Section 7: Anti-Patterns and Gaming Detection

What Not to Measure:

Anti-Pattern	Why It Destroys the Framework	What to Measure Instead
Lines of code shipped	Incentivizes volume over value	Lead time; Experiment velocity
Story points completed	Gamed; no correlation to customer value	Validated learning; Time to first feedback
Employee utilization rate	Incentivizes busyness over learning	Exploration time protected; Experiment velocity
Uptime as sole metric	Incentivizes risk avoidance	Resilience metrics (MTTR); Lead time
AI model accuracy alone	Ignores fairness and oversight	Accuracy + fairness + explainability + override rate

Gaming Detection Signals:

Experiment velocity rises, but experiments shrink to triviality → Add "validated learning" quality gate.
Self-service rate rises, but platform quality drops → Add internal NPS and error rate.
Failure rate drops below 30% → Teams avoiding risk. Add "exploration time protected."
Postmortems published but no system changes follow → Add "postmortem → improvement" tracking.