## Goal
<what change we are making and why>

## Preconditions
- Access: <roles/permissions>
- Feature flags: <names/defaults>
- Observability: <dashboards/alerts>

## Rollout Steps
1. Dry run: <how to validate in staging>
2. Enable flag (1%): <command/UI>
3. Validate: <SLOs, error rates, key logs>
4. Ramp (10% → 50% → 100%): <timings + checks>

## Verification Checklist
- [ ] Error rate stable
- [ ] Latency stable
- [ ] Key endpoints return expected values
- [ ] No new alerts

## Rollback
1. Disable flag: <command>
2. Revert PR: <how>
3. Verify recovery: <what to check>

## Notes
<gotchas, quirks, links to PR + incident report>