Is Kalshi Rigged or Calibrated? Independent 1,316-Market Brier Score Audit
We audited 1,316 continuous settled KXBTC15M windows (19,740 1-minute candles) to verify whether market prices match real-world outcomes.
- Across 1,316 settled windows, Kalshi's minute-4 mid-price produces an average Brier score of 0.2001.
- The market price decisively beat our internal fair-value quantitative pricing model (0.2063) at every single checkpoint (minutes 4, 7, 10, 13).
- Decile calibration is remarkably linear: contracts priced between 40¢ and 50¢ resolved YES 46.2% of the time, adhering closely to the ideal 45° calibration line.
- Naive discrepancy trading ('buying when model says 55% and market is 50%') lost an average of -2.15¢ per contract after taker fees.
Brier Score = (1 / N) × ∑[i=1 to N] (Forecast_Probability_i - Actual_Outcome_i)² Lower is better (0.0000 = perfect clairvoyance; 0.2500 = uninformative 50/50)
1. The 10-Decile Settlement Distribution
A market is well-calibrated if events priced at probability P occur with relative frequency P. We grouped 19,740 minute-level observations across 1,316 settled markets into 10 probability deciles.
As shown in our published findings, Kalshi's pricing exhibits near-textbook calibration across almost every bucket.
| Decile (Price Range) | Sample Size (n) | Mean Market Probability | Actual Empirical Win Rate | Calibration Deviation |
|---|---|---|---|---|
| 01 (0.00 – 0.10) | 2,140 | 5.1% | 4.8% | -0.3% |
| 02 (0.10 – 0.20) | 1,890 | 14.8% | 15.2% | +0.4% |
| 03 (0.20 – 0.30) | 1,920 | 24.9% | 24.1% | -0.8% |
| 04 (0.30 – 0.40) | 1,880 | 35.2% | 36.0% | +0.8% |
| 05 (0.40 – 0.50) | 2,050 | 45.1% | 46.2% | +1.1% |
| 06 (0.50 – 0.60) | 2,010 | 54.8% | 54.1% | -0.7% |
| 07 (0.60 – 0.70) | 1,890 | 64.9% | 63.8% | -1.1% |
| 08 (0.70 – 0.80) | 1,950 | 75.1% | 76.2% | +1.1% |
| 09 (0.80 – 0.90) | 1,910 | 85.2% | 84.9% | -0.3% |
| 10 (0.90 – 1.00) | 2,100 | 95.0% | 95.4% | +0.4% |
2. Why Trading Model Discrepancies Loses Money
Many quant retail traders assume that when their Black-Scholes or barrier option model differs by 3¢ or 5¢ from Kalshi's mid-price, the market is 'mispriced.'
Our canonical backtests proved the opposite: the market price incorporated microstructure order flow and spot volatility faster than theoretical models. When the model disagreed with the market, the market was right more often than the model, and after paying 1.75¢ in taker friction, traders systematically lost capital.
Frequently Asked Questions
Q1: Can I inspect the raw backtest code and data?
Yes, the complete script is in src/btc15m-predictor-backtest.ts and report output is committed in reports/btc15m-predictor-backtest-2026-10-03.txt.
Q2: What is Murphy Brier decomposition?
Murphy decomposition breaks Brier score into Reliability (calibration error), Resolution (ability to separate outcomes), and Uncertainty (inherent market entropy).
Related Guides & Teardowns
• The Real Prediction Market Breakeven Formula: Why 50¢ Contracts Need a 52.75% Win Rate → • How Kalshi 15-Minute Bitcoin Contracts Settle: CME CF BRTI 60-Second TWAP Demystified → • Kalshi Taker Fee Formula: How the $0.07 × p × (1-p) Parabolic Curve Really Works →- QuanterraOS Canonical Report: reports/btc15m-predictor-backtest-2026-10-03.txt (n=1,316)
- QuanterraOS findings.md §1 & §9b: 1,316-Market Evaluation Methodology
- Vanderbilt University: Clinton & Huang (2026) Prediction Market Analysis