Research Report · EAFC26 Wave 4 · Live Tournament

Predicting the 2026 World Cup

pointing a club-football simulation engine at a live 48-team tournament, and grading it against reality

No new model. The Wave-3 attribute engine that priced players and simulated club leagues is handed national squads, stripped of home-field bias, calibrated to tournament scoring, and asked the only question that matters in June: who wins.

June 2026·~18 min read·Pok Yeung Lee
World Cup 2026Dixon-ColesEA FC 26Monte CarloBacktest

Abstract

This is Wave 4 of the EAFC26 series: a live-tournament application of the Wave-3 simulation engine. National starting XIs are reconstructed from EA Sports FC 26 attributes, the club-league home-field term is zeroed for neutral venues, and the goal rate is calibrated down 10% to match the modern World Cup average of 2.52 goals a game. Each match is the average of 10 simulations to suppress single-run noise. On the 12 group games already played the engine called 5/12 outcomes and the exact scoreline of 3, with probabilistic skill +13% over a no-information prior; its only systematic miss is the draw, an artifact of ranking outcome classes rather than scorelines. Projected forward across all 72 group games and a 3,000-run Monte Carlo of the 48-team knockout bracket, the model's most likely champion is Argentina at 17.07%, inside a tight four-way cluster with England, France and Brazil.

LIVE READOUTpaper wallet + model-vs-market, fetched live from Polymarket each time this page loads
Loading live Polymarket data…
Paper wallet (no real money). Bets placed 2026-06-15 under a disciplined filter (clean EA-covered teams only, 35% edge haircut, quarter-Kelly). This panel is read-only and recomputes from live prices on every load; an hourly cloud agent tracks the same wallet independently.

01A second stream of prediction

The EAFC26 project began by turning EA's player attributes into real-world market values, then built a match-simulation engine validated on the top-five club leagues. That engine never saw international football. The 2026 World Cup is the cleanest possible out-of-sample test: a brand-new competition, neutral venues, national teams the model was never tuned on, and results arriving in real time to grade against.

So this wave asks two things. First, when the engine is pointed at games that have already been played, how close is it? Second, run forward, what does it predict for the rest of the group stage, the bracket, and the trophy?

The discipline here is that the answer to the first question is checkable today, and the answer to the second becomes checkable over the next month.

02Method: a club engine, re-pointed

Nothing about the core engine changed. It still consumes a starting XI as a vector of EA attributes, derives an attack and a defence strength for each side, and samples a scoreline from a Dixon-Coles bivariate-Poisson model. Three adaptations make it tournament-appropriate:

BRIDGE
Squads to attributes

Every player in each national squad is matched to his EA FC 26 record (184 of 264 backtest XI slots matched exactly, the rest by surname-anchored and diaspora matching). Per-90 form is synthesised from attributes with the same Wave-3 coefficient bank, so a national XI is just another XI.

VENUE
No home edge

The club model's home-field term (about +15% scoring) is zeroed for neutral World Cup grounds, with a half-strength bonus kept only for the three host nations playing at home.

CALIBRATION
Tournament scoring

Club leagues run hotter than tournament football. Expected goals are scaled by 0.90 so the engine averages 2.52 goals a game, matching the modern World Cup exactly.

STABILITY
Ten-run average

Each fixture is simulated ten times and averaged, so a single freak scoreline never drives a standing or a bracket slot. The knockout ties resolve draws by a strength-weighted coin flip, the model's proxy for extra time and penalties.

The calibration was not guessed. It was set so the engine's aggregate output matches 96 years of World Cup scoring, derived in section 7.

03Backtest: the engine against games already played

Twelve group games had been played at the time of writing (Groups A through F). Each was re-simulated using the actual starting XIs both teams fielded, then compared to the real result. The probability triple is the engine's pre-game home / draw / away split.

Predicted vs actual, 12 played group games. Green = correct outcome class; a tick in the Modal column means the single most-likely scoreline was exactly right.
DateGrpHomeAwayH/D/APickModalFinal
06-11A MexicoSouth Africa 65%/23%/12% Mexico 65% 1-0 2-0
06-11A South KoreaCzechia 34%/26%/40% Czechia 40% 1-1 2-1
06-12B CanadaBosnia and Herzegovina 46%/26%/29% Canada 46% 1-1 ✓1-1
06-12D United StatesParaguay 45%/27%/28% United States 45% 1-1 4-1
06-13B QatarSwitzerland 21%/26%/52% Switzerland 52% 1-1 ✓1-1
06-13C BrazilMorocco 52%/28%/20% Brazil 52% 1-0 1-1
06-13C HaitiScotland 24%/24%/52% Scotland 52% 1-1 0-1
06-14D AustraliaTurkey 17%/26%/57% Turkey 57% 0-1 2-0
06-14E GermanyCuraçao 74%/17%/8% Germany 74% 2-0 7-1
06-14E Ivory CoastEcuador 31%/28%/41% Ecuador 41% 1-1 1-0
06-14F NetherlandsJapan 39%/28%/32% Netherlands 39% 1-1 2-2
06-14F SwedenTunisia 51%/26%/22% Sweden 51% 1-1 5-1
OUTCOME
5 of 12 correct

