This started as a quarterfinal preview built on 123 matches of StatsBomb event data, 964 historical World Cup matches, and a Poisson model — published before a ball was kicked, previews preserved below, receipts and all. Since then: six knockout matches, six pre-registered forecasts, a model scoreboard with a winner, and the ending the simulation called its single most likely — Argentina vs Spain, tomorrow, 3pm ET at MetLife. The full breakdown of the Final is below. Predictions are for fun; the numbers are real.
Every shot gets a probability of scoring based on where and how it was taken. Add them up and you get xG — a measure of the chances a team created, stripped of luck. A team that wins 2–0 on 3.04–0.14 xG (France, QF1) earned every bit of it.
Six forecasters live on this page: the launch Poisson blend (Model A) plus a five-model ensemble — dynamic state-space, tournament form, Elo, and a club-form prior, weighted by measured backtest skill. Every forecast was published before its match. The scoreboard is in the Forecast Lab.
In every chart, the first-named team is blue and the second is orange — always, in every match. Gold marks the model's verdicts. Hover anything that looks like data.
Lift the trophy (incl. extra time & penalties): Spain 59.3% · Argentina 40.7% — but the ensemble is hiding a genuine argument. Four of the six models sit within five points of a coin flip; only tournament form screams Spain (79/21). The 28.8% draw probability is the highest of the entire knockout stage — extra time is live, and so is Dibu Martínez. Full model fight in the Forecast Lab.
Spain have played seven matches at this World Cup and conceded one goal — De Ketelaere's, in the quarterfinal — with six clean sheets, the first team ever to keep six at a single World Cup. They are unbeaten in 37 straight matches, have won 8 consecutive knockout games at major tournaments, and just held the French front four that terrorized everyone to 0.31 xG in a semifinal that was never close. The reigning European champions have been the best team in this tournament from the opening week, and it isn't particularly arguable.
Argentina's counter-argument is not an argument; it's a fact pattern. A tournament-high 19 goals, 12 of them after the 75th minute. Two knockout wins secured by goals in second-half stoppage time — no team had ever done that twice in one World Cup. Fourteen consecutive wins, 19 straight major-tournament matches unbeaten, and a 39-year-old captain with 8 goals and 4 assists — 33 career World Cup goal contributions, the all-time record — playing, by his own account, the last World Cup match of his life. They have trailed in the 75th minute or later in three knockout games this month. They won all three.
Tactically it's the cleanest contrast a final has offered in years: control against chaos. Both teams are expected at full strength in mirrored 4–3–3s — Rodri, Fabián Ruiz and Olmo trying to strangle the midfield Argentina likes to flood; Oyarzabal ahead of Yamal and Baena against the tournament's stingiest back line by xG; Messi central between Álvarez and Simeone, hunting the one transition Spain's press concedes per half. The stakes ledger, for history: Argentina are one win from a fourth consecutive major title (no nation has done it); Spain are one win from holding the men's and women's World Cups simultaneously (no nation has done that either). Something unprecedented happens at MetLife regardless.
xG rows from live match data plus reported figures (available for 6 of Argentina's 7 matches, 4 of Spain's 7). Elo recomputed from the full results history through the semifinals — Spain took the #1 spot from Argentina in Arlington; nine points now separate the two best-rated teams in the world.
Meanwhile, today in Miami Gardens: the bronze final, France vs England, 5pm ET. The ensemble says France 60.6/39.4 (form 72.2, the filter 52.6), likeliest score 1–1. The stakes are pride, seeding — and one last Golden Boot audition for Mbappé, Kane and Bellingham before Messi gets the final word on Sunday.
Advance (incl. extra time & penalties): France 54.7% · Morocco 45.3% — Model A's closest quarterfinal, much tighter than the pundit consensus. Ensemble v2: France 65.1% — but see the Forecast Lab: the model with the longest memory calls this a coin flip.
Fourteen years of history compressed into one fixture: the 2022 semifinal, replayed in Foxborough — except Morocco is no longer the team that defended its way to fourth place. At Qatar 2022 they held 38.8% possession and threw bodies in front of everything. At AFCON 2023 that number jumped to 54.9%. This tournament it's 60.4% — functionally identical to France's 60.5%. Walid Regragui's successors didn't just keep the wall; they learned to hold the ball. This is a genuine stylistic mirror match now, and the event data saw the transformation coming three years ago.
