Failure teaches faster than success, but only if you've built the right kind of structure to catch what it's showing you. The Scottish Enlightenment philosopher Dugald Stewart — less celebrated than his Edinburgh colleague Adam Smith, but arguably sharper on how minds update — argued that most people don't learn from experience; they experience things through their prior conclusions. The result is that your mental model absorbs new data like a river absorbs rain: it grows, but it doesn't change shape. What actually changes a model's shape is what Stewart called 'arrested attention' — the moment when an outcome so thoroughly violates your expectation that you cannot route it through existing categories. The practical implication is counterintuitive: if you want your models to stay genuinely useful, you should be cultivating specific predictions before decisions, not explanations after them. Not vague directional hunches — actual falsifiable forecasts: 'This feature will increase 30-day retention by 8%, or the model I'm using to understand this user behavior is probably wrong.' The wrongness is the data. Most practitioners do the opposite — they treat their models as lenses for interpretation rather than machines for generating bets that can lose, which means the model never gets stress-tested in the only way that matters.

Name the last product or leadership outcome that surprised you. Did you actually revise the model that failed to predict it, or did you absorb it as an exception?

Drawing from Scottish Enlightenment / Philosophy of Mind and Learning — Dugald Stewart (Elements of the Philosophy of the Human Mind, 1792–1827)