Calibration Platform
How accurately do prediction market prices track real-world outcomes? This page tracks Brier scores, reliability diagrams, and cross-venue accuracy for all resolved markets indexed by ForecastMind.
Total Resolved
—
markets with known outcome
Brier Score
—
— · lower is better (0 = perfect)
Mean Absolute Error
—
average |price − outcome|
No resolved markets tracked yet.
Calibration data accumulates as markets close. Check back after the next resolution cycle.
Methodology Note
Brier Score = mean(price − outcome)² across all resolved markets where the final price was recorded before resolution. Range: 0 (perfect) to 1 (perfectly wrong). A score of 0.25 corresponds to random guessing on binary outcomes.
MAE = mean|price − outcome|, where outcome is 1.0 for YES and 0.0 for NO. Prices are the last recorded Polymarket snapshot before resolution.
Reliability diagram: markets are binned by their final price (0–10%, 10–20%, …, 90–100%). The bar height shows what fraction of markets in each bin resolved YES. A perfectly calibrated market would show bar heights matching the bin midpoints.
Brier Decomposition (Murphy, 1973): Brier = Uncertainty − Resolution + Reliability. Uncertainty is base_rate × (1 − base_rate) — the inherent unpredictability of the questions. Resolution measures how well forecasts discriminate between events that will and won't happen (higher is better). Reliability measures calibration error — the gap between predicted probabilities and observed frequencies (lower is better). Skill score = 1 − Brier/Uncertainty; values above 0 indicate performance better than always predicting the base rate.