Skip to main content

Framework Structure

The 4 Quality Gates

Four validation checkpoints — M, P, E, L — sit between the COMPEL stages and decide whether work can progress. They are the mechanism that turns COMPEL from a framework diagram into a disciplined operating rhythm.

Quality gates are inspired by classic stage-gate delivery models but adapted to the realities of AI: they enforce that evidence, controls, and accountability are in place before a model, policy, or change is allowed to advance. For the full context see What is COMPEL?, the 6 Stages reference, and the Controls catalog.

M
Between Model and Produce

Model Gate

Validates that target-state architecture, policies, risk controls, and data governance are designed and approved before any production delivery begins.

Exit criteria
  • • Target operating model signed off
  • • Risk register and control catalog complete
  • • Data governance and residency policies approved
  • • Architecture review passed
P
Between Produce and Evaluate

Produce Gate

Validates that delivered AI systems have the required controls, evidence, monitoring, and documentation before they are considered operational.

Exit criteria
  • • Model cards and system cards published
  • • Monitoring and logging in place
  • • Controls operational and tested
  • • Incident response playbooks live
E
Between Evaluate and Learn

Evaluate Gate

Validates that evaluation evidence — performance, fairness, safety, value realization — has been collected, analyzed, and reviewed.

Exit criteria
  • • KPIs measured against baseline
  • • Audit evidence complete
  • • Bias and fairness testing documented
  • • Value realization report filed
L
Between Learn and the next Calibrate

Learn Gate

Validates that lessons learned have been captured, prioritized, and converted into concrete inputs for the next transformation cycle.

Exit criteria
  • • Decision log updated
  • • Retrospective complete
  • • Improvement backlog prioritized
  • • Policy and control updates tracked

Related references