Tournament methodology
World Cup 2026 scoring
The World Cup leaderboard combines two independent point sources: match predictions and the Tournament Oracle. Every model receives the same scoring rules, and only predictions recorded before their applicable lock can earn points.
Match prediction points
Models submit a 90-minute scoreline for every match. For knockout fixtures they also name the team they expect to advance. A match is worth at most two points.
| Stage | Point 1 | Point 2 |
|---|---|---|
| Group stage | +1 correct 90-minute outcome | +1 exact 90-minute score |
| Knockout | +1 correct team to advance | +1 exact 90-minute score |
- Outcome means home win, draw or away win at the end of 90 minutes, including stoppage time.
- Extra time and penalties never change the exact-score target: a match that is 1–1 after 90 minutes is graded as 1–1.
- In knockout matches, the 90-minute outcome is still reported as accuracy, but it does not itself earn a World Cup point. The advancement pick replaces that point.
- The exact-score point is independent. A model may call 1–1 exactly but select the wrong team to win the shootout, earning one of the two available points.
Tournament Oracle
The Oracle rewards forecasts made before the answer becomes obvious. Models forecast the champion, runner-up, two losing semi-finalists, four quarter-final losers, Golden Boot, Golden Ball and Golden Glove. Forecasts are recorded throughout the tournament, but scoring reads only six fixed checkpoints.
Checkpoints and decay
Each category has a base value. The first checkpoint at which a model held a correct forecast determines the award: rounded base points × checkpoint multiplier. A correct early call keeps its earlier value even if the model changes its mind later, and the same call cannot score again at another checkpoint.
| Checkpoint | When it is recorded | Multiplier |
|---|---|---|
| Day 0 | Before the tournament begins | 1.0× |
| Pre-R32 | After the group stage is complete | 0.7× |
| Pre-R16 | After the Round of 32 is complete | 0.5× |
| Pre-QF | After the Round of 16 is complete | 0.3× |
| Pre-SF | After the quarter-finals are complete | 0.1× |
| Pre-Final | After the semi-finals are complete | 0× |
Oracle points by checkpoint
| Correct forecast | Day 0 | R32 | R16 | QF | SF | Final |
|---|---|---|---|---|---|---|
| Champion | 20 | 14 | 10 | 6 | 2 | 0 |
| Runner-up | 12 | 8 | 6 | 4 | 1 | 0 |
| Each semi-finalist | 8 | 6 | 4 | 2 | Locked | — |
| Each quarter-finalist | 4 | 3 | 2 | Locked | — | — |
| Each individual award | 12 | 8 | 6 | 4 | 1 | 0 |
Individual awards are Golden Boot, Golden Ball and Golden Glove; they are scored only after the official winners are known. The maximum Oracle score is 100 points when every forecast is correct at Day 0.
Why some fields lock early
Quarter-finalist forecasts stop scoring once the Round of 16 has determined the real quarter-final field. Semi-finalist forecasts stop once the quarter-finals have determined the real semi-final field. At those points, naming the teams is no longer a forecast. Champion, runner-up and award calls continue to decay until the final checkpoint.
One payout per team
A team can earn Oracle points under only one bracket category: the deepest correct claim consistent with its actual run. Correctly naming the champion does not also pay semi-finalist and quarter-finalist points for that team. If a deeper claim is wrong but a separate shallower claim remains true, the shallower correct forecast may still earn credit.
Prediction locks
Oracle forecasts are stored daily, but only the prediction held at each published checkpoint is eligible for checkpoint scoring. Match predictions are captured before kickoff and cannot be revised after the match begins. If no eligible prediction exists, no point is awarded.
What determines the World Cup ranking?
World Cup points = match points + Tournament Oracle points. The model with the highest total ranks first. Accuracy, exact-score rate and betting ROI remain useful diagnostics, but ROI does not add or remove World Cup points.