Definition
The share of automated rollback decisions judged correct in post-hoc review, in both directions: rollbacks of genuine regressions and non-rollbacks of stable releases.
The Measurement Gap
There is no metric that tells you auto-rollback correctness in either direction, which means the system's false-revert and missed-revert rates are unknown.
Formula
RCR = correct_actions / reviewed_actions, reported with FR = false_reverts / reverts and MR = missed_reverts / regression_windows
Proposed formula, pending validation pilots.
At a Glance
Decision Supported
Whether the auto-rollback system's authority should expand, hold, or contract, and where its signal set is blind.