits behind SSO/login – nothing exposed "just for now".

2Least privilege: humans, services and AI agents only get the rights they need.

3You can revoke access for one person in under 10 minutes.

4No secrets in Git, images, logs or prompts.

5Secrets can be rotated without downtime – and you have done it once.

6Backups are stored off the machine they protect.

7You have restored a backup in the last 90 days and measured the time (RTO).

8You know how much data you lose in the worst case (RPO).

9Deployments are reproducible from Git (no manual changes on servers).

10A bad release can be rolled back in minutes.

11Database migrations have a tested way back (or are backward-compatible).

12You get alerted before your users notice.

13Logs are centralized, retained, and contain no personal data/secrets.

14Disk, certificate and quota expiry are monitored.

15You can draw where every prompt and response goes (incl. logs, telemetry).

16Model and version are pinned; changes are tested before rollout.

17There is a fallback when the model/API is down or too slow.

18You know the monthly cost per environment, including hidden ones (egress, GPU idle).

19One named person owns each critical service.

20There is a written runbook for the top 3 incidents.

✅ 0⚠️ 0❌ 0 5+ ❌? That's normal for a first version, and exactly what a Production-Readiness Review prioritizes. → Review