How the scoring works

The AI World Cup XI is picked by a statistical model, not by opinion. The same model powers both the live World Cup 2026 XI and our historical retrospectives like Qatar 2022.

The basics

Every player gets a single Impact Score out of 100, built from what they've actually done on the pitch. A striker is judged mostly on attacking output; a centre-back mostly on defending. The maths picks the XI — the AI only writes the explanations you read.

The number on each card is calibrated against others in the same position so the pitch reads at a glance. Compare a defender to other defenders, not to strikers — and every player carries their rank within their role ("#1 CB of the tournament").

What moves the needle

• Knockout matches count for more. A standout quarter-final is worth more than a quiet group game.
• You have to play to qualify. A minimum amount of pitch time is required before a player is eligible at all.
• Early on, we flag low confidence. Selections from a small sample are marked as provisional until more matches are in.

Team Strength

The Team Strength score (0–100) combines the XI's position-weighted Impact Scores into a single composite, with small contributions from the manager and bench. We weight Attack and Defence equally at 30% each, Midfield at 25%, with Manager and Depth providing smaller adjustments. The five sub-scores show where the XI's strength sits, so a 78 with an elite attack and shaky defence reads differently from a 78 with balance across the line.

Manager of the tournament

We rate each nation's manager on how their results compare to expectation, not just raw points. For every match we work out the points their team would be expected to take given the two sides' world rankings, then compare it to what they actually got — and we reward the quality of the result (a 4–2 over a strong side beats a 1–0 over a minnow). The base score is 55% overperformance and 45% results. On top of that, knockout matches count for more, a tie won on penalties counts as a win, and reaching the latter stages earns a deep-run bonus (with extra for lifting the trophy) — so a manager who takes an unfancied side on a long run outranks a one-off giant-killing that ended in the group stage. The same model scores both the live tournament and our Qatar 2022 retrospective. The manager shown is the current leader.

Match predictions

For a fixture we rate both nations with a Nation Strength (0–100). We start from the squad — their best XI's scores — and then learn from results: a nation's actual performances (points and goal difference) update the rating, and the more they've played, the more those results count. Early on, squad quality leads; by the latter stages, what's happened on the pitch dominates. Because the rating is built this way, every nation also carries an uncertainty band (the ± you'll see next to the number) that tightens with every game — so predictions for sides we've barely seen are honestly less confident.

From the two strengths we build a scoreline model. Each team's strength becomes an expected number of goals, and from those we work out the probability of every scoreline — which gives us the most likely result, the win/draw/win split (consistent with that scoreline by construction), and extras like over/under and both-teams-to-score. The model's few dials are fitted to the tournament's actual results as they come in, so it sharpens as the World Cup unfolds.

It's our own read on the match — not a bookmaker's line and not based on xG. A nation has to have played to be rated, and predictions are flagged provisional until both teams have a couple of games behind them.

Tournament forecast

To answer "who wins the whole thing?" we simulate the rest of the tournament 10,000 times. Each simulation plays out every remaining group game (using the scoreline model above), works out who qualifies, seeds the knockout bracket and plays it to a winner — with a penalty shootout decided by a strength-weighted coin when a knockout tie is level. Counting how often each nation tops its group, reaches the latter stages and lifts the trophy gives our forecast: advance, win-group and championship probabilities, and a simulation-based predicted top 3. Chances to advance and win the group are exact for our model; the deeper "who lifts it" odds depend on how the bracket falls, so we treat those as indicative.

The same engine drives a Golden Boot race: we take each scorer's share of their team's goals and carry it across the games their team is expected to still play, then simulate the race to a projected final tally and a chance of finishing top scorer.

What we don't do

We don't use Expected Goals to score players, or any tracking data we'd have to buy in — and we don't pretend to. Where our provider gives us team-level context (a match's xG, possession, shots), it's shown as labelled context, never as part of a player's score. We'd rather be honest than fake an "official" number.

---

Built for the love of the game. The opinions are the model's.