For the first time in World Cup history, the model projects an Asian side as joint group favorite alongside a European nation. Japan (75.8% qualification probability) edges Netherlands (74.3%), and the opener Netherlands vs Japan in Dallas (Arlington) emerges as the most balanced match of the 96 group-stage fixtures in the tournament: 32% / 36% / 32%. Sweden returns after missing 2022 with a Δ xG of −0.85 (worst in the group) and Tunisia continues its classic group-stage curse. But the math is so tight at the top that the upset is real: either co-favorite could miss out.
The Ensemble v1 (DIBP + Dual SPI + Elo + Bayesian Poisson, calibrated with temperature T=0.85 over 364 validation matches, post-calibration ECE 3.3%) produces probability predictions for each of the 103 World Cup fixtures. The table below shows the average per-match win probability for each nation within the group (not the probability of finishing 1st — that is covered in section 4):
| Nation | Per-match win prob (avg) | xG for | xG against | Δ xG |
|---|---|---|---|---|
| 🇯🇵 Japan | 49.3% | 1.87 | 1.21 | +0.66 |
| 🇳🇱 Netherlands | 48.0% | 1.86 | 1.12 | +0.74 |
| 🇹🇳 Tunisia | 21.1% | 1.01 | 1.56 | -0.55 |
| 🇸🇪 Sweden | 19.8% | 1.06 | 1.91 | -0.85 |
Japan and Netherlands are practically tied in avg_win (49.3% vs 48.0%) and xG for (1.87 vs 1.86). Netherlands wins the defensive comparison (xG against 1.12 vs 1.21), but Japan compensates with better easy-match distribution. Sweden and Tunisia fall clearly behind: both with negative Δ xG and avg_win below 22%.
The model's quiet revelation. Japan arrives at its 8th World Cup with the most complete squad in its history, blending the European generation (Mitoma, Kubo, Endo, Itakura) with mature J.League talent. The ensemble has them as group co-favorite with 75.8% qualification probability — the highest ceiling any AFC side has ever entered a World Cup with.
Model strengths — transition-machine attack (xG for 1.87, top of the group); European generation now established in Premier League, Bundesliga and Serie A; defensive cohesion built under Moriyasu since 2018; psychological edge from the 2022 last-16 against Croatia.
Risks — attack relies too heavily on Mitoma and Kubo (without them xG drops ~0.6); no elite 9 to guarantee goals in tight games; the Dallas heat (June) may punish their pressing — all three games in hot venues are problematic.
Realistic path — open with a draw or narrow win vs Netherlands (combined 1X = 68%), beat Tunisia (54%) and close out Sweden (62%). Cleanest path in the group if injuries cooperate: 7 points lead to first place in 41% of simulations.
The Oranje land at their 12th World Cup looking to wash off the failed 2010 final and Qatar disappointment. The model has them slightly below Japan in qualification (74.3% vs 75.8%), but with the best defense in the group (xG against 1.12) and a Δ xG of +0.74 that makes them the most reliable side here in high-pressure fixtures.
Model strengths — best defense in the group (xG against 1.12); balanced attack with Gakpo, Xavi Simons and Frenkie de Jong's transition; Δ xG +0.74 (second-best in the group after Belgium in adjacent groups); World Cup pedigree with three finals played.
Risks — right-back still without a reliable post-Dumfries replacement; keeper battle between Verbruggen and Bijlow unresolved; the MD1 game vs Japan is 36% draw and the model treats it as a coin flip — losing that kickoff complicates the whole group.
Realistic path — don't lose to Japan (combined 1X = 62%), beat Sweden (60%) and Tunisia (52%) → 7 points lead to first place in 39% of simulations. The cleanest Dutch narrative is classic: draw the opener and roll over the next two.
Tunisia plays its 7th World Cup — and has never made it past the group stage. The model sees no reason to break the streak: 26.5% qualification probability, Δ xG −0.55 and a calendar that requires at least 4 points against Netherlands or Japan. Realistic path: their only window is beating Sweden in MD1.
Model strengths — defensively compact when sitting deep; willing to run all day; World Cup experience through Khazri, Sliti, Skhiri; always a cohesive unit under Kadri.
Risks — limited attack (xG for 1.01 — worst in the group); produce little in build-up play; the model gives them less than 20% in each of their three matches individually; Δ xG −0.55 indicates a net goal deficit across all scenarios.
Realistic path — steal MD1 vs Sweden (28% away, only match with >25% chance), hold draw vs Japan in MD2 (29%) and scrap a point vs Netherlands in MD3 (29% draw). Only qualifies in 26% of simulations — most require a surprise point against one of the two co-favorites.
After missing Qatar 2022, the Swedes return with the weakest statistical profile in the group: Δ xG −0.85, xG against 1.91 per game (worst in Group F) and 19.8% avg_win. A squad with Alexander Isak, Lucas Bergvall and Anthony Elanga, but with defensive cracks the model does not forgive.
Model strengths — Isak (one of Europe's most efficient 9s), Bergvall (Spurs' rising gem) and Elanga (Newcastle); young squad with upside; Andersson's bench experience; sometimes pressure-free matches bring out the team's best.
Risks — weak defense (xG against 1.91 — worst in the group); no defensive leader post-Lindelöf; Δ xG −0.85 is a clear signal of a team needing more goals to offset what it concedes; calendar forces them to nail MD1 vs Tunisia.
