You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
docs/release/runbook.md §2c is stale: legacy channel names, force-push tag moves the ruleset rejects, and no mention of autocut/scheduled promote #1178
docs/release/runbook.md §2c "Staged canary / ring rollout" no longer matches how rollouts work:
Legacy channel names. It moves dev-lead/next, dev-lead/ring0, dev-lead/ring1 and dev-lead/stable. Live channels are major-scoped (Org Status — 2026-09-25 #1184): dev-lead/v139-next, v139-ring0, v139-ring1, v139-stable.
Its tag-move commands are rejected. It uses git tag -f … && git push --force origin <channel>. cut-release.sh's own header says a local force-push is not granted the ruleset bypass for a tag update and GH013s. Only the gh api path, as used by cut-release.sh and canary-rollout.sh, is a bypass actor.
It predates the automation. It says "Don't script the whole loop unattended" and describes promotion as purely human-driven. But canary-rollout.yml now autocuts (💡 Canary Pipeline Self-Health Observability & Reliability Standard #1069) and auto-promotes gated rings every 4h (#1045b). For example, dev-lead/v139-next and v139-ring0 moved to 1d6b38b5 automatically on 2026-09-24. It also says "pr-review is stable-only for now (feat(standards): fleet caller stubs for the idea→initiative pipeline #499)", yet pr-review/v1-next exists and pr-review/v1-ring0 was created on 2026-09-25.
Acceptance criteria
Rewrite §2c around the real process:
the ring registry (standards/canary-rings.json);
the scheduled promoter and its gate verdicts;
the blocker issues and their PRE_EXISTING / SUSPECT / REGRESSION handling;
how to evaluate and promote by hand via workflow_dispatch, including dry_run defaulting to true;
Activity