Incident Workflows
Docs/Backend Core/Incident Recovery Workflows

Incident Recovery Workflows

6-STEP

This guide details the closed-loop recovery state machine from initial error alert to verified restoration.

4 min read
•
Production Verified

End-to-End Execution Sequence

Sequence Flow Diagram

Chronological Sequence: Webhook Ingest, Multi-Factor Bisection, Human Approval, Rollback, and Canary Verification

The 8-Factor Commit Blame Scorer

Every commit shipped in a broken release bundle is evaluated against the runtime error and stack trace using an 8-factor mathematical scoring model:

FactorScoring RulePoints Awarded
1Exact File Match in Stack Trace+40 pts
2Line Number in Diff Hunk Range+30 pts
3Function / Symbol in Diff Patch+25 pts
4Symbol Mentioned in Commit Message+25 pts
5Same Directory Match+10 pts
6Error Token in Commit Message+10 pts
7Error Token in Diff Patch+8 pts
8Recency Decay (newer commits favored)Up to +10 pts

The 6 Preflight Rollback Safety Gates

Before an operator can approve a rollback, 6 safety assertions are evaluated:

CheckTypeSafety Assertion
target_readyBlockingTarget deployment on Vercel must be in READY state.
target_differsBlockingTarget deployment ID and SHA must differ from current broken production.
target_not_known_badBlockingTarget deployment has not caused an earlier recorded incident.
no_driftBlockingLive production has not changed since the proposal was generated.
not_expiredBlockingProposal must be approved within its 15-minute Time-to-Live.
still_failingWarningAdvisory check confirming service is still failing (turns yellow if recovered).