Updated 2026-06-03
World Cup Prediction Accuracy
No model is a crystal ball. Here is the measurement framework KickOracle uses for World Cup predictions: probabilistic scoring, calibration, benchmark comparison and limitations stated plainly.
$49 · one-time, all 104 matches
Accuracy report refresh
Walk-forward backtest results are being recalculated
We are replacing the retired leave-one-out retrospective with a walk-forward evaluation that better matches how the World Cup model is used before kickoff. Until that run is published, this page describes the measurement contract rather than showing placeholder accuracy numbers.
The next report will publish tournament sample size, Brier score, top-1 accuracy, log loss, calibration buckets, benchmark comparisons and the limitations needed to read those figures honestly.
Report status
Refreshing
Walk-forward evaluation in progress
Core metrics
4
Brier, calibration, top-1 accuracy and log loss
Benchmarks
Market
Compared with strong public baselines where available
How we measure accuracy
Predicting football is probabilistic, so a single right-or-wrong score is misleading. We grade the model the way forecasters do: Brier score (the mean squared error between predicted probabilities and the actual home/draw/away result — lower is better), top-1 accuracy (how often the highest-probability outcome happened), log loss and a calibration curve (do the things we call 60% actually happen about 60% of the time?). Together these show not just whether picks land, but whether the probabilities can be trusted.
- Brier score: probability error, lower is better.
- Top-1 accuracy: share of matches where the favoured outcome won.
- Calibration: predicted probability vs observed frequency.
- Scored out-of-sample on completed tournaments.
Benchmarked, not graded on a curve
Accuracy only means something relative to a strong baseline. The refreshed report will compare KickOracle against strong public references and market-derived pricing where reliable data is available. The aim is honest comparison: matching a market baseline on calibration is meaningful; claiming certainty from a small tournament sample is not.
What this does not prove
A backtest is not a guarantee about 2026. High-confidence buckets can have small samples, methodology changes must be separated from evaluation data, and a single tournament carries real variance. We publish limitations alongside the numbers because a forecasting model is only as credible as the caveats it is willing to state.
Frequently asked questions
How accurate are KickOracle’s World Cup predictions?
The public accuracy report is being refreshed with a walk-forward evaluation. Until that report is published, KickOracle describes the metrics and limitations instead of presenting retired placeholder numbers.
What is a Brier score and why does it matter?
A Brier score is the mean squared error between predicted probabilities and the actual home/draw/away outcome. Lower is better, and it rewards probabilities that are both confident and correct — a fairer accuracy measure than win/loss alone.
How does the model compare to FiveThirtyEight and bookmakers?
The refreshed report will compare KickOracle with strong public baselines and market-derived pricing where reliable data is available. Those comparisons will be presented as evidence, not as guarantees.
Can you predict the World Cup 2026 with certainty?
No. Football is high-variance and single-elimination knockouts amplify it. The model outputs calibrated probabilities, not certainties — useful for weighing likelihoods, not for guaranteeing a result.
What are the limitations of this backtest?
Tournament samples are small, high-confidence buckets can be thin, and evaluation data must be separated from tuning decisions. Treat any single backtest as directional evidence rather than a guarantee.
See the live model behind the backtest with Scout Pass
The accuracy page shows how model quality is measured; Scout Pass shows the current model working in real time — calibrated probabilities that update as squads, form and results change.
- Live, calibrated win probability on every fixture
- Tactical and squad-change alerts as they land
- Full player intel and scout reports behind each number
- Export probabilities and results to CSV