Three Blank Columns in a Scouting Report: The Limits of Football Data and the Integrity of the Analyst

One morning in February 2026, in Liverpool, I opened a forty-page scouting file sent over by a colleague in Lisbon. The player's name column was filled in. The date-of-birth column was filled in. The minutes-played-in-the-domestic-league column was filled in. But the last three columns — touches inside the opposition penalty area, chance-conversion rate, and the estimated value produced by our internal valuation model — were completely blank. Forty pages of paper, and on page thirty-seven there was a silence. My assistant asked whether we should fill in estimates so the file would look complete before it went to the recruitment board. I told him to leave it blank. He asked again, a little irritated: a file that goes upstairs missing three columns looks like unfinished work. I said that a file that goes upstairs with three columns filled in by guesswork is far more dangerous. That answer sounds simple. It took me nearly thirty years in this trade to say it without my hands shaking. Those three blank columns are the subject of this piece. Not a specific match, not a specific transfer, but the gap between what we know and what we think we know. It is the terrain every football analyst must cross, and it is where most of us fail in silence. Football analysis has travelled a long way in fifteen years. Cameras track the movement of twenty-two players at twenty-five frames per second. Every pass, every duel, every sprint is logged as a data row. A single Premier League match generates millions of spatial data points, and all of it sits in the hands of any club willing to pay a data provider. Looking at that volume, it is easy to believe football has become a solved problem. But there is one kind of data no provider sells. It is data about what has not happened yet. And in a season where the transfer market runs on expectation, that is the most expensive commodity of all. xG was a revolution, but every revolution needs time before people accept it. Expected goals entered public consciousness around 2026, and it took nearly a decade to move from personal blogs into the analysis rooms of Europe's biggest clubs. Today no sporting director signs an eight-figure contract without at least one page of xG in the file. What fewer people say is that this very popularity has created a new layer of gaps. When every club owns the same set of metrics, the advantage no longer lies in owning data — it lies in knowing where the data says nothing. In my files, gaps appear in three different forms, and each demands a completely different response. The first is the small-sample gap. A twenty-year-old with thirty senior appearances has metrics that sit inside the noise. You can calculate a number, but the number carries no statistical meaning. This is where the transfer market makes its most expensive mistakes. The second is the unmeasurable gap. Match psychology in front of seventy thousand people. A defender's tolerance for pressure in the last ten minutes when his team is a goal down. The understanding between two players that only forms after three years side by side. No sensor measures these things, and none will in the next decade. The third is the erroneous-data gap. This is the most dangerous kind because it does not announce itself. Two different providers can publish two different figures for the same match, and both are confident they are right. I learned the difference between these three gaps at Anfield, in the 2026-17 season, when I was still working inside a club's transfer-market department. That season Juergen Klopp took Liverpool to fourth place with seventy-six points. I sat up night after night calculating PPDA — passes allowed per defensive action — and came out with an average of 8.2 for Liverpool. That was the lowest in the league. Jose Mourinho's Manchester United finished the same season at 15.7. A gap of nearly double said something the league table could not: Liverpool did not defend in their own half. They defended in the opponent's half. I wrote a long piece on gegenpressing and published it on my personal blog. The reaction was not what I expected. I was attacked hard for being mechanical, for turning football into a spreadsheet, for understanding nothing about the soul of the game. One reader wrote that I was using numbers to prove I was smarter than people who watch football with their hearts. I kept my position, but I remembered the reaction. It taught me that correct data can still be rejected for emotional reasons, and that an analyst must be right about timing as well as arithmetic. On 14 January 2026, Liverpool beat Manchester City 4-3 at Anfield, ending the visitors' unbeaten run. I stayed in the office late that night, reviewing every passing and ball-recovery map. Data whispers, and those who listen hear magic in it. In the first half Liverpool recovered the ball in the opposition half fourteen times. That figure appeared in no match report the next morning. Yet it explained precisely why Manchester City — the best possession side in Europe at the time — could not impose their game. At Anfield I learned that belief is also a variable. The players' belief in the system, the crowd's belief in the team, and the analyst's belief in his own model. When those three resonate, they produce results no spreadsheet predicts. But all three can invert, and when they invert, no metric gives warning. In June 2026 I agreed to write a World Cup series from Russia for a sports outlet. I built a homemade xG model, ran it across all sixty-four matches, and concluded France would win because their chance creation was the highest in the tournament, at 2.4 xG per game. I also wrote that Croatia's run came from luck rather than chance quality. France won. But Croatia reached the final, and I became a joke in the comments. The person who is right before his time always pays in loneliness. In this case, though, I was not right before my time. I was simply confidently wrong. I hid in the city library for two weeks, re-watching every Croatia match, and found the error. My model ignored set pieces. It counted only open play, which erased the single most important chance source for a team that profits from corners and direct free kicks. A design error in the data, not an error about football. From then on I added a short section at the end of every piece titled Limits of This Analysis. It is usually three or four sentences long, but it changed how I read my own work. I no longer give absolute forecasts, only probabilities, with the conditions under which they hold and the conditions that void them. In March 2026 world football stopped because of the pandemic. Liverpool were twenty-five points clear of Manchester City and all but certain champions, but the season was suspended. I wrote three drafts and deleted all three. If data cannot forecast a pandemic, what use is data. When football returned in June with empty stands, I