Calibration Methodology
Crowd probability estimates are systematically biased: markets are generally underconfident over longer time horizons and across political domains. Field Estimate corrects for these biases using logistic recalibration constants derived from Le (2026), “Decomposing Crowd Wisdom” — a study of 292 million trades across 327,000 contracts.
The Recalibration Formula
The Le (2026) model uses a power-law transformation in probability space. For a raw market price p and a domain × horizon calibration slope b:
This is equivalent to applying a linear adjustment in logit space: logit(p*) = b × logit(p). When b> 1, the market is underconfident — prices are compressed toward 50% — and recalibration pushes the estimate toward the extremes. When b< 1, the market is overconfident, and recalibration pulls the estimate toward 50%.
The key insight from Le (2026) is that b varies substantially by domain and time horizon. A politics contract resolving in 2–4 weeks uses a slope of 1.83 — meaning the crowd is significantly underconfident and the true probability is more extreme than the market price suggests. A weather contract resolving in 0–1 hours uses a slope of 0.69 — the crowd is overconfident, and the true probability is closer to 50% than the market implies.
Calibration Slopes by Domain × Time Horizon
These slopes are hardcoded from Table 3 of Le (2026). Slopes are colored: blue = strong underconfidence / gray = near-neutral / orange = overconfidence.
| Domain | 0–1h | 1–3h | 3–6h | 6–12h | 12–24h | 24–48h | 2d–1w | 1w–1mo | 1mo+ |
|---|---|---|---|---|---|---|---|---|---|
| politics | 1.34 | 0.93 | 1.32 | 1.55 | 1.48 | 1.52 | 1.83 | 1.83 | 1.73 |
| economics | 0.96 | 1.07 | 1.03 | 0.97 | 0.98 | 0.82 | 1.07 | 1.42 | 1.20 |
| crypto | 0.99 | 1.01 | 1.07 | 1.01 | 1.01 | 1.21 | 1.12 | 1.09 | 1.36 |
| sports | 1.10 | 0.96 | 0.90 | 1.01 | 1.05 | 1.08 | 1.04 | 1.24 | 1.74 |
| weather | 0.69 | 0.84 | 0.74 | 0.87 | 0.91 | 0.97 | 1.20 | 1.20 | 1.37 |
| entertainment | 0.81 | 1.02 | 1.00 | 0.92 | 0.89 | 0.84 | 1.07 | 1.11 | 0.96 |
| science | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 | 1.17 | 1.25 |
| other | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 | 1.00 | 1.13 | 1.20 | 1.33 |
Source: Le, N.A. (2026). “Decomposing Crowd Wisdom: Domain-Specific Calibration Dynamics in Prediction Markets.” arXiv:2602.19520v1.
Brier Score Tracking
The Brier score measures probabilistic forecast accuracy: BS = (p − o)², where p is the forecast probability and o is the binary outcome (1 = YES, 0 = NO). Lower is better. A Brier score of 0.25 corresponds to random guessing; a score of 0.00 is perfect.
Field Estimate tracks Brier scores per user, per domain, and per time horizon — building a personalized calibration profile that shows whether you are systematically overconfident or underconfident in specific areas. This enables targeted improvement: if your economics Brier scores are worse than the market average, your economic forecasts are less calibrated than the crowd's and you should defer to the market price in that domain.
Five-Step Structured Forecasting Workflow
Calibration corrects for known biases in the crowd estimate. The five-step workflow produces the estimate itself — a structured process adapted from analytic tradecraft used in professional intelligence analysis.
Identify the key drivers that could push the outcome toward YES or NO. Each factor is weighted and tagged by direction. This prevents anchoring on a single scenario.
Find historical base rates for analogous situations. How often have similar questions resolved YES? Reference classes ground the estimate before any inside-view analysis.
Adjust the reference class base rate based on specific features of this case: idiosyncratic factors, timing, actor behavior, and information the market may be underweighting.
Compare the current estimate against Metaculus, Manifold Markets, and the crowd-implied probability. Note whether the deviation is intentional or whether new information should update the consensus.
Record the final estimate, the calibrated probability, the key assumptions, and the conditions that would change the estimate. The log creates accountability and enables retrospective scoring.
See how this methodology applies to a real geopolitical question in the sample analysis.