Random is 4 of 12. On the eight games that produced a winner, the engine went 5 of 8: it reads talent gaps well (Mexico, USA, Germany 7-1, Scotland, Sweden all called).

SCORELINE
3 exact, RPS +13%

The single most-likely scoreline was exact in 3 of 12, and probabilistic skill beat a no-information prior by 13%. Log-loss 1.0107 matches the engine's league calibration.

THE DRAW IS A SCORECARD ARTIFACT, NOT A MODEL ERROR

Five of the twelve games were draws, and the engine never made a draw its single top pick, so the outcome scorecard marks them wrong. But in Canada 1-1 Bosnia and Qatar 1-1 Switzerland the engine's single most-likely scoreline was 1-1. The reason is structural: win-probability is fragmented across many scorelines (1-0, 2-0, 2-1...) while a draw concentrates on 1-1, so a 1-1 can be the likeliest exact score even when "a win" is the likeliest class. Checked against history, the engine's mean draw probability of 24.0% almost exactly matches the modern World Cup draw rate of 23.6%. The draw probabilities are well calibrated; only the argmax display is harsh.

04Group-stage projection

Forward from here, the remaining 60 group games are simulated (10-run average), combined with the 12 real results, to produce final standings. Top two in each group plus the eight best third-placed teams advance to a 32-team knockout. Teams EA under-covers (Liga MX, the Qatari and several domestic leagues) are marked with a star; read those rows with more caution.

Group A
TeamPtsW-D-LGF:GA
Mexico7 2-1-0 5:2W
South Korea6 2-0-1 4:3RU
Czech Republic4 1-1-1 4:43rd
South Africa 0 0-0-3 1:5
Group B
TeamPtsW-D-LGF:GA
Switzerland7 2-1-0 5:3W
Canada4 1-1-1 4:3RU
Bosnia and Herzegovina4 1-1-1 4:43rd
Qatar 1 0-1-2 2:5
Group C
TeamPtsW-D-LGF:GA
Brazil7 2-1-0 5:3W
Scotland4 1-1-1 3:3RU
Morocco3 0-3-0 3:33rd
Haiti1 0-1-2 2:4
Group D
TeamPtsW-D-LGF:GA
United States7 2-1-0 7:3W
Turkey4 1-1-1 3:4RU
Australia3 1-0-2 3:33rd
Paraguay3 1-0-2 3:6
Group E
TeamPtsW-D-LGF:GA
Germany7 2-1-0 10:3W
Ivory Coast7 2-1-0 4:2RU
Ecuador3 1-0-2 3:43rd
Curaçao0 0-0-3 3:11
Group F
TeamPtsW-D-LGF:GA
Sweden5 1-2-0 7:3W
Japan5 1-2-0 5:3RU
Netherlands5 1-2-0 5:43rd
Tunisia0 0-0-3 2:9
Group G
TeamPtsW-D-LGF:GA
Egypt 7 2-1-0 6:4W
Belgium5 1-2-0 5:4RU
Iran 2 0-2-1 3:43rd
New Zealand1 0-1-2 3:5
Group H
TeamPtsW-D-LGF:GA
Spain7 2-1-0 5:3W
Uruguay5 1-2-0 4:2RU
Saudi Arabia2 0-2-1 3:43rd
Cape Verde1 0-1-2 2:5
Group I
TeamPtsW-D-LGF:GA
France9 3-0-0 7:2W
Norway4 1-1-1 4:5RU
Senegal3 1-0-2 5:53rd
Iraq1 0-1-2 2:6
Group J
TeamPtsW-D-LGF:GA
Argentina7 2-1-0 6:4W
Algeria5 1-2-0 5:4RU
Jordan 2 0-2-1 4:53rd
Austria1 0-1-2 2:4
Group K
TeamPtsW-D-LGF:GA
Portugal9 3-0-0 6:2W
DR Congo2 0-2-1 3:4RU
Uzbekistan 2 0-2-1 3:43rd
Colombia2 0-2-1 2:4
Group L
TeamPtsW-D-LGF:GA
England9 3-0-0 6:3W
Croatia6 2-0-1 6:3RU
Ghana3 1-0-2 4:53rd
Panama 0 0-0-3 2:7

The Spain game you would check first: Spain vs Cabo Verde, the model gives Spain 68% to win, draw 21%, Cabo Verde 11%, most likely 2-0, with Cabo Verde shut out about half the time. That is exactly the historical signature of a clear mismatch.

