Nobody at the ramp is keeping score

A teacher named George Dillard published a piece connecting student cheating to AI models hacking their own tests. Both are rational responses to bad incentives, he argues, Goodhart’s Law in action, in its familiar phrasing: “when a measure becomes a target, it ceases to be a good measure.” He’s right, and his essay sent me somewhere adjacent, to a contrast between two very different sports.

A GPA works like a golf score. Every failure gets baked into the cumulative, permanently, so the rational student learns to protect the scorecard, no bold plays or bad shots allowed. Skip the AP class you might blow. Attempt nothing you can’t already do. The scorecard is honest in golf, where a round really is the sum of its shots. Learning doesn’t sum. It ratchets: once a thing is learned it stays learned, the attempts that got you there were scaffolding, and a ledger that counts them against the total is measuring the scaffolding after the building is up.

Skateboarding settled the same question for me the other way. When I roll into a session, my skill gets read as it stands. Nobody is computing my career make-percentage. My favorite vert skater, Mike Frazier , is known for knee surgeries numbering in the teens, which is a way of saying he has failed bigger and more often than almost anyone, and he’s one of the greats. The session doesn’t average his falls against him. It doesn’t celebrate them either. It stays indifferent to them in both directions, because the only thing it measures is the make, and a make can’t be faked.

School could read the state instead of the path. Some places already do; everywhere else, kids learn the golf lesson and play it safe. And that’s only half of what’s broken about testing. The other half is deciding what the test should measure at all, a question that turned out to have been settled in math class decades ago: