How the simulator did at the 2026 World Cup

The favourites: the right four, no clear winner

Before the tournament the simulator could hardly separate three teams. France won it in 17.0% of 50,000 simulated tournaments, Argentina in 16.9% and Spain in 16.7%, with England next at 11.1% and nobody else above 4.8%. Those four are exactly the four that reached the semi-finals.

The final was Spain against Argentina, and the simulator rated it a coin toss: 49.8% for Spain, who won 1–0 (aet).

The group stage

The single likeliest result came up in 43 of the 72 group matches. Its own probabilities pointed to about 46, and the shortfall is the draws, below. Its 32 likeliest teams to get out of the groups took 25 of the 32 places in the round of 32.

The seven that missed out: Uruguay (91.5%), IR Iran (87.2%), Korea Republic (81.7%), Türkiye (74.3%), Panama (61.6%), Czechia (59.9%) and Scotland (59.9%). The figure is the share of simulated tournaments in which each got out of its group. In their place went Ghana (13.7%), Cabo Verde (24.3%), South Africa (35.4%), Bosnia and Herzegovina (35.6%), Congo DR (41.7%), Paraguay (44.2%) and Sweden (48.1%).

Uruguay were the biggest miss of the tournament: through in 91.5% of simulations, out in the real one.

Too few draws

The scoring was right: 2.99 goals a game in the group stage, against an expected 3.02. The draws were not. There were 20 in 72 group matches where the engine expected about 14, and in no match at all did it make a draw the likeliest result. Spain 0–0 Cabo Verde (9.4%) and England 0–0 Ghana (9.9%) were the least expected.

The cause is in how the engine scores a match. Like most simple football models, it draws each side's goals independently, and that produces too few level scores. When the 2030 engine was tuned against real international results, the same gap showed up, so it now turns 7% of one-goal games into draws: the leader's last goal never comes. Knock-out deciders are left alone, because there a draw means penalties.

The knock-outs

Here the favourite went through in 26 of 31 ties, more than the 22 its probabilities implied, so if anything it was too cautious. The two biggest upsets: Paraguay knocked out Germany on penalties in the round of 32 after a 1–1 draw (a 21.6% chance beforehand), and Norway beat Brazil 2–1 in the round of 16 (23.5% beforehand).

Norway went furthest against the odds: before the tournament the simulator had them reaching the quarter-finals in 7.1% of tournaments.

Home advantage

The 2026 engine gave no team any home advantage. All three hosts reached the round of 16 anyway, though beforehand it had given them only these chances of getting that far: Mexico (62.7%), USA (55.4%) and Canada (38.6%). Three teams prove nothing on their own, but they agree with the results data: the 2030 engine gives a side playing in its own country a 50-point boost to its rating, which also covers home qualifiers.

The least expected results

How this was measured

Every number here comes from the 2026 edition of the simulator as it stood on 5 June 2026: its ratings, taken from the FIFA ranking of June 2026, its match engine and its settings, none of which has changed since. For each of the 104 matches we took the probabilities the engine gives that pairing, computed exactly from its expected goals; for the tournament-wide figures we played the 2026 World Cup 50,000 times from the real draw. The results are the ones the 2026 edition shows, recorded from ESPN when the tournament ended on 19 July. That edition is still playable at wc26.org.

The 2030 engine was retuned afterwards on data that includes these matches, so running it on the same games would not be a fair test, and we have not done so. Its predictions for 2030 are on the predictions page, every nation's 2026 matches and its record are on its team page, and how it works explains the engine in full.