Realistic path — beat Tunisia in MD1 (32% home), defend vs Netherlands in MD2 (14% upset chance) and pray vs Japan in MD3 (13%). Qualification probability: 23.4% — they need 5 points winning MD1 and drawing one of the next two.
The six group games are spread across FIFA-assigned host cities. Probabilities reflect the model's 1X2 distribution for each match, expressed as percentages.
Up to this point we have shown the average per-match win probability (how strong each team is in any given individual match). But "who finishes 1st in the group?" is a different question: it requires winning ENOUGH matches AND scoring MORE points than the other 3 by the end of matchday 3. For that we simulate the entire group 100,000 times and count how often each team finishes 1st, 2nd, 3rd or 4th.
Netherlands and Japan are separated by 1.5 percentage points in qualification probability (74.3% vs 75.8%). Across three matches, that gap is mathematically irrelevant. The narrative "Netherlands big group favorite" does not survive the ensemble comparison: this group is 50/50 at the top. Either co-favorite can miss out if the other catches fire and Sweden/Tunisia steal points.
Simulating the six group matches 100,000 times under the model's probability distributions, we obtain the following probabilities of finishing the group in each position. These numbers are the most relevant probabilistic input for informed Battle Royale decisions.
| Nation | P (1st) | P (2nd) | P (qualify) | P (eliminated) |
|---|---|---|---|---|
| 🇯🇵 Japan | 43% | 33% | 76% | 24% |
| 🇳🇱 Netherlands | 41% | 33% | 74% | 26% |
| 🇹🇳 Tunisia | 8% | 18% | 26% | 74% |
| 🇸🇪 Sweden | 7% | 16% | 23% | 77% |
Key reading: Japan and Netherlands are genuine co-favorites — their gap over Sweden and Tunisia is enormous (more than 45 percentage points in qualification), but between them the model barely differentiates. Sweden (23.4%) actually has a worse chance than Tunisia (26.5%) of qualifying — the model penalizes Sweden's Δ xG more than it rewards their World Cup pedigree. For Battle Royale: Japan is the most statistically solid pick of the group, by minimal margins.
Another powerful metric: the expected number of points each nation will accumulate across its three group fixtures, weighted by the model's probabilities. Results:
| Nation | XP | Expected goals (for) | Expected goals (against) |
|---|---|---|---|
| 🇯🇵 Japan | 5.33 | 4.75 | 3.09 |
| 🇳🇱 Netherlands | 5.25 | 4.56 | 2.98 |
| 🇹🇳 Tunisia | 2.88 | 2.63 | 3.91 |
| 🇸🇪 Sweden | 2.70 | 2.86 | 4.81 |
Gap between Japan (5.33 XP) and Netherlands (5.25 XP): 0.08 points out of 9 possible. Statistically indistinguishable. Sweden (2.70) and Tunisia (2.88) fall below the 3-point expectation threshold — historically not enough to qualify.
The opener Netherlands vs Japan in Dallas (Arlington) is the most balanced of the 96 group-stage matches of World Cup 2026. The ensemble projects it as 0.3227 / 0.3611 / 0.3162 — a near-perfect coin flip. No other match in the tournament has a max-min spread as low as this (0.0449).
The statistical reason is clear: both sides have an identical avg_win (49.3% Japan vs 48.0% Netherlands), nearly identical xG for (1.87 vs 1.86) and the defensive gap (0.09 xG in Netherlands' favor) is not enough for the model to tip one as favorite. The tiebreaker would be playing style: Netherlands dominance + structure, Japan transitions + coordinated pressing.
The team that does not lose this match has a 71% chance to finish top of the group. The loser drops to 31%. This is a World Cup fixture with the weight of a knockout final, played on matchday 1.
Sweden qualifies (23.4%) Needs to beat Tunisia in MD1 (32% probability), avoid loss vs Netherlands in MD2 (26% draw) and scrape something vs Japan in MD3 (38% combined draw+win). Most probable path: 6 points with two wins over Tunisia and Netherlands. Unlikely but possible if Isak enters scoring form.
Tunisia qualifies (26.5%) Needs to steal MD1 vs Sweden (28% probability), hold draw vs Japan in MD2 (29%) and scrape vs Netherlands in MD3 (29% draw). Most likely scenario is 5 points: beat Sweden + two draws. Historically Tunisia has never advanced past groups in 6 World Cups — the model reflects that structural burden.
The MatchScan Ensemble v1 combines four independent predictive models weighted by their historical accuracy over the consolidated 87,000+ match database:
DIBP (37.5%) — Dynamic Inter-team Bayesian Poisson. Goal-generation model with inter-team rating adjustment.Dual SPI (25%) — Soccer Power Index with separate attack/defense adjustment.Elo v0 (37.5%) — Dynamic Elo system with opponent-strength adjustment.Bayesian Poisson (0% in final blend) — Bayesian model showing instability on asymmetric matches.
The resulting probability set is calibrated via scalar temperature T=0.85 (Platt scaling) with Bayesian correction b_H=-0.24 and b_D=+0.38. Calibration is verified over 364 cross-validation matches, yielding pre-calibration ECE of 4.68% and post-calibration ECE of 3.26% — very low expected calibration error, indicating published probabilities are reliable as frequentist expectations.
World Cup 2026 Battle Royale on MatchScan. Pick 8 teams for the tournament. One per matchday. If your team loses or draws, eliminated. Last survivor wins €100 + 1 year MatchScan VIP. Free entry, no card, no spam.
Sign up →