Wembley, Minute 104: My Model Was Right in Every Column and Wrong on One Variable

Minute 104, Wembley, 7 July 2026. Kasper Schmeichel dived the right way, got a hand to the ball, and the ball still ended up in the net off Harry Kane's follow-up. In Shanghai I sat in front of a screen, hands still on the keyboard, one line running through my head: my model just lost.

Before the match I had said on a radio broadcast that Denmark would reach the final. I said it in the exact tone I use when the spreadsheet is on my side. Denmark covered an average of 118.7 kilometres per match; England only 112.3. Denmark produced roughly 18 shots per match; England roughly 11. Those two lines of data, plus the lesson about empty stadiums I had collected a year earlier, were enough for me to push everything else aside.

I was wrong. This piece is the record of that error, written the way I always write it: no deletion, no quiet edit, no burying it at the bottom of the page.

Data context

This is the part I place at the top of every calculation, and this time I placed it in the wrong slot.

Euro 2026 was played in 2026, a year late because of the pandemic. The semi-final between England and Denmark was played at Wembley Stadium, London, on 7 July 2026. The stands were not empty. Wembley opened around 60,000 seats for the semi-finals and the final, after a group-stage limit of 22,500. I had read that detail, written it in my notebook, and then convinced myself it weighed less than kilometres covered.

Denmark's journey through this tournament is the thing a spreadsheet has no column for. On 12 June 2026, in the 42nd minute against Finland in Copenhagen, Christian Eriksen collapsed on the pitch. The match stopped, the team wept, and Denmark lost 0-1 when the game resumed. Three days later they lost 1-2 to Belgium. That team advanced on the back of a 4-1 win over Russia, then a 4-0 demolition of Wales in Amsterdam, then a 2-1 win over the Czech Republic in Baku. A team living on emotion, running on collective grief, and running a great deal.

England's path was cleaner: 1-0 against Croatia, 0-0 against Scotland, 1-0 against the Czech Republic, 2-0 against Germany at Wembley, 4-0 against Ukraine in Rome. They reached the semi-final without conceding a single goal. I knew that figure. I simply did not give it enough weight.

One more fact sits outside the spreadsheet: England reached a major tournament final for the first time since the 2026 World Cup, and that final would also be played at Wembley. A national team playing its most important semi-final in 55 years in front of nearly 60,000 of its own supporters is a variable I had no way to model in kilometres.

What my model contained

I built this model after the 2026 World Cup, when Germany's PPDA, an average of 11.3, well above the 8.5 to 9.5 range of the leading pressing sides, gave me a prediction that landed. Since then, before every major tournament, I publish a list of slow-burning bombs, ranked by PPDA and shots faced per match.

The model has four inputs: PPDA, which measures how aggressively a team contests possession; total distance covered; shots and shots on target, which measure attacking intent; and home win rate, which measures contextual advantage. Those four inputs give me a probability. They do not give me a result.

With Denmark, three of the four inputs tilted their way. Only the fourth, home advantage, tilted towards England, and I assigned it the smallest weight because my 2026 study of empty stadiums said home advantage was shrinking.

That is where I fooled myself: a phenomenon shrinking under abnormal conditions does not mean the variable becomes meaningless under normal ones.

The evidence chain on the pitch

Denmark took the lead in the 30th minute. Mikkel Damsgaard, 21 years old, struck a direct free kick from outside the box, the ball clearing the wall and finding the net. It was the first goal England had conceded all tournament.

In the 39th minute Bukayo Saka crossed from the right, Simon Kjaer stuck out a leg to intercept and turned the ball into his own net. 1-1. An own goal in the first half sits in no branch of any probability tree I have ever built.

Structurally, Denmark played with a back three, with Joakim Maehle pushing high like a full-back converted into a wing-back. England played a back four, with Declan Rice and Kalvin Phillips screening in front, Mason Mount free ahead of them, and Kane dropping deep as a pivot. Denmark's idea was to drag the game into quick transitions and set pieces. They scored their opener from exactly that.

Then the match entered the zone I cannot measure: extra time. Denmark covered 118.7 kilometres per match, but they covered those kilometres inside 90 minutes. By the 100th minute, the price of outrunning your opponent stops being an advantage and becomes an invoice. England entered extra time with a bench Denmark did not have. Jack Grealish, Phil Foden, Jadon Sancho, Marcus Rashford, four names good enough to start for most teams at the tournament. In the data I log, they are a binary 0/1 variable labelled substitute. At Wembley, they were the final 30 minutes of the match.

In the 104th minute the referee awarded England a penalty after contact between Maehle and Raheem Sterling. A contentious call, I know, and I will not use it as an excuse. Schmeichel saved Kane's kick. The ball came back. Kane converted the rebound. 2-1.

