A Football Tag With No Football Inside: The Crack in Sports Data

2:40 a.m. in Marseille. I open a data file with fourteen points inside. The label is explicit: football. No club. No league. No player, no coach, no scoreline, no stoppage time. Only a list of reality-show finalists, a briefcase holding a cash prize, a gala about to air, and one influencer's remark about one singer. I read it three times. Then I understood: the problem is not the record. The problem is the label. That label is a lie, and it is not alone. Every day, thousands of records like it flow into football databases, into prediction models, into the bulletins you read over breakfast. We are building the house of football on sand made of bad tags. CONTEXT: A COMPETITION WITH NO FOOTBALL IN IT The source story, plainly: a Spanish-language article about the closing stretch of a reality show. The finalists' list was almost complete, and it included Yahír, a singer. Ese Pérez, an influencer, was said to have told Gema whom he thought should reach the final. He named La Guardia (Ernesto) and Mariana, with a merit-based reason: she has entered the game. Karina Torres was also a finalist. The headline asked a question: does Ese Pérez envy Yahír? The body recorded, verbatim, his denial of envy, followed by a distancing remark: he did not want to generate bad vibes. That is the entire raw material. And someone, either a machine or a human working too fast, pasted a football label on top of it. Absurd on its face. But look at the structure. A competition. A large prize. A group of people who reach the last round. An elimination mechanism. A final night. Rename the parts and you have a season: the league is the competition, the trophy is the briefcase, the finalists are title contenders, elimination is relegation, the gala is the last matchday. The schema fits so neatly that an automated classifier has no reason to hesitate. Here is where I stop: the error is not sloppiness. The machine was built to recognise the shape of a contest, and football is only one of many things that share that shape. We check content. We almost never check shape. THE CORE The machine does not read football, it reads schemas Fourteen data points. I counted again: not one football entity, no club, federation, league, player, coach, sporting director, contract, transfer, tactic or scoreline. The only monetary figure is the show's prize. Yet the label still says football. If you run a database, you know what happens next. The record passes the first gate, because that gate asks a single question: is this a competition? It enters a training set, or a feed, or an aggregate index. And there it teaches the system something false: that a spat between two celebrities is a form of sporting competition. The headline asks, the body denies This is the part that worries me most, and it belongs to my trade rather than to machines. The headline asks about envy. The body records a denial of envy. Technically the two do not contradict each other; people are entitled to ask questions. Functionally they do: the headline manufactures what the body has just dismantled. The entire Perez-Yahir friction rests on a single quote, plus a conversation with a third party known to us only through description. One sentence. That is the whole sample size. And when I read the quote itself, it is comparative rather than hostile: he names two people he prefers, with a reason grounded in performance. That is a judgement about work, not a personal strike. I have worked in this trade for thirty-one years. I have seen thousands of headlines like it in football. Star X is unhappy. Body: the agent declined to comment. Club Y is about to lose a key man. Body: another outlet said it might. The machine did not invent this journalistic move. It learned it, duplicated it, and dropped it into a file tagged football. What I learned while being laughed at In late 2026, at thirty-eight, I wrote that Monaco would collapse after selling Mbappe. I cited the numbers: in 2026-17 Monaco scored 107 goals in Ligue 1, with Mbappe contributing 15 plus a string of decisive assists. My contrarian call was mocked for three months. Then Mbappe moved to PSG for 180 million euros. The piece was shared more than 50,000 times. I am not telling that story to flatter myself. I tell it because it explains why I hold to one rule: data takes me to the stadium gate, my eyes take me into the dressing room. A spreadsheet told me Monaco scored 107 goals. It did not tell me what that team would look like without the man who set the rhythm. To know that, I had to be at the ground, watching players' runs, watching how the bench reacted after every dead ball. That is exactly why the record irritates me. It has the shape of a competition but not one second of field observation. It is an empty shell with a tag on it. In 2026, in Moscow, before the Croatia-England semi-final, I wrote that Croatia would reach the final because Modric and Rakitic would control midfield. I sat and watched them circulate the ball, stretch England's shape, and I called Croatia 2-1 after extra time. The piece reached two million views and I was invited onto national television. The lesson was not that I am clever. The lesson is: a shock opinion is only worth something when it stands on a detail everyone else skipped. If I had simply said Croatia will win without showing where Modric received the ball or which gap Rakitic sealed, that would have been a guess, not analysis. One funnel, two labels Reality television: producer, broadcaster, social engagement, contestants' personal brands. Revenue from advertising and sponsors. Conflict lifts viewing figures. Football: club, broadcaster, aggregator, your feed. Revenue from rights, shirt sponsorship, global partners. Conflict lifts viewing figures too. The two funnels share a shape, a motive and a reward for pushing emotion upward. That is why I believe the mislabel is no rare accident: football and reality television differ in content but increasingly resemble each other in economic form. When two things share a shape inside an automated taxonomy, they will be confused. The question is no longer whether confusion happens, but how much of it there is and who catches it. The cost does not fall on the celebrities In that story, someone gets branded the jealous one, and someone benefits from sympathy. Perez had to issue a soothing statement, a textbook response showing he knew what was coming. Yahir is barely damaged. The production is neutral, because noise is good. But the cost exists, and it falls on none of them. It falls on the record. The narrative heat is high while the informational content is close to zero. If I score it the way I score football data daily: reference value one star out of five; timeliness three out of five, because the gala is imminent so the window is measured in days. No outlet, no author, no publication date, which means it cannot be verified against anything. A record like that does not correct itself. It sits there, waiting to be counted in some table nobody re-sources. Three questions to screen a record Based on my experience tracking and cross-checking match data, I use three questions whenever a record enters my table. First: is there at least one verifiable football entity, even just a club, a league, a player or a federation? If not, stop. Second: does the source carry an outlet name and a publication date? If not, drop the reliability rating to its floor and exclude it from every aggregate. Third: does the headline assert something the body never proves? If so, take the body as the fact and discard the headline. The record fails all three gates. THE CONTRARIAN ANGLE Where could I be wrong? First, this might be small. One record among millions. A wrong tag gets fixed, done. If the system already had a domain gate, requiring at least one valid football entity before progressing, this would be a single line in a log. I do not buy that reading, because a classification error does not behave like a typo. It does not vanish. It spreads into filters, then into training sets, then into composite indices. Three hops later, nobody can trace the origin. Second, I might be inflating this. A headline that says one thing while the body says another has been tabloid craft for a century; there is nothing new in it. True. But there is a difference in scale and in who audits. Once, an editor had to sign that headline. Now a pipeline pushes it through and nobody signs. Third, and this is where I distrust myself most: I like a strong verdict. It is how I make a living, and it is also my weakness. When I see a mislabelled record, my first reflex is to turn it into a crisis. That may be right. It may also be that I am using one scrap of data to tell a story bigger than itself. So I set myself a test that can be checked. If, over the next twelve months, no sports data provider publishes a domain-verification gate, my hypothesis weakens. If one does, it strengthens. And if all I have is a single record, I must say plainly: my sample is too small. I will still say it. This is where I part with the crowd: I would rather issue a verdict that can be rejected, and be rejected, than stand in the middle and say nothing. TAKEAWAY My prediction, checkable: within eighteen months, at least one major sports data provider will have to withdraw or revise a public index because its inputs were contaminated by non-football content. When that happens, they will call it a data-quality issue, not by its proper name: a domain-classification failure. And the question I leave behind: if a record with no club, no player and no scoreline can pass through a football gate, how much else has gone through that gate that you currently believe? The match is decided where the crowd is not looking. This time, that place was a label column.