When a measure becomes a target, it stops measuring.

We measure what we can and reward what we measure. A school's exam results stand in for how much its pupils have learned; a hospital's waiting times stand in for the quality of its care; clicks stand in for whether anyone found the article worth reading. These proxies work — at first — because they genuinely correlate with the thing we actually care about. The trouble begins when we forget they are proxies and start treating the number itself as the goal.

The moment a measure carries real consequences — a bonus, a ranking, a budget, a job — the people being measured have every reason to raise the number by the cheapest route available. And the cheapest route is almost never “improve the underlying reality.” It is to find the gap between the proxy and the goal and climb through it. The slider above is that gap opening up: the metric soars, the reality it was standing in for does not move an inch, and the bond between them quietly dissolves.

The law is named for the British economist Charles Goodhart, who in a 1975 paper on monetary policy made a dry observation about central banks: “Any observed statistical regularity will tend to collapse once pressure is placed upon it for control purposes.” Central banks had noticed that certain measures of the money supply moved reliably with inflation, and reasoned that if they controlled the measure, they would control inflation. The relationships promptly fell apart — because targeting them changed the behaviour that had produced them in the first place.

The pithy version everyone quotes — “when a measure becomes a target, it ceases to be a good measure” — came later, paraphrased by the anthropologist Marilyn Strathern in 1997, writing about audit culture in British universities. And the same insight had already been stated, arguably first, by the social psychologist Donald Campbell, whose Campbell's law warns that the more any indicator is used to make decisions, “the more subject it will be to corruption pressures and the more apt it will be to distort and corrupt” the very thing it measures. Three thinkers, three fields, one law.

The most-told illustration is the cobra effect. As the story goes, colonial authorities in Delhi, alarmed by venomous snakes, offered a cash bounty for every dead cobra. It worked — dead cobras poured in, and the metric looked wonderful. Then enterprising locals did the rational thing: they began breeding cobras to kill for the reward. When officials realised what was happening and scrapped the bounty, the now-worthless snakes were released, and the wild population ended up higher than when the scheme began.

Once you have the shape of it, you see that the gaming is usually not even dishonest — it is obedient. People do precisely what the number asks of them. A Soviet-era parable makes the point: a nail factory judged on the number of nails floods the country with useless tacks; judged instead on the weight of nails, it forges a single monstrous spike. Both factories hit their target perfectly. Neither makes a useful nail.

Goodhart's law is not a curiosity of central banking or colonial India. It is one of the most reliable forces in any system run by numbers, and the modern world runs on numbers. The same machine grinds away in classrooms, hospitals, boardrooms, laboratories, war rooms and feeds.

Its sharpest historical form has a name of its own: the McNamara fallacy, after Robert McNamara, the US defence secretary who tried to run the Vietnam War by quantitative indicators, his favourite being the enemy body count. The number rose and rose; the war was lost regardless. The fallacy is the second half of Goodhart's curse — not just that targeted measures get gamed, but that the things which can't be counted at all get treated as if they don't exist. As one telling of the McNamara story has it: what cannot be measured is first set aside, then presumed unimportant, and finally presumed not to exist. The body count was countable; whether the Vietnamese people wanted the Americans there was not, so it fell out of the analysis entirely.

It is structural, not moral. The most important thing to understand about Goodhart's law is that it is not a complaint about cheaters. It bites even when everyone is acting in good faith, because the flaw is in the proxy, not the people. That is why “hire better staff” or “punish the gaming” rarely fixes it — the pressure simply finds the next gap. Treating it as a character problem is itself a way of missing the point.

What actually helps. None of the partial cures are magic, but they share a logic: stop letting a single number stand in for a rich goal. Watch several metrics at once, so that gaming one shows up as distortion in another. Pair every quantity with the judgement of someone who can see the thing itself. Measure outcomes that are close to what you truly want rather than convenient proxies far upstream. Keep some discretion human and unautomated. And hold the metric loosely — the instant it becomes the goal, it has already started to rot.

So the question to carry is the one in this entry's title bar: what does this number reward, now that it's a target? Ask it of your own dashboards before you ask it of anyone else's. The rest of the compendium is about numbers that mislead by accident; this is the one where the misleading is baked into the act of caring about the number at all. Measure what matters — but never forget that the measure is not the matter.