Every AI team put its eval pass rate in its OKRs. Six months later every score is up and nothing users see has changed. What went wrong?

When a measure becomes a target, teams tune to it. Take eval scores out of targets, keep a sealed holdout teams cannot tune against, separate capability from regression evals, and tie OKRs to user outcomes.

Updated Nov 2026 · Original practice questions with worked answers and teaching rubrics. Our editorial policy explains the sources, review process and limits.