Prediction markets trade on the proposition that crowds can forecast the future. Polymarket, the largest crypto-native prediction market, has now resolved over — markets with real money on the line. This page asks two questions: How accurate are those predictions? And where do the mispricing opportunities live? All data is fetched live from Polymarket's public API.
§1Calibration: Are the Markets Accurate?
A market is well-calibrated if events it prices at 30% actually happen ~30% of the time. The chart below groups all resolved markets into probability bins, then plots the predicted probability (market price ~1 month before resolution) against the actual outcome rate. Points on the diagonal = perfect calibration.
Per-bin breakdown
| Predicted Bin | n | Actual Yes Rate | Gap |
|---|
§2Where the Market Was Wrong
These are the resolved markets with the largest gap between predicted probability and actual outcome — the moments where the crowd was most surprised. For a trader looking for mispricing patterns: these are the failure modes.
Pattern Analysis
Analyzing patterns…
§3Active Mispricing Opportunities (Live Spreads)
The bid-ask spread is the market's uncertainty tax. Wide spreads = disagreement = opportunity. Tight spreads = consensus = priced in. Below are the widest and tightest active spreads on Polymarket right now.
Widest spreads (disagreement)
Tightest spreads (consensus)
§4Methodology
All data is fetched client-side from Polymarket's public Gamma API and CLOB API. No backend, no API keys.
Calibration method
For each resolved market, the pre-resolution probability is reconstructed using the
oneMonthPriceChange field:
predicted = finalPrice − oneMonthPriceChange, where
finalPrice = 1 (resolved Yes) or 0 (resolved No).
This gives the market's probability estimate ~1 month before resolution,
avoiding the trivial case where the price has already converged to 0 or 1.
Markets with < $5K volume are excluded for quality.
Brier score
Brier = (1/n) Σ (pᵢ − oᵢ)², where pᵢ = predicted probability, oᵢ = outcome (0 or 1). Lower is better. 0.25 ≈ random chance.
Spread data
Spreads are the spread field
from the Gamma markets endpoint, representing the current bid-ask gap.
Wider = more disagreement / lower liquidity.
Limitations
The oneMonthPriceChange proxy may not perfectly reflect the market's probability at a fixed point before resolution — it captures the price delta over a trailing month, which may include the resolution event itself for recently closed markets. Markets with no meaningful price change are excluded (they add no calibration signal). Multi-outcome events (e.g. "Who will win the election?") are decomposed into individual binary markets.