found something that later became the foundation of my whole method. The home win rate in the Premier League fell from roughly 46 per cent to roughly 39 per cent. Empty stadiums do not distort the data, but they make the truth feel hollow. Home advantage, treated for more than a century as an immutable constant, turned out to be a variable dependent on human noise. From that finding I built the concept of data context. Every metric must be read together with the environment that produced it. A pass that becomes an assist in front of a full stadium and one in front of an empty stadium are not the same pass, even though the system logs them identically. A spreadsheet cannot tell the two apart. An analyst must. In July 2026, during a Euro series, I connected online with an Italian tactical analyst. He shared internal training data from the Italy squad: they covered about 112 kilometres per match, not the highest in the tournament, but their ball-circulation index was markedly superior. I wrote a piece arguing that Italy were not a defensive team but a movement machine. It travelled widely. What I remember most is not the share count but the feeling of working inside a community that reads to the end. After that I dropped the habit of hiding. I invited readers to send me the data they had, and we checked it together. One reader in Hai Phong once sent me his handwritten tracking sheets for a lower-division league, accurate enough that I had to double-check them twice. But back to the transfer market, where data gaps get filled with money. In January 2026 Chelsea paid around 121 million euros for Enzo Fernandez from Benfica, after the player had spent roughly six months in European football. In the same window Mykhailo Mudryk arrived from Shakhtar Donetsk for a reported 70 million euros plus add-ons potentially reaching 100 million, with very limited elite European minutes behind him. Earlier, in 2026, Atletico Madrid paid 126 million euros for Joao Felix at nineteen. Every number in a transfer table is a life waiting to be written. The problem with these numbers is not that they are wrong. It is that they are calculated from a sample too small to be called evidence. When a player has fewer than fifty senior appearances, every advanced metric sits inside an interval far too wide. You can draw a beautiful chart, but the chart forecasts nothing. This is why I believe the young-player price bubble is deflating — not with a crash, but through a quiet correction. Clubs have begun to accept that paying a hundred million euros for a player with fewer than fifty elite matches is a naked gamble, and that gamble does not always win. Here, though, I must argue against myself. Correlation is not causation. The fact that many expensive young signings fail does not prove that youth caused the failure. It proves only that in a large enough set there will always be failures. To conclude properly I must compare against a control group: players of the same age, with the same minutes, bought for less, and their success rate. I ran that comparison on public data from Europe's five major leagues between 2026 and 2026. The result was less clear than I expected. A high fee correlates with high expectation, and high expectation correlates with high pressure, but the fee itself predicts nothing. The variable with real explanatory power is elite minutes played before the transfer. That is a far less attractive conclusion than a bursting-bubble story. But it is the conclusion the data permits me to state. The same problem appears in officiating. When VAR spread, it was marketed as a tool that removes error. But the intervention threshold is clear and obvious error. The phrase itself is an ambiguous clause. Clear to whom, obvious by how much, from which camera, in which frame. There is no mathematical definition of clear. So VAR does not replace subjective judgement; it relocates it from the referee on the pitch to the referee in a quiet room. That subjective space is wider than spectators imagine. When a decision arrives after four minutes of review, the public assumes technology has verified it. Most of those four minutes are a human weighing whether an error was genuinely clear and obvious. This is where VAR and my scouting file intersect. Both are systems designed to manage information gaps, and both fail in the same way: we fill the gap with process, then believe the process has converted a guess into a fact. The football industry rewards certainty. An analyst who says I do not know does not get invited on television. A pundit who says this player will definitely succeed gets more views. That incentive structure produces a system in which people are pushed towards always having an answer, even when the answer does not exist. I have met many colleagues who are good, careful, and who must repeatedly choose between being right and being heard. That is a choice nobody should have to face in an analytical job. So what should an honest analyst do with a gap? He must flag it. An empty cell in a spreadsheet has higher diagnostic value than a cell filled with the league average. When I see a model with no set-piece column, I immediately know which direction its conclusions will lean. A gap does not ruin a model. A hidden gap ruins a model. He must name the type of gap. Small sample is one thing, unmeasurable is another, erroneous data is a third. These three require three different responses, and the worst possible response is to treat them as one. And he must accept that some questions will have no answer within a given timeframe. For the 2026 World Cup across the United States, Canada and Mexico, expanded to forty-eight teams and more than a hundred matches, the number of variables grows exponentially. Deeper squads, more travel between three countries, large climate differences between host cities, and a schedule whose density will generate entirely new data patterns. Anyone claiming to have finished modelling this tournament at this stage is selling you a belief, not an analysis. Semi-automated offside detection, together with high-frequency sensor balls, will add a new layer of data at national-team level. Precisely for that reason, the distance between what is measured and what matters will widen. There will be data on the position of every boot, and no data on what the player was thinking in that moment. In a world of seasons that stretch on, the awakened can only rely on their own spreadsheet. But that spreadsheet must be one whose empty cells are clearly marked, not one padded to look finished. I returned to the forty-page file this morning. The three columns are still blank. I sent my colleague in Lisbon a short note: we need club-level tracking data, at least two thousand minutes, before we can value him. The note is not exciting. It will not make a headline. But it is true, and that is everything I can hand the recruitment board right now. Data never lies. But silence, sometimes, is the most honest answer a professional can give.