Release the Backend to Prod

Symptoms

A commit on main has run on staging and the team agreed to ship it.

The prod api restarts (rolling). DB migrations in the release run at its boot and are one-way.
If the broker changed since the last release, the prod broker restarts first: every WebSocket drops and clients reconnect. Announce it.

Pick the commit: the newest commit on main whose staging deploy ("ci: push image and deploy to Northflank (staging)") is green. A docs-only merge deployed nothing and can't be promoted.
Pick the version:
- a major for any breaking API change (the promote checks it against the published contract);
- otherwise a minor for new features, a patch for fixes.
- It must be above every published version.

Check the migrations the release adds, and plan their effect on prod data:

git diff --stat v<last version> <commit> -- internal/database/migrations

Dry run. Actions → "Prod - Promote Staging Image to Prod" → Run workflow:

Field Value
Use workflow from main
staging_sha the commit (7 to 40 characters)
version the version
dry_run checked
release_summary empty

It runs every guard and changes nothing. Check the preflight job's logs (the staging proof names the qubital-infra commit and says verified; the scans are also in the run summary) and the contract-guard job's last step (No breaking changes …, or a breaking change that needs a major).
Release. Run the workflow again with the same inputs, dry_run unchecked, and a release_summary of one to three sentences.
If it fails, read the failed step, fix the cause, then Re-run all jobs. Never "Re-run failed jobs": the deploy jobs refuse it, because it would reuse an earlier attempt's guards.

The run is green; the GitHub Release v<version> exists and its notes name the qubital-infra commits.
curl -fsS https://api.qubital.space/health answers.
Sign in to the desktop app against prod, send a chat message, change your presence.

The team owner (Northflank and repository owner).