The record — calibration

Graded against reality.

Every probability we publish comes from a sensor we can score: when a prediction market resolves, reality hands us an answer key we did not write. This page is the referee’s ledger — the crowd consensus we cite, and the deterministic engine we run on top of it, graded with proper scoring rules. We grade to keep the probabilities we publish honest — not to outguess the crowd. Markets are a sensor we cite, never an opponent we score against; a tie with the crowd here is the system working, not an edge failing. As of 2026-07-13. Case-selection methodology →

The living benchmark

Forecasts are logged while each market is still open, then graded only after reality resolves it — the record cannot be back-filled, because the forecast exists before the answer does. So far 26,886 open-market forecasts have been captured and 13,472 have matured and been scored.

Scores on matured markets. Brier: mean squared error of the probability, 0 is perfect (with a 95% confidence interval, resampled by market). Log loss: same idea, punishing confident misses harder. Skill: improvement over always guessing the base rate, 1 is perfect, 0 is no better.
forecasterBrier (95% CI)log lossskill vs base ratemarkets
Crowd consensus (the sensor we cite)0.0705 [0.0678, 0.0734]0.2260.445513,472
Our engine (crowd + deterministic correction)0.0704 [0.0678, 0.0734]0.22670.44613,472

Calibration curve, as a table

A calibrated forecaster’s 30% calls should come true about 30% of the time. Each row groups the crowd’s matured forecasts by stated probability and shows what reality delivered — this is the resolved-outcome table, bin by bin.

stated probabilityresolved marketsmean predictioncame trueread
0–10%8,2522%1.4%calibrated
10–20%1,25414.5%12.4%calibrated
20–30%1,04724.6%20.3%overconfident by 4.3%
30–40%89034.8%31.7%overconfident by 3.1%
40–50%70044.7%32.1%overconfident by 12.6%
50–60%38053.4%48.4%overconfident by 5%
60–70%14364.2%57.3%overconfident by 6.8%
70–80%13774.7%74.5%calibrated
80–90%11285%81.3%overconfident by 3.7%
90–100%55797%99.3%calibrated

The out-of-sample check

A second benchmark guards against fitting the past: 34,325 resolved markets split by resolution date, the engine fitted only on the earlier half and scored only on the later half, against a base-rate null.

forecasterBrier (95% CI)log lossskill vs base ratemarkets
Base rate (the null)0.1063 [0.103, 0.1097]0.3699017,163
Crowd consensus0.0705 [0.0658, 0.0756]0.23390.55554,069
Our engine0.07 [0.0652, 0.0751]0.23180.55894,069

The honest reading: the crowd is a strong, well-calibrated sensor — which is exactly why we cite it — and our engine’s deterministic correction adds a small, measured improvement on top. We publish the grade either way; the referee runs whether or not we like the score.

Reproducibility

The full dataset behind this page — every score, confidence interval, and calibration bin — ships as machine-readable JSON, regenerated daily as new markets mature.

Informational only — not financial, legal, or investment advice. Prediction-market prices are shown as a signal of what the crowd believes.