Goodhart's law

Goodhart's original wording was technical: "Any observed statistical regularity will tend to collapse once pressure is placed upon it for control purposes." He was writing about monetary policy, where relationships between measures of money and the economy that had looked stable broke down once the authorities used them as targets. Strathern, an anthropologist writing about audit in British universities, gave it the form that spread: "When a measure becomes a target, it ceases to be a good measure." The common reading is that a proxy is informative only while nobody is being judged by it. The moment the proxy pays, people produce the proxy.

It holds because a metric is a compressed signal. It tracked the thing you cared about because, in the past, the cheapest way to move the metric was to do the real thing. Attach a reward and you change what the cheapest way is: reclassify the case, pick the easy patients, teach to the test, close the ticket without fixing the bug. The correlation that justified the metric was conditional on nobody gaming it, and the target removes that condition. In practice you see it in call-centre handling times, exam scores, lines of code, citation counts and, as the example shows, in a bank's cross-selling numbers.

Not every metric collapses. The law bites when the people measured can move the number cheaply and the real outcome is hard to see; a metric that is close to the thing itself, such as revenue or uptime as users experience it, resists better. The answer is not to stop measuring but to use several measures, keep some unannounced, rotate them, and treat a metric that improves while complaints rise as a warning. On credit: Goodhart stated the collapse, Strathern wrote the sentence, and the two are often swapped. Second-order thinking is the habit that catches this before launch: ask how people will respond to the target, not just what the target says.