Making the error smaller

This is not a wine thing. You already do it.

Think of joining a queue at the supermarket and guessing how long the wait will be. Six people ahead, so you guess five minutes. It takes nine. The next queue you join, you guess a bit higher than you would have. That one takes seven. Slowly you get decent at it.

You never learn "the answer," because every queue is different. What you build instead is a better sense of how much a long queue counts, or a slow cashier, or a full trolley. And you build it by seeing how wrong each earlier guess was.

Notice what you never arrive at. You do not become correct about queues. Your guesses just stop being badly wrong, and that is enough to be useful.

Smaller is the whole ambition. Not zero.

A human may improve in several ways. To make one possible mechanism visible, this chapter asks a narrower question:

What can error change inside a deliberately small copy of the pricing task?