Audit decision-driving metrics every 3 months

When This Fires

• Every 3 months for any metric that influences your decisions or behavior
• When a metric looks good but the underlying reality feels wrong — Goodhart's Law may be operating
• When someone proposes a new metric to drive decisions — plan the first audit at the 3-month mark
• During system reviews when evaluating measurement quality

Common Failure Mode

Trusting long-running metrics because they've "always worked": "We've tracked lines of code for years." The longer a metric has been used without audit, the more time gaming behavior has had to exploit the metric-outcome gap. Long tenure is a risk factor, not a safety factor. Audit frequency should increase, not decrease, with metric age.

The Protocol

(1) For each metric that drives decisions, schedule a 3-month audit. (2) Ask three questions: Incentive check: what behavior does this metric actually incentivize? Is that behavior the one you want, or has it drifted? "Words per day" incentivizes word production — but you want quality writing, which words-per-day doesn't measure. Proxy check: is the metric still correlated with the outcome you care about? Track both the metric and the outcome for a sample period. If they've diverged → the proxy has decoupled. Gaming check: describe what gaming this metric would look like. Compare that description to your current behavior. If they overlap → you're gaming without realizing it. (3) If any check fails → revise or replace the metric. (4) If all pass → retain for another 3 months, then re-audit.