The number gets better; the thing gets worse.

Every metric on your dashboard is a lie you’ve agreed to tell yourself, and a useful one. You care about whether people find your product valuable — but “value” doesn’t have a number, so you pick a stand-in that does: weekly active users, activation rate, sign-ups, time-in-app. The stand-in is a proxy. It correlates with the real thing well enough that watching it tells you something true. That correlation is the whole reason the metric is worth having. It is also, the instant you start steering by it, the thing that begins to break.

This has a name, from of all places monetary economics. In 1975 Charles Goodhart, watching central banks try to manage the money supply, noticed a pattern grim enough to become a law: any statistical regularity you rely on tends to collapse once you put pressure on it for control. The anthropologist Marilyn Strathern later gave it the phrasing everyone quotes — when a measure becomes a target, it ceases to be a good measure. The social scientist Donald Campbell had arrived at the same wall from another direction: the more any indicator is used to make decisions, the more it gets gamed, and the more it distorts the very process it was meant to monitor. Three people, three fields, one finding. A proxy watched is a proxy. A proxy optimized is a target, and targets get gamed — sometimes by bad actors, but far more often by well-meaning people doing exactly what you asked.

The failure has two flavors and you’ll meet both. The loud one is gaming: make “activated users” the number that matters and someone finds a way to mark users activated that has nothing to do with them getting value — a default toggle, a mandatory step reframed as an achievement, a definition quietly loosened at quarter’s end. The quiet one is tunnel vision: the team pours itself into the sliver of reality the metric can see and lets everything it can’t see rot. Optimize session count and you may be rewarding people for failing to find what they came for and having to come back. The number climbs. The experience it was standing in for gets worse in exactly the places the number is blind to. Nobody lied. The proxy just did what proxies do when you lean on them.

Worth being precise here, because this is a cousin of a mistake I’ve written about before and not the same one. The vanity-metric problem is a number that never measured what mattered — comfort food that goes up while you starve. Goodhart’s is sadder and sneakier: a number that measured the real thing honestly, right up until you made it the goal, and then curdled because you started chasing it. The better the proxy, the more tempting to target, the more it has to lose.

⁂

You can’t run without proxies; the answer isn’t to throw away the dashboard. It’s to hold the number in the right grip. Treat every headline metric as a measurement, never the mission — the mission is the thing it stands for, and you say that thing out loud so the team never forgets what it’s a proxy for. Pair it with a guardrail that moves the opposite way when it’s being gamed, so a climb that’s really rot has a place to show up (activation paired with week-four retention; engagement paired with reported satisfaction). And every so often, go back and re-check that the proxy still tracks the real outcome, because the day it stops is the day it looks best. The number getting better is not the goal. The number staying honest while it gets better is.