Goodhart's Law

mental-model Also known as Goodhart

Goodhart's Law begins when a measurement stops being a window onto reality and becomes the object of reward. Once the score carries consequences, people reshape their behavior around the proxy, making it progressively less trustworthy.

A conference can proclaim its commitment to rigorous research while accepting fake papers—because every accepted paper brings in fees. The public value is quality; the operating incentive quietly rewards volume.

E1

When the scoreboard starts steering the game

A measure works while it merely samples an underlying reality: papers can indicate research quality, weight can indicate health, and a sleep score can indicate recovery. Promotion to a target changes the causal system. The people being measured now gain a reason to manipulate whatever feeds the number, including shortcuts that sever it from the outcome it once represented.

This need not involve dishonesty. Attention itself shifts: what is legible and rewarded crowds out what is important but harder to count. The measure can therefore become more precise at exactly the moment it becomes less truthful.

E6 E8

Where it shows up

The 66-kilogram trap

Make 66 kg the definition of success and someone can hit it through unhealthy behavior. The scale records compliance while concealing the damage done to achieve it.

E3

Learning by league table

Calling a university “best” because of its ranking substitutes the ranking system's criteria for what a particular student will actually learn there.

E5

Sleep as a nightly exam

A sleep score meant to inform you can become something you anxiously optimize. The dashboard starts commanding attention that should remain on how you sleep and function.

E7

Quantity that creates quality

Grading photography students by the number of pictures makes them chase the assigned number. Yet this case complicates the law: producing more attempts can also generate more practice and better work.

E2

Targets are not automatically corrupting

Numbers remain useful when they preserve feedback about the outcome and cannot be cheaply improved by making that outcome worse. Even a blunt quantity target may help when repetition is genuinely part of learning. The danger is not measurement itself, but treating one manipulable proxy as a complete definition of success.

E2 E4

Add a counter-metric before attaching a reward

For your next numerical target, write down one way it could rise while the real outcome deteriorates, then track that failure mode alongside it—for example, pair a weight target with strength and energy, or an output quota with a quality review. If you cannot name a counter-metric, keep the number diagnostic rather than rewarding it.

Episodes that teach this