Catch a bad release before it ships

Who: release managers, on-call. Goal: don’t let an under-evidenced or risky release reach production.

The workflow

1. Set a policy: minimum confidence (e.g. 0.7), require approval, block on orphans.
2. On reconcile, a risky release lands in ready_with_exceptions or blocked_by_policy.
3. Inspect the Release 360: low confidence tier, orphaned changes, a force-push, or a missing deployment jump out.
4. Fix the gaps (curate evidence, wire the missing connector) and re-reconcile, or reject the approval.
5. If an already-approved release regresses (force-push collapses confidence), G3 auto-suspends it.

Why it works

Confidence + exceptions turn a gut-feel “is this safe?” into an explainable, enforceable gate — and the guardrails keep protecting the release even after approval.