Watching the footage back, I counted roughly 12 shots for England and roughly 6 for Denmark across 120 minutes, with possession leaning to England at around 58 percent. I counted those by hand rather than pulling them from a data provider's API, so the margin of error may be one shot. The direction is clear: the team that ran less held the ball more, and my model did not forecast that.

Harry Kane finished the tournament with four goals, and the one in the 104th minute was the most important and the least beautiful: a follow-up from seven metres after the goalkeeper had already made the save. In the data, that goal and a 25-metre volley are worth the same single unit.

Transfers are a fertile gamble, but I count cards before placing a bet. In this match, the combined transfer value of England's starting eleven was several times that of Denmark's, and the gap was not in the eleven, it was on the bench. That is the kind of gap my model records as money and then dismisses as an administrative detail.

The blind spot: I learned from empty stands

In 2026, when football resumed in Germany without spectators, I collected 250 Bundesliga matches and found two figures. Home win rate fell from 43 percent to 31 percent. Goals per match dropped by 0.4. I wrote a study titled Silent Stands Is a Metric. No crowd, and football transforms. I found that, and I was rejected for it. The desk wanted me to add an upbeat message about recovery; I refused, and lost my separate contract.

The lesson I drew was that crowd density is a variable. But I drew it in one direction only. I learned that empty stands dilute home advantage, and then quietly treated the presence of a crowd as a neutral baseline. Wembley with nearly 60,000 people is not a neutral baseline. It is a different, stronger variable, and I had no data for it.

This is the most serious technical error in the whole exercise: a model trained under abnormal conditions, then applied to normal ones. I took an exception and called it a rule.

After the match I spent two days rereading what players on both sides said and watching the tape from the 60th to the 104th minute three times. No new fact appeared in the match record. What appeared was the thing the record does not capture: Denmark's defensive line retreating more slowly after the 75th minute, the gaps between the three centre-backs widening, and midfield duels starting to be won by England by half a step. Half a step is not a metric. It is the consequence of one: 118.7 kilometres inside 90 minutes, multiplied by a semi-final, divided by a thin bench.

Where my assumptions could be wrong

I have been forced to write this section since Euro 2026, and it still sits at the end of everything I publish.

I assumed distance covered is a measure of strength. It may only be a measure of helplessness: the team that runs most is usually the team chasing the ball. Denmark ran because they had to contest, not necessarily because they were fitter.

I assumed extra time is a linear extension of 90 minutes. In my data, extra time is 90 plus 30. In reality it is a different game, where the bench and the number of substitutions decide more than fitness.

And this is the assumption I doubt most: I assumed that a prophecy coming true validates the method. In March 2026 I wrote a prophecy. The whole of Germany laughed. On 27 June 2026, Germany lost 0-2 to South Korea and finished bottom of Group F, exactly as their PPDA of 11.3 had hinted. I was right. But being right once is not a method. It is just being right once.

Why I am still publishing this

On the night of the Shanghai derby, I chose the numbers over an entire city. That night I was right, and being right opened up my Data Reading column. Today I am wrong, and being wrong has to be published too, in the right place, without being softened.

People have asked whether I regret speaking so confidently on air. I do not regret speaking. I regret speaking while my spreadsheet was missing three columns: stand density, bench depth, and the pressure on the man with the whistle in extra time. The spreadsheet is an altar, and I give myself to every number on it. But an altar also needs room for the numbers I have not yet learned to write down.

In the field I follow alongside football, esports, this lesson shows up more clearly. In a League of Legends or Counter-Strike match, mid-series substitutions essentially do not exist; the lineup that starts is the lineup that finishes. When I carried data thinking from esports into football, I carried a blind spot with it: I forgot that football is a sport of momentum shifts delivered from the bench, and that every substitution is a variable you cannot measure in minutes played. Denmark won Euro 2026 after being called up as a late replacement for Yugoslavia, and that is the kind of story that makes people believe in the unmeasurable. I do not believe in the unmeasurable. I believe the unmeasurable is simply the not-yet-measured, and my job is to find the ruler.

From the Bundesliga to Worlds, I am looking for the same thing: a truth that can repeat itself. The problem is that truth repeats more slowly than I expect, and every tournament rewrites the rules of its own game.

Signals for the next cycle

Wembley handed me three new variables, and I will fold them into the model before the next major tournament.

Crowd density must be weighted by match tier, not by tournament. An empty group-stage ground and a full semi-final are two different tournaments inside one competition.

A bench-depth index must measure the quality minutes a team can introduce after the 70th minute, not the number of substitutes available. Denmark had a bench. They did not have 30 quality minutes. England had both.

A referee-pressure index in extra time is something I still do not know how to measure, but I know it exists, and I know it tilts with noise.

Every crowd is wrong. The only thing that is not wrong is probability, but probability is only right when I admit I have not listed every variable. They said I was stirring trouble. I was only reading the ending a few months early. This time I read it wrong, and I am leaving those wrong words here as the baseline for the next read.