SpainversusArgentina

What was collected, from where, and what could not be obtained. The gaps are stated because a forecast that hides them is not reproducible.

1.1 Sources used

1.2 The raw data question

Whether ball by ball event data or player tracking data exists publicly for this tournament, which determines the ceiling on any analysis of it.

1.3 Data conflicts left unresolved

Where independent providers disagree, both figures are carried rather than one being silently chosen.

1.4 Validation

Every model below is fitted on the same 102 matches so that comparison between them is fair.

2.1 The core specification

Goals scored by team A against team B are modelled as a count with rate:

Ratings are estimated by maximum likelihood with ridge shrinkage. Extra time goals are excluded, because extra time is a different game state and the final is modelled from kickoff.

2.2 Choosing the shrinkage by cross-validation

With seven matches per team, an unpenalised fit would give Spain an absurd defence rating on the strength of conceding once. The penalty strength was selected by held out likelihood, not chosen by hand.

2.3 The nine models

2.4 Model comparison

Cross validated accuracy on the actual match outcomes. Lower is better for both measures.

2.5 Contextual adjustments

Applied multiplicatively to expected goals. Each carries its justification, so a reader who disagrees with one can see exactly what it is worth.

2.6 Ensemble construction

A seven match sample is thin. The final forecast blends our models with two external estimates rather than trusting one method.

3.3 Scoreline distribution

Rows are Spain goals, columns Argentina goals, after 90 minutes. Red cells favour Spain, blue favour Argentina, grey is a draw.

3.4 When goals arrive

Argentina scored 12 of 19 goals after the 75th minute, the most by any team at a single World Cup. A model of final scores alone would miss this entirely.

3.5 Team ratings, all 48 teams

Opponent adjusted, so beating France counts for more than beating Cape Verde. Teams with fewer matches are shrunk harder toward average.

3.6 Elo, walked forward through the tournament

Team models say how many goals. This section allocates them to individuals, using expected goals per 90 minutes weighted by projected minutes in the final.

4.1 Who scores

4.2 First goalscorer

The probability a given player scores the opening goal, which combines their share of their team's goals with the chance their team scores first at all.

4.3 The Golden Boot

4.4 Physical profile

Distance, sprints and peak speeds per player. This is the closest publicly available data to the tracking feeds that clubs use internally.

4.5 Full squad data

Each run draws a fresh result from the same calibrated model behind the headline number, so repeated watching reproduces the stated probabilities. Goal timings follow each team's real scoring profile. Drag to orbit the camera, scroll to zoom.

Head to head, measured

Stated in full, because a forecast is only as credible as its account of what it cannot do.

7.1 What would change the answer

References and data sources

8.1 Authenticity, with timestamps

A prediction is worthless if it can be edited after the event. This forecast lives in an immutable git history: the commit hash is a cryptographic fingerprint of the exact text, so any later change produces a different hash. The forecast commit predates kickoff by roughly 36 hours. The whole chain is reproducible with git log.

8.2 Forecast versus outcome, scored

Each pre registered probability is graded by its Brier score, the squared miss between the odds put on the realised event and 1, where lower is better. A blind coin scores 0.250. Bars show the probability the model assigned to what actually happened.

8.3 The match

Spain out created Argentina on every axis. The expected goals gap, 2.12 to 0.37, is a rout the single goal barely reflects. Argentina managed zero shots on target until the 117th minute, and the score stayed close only because Emiliano Martínez produced a World Cup final record of 11 saves.

8.4 Minute by minute

The turning point was disciplinary, not tactical. Enzo Fernández collected a second yellow deep in stoppage time, and Argentina played all of extra time a man short. Two Spanish substitutes combined for the only goal.

• 4' Yamal is played through, Martínez saves, Spain's first warning.
• 41' Lisandro Martínez booked. 44' he goes off injured, Otamendi on, Argentina's plan disrupted.
• 62' Spain introduce Ferran Torres and Pedri, the goalscorer enters.
• 75' Spain introduce Nico Williams and Merino, the assister enters.
• 82' Enzo Fernández booked. 90+3' second yellow, Argentina down to 10 for all of extra time.
• 106' GOAL. Ferran Torres sweeps home Nico Williams' header. Spain 1-0.
• 117' Argentina's first shot on target of the match, too late.
• FT Spain 1-0 Argentina after extra time, La Roja's second world title.

8.5 Player ratings

ESPN marks out of 10. Rodri took the Golden Ball, Unai Simón the Golden Glove, Cubarsí the young player award, and Torres the match. Enzo's 4.0, the lowest on the pitch, marks the red card that decided it.

8.6 Case study: right answer, but right for the right reasons?

A correct forecast can still rest on a broken mechanism. The pre registered paper made specific causal claims, not just a number. Grading them against the tape separates genuine signal from luck.

Glossary

Every technical term used in this paper, in plain language. Terms are also underlined in the text, where tapping one shows its definition in place.