Calibration Tracker
Measure whether your probability estimates match reality. Log predictions, track resolutions, and see your reliability diagram update over time. Calibration is the single most important skill in probabilistic forecasting.
A well-calibrated forecaster who says “70% chance” is right about 70% of the time. This tool tells you whether you are.
Historical Calibration
How well do Polymarket prices predict actual outcomes? A perfectly calibrated market prices 70% probability events at 70¢ — and they win exactly 70% of the time. Explore the data below.
Near-Perfect Calibration
Polymarket events priced at 70% resolve YES approximately 68–72% of the time — a calibration error of under 3 percentage points. This makes it one of the most accurate publicly available forecasting platforms, rivaling professional forecasters.
Loading calibration data...
Calibration FAQ
- What is forecaster calibration?
- Calibration measures whether your stated probabilities match observed frequencies. A well-calibrated forecaster who says '70% chance' is right about 70% of the time across many predictions. Calibration is the single most important skill in probabilistic forecasting; raw accuracy is worth less than knowing exactly how confident to be.
- How does the calibration tracker work?
- Log each prediction you make with the probability you assigned and the market it relates to. When the market resolves, the tracker records the outcome. Over time, it builds a reliability diagram: for each probability bucket (10%, 20%, ..., 90%), what fraction of your predictions actually came true. Perfect calibration is a straight diagonal line.
- Why do most traders lose despite being right about events?
- Because they are poorly calibrated. A trader who is right 60% of the time but thinks they are right 85% of the time will systematically over-bet and go broke. Kelly criterion assumes calibration. Without it, Kelly can ruin you faster than any other strategy.
- How many predictions do I need before the tracker is meaningful?
- Statistical significance per probability bucket requires about 30 to 50 resolved predictions per bucket. For full calibration across 9 buckets (10%-90%), that is 300+ total predictions. Start logging now; six months of serious trading will get you there.
- Are prediction markets themselves well calibrated?
- Yes, on average. Published studies of Polymarket and PredictIt show markets are well-calibrated within about 2 to 3 percentage points across most probability ranges. The interesting exception is extreme probabilities: markets at 95%+ underestimate tail events, and markets at 5% overestimate them. This is the 'long-shot bias' seen in sportsbooks too.
- Can I import past predictions from elsewhere?
- Yes. The tracker accepts CSV imports from Metaculus, PredictionBook, or manual spreadsheets. Include date, question, predicted probability, and (if resolved) outcome. Once imported, the reliability diagram updates instantly.
How to read a reliability diagram
The X axis is the probability you stated (e.g. “60%”). The Y axis is the fraction of those predictions that actually came true.
Perfect calibration is the 45-degree diagonal line. Points above the line mean you under-confident (events happened more often than you predicted). Points below mean overconfident (events happened less often than you said).
The most common mistake: traders cluster their predictions around 60-80% and are almost always overconfident in that range. Improving calibration there has more impact on long-run bankroll than any other single change.