The Empty Dataset and the Temptation of Conclusion in Football Analysis

That night in Lyon, the screen in front of me was blank. The tracking feed for a European qualifying match would not load: no player coordinates, no touch counts, not a single line of PPDA. And yet, in the room, three people were still talking. One spoke about a high press. One spoke about a defence stretched out of shape. The third insisted the away side had lost the midfield from the 20th minute. None of them looked at the empty screen. I stayed silent, and remembered another night, at Saint-Denis, in the autumn of 2026. That night I was commentating live on France versus Sweden in the 2026 World Cup qualifiers. In the first half, I mispronounced the name of the midfielder Ola Toivonen three times. The director had to correct me through the earpiece, his voice urgent. Getting a name wrong sounds small. But I understood what it exposed: when I did not fully grasp a detail, I tended to fill the gap with a version that sounded plausible. The mispronounced name was only a symptom. The disease was confidence built on an empty foundation. After the match, I spent a month reviewing the footage, noting the correct pronunciation of 200 European players, and building my own phonetic table by language of origin. Since then, every analytical draft of mine carries phonetic annotations beside player names, and every piece is cross-checked before publication. When I mispronounced a player's name, I learned to listen to the rhythm of a match. This article is about that very gap: what happens to football analysis when the input dataset is empty, and why the human reflex is to invent a conclusion rather than admit the emptiness. Professional football today runs on a layer of data thicker than ever before. Optical tracking systems record the coordinates of 22 players and the ball at up to 25 frames per second. Event-data providers label every pass, every duel, every off-ball run. Expected goals has travelled from the internal analysis room to the scoreboard graphic on television. Some clubs have built an entire recruitment model around these numbers. Three examples are real and verifiable. Brentford, owned by Matthew Benham, who also owns the sports-data company Smartodds, uses quantitative analysis to run its transfers. According to public data, the club was promoted to the Premier League after a 2-0 win over Swansea in the Championship play-off final on 29 May 2026, marking its first return to the English top flight since the 2026-47 season. Liverpool, since Fenway Sports Group appointed the physicist Ian Graham in 2026, has built a research department that plays a major role in many deals. Brighton, owned by Tony Bloom, founder of the betting-data company Starlizard, has followed a similar path. Based on my experience watching matches over 21 years, this is the first time in the history of the sport that a data analyst can hold a role as important as an assistant coach. I entered the profession in 2026, in the sports department of a television station in Belgrade. Since then, I have covered 8 Olympic Games, 8 World Cups, and many editions of the Giro d'Italia and the Tour de France. Cross-discipline experience taught me something football sometimes forgets: every sport has its own data layer, and every layer has its own kind of gap. In cycling, power data is almost impossible to fake, so teams learn to stay silent when a meter is missing. In football, where the data still has many holes, we have not learned that silence. There is a hidden side rarely discussed. The more data there is, the more pressure there is to conclude. A dense table of numbers creates the feeling that the answer must exist somewhere inside it. When it does not exist, when the feed fails, when the sample is too small, when a variable is mislabelled, that gap is not left empty. It is filled. I have sat in rooms like that. And I noticed one thing: data gaps are rarely openly admitted. They are wrapped in language. In my feeling. Look at the shape of the game and it is obvious. The whole team is playing with a different focus. Those sentences are not wrong rhetorically. They merely pretend to have a basis. This industry has yet another layer that makes the problem worse: the media cycle. The Premier League adopted VAR from the 2026-20 season; Serie A and the Bundesliga from 2026-18. The technology was introduced to reduce errors, but it also produced a side effect: every decision now comes with an image, and every image comes with an explanation. When the image is not clear enough, the explanation is still given. The gap is filled again, this time by a referee, a commentator, or a line-drawing algorithm that is itself guessing. I want to separate that moment into four layers, because I believe it is a mechanism that can be recognised, not an isolated accident. At the human layer, the brain cannot tolerate a gap. When a fact is missing, it automatically fills it with the nearest, most familiar, most easily imagined fact. In my profession, the clearest expression is the scout who does not watch the match. A person is asked to report on a player but has only seen two highlight clips. He still submits a report. The report does not say I only had two clips. The report talks about positional awareness and fighting spirit. That is the nearest fact pushed into the gap by the brain. At the model layer, a forecasting system is only as good as its input. When the input is empty, a decent model returns null, and that is correct behaviour. The problem is that in football, most models are not machines of that kind, but people with authority. And people with authority rarely return null, because returning null is seen as weakness. I have witnessed meetings where an assistant presented a tactical conclusion based on a five-match sample, while nobody asked how large the sample was. The numbers were not wrong. The way they were used was the flaw. This is the point I want to stress: the danger of modern football analysis lies not in a lack of data, but in thin data being presented as thick data. A conclusion drawn from five matches, if spoken in the tone of a conclusion drawn from fifty, will go straight into a transfer decision, into a starting line-up, into a three-year contract. At the market layer, transfers are where data is bent hardest, because that is where there is money, agents, and deadlines. An agent does not need correct data; he needs a story plausible enough to inflate a price. When a player is advertised with an exceptional progress metric, my first question is always: how many matches in the sample, and who provided it? If the answer is a source close to the situation, I treat it as empty data in packaging. At the broadcast layer, the gap is filled with something else: sound. With no images to hold onto, the commentator leans on his voice to keep the rhythm. That is exactly where I learned my lesson. A passage of play can be painted with three adjectives without a single fact. And the listener, trusting the tone, will remember the adjectives rather than the truth. A few seconds of silence, in this profession, is worth more than an empty sentence of commentary. And here is the layer I impose on myself, after