France, meanwhile, are the tournament's steamroller: 14 goals, a tournament-best 17.8 shots per game, 10.7 xG, and a front four — Mbappé, Dembélé, Olise, Barcola — that Norway's coach flatly called "the best in the competition by far." The one crack: Aurélien Tchouaméni's adductor is a game-time question, and Morocco's biggest blow is bigger — breakout star Ismael Saibari (3 goals, new €50m Bayern signing) is out with a hamstring injury. Morocco's false-nine system was built around him.
"Last major" = Euro 2024 (France) / AFCON 2023 (Morocco), computed from StatsBomb event data. Morocco's AFCON attack numbers came against weaker opposition — one reason the model is friendlier to them than the market.
Advance: Spain 71.0% · Belgium 29.0% — the model's biggest favorite of the round. Ensemble v2: Spain 71.4% — five models, one verdict; the most robust call of the weekend.
Spain have not conceded a goal at this World Cup. Not one. Unai Simón's shutout streak stands at 609 minutes — a men's World Cup record, past the 517 of Italy 2006. And it's not keeper heroics: this is the same suffocation machine our Euro 2024 event data measured winning that tournament with a PPDA of 10.9 (passes allowed per defensive action — lower = more aggressive) and the highest press line in the field (65.7). They smother you where you build.
Belgium are the opposite: gloriously, chaotically alive. Down 2–0 to Senegal with 20 minutes left — won 3–2 in the 125th minute. Benched De Bruyne and Doku in the round of 16 and hammered the USA 4–1 anyway, with Charles De Ketelaere (2G+1A) looking like the heir the golden generation never got. But the ledger is grim: Amadou Onana's ACL rupture removes their only true midfield destroyer, and Spain's whole game is midfield asphyxiation. The last time these teams met in a World Cup knockout — 1986 — Belgium won on penalties. Belgium have never lost to Spain at a World Cup. That's the thread they're hanging by.
Euro 2024 rows computed from StatsBomb event data — Spain played 7 matches (won the title), Belgium 4 (out in the R16).
Advance: England 63.7% · Norway 36.3% — and the highest combined expected goals (3.02) of the round. This is the one to watch with friends. Ensemble v2: England 66.8% — but Norway carries the widest uncertainty band in the field, and their Elo rating (2003) is the highest in Norwegian history.
Norway's men had played one knockout match in their entire World Cup history — 1938, a 2–1 extra-time loss to Italy — before this summer. Now they've won two in twelve days, the second of them eliminating Brazil, with Erling Haaland scoring in the 79th and 90th minutes. It is Norway's first men's quarterfinal ever, their first World Cup at all since 1998 (a tournament where, incidentally, they also beat Brazil — the men are still unbeaten against them). Haaland has 7 goals on 3.7 xG: the most violent finishing overperformance in the field, from a man who does this every week in the Premier League.
England under Thomas Tuchel are the round's stylistic inverse: 65% possession, 15.6 shots a game, and the knockout resilience they historically lacked — they beat Mexico at the Azteca, with ten men, on Bellingham's two goals in 98 seconds. But the back line is patched: Quansah is suspended, Reece James is a hamstring doubt, Saka's Achilles is being managed. And Norway put at least one goal past everyone, including France. Both teams have scored and conceded in every Norway match this tournament. The nil-nil is nearly off the board (4.9%).
Norway is the one quarterfinalist with no StatsBomb event history (no major since 1998 — the data gap is the story). Their rows are 2026-tournament only; England's Euro 2024 profile from our data: 0.94 xG / 0.87 xGA per match under Southgate — Tuchel's version is a different animal.
Advance: Argentina 63.3% · Switzerland 36.7% — with the round's second-highest draw probability. Extra time feels live here. Ensemble v2: Argentina 66.7% — though the tournament-form model still calls it a coin flip (48.0%); Switzerland remain the round's live upset.
Lionel Messi is 39 years old, playing his sixth World Cup, and leading the Golden Boot with 8 goals. Against Egypt he missed a penalty, watched his team fall 2–0 behind, then delivered the cross for Romero, scored the equalizer himself, and watched Enzo Fernández win it in stoppage time — the first team in World Cup history to trail by two in the 75th minute of a knockout and win in regulation. He has now scored in six consecutive World Cup knockout matches, which no one had ever done. Every match may be his last. He keeps refusing to let it be.
Switzerland's counter-argument is 132 years of patience cashing in at once: first knockout win in 88 years (2–0 over Algeria), then their first shootout survival (4–3 over Colombia, Kobel's save the difference), and a defense that has not trailed for a single minute all tournament. Xhaka, at his fourth World Cup, is now Switzerland's most-capped player ever. The blueprint is obvious — 0–0 as deep as possible; they generated 0.35 xG against Colombia and celebrated like it was a plan, because it was. One problem: penalties mean Dibu Martínez, and nobody on earth wants penalties against Dibu Martínez.
