Don't Decide on Rollbacks by Gut Feeling
When error logs spike right after a deploy, engineers always face the same question: roll back now, or wait a bit longer? Emotional judgment calls can lead to either over-rollback (reverting a deploy that was actually fine) or under-rollback (leaving an incident unresolved too long). This tool takes three inputs — error rate increase, affected user percentage, and incident severity — and returns a 0-100 rollback risk score with a recommended action.
How the Score Works
| Score Range | Recommended Action |
|---|---|
| 70+ | Roll back immediately |
| 40-69 | Increase monitoring, reassess in 30-60 min |
| 0-39 | Keep deployment, monitor routinely |
Severity Can Override the Score
If core functions like payments or login are completely down, it's safer to roll back immediately even if the score comes out lower than expected. This tool is a reference framework — the final call should still factor in your team's on-call policy.
Frequently Asked Questions
It combines the error rate increase, the percentage of affected users, and incident severity into one score. A higher score means rollback is more urgent.
Yes, if core functions like payments or login are down, roll back immediately regardless of the score.
Re-check the error rate trend every 30-60 minutes. If it does not improve, reconsider a rollback.