Interactive · the forecast in your hands
Explore the field
Two instruments, one locked forecast. Put any two of the 48 teams head-to-head on a neutral field, or pull a single slider to watch the draw reshape the title race. Every number here was computed once, in the locked pipeline, and is read back exactly — nothing on this page re-runs the simulation.
§ 01
Team versus team
Pick two nations. The head-to-head is the locked ensemble's neutral-venue forecast for that exact pairing — the same engine, on the same scale, as the champion and group probabilities everywhere else on the site. Argentina and Spain are shown to begin.
Fig. X1 Neutral venue · locked ensemble
Head-to-head, on a neutral field
Win / draw / win, expected goals, the most-likely scorelines, the goal markets, a side-by-side strength read, and the soonest the bracket could put these two together. Change either side to recompute.
Neutral-venue result
Most likely scoreline 1–1 (14.6%), Argentina first.
Expected goals & goal markets
Most-likely scorelines
Top five exact scorelines from the neutral 11×11 grid. Bars shaded by result: Argentina win · draw · Spain win.
Strength profile, side by side
Bars are each metric scaled to the larger of the two sides; raw values are labelled. Source · Oxford Football Forecasting model — per-team dossiers.
Where the draw could pit them together
The earliest the bracket can pit Argentina against Spain is the Round of 32 — and only if the draw breaks that way. This is the soonest their two possible paths can cross, not a prediction that they will.
1st(1J) vs 2nd(2H) (R32 tie 13: 1J vs 2H)
A neutral-venue meeting strips out home advantage — it is the cleanest 'who is better, right now' question, and the same neutral grid the simulator uses for every knockout tie.
What this comparison is — and what it is not
Each cell is the locked ensemble's forecast for that one pairing at a neutral venue: the per-cell log-opinion pool of the Dixon-Coles bivariate-Poisson model, the global LightGBM-Poisson and the Bayesian hierarchical model, marginalised from the same 11×11 scoreline grid the tournament simulator consumes. Because it is neutral, it deliberately drops host advantage — so for a group game a host actually plays at home, the live fixture card will differ by exactly that bump. The strength rows are read straight from each nation's dossier; “where they could meet” is pure bracket geometry — the earliest stage the actual draw allows these two to face off, taken over every way each side can qualify (1st, 2nd or one of the eight best thirds). It is a possibility, not a probability. For the realised odds of reaching a round, see the team’s own forecast funnel.
§ 02
Power versus the draw
One slider, two genuine endpoints. Slide from raw Power — each team's champion odds before the bracket existed — to fixture-aware Reality, the odds given the actual draw. Watch who the draw lifts and who it weighs down.
This is an illustration, not a re-simulation. The two endpoints are real, locked outputs; the in-between is a straight interpolation, p(w) = (1−w)·Power + w·Reality, renormalised to sum to 100%. It shows how far the draw moves each team — it does not re-run the 1.1M tournaments, and no blend other than the two endpoints is a model output.
Fig. X2 Champion odds · interpolated between two locked rankings
From Power to Reality
At the far left, the field as raw strength alone would seed it. At the far right, the same field after the actual draw — softer or harder paths, balanced or stacked halves. The arrows mark each team's rank move between the two.
100% Reality / 0% Power — pure fixture-aware Reality (the locked forecast)
Showing the top 16 of 48; the long tail is at or near zero either way. Arrows compare each team's Reality rank with its Power rank.
The draw is close to fair, so most teams barely move — but it is not perfectly fair: England climbs 1 place on a kinder path while France slips 1. That reshuffle is the whole point of keeping two rankings, not one.
Why two rankings, and why only interpolate them
Power is the champion probability from simulating the bracket with every team in its own group but the draw re-randomised a million times — strength with the luck of the draw averaged out. Reality is the locked forecast: the actual 2026 draw, simulated as it stands. The gap between them is draw-luck — positive for a team the bracket favours, negative for one it punishes. Interpolating between the two is a faithful way to see that gap grow from zero; it is not a third model, and the page never presents the midpoint as a forecast — only the two endpoints are model outputs. The real re-draw experiment — a genuine million-simulation resampling — already happened in the pipeline and is what produces the Power column you are sliding to.