"Last major" = Copa América 2024 (Argentina won it, conceding once in six games) / Euro 2024 (Switzerland, who were 3 penalty kicks from a semifinal). StatsBomb event data. Switzerland's reported 1.80 xG/match ’26 runs hot against their 0.35 vs Colombia — treat with salt.
Nobody was brave here. All six forecasters put the match within a whisker of even — the ensemble said Spain 54.7%, the widest any model went was tournament form's 62.1%, and the club prior actually leaned France (52.4%). Spain's win was the sixth straight match in which the eventual winner was the model favorite — but this was the round where confidence went to die: a six-model average of ~55% for a match that finished as a two-goal suppression job.
The result did violence to the title race: France had been the simulation's championship favorite (32.4%) since the night they dismantled Morocco. Ninety minutes later they were playing for bronze, and the crown passed back to the team the launch model had favored on day one.
Spain scored with both shots on target. France–Spain was the first semifinal since 2010 in which the losing side failed to reach 0.4 xG. Mbappé stays on 8 goals — the Golden Boot holds its breath for the bronze final.
Unanimous, and mildly confident: Argentina 59.5% from the ensemble, with Elo boldest at 63.8% and the club prior — still blind to Messi's Inter Miami form — barely over the line at 50.8%. It's the fourth time in six matches that Elo was the most aggressive forecaster on the eventual winner, a pattern that decides the model scoreboard below.
The deeper pattern is Argentina's: across six knockout rounds of this and the last World Cup, they have now trailed in the second half four times and lost none of them. The dynamic filter, which watches a decade at a time, has quietly raised Argentina's latent attack state to its highest value since data begins in 2012 — late goals are goals too, and 12 of their 19 have come after the 75th minute.
Messi's two assists take him to 8G + 4A — 33 career World Cup goal contributions, the all-time record. England to the bronze final; Argentina to a date with the only defense of the tournament better than their own.
Running every remaining path through the launch model, as published July 9. Spain's zero-conceded defense makes them the quiet favorite; Argentina's trophy-hardened floor keeps them close; and the France–Morocco winner walks into the bracket's softest semifinal slot. (The ensemble built afterward disagrees in interesting ways — keep scrolling.)
Most likely final: Spain vs Argentina (≈17%) — the Euro champion against the World champion, Rodri against Messi, and a rematch of the "Finalissima" lineage. The model's dream scenario is also its median one, which almost never happens. Except this time it did. Three of four quarterfinal favorites advanced (Norway's fairy tale fell in extra time), both semifinal favorites advanced, and the one-in-six median final is the one being played at MetLife tomorrow. The launch model's title podium — Spain, then Argentina — is now, literally, the matchup.
Still a dead heat — and now a scheduling duel. Neither scored in the semifinals (Spain's defense saw to Mbappé; Switzerland ended Messi's nine-game scoring streak the round before). Mbappé gets his last audition today against England's leaky back line in the bronze final; Messi answers tomorrow against the best defense of the century. Projections below use each team's remaining fixture at ensemble scoring rates. If they finish level, the tiebreaker is assists — where Messi's four loom large.
A week ago, two players projected to break Gerd Müller's post-1970 barrier of 10. Then the knockout rounds remembered what defenses are for: the top four scorers have combined for exactly one goal since the quarterfinals began (Bellingham's brace aside). Müller's 10 and Fontaine's 13 sleep safely for another cycle — but an 8-goal, 4-assist tournament at age 39 may be the more permanent kind of record.
The launch-day model (call it Model A) was one equation. This is a real ensemble — four more forecasters, each imported from a field that isn't soccer, combined the way meteorologists combine weather models: weighted by measured skill on past tournaments, not by which one sounds cleverest. Every knockout forecast on this page was published before its match kicked off; the scoreboard below settles who was right. Version 2 data: the full international results history (every match since 1872), real per-match xG for the 2026 games, and a classic Elo forecaster.
60% 2026 form, 25% last major, 15% WC 2022. Simple, transparent, no opponent adjustment — its known flaw. Frozen on July 9 and never updated: the control group.
A Kalman-filtered Poisson (a dynamic GLM, West & Harrison style). Team strength is a latent state that drifts through time; every international since 2012 updates it, friendlies at half gain. It has watched Morocco's entire decade-long rise — and it's the one model that called France–Morocco a coin flip. It regrets that now.
