TB: Goodhart's Law

"When a measure becomes a target, it ceases to be a good measure." Once people know they are being evaluated on a specific metric, they optimize for that metric — often at the expense of the underlying goal the metric was designed to proxy.

Goodhart's Law is a fundamental problem in any system that uses metrics for evaluation: education (teaching to the test), business (gaming quarterly earnings), AI alignment (reward hacking), government policy (Soviet factory quotas). The law doesn't say measurement is useless — it says that measurement alone, once made into an incentive, corrupts itself. The solution is to maintain multiple diverse metrics, change metrics regularly, and to also evaluate the underlying goal directly rather than relying solely on proxies.

• No reading examples yet — will populate as books are ingested.

• Second-Order Thinking — Goodhart's Law is a second-order effect: optimizing the metric changes the system so the metric no longer measures what it measured before
• Goodhart's Law ← this page
• Chesterton's Fence — before removing a metric or evaluation system, understand why it was put in place