Show all 60 remaining group-fixture predictions
DateGrpHomeAwayH/D/AFavouriteModal
06-15GIranNew Zealand 53%/27%/20%Iran 53%1-0
06-15GBelgiumEgypt 43%/27%/30%Belgium 43%1-1
06-15HSpainCape Verde 68%/21%/11%Spain 68%2-0
06-15HSaudi ArabiaUruguay 21%/28%/51%Uruguay 51%0-1
06-16IFranceSenegal 52%/27%/21%France 52%1-0
06-16IIraqNorway 13%/22%/65%Norway 65%0-2
06-16JArgentinaAlgeria 56%/22%/22%Argentina 56%1-1
06-16JAustriaJordan 34%/32%/34%Austria 34%0-0
06-17KPortugalDR Congo 60%/25%/16%Portugal 60%1-0
06-17KUzbekistanColombia 23%/30%/47%Colombia 47%0-1
06-17LEnglandCroatia 51%/25%/24%England 51%1-1
06-17LGhanaPanama 54%/27%/19%Ghana 54%1-0
06-18ACzechiaSouth Africa 58%/25%/17%Czechia 58%1-0
06-18AMexicoSouth Korea 45%/26%/29%Mexico 45%1-1
06-18BSwitzerlandBosnia and Herzegovina 46%/29%/25%Switzerland 46%1-1
06-18BCanadaQatar 67%/21%/12%Canada 67%2-0
06-19CScotlandMorocco 28%/30%/42%Morocco 42%1-1
06-19CBrazilHaiti 69%/20%/10%Brazil 69%2-0
06-19DUnited StatesAustralia 57%/25%/18%United States 57%1-0
06-19DTurkeyParaguay 44%/27%/29%Turkey 44%1-1
06-20EGermanyIvory Coast 45%/29%/26%Germany 45%1-1
06-20EEcuadorCuracao 48%/29%/23%Ecuador 48%1-0
06-20FNetherlandsSweden 48%/28%/24%Netherlands 48%1-1
06-20FTunisiaJapan 20%/26%/54%Japan 54%1-1
06-21GBelgiumIran 45%/29%/26%Belgium 45%1-1
06-21GNew ZealandEgypt 17%/22%/61%Egypt 61%1-1
06-21HSpainSaudi Arabia 66%/22%/12%Spain 66%1-0
06-21HUruguayCape Verde 54%/28%/18%Uruguay 54%1-0
06-22IFranceIraq 80%/14%/6%France 80%2-0
06-22INorwaySenegal 35%/29%/36%Senegal 36%1-1
06-22JArgentinaAustria 66%/21%/12%Argentina 66%2-0
06-22JJordanAlgeria 23%/26%/51%Algeria 51%1-1
06-23KPortugalUzbekistan 59%/26%/15%Portugal 59%1-0
06-23KColombiaDR Congo 47%/29%/24%Colombia 47%1-1
06-23LEnglandGhana 63%/22%/16%England 63%2-0
06-23LPanamaCroatia 13%/23%/64%Croatia 64%0-1
06-24ACzechiaMexico 31%/28%/41%Mexico 41%1-1
06-24ASouth AfricaSouth Korea 17%/24%/59%South Korea 59%0-1
06-24BSwitzerlandCanada 40%/30%/30%Switzerland 40%1-1
06-24BBosnia and HerzegovinaQatar 58%/25%/17%Bosnia and Herzegovina 58%1-0
06-24CScotlandBrazil 18%/26%/56%Brazil 56%0-1
06-24CMoroccoHaiti 55%/26%/19%Morocco 55%1-0
06-25DTurkeyUnited States 40%/29%/31%Turkey 40%1-1
06-25DParaguayAustralia 53%/26%/22%Paraguay 53%1-1
06-25ECuracaoIvory Coast 18%/26%/56%Ivory Coast 56%0-1
06-25EEcuadorGermany 20%/28%/51%Germany 51%0-1
06-25FJapanSweden 43%/27%/30%Japan 43%1-1
06-25FTunisiaNetherlands 16%/26%/58%Netherlands 58%0-1
06-26GEgyptIran 40%/28%/32%Egypt 40%1-1
06-26GNew ZealandBelgium 13%/22%/64%Belgium 64%0-1
06-26HCape VerdeSaudi Arabia 32%/30%/39%Saudi Arabia 39%1-1
06-26HUruguaySpain 23%/29%/48%Spain 48%0-1
06-26INorwayFrance 21%/26%/54%France 54%1-1
06-26ISenegalIraq 65%/22%/13%Senegal 65%1-0
06-27JAlgeriaAustria 50%/28%/23%Algeria 50%1-1
06-27JJordanArgentina 12%/19%/69%Argentina 69%0-2
06-27KColombiaPortugal 24%/28%/47%Portugal 47%1-1
06-27KDR CongoUzbekistan 35%/32%/33%DR Congo 35%1-1
06-27LPanamaEngland 7%/16%/78%England 78%0-2
06-27LCroatiaGhana 47%/27%/26%Croatia 47%1-1