This World Cup only, on xG-blended pseudo-goals (60% xG, 40% goals) from the live match data — the "naive seasonal" benchmark every forecasting textbook warns is embarrassingly hard to beat. Won three of the six knockout rounds on the scoreboard.
Arpad Elo's chess system, computed fresh from the results history with football conventions (K by competition, goal-diff multiplier). Current field, post-semifinals: Spain 2259, Argentina 2250 — #1 and #2 in the world, nine points apart — then France 2156, England 2108. It backtested best (80% accuracy), and it is winning the live scoreboard too.
Like satellite parking-lot data for GDP: 1,800 club matches from the 2025/26 big-five leagues rate each squad's players' clubs, then tilt the forecast. Still can't see Messi (Inter Miami isn't big-five data) — and it has paid for that blindness on the scoreboard.
| Model | Brier score ↓ | Log loss ↓ | Picked the winner |
|---|---|---|---|
| Elo (1960s chess math) | 0.165 | 0.507 | 80% |
| Tournament form (xG) | 0.165 | 0.502 | 73% |
| Dynamic state-space | 0.182 | 0.546 | 70% |
| Coin flip | 0.250 | 0.693 | 50% |
Every model improved with the richer data (the dynamic filter most: 0.214 → 0.182), but the podium belongs to the two simplest ideas in the room — "how are they playing this month" and "a rating system invented for chess in 1960." Weights follow the evidence: 30% form, 30% Elo, 28% dynamic, 12% club prior. Nobody's model is magic; everybody's model beats a coin.
The widest model disagreement of the round — a 33-point spread — and the verdict was emphatic: tournament form was right (2–0, xG 3.04–0.14; Morocco's worst attacking output of the tournament by a mile). The dynamic filter's romance with Morocco's decade of data was the Lab's first public faceplant.
Five independent methods, one tight cluster, and the outcome the cluster predicted: Spain led the xG 1.96–0.34 and never trailed. Form (79.5%) was closest for the second match running. This is what a robust forecast looks like.
Every model said England; England needed a 93rd-minute Bellingham goal off a keeper error to prove them right. The club prior — which could see that Norway's squad plays for smaller clubs than its Elo suggests — posted the round's single best Brier (.067). The one-in-three fairy tale died at one-in-three odds, which is exactly how probability is supposed to feel.
The genuine model fight of the round, and both sides got to feel right: it was 1–1 in the 90th minute (form's case) before Argentina's extra-time class and Embolo's red card settled it (Elo's case, at Elo's confidence). The long-memory models won the argument — a pattern to remember for tomorrow, because they're the ones calling the Final even.
The whole Lab huddled within eight points of a coin flip for a match Spain controlled from the 22nd minute. Only the club prior picked France — its second miss, and its costliest. The lesson the ensemble can't learn mid-tournament: Spain's suppression doesn't show up in win probability until you watch it delete a 2.14-xG-per-game attack in real time.
All six models picked Argentina; England led from the 55th to the 85th anyway, which is the difference between a probability and a promise. Elo, boldest at 63.8%, banked its fourth scoreboard win. The xG (1.84–0.53 Argentina) says the models underrated the champions if anything.
Look at the shape of it: four models pinned to the coin-flip line, and form off alone at 79% — the exact inverse of QF1, where form was the lone wolf and was right. The scoreboard's two best models (Elo, form) disagree by 28 points, the largest gap they've had all tournament. Elo says these are the two strongest teams it has ever rated at a World Cup final; form says Spain's month has simply been better than everyone's, Argentina included. The ensemble splits the difference at 59.3 — and the dynamic filter's credible interval (23–77%) is the Lab's way of admitting that a one-off final between these two is barely a forecastable event.
| Match | Elo | Form | Ensemble | Model A | Dynamic | Club |
|---|---|---|---|---|---|---|
| QF1 · France 2–0 Morocco | .109 | .039 | .121 | .205 | .282 | .238 |
| QF2 · Spain 2–1 Belgium | .063 | .042 | .082 | .084 | .129 | .141 |
| QF3 · England 2–1 Norway aet | .124 | .102 | .110 | .132 | .129 | .067 |
| QF4 · Argentina 3–1 Switzerland aet | .051 | .270 | .111 | .135 | .086 | .075 |
| SF1 · Spain 2–0 France | .225 | .144 | .205 | .170 | .198 | .274 |
| SF2 · Argentina 2–1 England | .131 | .174 | .164 | .167 | .161 | .242 |
| Mean (6 matches) | .117 | .129 | .132 | .149 | .164 | .173 |
Form wins the battles; Elo wins the war. Tournament form took three of the six rounds outright but torched itself on the Switzerland coin-flip call; Elo never finished worse than fourth in any match and picked all six winners — as did, remarkably, Model A, the "flawed" launch blend, which beat two of its fancier successors. The ensemble did exactly what an ensemble should: never the best, never the worst, second overall. The backtest's forecast about the forecasters (Elo and form on the podium) has, itself, verified. One match left to settle it — and Elo's lead is larger than any single match can erase, so the Lab has a champion regardless of Sunday: the 1960s chess math.
