# What is Calibration? Calibration is the relationship between how confident you say you are and how often you turn out to be right. A well-calibrated forecaster has this relationship in alignment: when they say 80%, they're right 80% of the time. When they say 60%, they're right 60% of the time. And so on. It is the single most important skill in forecasting — and it is learnable. --- ## A Simple Example Imagine you make a hundred predictions, each at **80% confidence**. - If you're **well calibrated**, about 80 of them come true. - If you're **overconfident**, fewer than 80 come true — maybe 65. Your "80%" really meant 65%. - If you're **underconfident**, more than 80 come true — maybe 92. Your "80%" really meant 92%. This same check applies at every confidence level. Group all your 70% predictions; about 70% should come true. Group your 90% predictions; about 90% should come true. And so on. A picture of this relationship at every level is your **calibration curve**. --- ## Why It Matters Three reasons. **1. It changes how you trust your own thinking.** Once you know you're consistently overconfident at 90%, you can mentally lower the dial. Your future 90% becomes a more honest signal. **2. It distinguishes good thinking from lucky thinking.** Anyone can be right occasionally. A calibrated forecaster is right at the rate they claim — and that rate is something you can verify over time. **3. It's the only known way to actually improve.** Forecasting research consistently finds that the people who get better at prediction are the ones who keep score. The ones who don't, don't. --- ## What Calibration Is Not **It's not accuracy.** A 50% prediction that comes true is well-calibrated. A 99% prediction that comes true is more accurate but isn't necessarily better-calibrated. Calibration cares about the relationship, not the hit rate. **It's not intelligence.** Smart people are routinely badly calibrated. Knowledge of a domain doesn't automatically translate into calibrated confidence in it. **It's not the same as being right a lot.** A doomsayer who's always 95% confident the worst will happen and right about it 40% of the time is badly miscalibrated, even if you find them prescient. --- ## How Belief Tracker Measures It The app shows you calibration in three views, each on the Calibration tab: 1. **The calibration curve** — your hit rate at each confidence band, plotted against the perfect-calibration line. 2. **The Brier score** — a single number summarizing your overall calibration, where 0 is perfect. 3. **Hit rate** — the simple percentage of all your predictions that came true, regardless of confidence. Each of these has its own page in this section. --- ## How Long Does It Take? Calibration becomes statistically meaningful around 30–50 resolved predictions. Below that, your curve is too noisy to read confidently. Most people who keep up the discipline see measurable improvement within a few months. The standard pattern is: discover overconfidence, lower the dial, watch the curve straighten out over the next few dozen predictions. --- ## Next Step → [Reading the calibration curve](reading-the-curve.md)