📈 Model Performance

The Track Record — What the Model Got Right and Wrong

Every prediction on FootUps is published before kick-off. Every miss is published here, alongside every correct call. No prediction is deleted or adjusted after the fact. That is the policy, and this page is the evidence.

Danny, FootUps Editor

70.2%
Overall accuracy — FIFA World Cup 2026
73 correct out of 104 matches • Brier score: 0.488
Editorial policy: Probabilities are locked the moment a match kicks off. Results are recorded from official sources. A match is counted as "correct" if the outcome with the highest pre-match probability matches the actual result. Draws are included. No thresholds, no cherry-picking. Publishing misses is intentional. No content farm does this. If you want to know whether to trust a model, look at what it got wrong — not just what it got right.
Accuracy by Round
Group Stage
66.7%
48 / 72 matches
Round of 32
87.5%
14 / 16 matches
Round of 16
75.0%
6 / 8 matches
Quarter-Finals
100.0%
4 / 4 matches
Semi-Finals
50.0%
1 / 2 matches
Third Place
0.0%
0 / 1 match
Final
0.0%
0 / 1 match • Model picked Argentina; Spain won
Published Misses — Biggest Upsets the Model Missed

These are the five largest gaps between the model's pre-match prediction and the actual result. The model's stated probability is shown; the outcome is what actually happened.

Match Stage Score Model gave Outcome
Ecuador vs Germany Group Stage · 25 Jun 2–1 Germany 75.2% (model picked Germany) Ecuador win ✖
Spain vs Cape Verde Group Stage · 15 Jun 0–0 Spain 79.3% (model picked Spain) Draw ✖
Portugal vs DR Congo Group Stage · 17 Jun 1–1 Portugal 76.2% (model picked Portugal) Draw ✖
Uruguay vs Cape Verde Group Stage · 21 Jun 2–2 Uruguay 75.5% (model picked Uruguay) Draw ✖
South Africa vs South Korea Group Stage · 25 Jun 1–0 South Korea 69.7% (model picked South Korea) South Africa win ✖
The Final — What the Model Got Wrong at the Last Hurdle

The model gave Argentina a 39.2% chance of winning the Final against Spain (Spain 32.4%, Draw 28.3%). It picked Argentina. Spain won 1–0. A 6.8-point upset by the model's own measure. The semi-final was similar: the model backed France (42.2%) to beat Spain; Spain won 2–0.

These misses are documented because they happened, not hidden because they're inconvenient. The 70.2% overall figure includes them.

Brier score (lower = better probability calibration): overall 0.488. A perfectly calibrated model scores 0.000; predicting every match at 33%/33%/33% scores 0.667. See methodology for how probabilities are calculated.