A Crime Report Labeled Football: When the Data Pipeline Deceives Itself

Thursday, October 1. I opened my work queue at seven in the morning, Beijing time. The first item carried the label 'football.' I clicked in and found no match. No club, no player, no coach, not a line of tactics. There was a secondary school in Torreón, Coahuila, Mexico; a vice-principal killed; three people injured; two eighteen-year-old former students detained. I sat still for a while. This is the moment on which my profession lives or dies. A label says 'football,' the content says 'crime.' There is no gray zone between them. What happened in Torreón is the prosecutor's business. My business is where this label flows next, and who at the other end will believe it. Empty stands still have noise — that is the noise of wrong data. Before anything else, I have to say this: the source report was written carefully. The incident took place at Escuela Secundaria Número 13, in ejido La Unión, in the city of Torreón. The report states the victim, the number of injured, and the legal status of the two suspects — detained and placed at the disposal of the prosecutor's office. It leaves the motive open, leaves open whether a weapon was used, and preserves the presumption of innocence. As a piece of journalistic craft, it is clean. The problem is not in the report. The problem is in the line my pipeline attached to it: 'Domain Label: football.' Not a single football entity exists in the text. No league, no transfer, no financial fair play, nothing at all. There is only a label. And a wrong label is more dangerous than an empty field, because with an empty field people know they must go looking, while with a wrong label they think they have already found it. This is the transfer window, and in the transfer window noise always outruns signal. Readers are drowning in rumor. What they actually need is a reliability filter. A filter is only worth something if the label at the input can be trusted. And the label at the input has just handed me a murder case in Mexico. Misclassification rarely comes from ignorance. It comes from convenience. An automated tagger does not read the article; it counts entities. It sees 'Torreón,' 'Coahuila,' 'Mexico,' and somewhere in a database a link says that Torreón is home to a professional football club. So a school homicide becomes 'football.' Nobody meant to do it. It is just a skewed entity match, and a label is born. The frightening part lies elsewhere: the label carries authority. In our pipeline, an item tagged 'football' automatically enters the football analysis stream. It is not questioned. It is not cross-checked. It simply flows. And if the analyst at the far end is not lucid enough to stop, the pressure of the template will force him to fill it in: tactics, club finance, the transfer market, financial fair play. He will invent a club just to satisfy the template. That is the moment a small error becomes a large lie. I have seen it happen, not with a homicide, but with far smaller things. A transfer rumor with no source tagged 'confirmed.' A three-match stat sheet tagged 'season form.' The label always walks ahead of the facts, and the facts must run behind to justify the label. Numbers do not know how to lie, but whoever picks the numbers does — and whoever applies the label does too, even unconsciously. In the analysis I received, one phrase repeats again and again: 'insufficient information, cannot assess.' To an outsider, that is a weak answer. To someone in my trade, it is the strongest answer there is. It means: I would rather leave a cell empty than invent a number. The football analysis template has nine parts — tactics, club finance, results, league landscape, rules and governance, the dressing room, risk profile, media narrative, transmission chain. A crime report fills none of them. The good analyst is the one who dares to write 'cannot assess' nine times, rather than invent nine sentences to look useful. One detail shows the system contradicted itself from the start. The analysis labeled the item 'football' yet spent almost all its length saying the content is not football. It listed nine professional dimensions, then filled all nine cells with the same line: insufficient information. That is a pipeline that recognizes its own error and still does not stop, because the template must be filled. A system that knows it is wrong and continues anyway is worse than a system that knows nothing. Think in probabilities. If a tagging pipeline mislabels one percent of items, and each day it processes ten thousand items, then each day one hundred wrong items flow into the system. None of them does harm on its own. But each is a brick. After a year, you have thirty-six thousand bricks building a wall nobody ever drafted. When the wall is tall enough, people begin to cite it as a source. That is when wrong data becomes old data, and old data becomes accepted truth. One detail in the deconstruction made me stop. It held nineteen information points, and many carried the line 'Source: None' — no source. For a public-safety report, that is acceptable up to a point, because details such as the number of injured often come from the scene before the prosecutor speaks officially. But for a data pipeline, it is a red flag. An item with no traceable origin cannot have its reliability assessed. An item whose reliability cannot be assessed should not be reused. And an item that is both unsourced and mislabeled has already traveled too far to be saved. This leads to a larger question: if this content had been routed correctly, where would it go? It belongs to the public-safety stream, and there it would be treated as an incident under investigation, not a closed event. Public-safety readers want to know the motive, the condition of the injured, and the legal developments. Football readers need none of that. Blending the two streams teaches no one more about football, and teaches no one more about the case either. Drawing on my experience watching matches, I always ask two things before believing anything: how large is this sample, and how long does it run? I learned that from two failures of my own. In 2026, at seventeen, I wrote a prediction that Germany would beat Mexico 2-0, based on head-to-head history and the pedigree of champions. Mexico won 1-0. I had ignored their high-pressing data — nineteen presses in the attacking third in the first half, double Germany's average. I believed the label 'champions' instead of reading the facts. Two years later, when the Bundesliga returned to empty stadiums, I measured the home win rate fall from 43 percent to 36 percent. A lecturer objected that nine rounds is too small a sample. He was right to demand caution, and I answered by cross-referencing five previous seasons of data. Both times taught me the same thing: a label is never evidence. The most predictable reaction is to blame the algorithm. Fix the entity filter, add exclusion keywords, rerun everything. Done. But that looks at the wrong place. The fault is not in the algorithm. A tagger will always be wrong at some rate; that is its nature, just as a striker always misses some share of chances. The fault is that we design the pipeline on the assumption that the label is the truth. We build no checkpoints. We never ask what error rate is acceptable. We simply believe. And here is the sharpest irony. The source report is a model of good journalism. It preserves the presumption of innocence. It assigns no motive before confirmation. It builds no hero-versus-villain story, only correct and incorrect facts. The media sells dreams; I sell the dressing-room record — and this report, though from an entirely different field, lives up to that standard. Yet it was turned into 'football' by a label. The fault is not the article's. The fault belongs to the system that read it. History is a reference, not a verdict — and a label should be the same: the starting point of a question, not the ending point of a conclusion. The signal I track next lies elsewhere. The Torreón case is the business of the Coahuila prosecutor's office, and I have no authority to speak on it. The signal I track is the misclassification rate inside my own pipeline: how many items that are not football are carrying a football label, and how far they have flowed before someone stops them. One wrong label kills no one. But a thousand wrong labels, flowing through a thousand decisions, manufacture a false truth whose origin no one remembers. When people forget the starting point, they will defend the label the way they defend a belief.