Rollback Decision Correctness Rate (RCR)

GMI-MLOPS-005MLOps / AI Platform OperationsPilot-Ready

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

Category
Autonomous Change Governance

Decision Supported

Whether the auto-rollback system's authority should expand, hold, or contract, and where its signal set is blind.

← Retrain Net Economic Yield (RNEY)Drift Alert Actionable Yield (DAAY) →

Assess Your Organization

The AI-GPS Assessment evaluates your AI governance maturity and recommends KPIs like this one for your implementation roadmap.

Take the Free AI-GPS Assessment