World Cup 2026: the final verdict on our AI forecasts
Across 90 receipt-verified pre-kickoff forecasts at World Cup 2026, the GoalPulse model called 66.7% of favorites correctly with a Brier score of 0.482 — but the betting market’s closing consensus scored 0.439 and stayed ahead (uniform baseline: 0.667; lower is better). We publish that result as-is: every forecast was locked with a public sha256 receipt before kickoff.
Data through: · The tournament is still in progress; this page updates automatically as matches finish and freezes after the final.
Headline numbers
- Scored matches
- 90
- Favorite hit rate
- 66.7%
- Model Brier
- 0.482
- Market Brier
- 0.439
- Uniform Brier
- 0.667
Brier score measures probability accuracy; lower is better. Model vs market is compared only on matches with a pre-kickoff market closing consensus.
Accuracy by stage
| Stage | Matches | Hit rate | Mean Brier |
|---|---|---|---|
| Group stage | 64 | 62.5% | 0.516 |
| Round of 32 | 14 | 78.6% | 0.392 |
| Round of 16 | 8 | 62.5% | 0.471 |
| Quarter-finals | 4 | 100.0% | 0.264 |
Calibration: when we said X%, how often did it happen?
Buckets by stated favorite confidence. A well-calibrated model has "happened" close to "said" in every bucket.
| Confidence bucket | Matches | Said (mean) | Happened |
|---|---|---|---|
| 33%–45% | 13 | 42.5% | 15.4% |
| 45%–55% | 21 | 50.8% | 71.4% |
| 55%–65% | 21 | 60.5% | 71.4% |
| 65%–75% | 23 | 69.6% | 82.6% |
| 75%–100% | 12 | 79.5% | 75.0% |
Best calls
- Argentina 3–2 Cape Verdesaid: Argentina 83% · Brier 0.045
- Jordan 1–3 Argentinasaid: Argentina 82% · Brier 0.051
- Spain 4–0 Saudi Arabiasaid: Spain 82% · Brier 0.054
- Colombia 1–0 Ghanasaid: Colombia 79% · Brier 0.071
- Canada 6–0 Qatarsaid: Canada 78% · Brier 0.077
Worst misses
- Spain 0–0 Cape Verdesaid: Spain 84% · Brier 1.472
- England 0–0 Ghanasaid: England 79% · Brier 1.365
- Ecuador 0–0 Curaçaosaid: Ecuador 79% · Brier 1.358
- Portugal 1–1 DR Congosaid: Portugal 74% · Brier 1.267
- Uruguay 2–2 Cape Verdesaid: Uruguay 70% · Brier 1.206
Verify it yourself
Every forecast on this page was snapshotted before kickoff with a sha256 receipt. Download the ledger, hash any row’s snapshot, and compare against the published receipt — no trust required.
Downloads
prediction-ledger.jsonprediction-ledger.csvai-decision-methodology.json
What we take into 2027
The honest conclusion: a transparent public model with immutable receipts did not beat the market at World Cup 2026 — and publishing that is the point. The methodology (receipts, public scoring, market baselines) carries unchanged into the Women’s World Cup 2027.
Full calibration center →Open data & receipts →Champion odds & bracket →Road to 2027 →