the Saint-Denis night. I call it the minimum-information gate. Before writing anything, I ask myself: do I have at least one data point and one identified entity? If not, I do not write. I return null. The phonetic table of 200 players I built after the Toivonen mistake is exactly such a gate, but at the language layer: a name is only spoken once it has been verified. Let me tell one specific case, because I believe in examples more than in bare principles. In 2026, I watched an Atalanta versus Juventus match in Serie A. Gian Piero Gasperini's side made me stop. Not because they won. But because of the way they cut off every pass out of Juventus's defence. I sat and measured: around 62 high presses in 90 minutes. Nobody in the French media noticed. I wrote a 3,000-word analysis of what I called zonal pressing defence, and sent it to two editors. It was published. A channel invited me on air. I declined, to spend the time extracting movement data on Atalanta's 11 players across five matches. Why did I decline? Because I did not yet have enough data to say something I would have to defend. One match gives me an observation. Five matches give me a sample. And I need to know whether that sample repeats before I call it the essence. Since then, I use heat maps and pressure metrics in every tactical analysis. But I do not use them for decoration. I use them to know when I am not yet allowed to conclude. And I drew a line I still repeat: Atalanta does not press, they read the opponent before the referee blows the whistle. The difference between pressing and reading is not in the words. It is this: one is a label, the other is a measurable behaviour. Here the story touches the most sensitive part of my profession: prediction. In early 2026, when football was suspended by the pandemic, I had spare time I normally do not have. I went back over Marco Verratti's passing and noticed a detail: PSG lacked a genuine defensive midfielder for the match against Dortmund. When the competition resumed, I wrote three pieces warning about the gap between the two centre-backs whenever Marquinhos pushed up. PSG reached the Champions League final. They lost 0-1 to Bayern. The goal came from exactly the gap I had sketched in my June article. Colleagues began calling me a tactical prophet. I do not like that label. Not out of modesty. But because it misunderstands the nature of the work. I do not predict the future. I line the details up in a row and let them point to the breaking point themselves. Football has no luck, only details that have not yet been put in order. When someone calls me a prophet, they are assigning me an ability I do not have, and at the same time, they are forgiving themselves for not having put the details in order. And here is the part I must say, even though it is not pretty: after the nail-biting PSG call, I clearly felt the pressure of a track record. I wanted to recount the times I was right. I had little desire to recount the times I was wrong. I had to build a public prediction-tracking table and, at the end of each season, reconcile it myself. Without that table, I would slide toward being someone who publishes only the predictions that came true, that is, toward a liar with selective memory. A prediction is only worth something if it is verified, even when the verification is failure. There is another paradox I want to raise, because it shows the problem lies not in the data but in discipline. There are conclusions that the data genuinely supports, and we still ignore them. Fixture congestion is one example. No medical staff can save a team playing two matches a week across a whole season. This is measurable, repeatable, and verifiable across many seasons. And yet the common reaction is still to blame fitness or mentality. We fill the gap even when that gap has already been filled with real data. And there is a field where the gap is filled even more dangerously than in traditional football: esports. There, betting is eroding competitive integrity faster than in any traditional sport, simply because the regulatory system lags behind the speed of the market. When the legal framework is slow, people do not wait. They build their own story, believe it, and act on it. It is the same mechanism: an empty dataset, plus a reflex that refuses to stay silent. Here I want to go against a common belief in analytical circles. The common belief is: the more data the better, and a lack of data is the biggest problem. I think the order of priorities is reversed. The problem greater than a lack of data is thin data being allowed to produce a firm conclusion. An empty dataset is harmless, because it forces you to stay silent. A thin dataset is dangerous, because it gives you just enough material to fool yourself, and just enough credibility to fool others. I call this the trap of false information gain. Today's search algorithms reward content that offers a new angle. But new does not mean true. An article can give you an angle you have never heard, and that angle is built on three matches. It is new. It is also empty. The result is a paradox in my profession. The most honest analyst is usually the one who says I do not know the most. But I do not know does not sell airtime, does not make the front page. So the industry's incentive structure pushes the most careful people into silence, and the most confident people into positions of authority. That is why I built a personal rule: end every piece with a verifiable judgement, with a clear statement of the sample and the conditions of application. If I am not willing to bet on what I have just written, I have not finished writing. And I must admit one more temptation, subtler still. Probability is a safe zone. When I say likely, I cannot be proven entirely wrong. That is a way of dodging responsibility dressed up in scientific language. I have reminded myself many times: do not hide behind probability. Make a judgement specific enough to be refuted. An analyst who cannot be refuted is an analyst who says nothing at all. I once got a person's name wrong, but I have never got the essence of a match wrong. I want to keep that line as a professional oath, not a boast. Because between its two halves there is a distance: the first is a mistake I can fix with a month of tape review; the second is a commitment I must pay for every time I go on air. So what will separate the clubs in the coming cycle? I am betting on one possibility, and I state clearly that my sample is professional observation, not a quantitative study: the clubs that build a discipline for handling gaps, that know how to return null when the data is insufficient, will outperform the clubs that merely race for data volume. Because in a market where everyone has the same set of numbers, the advantage no longer lies in owning the data, but in knowing when it may not be used. When a team wins, I look at the bench before I look at the goal. And when a team loses, I look at the dataset before I look at the defence. Because a match, like a report, is only as honest as what was put into it. And you: the last time you reached a firm conclusion, did your dataset actually exist, or was it just a blank screen you had filled with your own imagination?