Segment Quality Divergence Rate (SQDR)

GMI-MLOPS-004MLOps / AI Platform OperationsIntegration-RequiredPattern O-22: Fairness Drift

Definition

The rate at which quality gaps between defined user or workload segments widen across successive model updates, measured against the aggregate quality trend.

The Measurement Gap

There is no metric that tells you segment-level quality shift across model updates, which means fairness regressions ship inside net-positive aggregate evals.

Formula

SQDR = (gap_after - gap_before) per update, where gap = q_aggregate - min_s(q_s)

Proposed formula, pending validation pilots.

At a Glance

Category
Evaluation Integrity

Decision Supported

Whether a net-positive model update is shipping a regression to specific segments, and which updates require segment-level rollback or remediation.

← LLM Judge Calibration Index (LJCI)Change Review Absorption Ratio (CRAR) →

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