Football accuracy

Every graded Football analysis, sorted by most recent. No cherry-picking; no filtered wins.

Graded

300

Accuracy

49.7%

Brier (↓)

0.606

Log loss (↓)

1.013

ECE (↓)

5.2%

ECE(expected calibration error) is the count-weighted average gap between confidence and outcome — < 5% is well-calibrated, > 10% means the model's probabilities can't be trusted at face value yet. Independent of accuracy: a model can be 65% accurate and 25% mis-calibrated.

🔥Current streak: 7 calls right in a row

Trend (30-match rolling)

30-match rolling

Brier 0.566↑ 0.017 vs start

Accuracy

60.0%

Brier (↓ better) Accuracy (↑ better)

Calibration (predicted probability → actual win rate)

0%50%100%

Predicted probability →

Diagonal = perfect calibration. Above it = under-confident, below = over-confident.

Dot size = sample share. Faint dots have fewer than 10 predictions; treat with care.

Above the dashed diagonal = under-confident on this sport; below = over-confident (the dangerous direction).

Per-bucket detail

Each dot is one decile of model confidence. Sit on the diagonal = well-calibrated. Below the diagonal = the model is over-confident at that confidence level (the worst regime for staking).

calibration dots (size ∝ sample count)perfect calibration

Where the model is sharpest

League	Graded	Accuracy	Mean Brier
Primera Division	58	43.1%	0.640
UEFA Champions League	58	51.7%	0.589
Premier League	55	45.5%	0.624
Serie A	40	55.0%	0.599
Bundesliga	40	40.0%	0.632
Ligue 1	38	60.5%	0.566
FIFA World Cup	11	72.7%	0.479

Sorted by sample count. Low Brier + high sample count = the model understands this league.

Match log (300)

Date	Match	Prob assigned	Brier	Result
Jul 2	Portugal vs Croatia	51.3%	0.355	✓ right
Jul 2	Spain vs Austria	54.4%	0.314	✓ right
Jul 2	United States vs Bosnia-Herzegovina	53.2%	0.329	✓ right
Jul 1	Belgium vs Senegal	53.4%	0.326	✓ right
Jul 1