Reliability
- Load-tested at expected + 2x traffic
- Rollback tested end-to-end
- Feature flag or kill switch in place
Observability
- Dashboards for the top 3 metrics
- Alerts wired to on-call
- Log queries for the top 5 likely failure modes
Security & privacy
- Threat model reviewed
- Data flows documented
- Privacy sign-off if user data is new
Comms
- Support & sales briefed
- Runbook linked from the incident channel
- Named DRI for the first 48 hours
Common pitfalls
- — Treating it as a rubber stamp. Read it, don't skim it.
- — Skipping rollback rehearsal. You will need it.
- — No named DRI for the first 48 hours after launch. That's when things break.