KrystalPredict Backtests & Model Calibration

Direct answer

This page publishes available backtest context and calibration metrics from the running model service. Historical results describe the tested sample only; they do not predict future outcomes or guarantee returns.

Informational and educational only — not betting, financial, legal, or investment advice. 18+; follow local laws.

Matches in backtest

156

Tournaments tested

Model

blend

Best historical strategy

1/2 Kelly

Published backtest assumptions

For backtest year Y, training uses only matches from years before Y; same-year and future data are excluded.

For each match, only the single 1X2 outcome with the highest positive Full Kelly value is selected.

Historical odds are not stored; market prices are a simulated proxy based on Elo probabilities plus an overround, not live odds.

Calibration and loss metrics

Calibration sample

156

Evaluation tournaments

Log loss

elo: 1.054 · poisson: 1.063 · blend: 1.055

ROI is not repeated here unless the underlying backend returns it with a defined sample and assumptions. No metrics on this page should be read as a promise of accuracy or profit.