AI Scorecard

Every prediction. Every result. Nothing hidden, nothing deleted.

Predictions

Correct

Wrong

30%

Accuracy

0.722

Average Brier Score

Lower is better. A random model scores ~0.33; a perfect model scores 0.

Brier score explained

Brier score measures calibration, not just accuracy. A model that says "60% home" when homes win 60% of the time scores perfectly on those calls even if accuracy looks flat.

Context: the best football prediction models sit at 55-65% accuracy across a full season. A Brier score below 0.22 is competitive. Anything above 60% on a large sample is strong. High-confidence picks should outperform low-confidence ones. If they don't, the model is miscalibrated.

Accuracy by confidence tier

High-confidence picks should win more often. If they don't, the model needs recalibration.

TierTotalCorrectAccuracy

High (≥70%)21

50%

Medium (50–69%)204

20%

Low (<50%)2810

36%

Calibration

When Marcus says 70%, does it actually win 70% of the time? Predicted Actual

30–40%8

-13.5%

40–50%20

-4.5%

50–60%17

-36%

60–70%3

-32.6%

70–80%2

-24.3%

Best call

Arsenal vs Coventry City

ARSENAL74% confident3–0

Worst miss

Chelsea vs Hull City

CHELSEA74% confident2–2

This is what honest AI looks like. We don't delete the misses.

The record is public and permanent. If you find the accountability useful, support Marcus on Ko-fi.

Support on Ko-fi