Gray bar = Model A on July 9, before a quarterfinal was kicked. Gold = the ensemble today, with every result folded into every model. Six teams' gold bars are now zero, which is the least forgiving chart this page has ever displayed. France's story is the cruelest: their QF1 demolition of Morocco made them the title favorite (32.4% after the post-QF1 rerun) — for exactly five days.
The two teams the launch model ranked #1 and #2 on July 9 are the two teams playing on July 19, in the final it called most likely. That's either a modest toy model having a very good month, or the rare tournament where the chalk held. (It's both. It's mostly the second one.) The engine lives in footycast/ — one command re-ran everything for tonight's update.
Local data. Team and player profiles come from StatsBomb open event data: 123 matches involving these eight teams across WC 2018, WC 2022, Euro 2020, Euro 2024, Copa América 2024, and AFCON 2023 — every pass, shot, pressure, and carry (~3,400 events per match). Historical head-to-heads and records come from the Fjelstul World Cup Database (all 964 men's matches, 1930–2022). 2026 tournament results, form, and injuries were compiled July 9, 2026 from public reporting; knockout results, xG, and Final team news were folded in on July 18 (sources below).
Model A (launch). Each team's attack/defense rating blends 2026 per-match xG and xGA (60%), their most recent major tournament from event data (25%), and WC 2022 (15%). Ratings produce Poisson goal expectancies per match; draws in the knockout resolve by a regulation-strength-weighted coin flip standing in for extra time and penalties. The bracket runs 100,000 Monte Carlo iterations. Frozen at launch, deliberately: it serves as the control for the fancier models — and, six matches in, it has picked every winner.
The ensemble (Forecast Lab, v2). Four further models: a dynamic Poisson state-space filter (extended-Kalman updates on latent attack/defense states across every men's international since 2012 — 49,505-match results history, friendlies at half gain, time-gap-scaled process noise); a tournament-form Poisson on xG-blended pseudo-goals (60% xG / 40% goals, per-match xG from the live 2026 dataset, validated against known results); classic football Elo computed from scratch (K by competition, goal-difference multiplier, no snapshot leakage); and a club-form prior fit on 1,800 matches of 2025/26 big-five-league results mapped through each squad's core players. Weights come from a genuine out-of-sample backtest — all models forecast the WC 2022 and Euro 2024 knockouts using only pre-knockout data (Brier: Elo 0.165, form 0.165, dynamic 0.182, coin flip 0.250, n=30).
The scoreboard (added July 18). Every finished knockout match is scored on forecasts refit from data strictly before its date — the dynamic filter, form model, and Elo are rolled back match by match, so no forecast ever sees its own future (results on the same calendar day are excluded both ways; the data has day, not hour, granularity). Knockout results after the datasets' snapshots enter through a manual results table with reported xG, which also now feeds the Elo updates. Brier = (forecast − outcome)² on the advance probability; extra-time and penalty wins count the same as regulation ones, as the forecasts intended.
Honest caveats. This is a toy model built in an afternoon and updated in an evening, not FiveThirtyEight. It doesn't adjust for opponent strength within tournaments, doesn't regress hot finishing, doesn't model red cards (Embolo's changed QF4), game states (Argentina's entire personality), or the difference between a final and any other match. The final forecast treats Sunday as a neutral fixture between two rating vectors; it is also Messi's last World Cup match, in front of 82,500 people, for the largest prize in the sport, and no Poisson process knows what that does to anyone. Where the model and the market disagree, the market is probably right and the model is more fun. Not gambling advice; the author's stake is joy.
Data © their respective owners · StatsBomb open data used per its license · Fjelstul World Cup Database © Joshua C. Fjelstul, CC-BY-SA 4.0 · Built July 9, 2026, before QF1 kickoff · Updated July 18, the eve of the Final. Whatever happens at MetLife: the model regrets nothing.