Calibration isn't 'being right' — it's matching your confidence to your hit rate

A good price estimator's real weapon isn't "always landing near the answer" — that's nearly impossible. The real weapon is calibration: a state where your felt confidence and your actual accuracy line up. Someone well-calibrated gets about 90 of 100 guesses right among the ones they tagged "90% sure," and roughly half right among the "it's a coin flip" ones. Most people run the opposite way — they feel 90% sure, land 60% of the time, and never notice the gap. That's overconfidence, and it's the most common failure pattern in price estimation.

Here's the catch: calibration is separate from knowledge. Memorizing the gold price won't calibrate you. Knowing in your gut which kinds of prices you're bad at will. Maybe you're confident on FX but keep botching the order of magnitude on commodities, or you nail stock prices but always lowball salaries. That self-awareness doesn't come from cramming more facts. It accumulates only through the cycle of guess, see the result, and confront how far off you were.

Why 'one problem a day' beats cramming — the spacing effect

Learning research keeps confirming one thing: spreading the same amount of practice out over time (spacing) beats doing it all in one sitting (cramming), both for long-term memory and for intuition. Fifty problems in one go versus one a day for 50 days — same total time, different results. Each time you have to "dust off and recall" after a gap, that small load becomes retrieval practice, and retrieval practice hardens the memory.

A second mechanism stacks on top. Daily rotates the category every day — gold today, FX tomorrow, salaries the day after. Solve the same asset 50 times in a row and the previous problem's number sits down as the anchor for your next guess: yesterday's $1,800 quietly tugs today's estimate toward it. But when the category shifts daily, that anchoring and recency bias get diluted on average, because you simply never get to dwell on a single number.

Low stakes is what makes it last — and lasting is the whole game

The brutal thing about calibration is that it doesn't grow from a one-off insight; it grows only from an accumulated record. That makes "just keep doing it" more than half the battle. One problem a day, six tries, a 5% error threshold — this small burden is the point. Unlimited retries and you stop taking any single problem seriously; force 100 a day and you quit within three days. A bite-sized task lets a light "did it again today" win carry the streak forward, and the longer the streak, the more data points pile up and the sharper your calibration gets.

Concretely: when the answer is $1,000, a 5% error means ±$50, so you have to land between 950 and 1,050 to pass. A miss gives only a higher/lower hint and hides the exact gap. So your six tries become a "narrow the range" drill — throw wide first, tighten the upper and lower bounds with the directional hints, then nail the magnitude and trailing digits at the end. A miniature of statistical estimation. Do it once a day and a month later you'll carry a per-domain feel like "I can tighten FX to ±3% but should leave commodities open to ±15%." That is exactly what being calibrated means.

✍️ Operator's note — One problem a day — when I first built this mode, even I thought "how is that supposed to improve anything?" But calibration is exactly the kind of thing that builds that way. The person who lets one round slip by each day for 50 days ends up with sharper instincts than the one who crams 100 into a single sitting. My own routine now is to jot a range in my head before the problem loads, and I can feel that one-line note fending off hindsight bias ("oh I totally had that").

To actually use this game as calibration training, set a range before you guess instead of a single point — something like "850–1,100, most likely 950." Then, separately from pass/fail, log whether the answer fell inside your range. If you're often outside it, you're overconfident, so widen it; if you always land comfortably, tighten it. And since the category rotates daily, note which domains you keep missing — a month later you'll have a complete map of your own weak spots.