Football analysis is inventing its own facts — and nobody rewinds the tape to check

In 2026, midway through an El Clásico, I switched off the live broadcast, rewound the tape twelve times and measured the ball's angle of travel with frame-analysis software. Camera A and camera B were 1.7 metres apart — enough for a Real Madrid goal to be confirmed by VAR when in reality it was not valid. My article spread to more than 200,000 shares within a day. But what kept me awake was not the figure of 1.7 metres. It was that millions of people had believed a conclusion built on corrupted data, and almost none of them bothered to rewind the tape and check for themselves. The television screen does not lie; only the person sitting behind it fools himself.

Today, every morning, thousands of "deep analyses" pour into the phones of Vietnamese fans. A two-thousand-word piece on how a big club's back four is hiding weaknesses on both flanks. A heat map of running distance. An xG table redrawn by hand in three colours. Readers read, nod, and share. But very few stop to ask one single question: where does this data come from?

This is the transfer window — the phase in which noise completely drowns out signal. Every hour brings ten more rumours, every rumour brings five more sources, and every source cites another source. News sites race for quantity of articles, not accuracy. Analytical models are trained to generate text that sounds professional, regardless of the input. And when the input is empty — when there is no data, no player, no match — the system still has to produce an output. So it invents.

I once saw an internal analysis from a sports-data company. Its input — the source document used for analysis — was completely empty. No title. No information. No entities. No time. No source. Yet the output was still a full nine-dimension framework: tactics, club finances, match results, league landscape, rules, dressing room, risk, media, and industry transmission. Every cell was filled in fluently. And if you did not read carefully, you would not notice that each cell actually said only one thing: "insufficient information".

That is the crux. A system designed to always answer will always answer — even when the most correct answer is silence. Modern football loves numbers, but numbers do not know fear. They feel no shame when they invent. They only know how to fill an empty cell.

I saw this coming. Not because I am a good prophet, but because I keep a notebook of failure patterns that repeat across World Cups. In 2026, when every expert was praising Pogba and Griezmann, I published a series pointing out that France's defence with Varane and Umtiti was hiding a weakness at full-back. Before the round of 16, I built a fourteen-minute video showing that Pavard needed to drop three metres deeper to neutralise Messi. The Argentina national team's coaching staff photocopied my article as meeting material.

Why was I right? Not because I had more data. But because I looked at the static structure — how the blocks of players stand still before the ball rolls — instead of trusting numbers generated afterwards. The defence tells the story before the attack. The distance between the lines, the retreating rhythm of the back four, the average position of the whole block — those are pieces of evidence that cannot be invented, because they live in the tape, not in the model.

By contrast, what I call "sourceless analysis" has three layers. Layer one is the origin: a rumour, a tweet, a sentence cut out of context. Layer two is the model: it receives a thin input, then pumps in probability of its own. Layer three is the output: a confident article with numbers, names, and a conclusion. The reader only sees layer three. They do not see that layer one is empty.

In the transfer window, the real story is the structure of release clauses and the wage bill. But nobody wants to read about release clauses, because it is boring. They want to read that a star "has agreed personal terms". That phrase appears hundreds of times a week, and in most cases it means nothing beyond an agent wanting to apply pressure to raise the price. I have watched enough transfer windows to know: noise is not information. It is only noise packaged more attractively.

Live data supplied to betting companies is the darkest side effect of the digitisation of sport. At the same time, shirt advertising is destroying the thread between clubs and their local communities — global sponsors care only about exposure ROI, not about the stands. And that is why I still keep my notebook.

VAR was created to fix human error, but in the end it creates machine error. The same logic applies to analysis: the tools were created to reduce the error of the eye, but in the end they create a new kind of error — the error of confidence. A referee who errs is punished. A model that errs is punished by no one, because no one knows it erred. It simply quietly produces the next article.

I may be wrong here. There is a serious argument for the data camp: models see things the eye misses. A lower PPDA can reveal high pressing that a single live viewing cannot capture. A metric like xG can show that a team won through luck, and that will hold true over the long run. I do not deny that. Every season, I spend twenty minutes reading an analytical report written by a young person, just to remind myself that my eye has blind spots too.

But there is a difference between a model and a mirror. A model with a source is a window — it lets you see what is out there. A model without a source is a mirror — it only shows you the writer's own assumptions. The problem with modern football is not too much data. It is too much data with no origin, presented with the same confident tone as real data.

If a model can write fluently about a match that never existed, then it can also write fluently about your match. The question is not whether you believe it. The question is: when did you last open the tape and check for yourself?