Estimation

Why calibration matters (the superforecaster method)

• Start from base rates (the outside view). Before reasoning about the specifics, anchor on how often things like this happen. Most forecasts go wrong by ignoring the reference class.
• Decompose (Fermi). Break a hard quantity into pieces you can estimate, multiply them out, and bracket the result. A structured guess beats a vibe.
• Use granular probabilities. Good forecasters don't round to 0 / 50 / 100. The difference between 65% and 80% is real information — say it.
• Watch your reliability curve. When you say 70%, it should happen ~70% of the time. The curve on this page is your honest self-assessment — straighten it and you're calibrated.

Honest note: this trains calibration with instant feedback — against the crowd (Polymarket) or a known value (markets, arithmetic). It is not long-horizon, real-outcome forecasting: you're not waiting weeks for events to resolve, you're tightening the gap between your confidence and a truth we can show you now.