[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"glossary-goodharts-law::en":3,"gloss-cluster-goodharts-law::en":23,"gloss-next-goodharts-law::en":9},{"slug":4,"category":5,"name":6,"definition":7,"meta_desc":8,"faq":9,"schema_markup":9,"related":10},"goodharts-law","analytics","Goodhart's Law","Goodhart's law is the observation that a measure stops being a good measure once it becomes a target. The number was worth watching because it moved with something that mattered; the moment a team is judged on it, the cheapest way to move it is usually to break that relationship rather than to improve the thing underneath. A support team held to average handling time closes conversations faster and reopens more of them. A team held to deflection publishes help articles that answer the question badly but keep the session out of the queue. Sales held to demos booked fills the calendar with people who were never going to buy. In each case the metric improves and the business does not, and nobody had to act in bad faith for it to happen — the incentive selects for the shortcut without anyone deciding to take it. That is why the usual advice, pick a metric nobody can game, is weak: a metric simple enough to be a target is simple enough to be gamed, and the more legible it is the faster someone finds the seam. What actually helps is treating a target as an estimate with a shelf life. Write down the outcome each target stands in for and check periodically that the two still move together, because a target climbing while its outcome flattens is the signature of the law operating. Declare guardrails in advance that would catch the side effects the shortcut produces, so the shortcut is recorded rather than rewarded. Prefer two or three imperfect measures to one clean one, since moving all of them dishonestly at once is harder and tends to be visible. And accept that you will need to change what you optimise for before the gap opens rather than after it has been reported to the board. None of this is an argument against measurement. It is an argument against leaving the same number in place for three years and reading it as though it still means what it meant when it was chosen. One distinction is worth keeping: a target used to coordinate work behaves differently from a target used to decide pay or headcount, and the same number can be safe in the first role and corrosive in the second.","Goodhart's law says a measure stops measuring once it becomes a target. Why ungameable metrics do not exist and what to pair every target with.",null,[11,14,17,20],{"slug":12,"name":13},"active-user","Active User (DAU, WAU, MAU)",{"slug":15,"name":16},"deflection-rate","Deflection Rate",{"slug":18,"name":19},"guardrail-metric","Guardrail Metric",{"slug":21,"name":22},"north-star-metric","North Star Metric",[24,28,29,32,35,38,42,43,46,49,50,53],{"slug":25,"category":5,"name":26,"updated_at":27},"ab-testing","A\u002FB Testing","2026-09-08T21:32:36+00:00",{"slug":12,"category":5,"name":13,"updated_at":27},{"slug":30,"category":5,"name":31,"updated_at":27},"autocapture","Autocapture",{"slug":33,"category":5,"name":34,"updated_at":27},"cost-per-resolution","Cost per Resolution",{"slug":36,"category":5,"name":37,"updated_at":27},"customer-data-platform","Customer Data Platform (CDP)",{"slug":39,"category":5,"name":40,"updated_at":41},"data-sampling","Data Sampling","2026-09-08T21:32:37+00:00",{"slug":15,"category":5,"name":16,"updated_at":27},{"slug":44,"category":5,"name":45,"updated_at":41},"double-counting","Double Counting",{"slug":47,"category":5,"name":48,"updated_at":41},"feedback-signal","Feedback Signal",{"slug":18,"category":5,"name":19,"updated_at":27},{"slug":51,"category":5,"name":52,"updated_at":27},"identity-resolution","Identity Resolution",{"slug":54,"category":5,"name":55,"updated_at":41},"incrementality","Incrementality"]