When Football Data Mislabels Itself

Last Tuesday I sat in front of a file containing several thousand content items, all of them labelled "football" by an automated classification system. My job was nothing grand: skim through, check whether the label matched the body text. Most of them went by quickly — a V.League match, a transfer story, a league table for the annual season. Then my hand stopped on the scroll wheel. The label read, plainly: "football". The body text told of a family incident in Guadalajara, in the state of Jalisco, western Mexico. The system had extracted sixteen information points, and across those sixteen points there was not a single team, player, coach, competition, passage of play, or metric belonging to this sport. Nothing. I read it three times — a habit I imposed on myself after misspelling the names of three Croatian players in a long analysis piece and being reminded of it, loudly, by the internet. Then I sat still for a while. The first feeling was not anger. It was confusion. And I have learned to trust that feeling. I began from confusion at the 2026 World Cup, and it turned out to be the only way to understand a match. I was twenty-two that year, still in my final year of university, dissecting how Croatia ran a 4-3-3 with Luka Modrić and Ivan Rakitić. I got a great deal wrong. But every time I noticed I did not understand something, I moved a little closer to actually understanding it. That mislabelled item was the same. It did not give me an answer. It gave me a question. The question was this: what happens when a football data system lies to itself? Most sports content platforms in Vietnam today — news sites, score apps, and the metric tables some outlets use to rank squads — run through a layer called topic classification. That layer does not read an article the way an editor does. It scans keywords and entities: competition names, club names, player names, place names. It sees "Premier League", "V.League", "Messi", and tags it football. It sees "Guadalajara", and tags it football too — because in its training corpus, "Guadalajara" almost always comes attached to a club. That is the first blind spot. A place name is not a club. Guadalajara is Mexico's second-largest city, the capital of Jalisco, and also the name of a team. In a sportswriter's head, those two things can be kept apart. In the head of a probability-based classifier, they overlap perfectly. The second layer is entity extraction. The system looks for names of people, organisations, numbers. It cannot distinguish a name in a football article from a name in an article about someone's private life. It only sees a string of characters matching a pattern. And because that item already carried the "football" label, it was pushed down exactly the pipe reserved for football. The third layer — the one I think about most — is the analysis layer. The very layer I was sitting inside. When an item walks into the analysis room with a "football" label stuck to its forehead, the analyst has a professional instinct that is very hard to resist: to find football meaning in it. I tried doing exactly what I do every weekend. I broke that item into eight analytical dimensions. Tactics and technique. Club finance and the transfer market. Results and the cycle of public opinion. League landscape and team positioning. Rules and compliance. Management and the dressing room. Risk profile. Media and expectations. All eight came back with the same result. Insufficient information to assess. For a writer, that is the greatest temptation. A blank page always calls on you to fill it. There is a city with a club. There is an age mentioned. There is a story. One could quite easily start weaving an argument that sounds plausible, sounds "expert", about something that does not exist at all. I know, because I have done it. Not because I wanted to deceive anyone, but because I was afraid of silence. But there is a principle data analysts still use: when there is no information, write plainly, "insufficient data, cannot assess". It sounds simple. Practising it is hard, because it forces you to accept that sometimes reality is smaller, duller, less exciting than what you could have invented. What is true here is simple and also rather blunt. Across those sixteen information points there is no football. No tactician is named. No transfer fee is exchanged. No table is affected. Any conclusion about a Mexican club, or about a particular human being, would be the product of imagination rather than of footage. And the consequences do not stop at one article. When a mislabelled item enters a system, it does not vanish. It flows downstream. A league's sentiment index absorbs it. An entity graph links "Guadalajara" to a club. An automated aggregation tool may pick it up and republish it in a football section. Months later, a reader looking for information about the weekend's match will meet an article with nothing to do with it, and may believe that it does. The annual season is a season of labels applied at breakneck speed. Every matchday, hundreds of articles pour in: previews, post-match statistics, transfer news, referee analysis. No newsroom in Vietnam has enough people to read each one by hand. Automation is not a choice; it is a condition of survival. That is exactly why a wrong label is more dangerous than we