Every lesson so far updated a belief you could check against a fact you already half-knew (disease odds, prosecutor’s fallacy). This one trains a different skill: calibration — knowing how much to trust your own belief, not just what to believe.
The tool is the estimate puzzle. You give two things: a best single guess, and a 90% interval — a low and high bound you believe has a 90% chance of containing the true value. That interval is not a hedge to make you feel safe by going wide. It’s a specific, checkable promise:
If you’re well-calibrated, a true value should fall outside your stated 90% interval about 1 time in 10 — no more, no less.
Too narrow (e.g. always guessing a tight range around your best guess) and you’ll miss constantly — overconfidence, the single most common calibration failure. Too wide (e.g. “between 0 and 100%”) and you’re always “right” but the interval carries no information — useless caution. Good calibration sits between: an interval honest about what you actually know, no more and no less.
Over many estimate puzzles across this track, your hit rate against your stated intervals gets logged — that log is the actual point of these puzzles, more than any single answer.
Today’s question: what percentage of Earth’s atmosphere, by volume, is nitrogen? Give a best guess and a 90% interval you’d be surprised to miss.