History & model performance

We publish the model's track record — judge it yourself before trusting it.

6/6

leagues where the model beats league averages

7.8%

lower log loss than benchmark (1X2)

2,011

matches in holdout test

Settled forecasts

No settled forecasts yet — they appear after matches the model forecasted are finished.

Backtest — validation · 2023/2024

poisson-v1-hl180-k3 · Nov 19, 2026 · N=1979

ModelBrierLog lossECE
Poisson 1X20.59520.99590.0120
Elo 1X20.59861.00130.0247
Benchmark (league frequencies) 1X20.65351.07940.0197
Poisson O/U 2.50.48270.67600.0184
Benchmark (league frequencies) O/U 2.50.49160.68470.0034
Poisson BTTS0.49380.68700.0253
Benchmark (league frequencies) BTTS0.49400.68710.0108

Backtest — test (holdout) · 2024/2025

poisson-v1-hl180-k3 · Nov 19, 2026 · N=2011

ModelBrierLog lossECE
Poisson 1X20.59240.99260.0113
Elo 1X20.59650.99820.0242
Benchmark (league frequencies) 1X20.65201.07640.0086
Poisson O/U 2.50.48980.68310.0368
Benchmark (league frequencies) O/U 2.50.49540.68850.0131
Poisson BTTS0.49840.69170.0367
Benchmark (league frequencies) BTTS0.49450.68770.0135

Walk-forward backtest: each forecast uses only matches played earlier. Lower Brier/log loss is better. Past results do not guarantee future ones.