think: it is not a rare exception, it is the inevitable by-product of a system running faster than any human can follow. Based on my experience following matches, a wrong label is like a player flagged offside before the ball has even been passed. The whole system runs on a false assumption, and by the time it notices, it is one beat too late. In football, one beat is enough to concede. In data, one wrong beat can survive for years. Take an example from history. In 2026, Manchester United beat Bayern Munich 2-1 in the Champions League final, with two goals in stoppage time. If a system recorded only the final score and ignored context, it would say United won comfortably. It would skip the fact that Bayern led for almost the entire match, skip the three substitutions in the last ten minutes, skip Sir Alex Ferguson reading the game at the right moment. A conclusion that is correct in its numbers can be wrong in its essence. The "football" label stuck on an article with no football is that kind of error — except it is wrong from the very start. Or take Euro 2026. Denmark lost Christian Eriksen in the opening match, and Kasper Hjulmand switched from a 3-4-3 to a 3-5-2, taking the team to the semi-finals. If someone only read the line "Denmark reach the semi-finals" without rewatching the footage, they would not understand why. Context is not decoration on data. Context is data. Return to the mislabelled item. Here there is no football context to understand. There is a human case, real and serious, and it was dragged by mistake into a trough it does not belong to. That was when I realised the problem does not lie in the label. The problem lies in the fact that the pipe is designed to always produce an answer. There is no room for silence. Such a system, by its nature, never says "I do not know". It simply picks the nearest approximation and presents it with certainty. That is the language of an expert who believes he has seen everything — the language I try to avoid in my own writing. A tactical diagram is not an answer, it is only a way of asking a question about space. A topic label is the same. It is only a way of asking a question about content. If we turn it into the final answer, we have lost the very thing we need. There is one more thing, and I want to say it plainly. That item referred to a named private individual and described acts of violence. Material like that needs its own place in a content system — a safety gate where sensitive content is held back, not aggregated automatically, not optimised for engagement, not put into a ranking. I did not see that gate in the pipeline I was checking. That is a more serious flaw than the labelling error itself. Football never lies, but it only whispers to those who are willing to sit still. Data is the same. It is only trustworthy to those who stop and ask: is this label actually right? People usually think the problem with data is a lack of data. In this case, I would argue the opposite. The problem is an excess of data and a shortage of caution. A classifier fed on millions of articles will be very good at finding patterns, and very poor at recognising that there are moments when no pattern should be sought. The better it is at finding patterns, the more easily it manufactures a pattern that is not there. That is the paradox of every powerful system: its strength and its blind spot are the same thing. Every pass is a choice, every press is a refusal. A labelling system also makes thousands of refusals every second — it refuses the possibility that "Guadalajara" might be a city, that an article might not belong to football, that the correct answer might be "I do not know". Those refusals are recorded nowhere. And what is not recorded cannot be fixed. In Vietnam I learned that a team can play with its heart before it plays with a formation. I think data is the same. A sports data system is only trustworthy when it is built on respect for people before it is built on algorithms. Because behind every line of data — even the wrong ones — there is always a real person. An empty stadium stripped bare the things we thought mattered, leaving only the sound of boots on grass. I once thought I understood that line during the pandemic years, when the leagues paused and I sat reconstructing classic matches. Now I understand it differently. When all the noise is scraped away, what remains are the most basic things: a name spelled correctly, a number with a source, a decision that knows when to stop. That mislabelled item taught me this. It is not a lesson about football. It is a lesson about how we treat the truth when the truth does not fit the mould we have already cast. I no longer believe in victory; I believe in the moments a match opens itself before me. The moment I want to keep from that Tuesday evening is not when I found the error. It is when I decided not to fill the gap with anything at all. I left it empty. I wrote a single line in my report: "Wrong label, content does not belong to football, please reroute for handling." Perhaps the right question for those building sports data in Vietnam is not how to make the system smarter, but how to make it know when to stay silent. Because a system that does not know how to stay silent will one day say something false about a real person, and no algorithm will ever be able to undo that.