05The knockout bracket

The official 2026 bracket sends group winners, runners-up and the best thirds into a fixed Round-of-32 tree. Running the projected qualifiers through it (10-run average per tie) gives one representative path to the trophy. Shown from the Round of 16; the winner of each tie is in blue.

Round of 16
Germany
France
South Korea
Sweden
Brazil
Norway
Morocco
England
Croatia
Algeria
United States
Egypt
Argentina
Belgium
Netherlands
Portugal
Quarter-finals
France
Sweden
Croatia
Egypt
Brazil
England
Argentina
Netherlands
Semi-finals
France
Croatia
England
Argentina
Final
France
England
🏆 England

In this single representative run, England came through, beating France in the final. But a single bracket understates how open it is, which is why the trophy question is better answered probabilistically.

06Who wins the World Cup

To turn one bracket into odds, the entire tournament was Monte-Carlo simulated 3,000 times: every unplayed group game and every knockout tie re-drawn each run, champions counted. This propagates the real uncertainty that a single averaged bracket hides.

Figure 1. Championship and final-reaching probability, 3,000 full-tournament simulations. Source: tournament_forecast.json.
THE MODEL'S CALL

The most likely champion is Argentina (17.07%), in a tight top cluster with England (14.97%), France (11.8%) and Brazil (9.97%). No team is above one-in-five: this is a genuinely open tournament, and the favourites are exactly the squads EA rates highest, led by Argentina's attacking strength. Treat the single-bracket champion (England) as one plausible story inside that distribution, not a contradiction of it.

07What 96 years of World Cup history say

The tournament calibration is grounded in every men's World Cup match ever played (1930-2022, 964 games). Two facts shaped the model and one folk-belief got tested.

Figure 2. Goals per game and the share of games where at least one team is shut out, by tournament. Source: martj42 international-results dataset.

Scoring collapsed from the goal-fests of the 1950s (5.4 a game in 1954) to a stable ~2.6 from the 1960s onward, with no drift since. About half of all World Cup games (49% all-time, 52% in the modern era) see at least one team fail to score, which is why the engine's goal rate was tuned down to match. And the popular claim that a team finishing on zero is near-certain? It is a coin flip, not a lock, and it is the weaker team that supplies the zero: by Elo the underdog scores in only 58% of games versus 82% for the favourite, so shutouts cluster in mismatches, exactly where this model is most confident.

08Limits

The honest caveats. EA league licensing leaves several squads under-covered (Mexico, Qatar, South Africa, Curaçao, Tunisia, and others marked with a star); their missing players are filled at the team's own median rating, so their projections are softer. The engine has no information on form, injuries, or in-tournament momentum, only squad talent as EA rates it. The draw is correctly priced but never the top single pick. And the knockout coin-flip for drawn ties is a proxy for penalties, which are close to random. None of this is hidden in the numbers; it is why the trophy is given as a distribution, not a certainty.

Everything above is checkable. The backtest already is; the forward book and the bracket grade themselves over the next month.

09References & provenance

  1. Dixon, M. & Coles, S. (1997). Modelling Association Football Scores and Inefficiencies in the Football Betting Market. Applied Statistics 46(2).
  2. EAFC26 Wave 3. A Football Simulation Infrastructure. Companion methodology report.
  3. EA Sports FC 26 player attribute database (men's, ~16k players).
  4. martj42, International football results 1872-present (49,478 matches). Used for World Cup history and Elo.
  5. FIFA, 2026 World Cup knockout-stage bracket structure (48-team format, matches 73-104).

EAFC26 Wave 4 methodology note v1.0, June 2026. Every numeric claim traces to a CSV/JSON artefact in the project's wave4_worldcup/3_artifacts directory or to a reference above. Backtest as of 14 June 2026, 12 group games played.

Pok Yeung Lee
EAFC26 · Wave 4 · June 2026