Mislabeled and Its Cost: When Football Data Gets Contaminated

In my study in Nha Trang, I opened a data file that the system had labeled "football." There was no team inside. No player, no coach, no match, no goal, no contract, no federation. All I received were thirty information points about Ring Road 3.5, Me So Bridge, Ngoc Hoi Bridge, the Hanoi–Hai Phong Expressway, National Highway 5, metro lines 8 and 14, and a low-rise housing project in East Hanoi. The whole file was about real estate. The label said football. The two never met at any point. I sat still in front of the screen for a long while. For someone who has spent 41 years observing the industry and nearly a decade working with movement data, errors in data are nothing new. But this was the first time I encountered an error of classification itself: content from an entirely different field filed into the football vault. What made me stop was not the error itself, but its consequence. A single bad data point, if unchecked, will quietly flow into every aggregate, every model, every conclusion built from that vault. And in football, where people are used to trusting goals faster than they trust process, such an error can live a long time without anyone noticing. This story does not lie in a real-estate project. It lies closer to my trade: the way Vietnamese football is learning to use data. Over the past seven or eight years, V.League clubs have begun hiring analytics firms, buying movement-data packages, installing cameras to track players. Youth academies have begun recording physical indices. Sports media have begun citing numbers as a way to build credibility. An entire young ecosystem has formed, and it has grown faster than its speed of learning to check itself. When an industry has just acquired data, people tend to celebrate having data before asking whether that data is clean or dirty. I was once in exactly that place. In 2026, when I received the 14-movement-metric dataset of 22 players in the Sanna Khanh Hoa – Hanoi FC match, I was stunned to see for the first time what the naked eye misses. Hanoi FC held 68 percent possession but managed only 4 shots on target; Khanh Hoa won thanks to 18 high-pressing actions aimed at the opponent's left-back. My 2,500-word article that followed drew more than 100,000 views, a figure unprecedented in my twenty years in the trade at the time. I thought I had found the path. I did not yet know that the path is only safe when the ground beneath it is packed firm. That ground is data quality. A football data vault is not correct by nature. It is correct when each item is labeled accurately, sourced, and classified into the right field. When an article about Ring Road 3.5 and metro line 14 slips into the football vault, the fault is not in the article. The fault is at the gatekeeper — the classification process that let it through. That process, in principle, has several layers. The first layer determines which field the content belongs to. The second determines which topic within that field. The third determines the reliability of the source. An article about real estate slipping into the football vault means the first layer has failed. And when the first layer fails, the other two become meaningless, because they are classifying and rating something that does not belong where they stand. There is a paradox I realized after many years of work: the more data there is, the more readily people believe. It should be the opposite. When I worked mainly by eye, every judgment carried a natural suspicion, because I knew my eyes could deceive me. When data arrived, that suspicion vanished, even though data can deceive just as much, only in a harder-to-see way. This is where the story touches the transfer window, a period in which the whole football world lives amid noise. Every day brings hundreds of rumors, dozens of names, countless transfer-fee figures thrown out without a source. If a single bad data point can slip into an analytics vault unchecked, then hundreds of unsourced rumors are slipping into fans' heads in exactly the same way. The only difference is that the analytics vault has a process, while fans' heads do not. What is worth noting is that this classification error is not rare. It is only rarely seen. When I built the 47-Situation Code — a system classifying attacking, defensive, and transition phases, numbered 01 to 47 — I learned something I later applied to data as well: one wrong label drags a whole chain of error. Code 23 is "counterattack after losing the ball in the opponent's third." Code 35 is "offside-trap pressing in midfield." If I misassign a passage of play to Code 23 when it is really Code 35, then not only is that passage wrong. Every statistic on that team's counterattacking frequency is wrong. Every conclusion about whether that team counterattacks much or little is skewed. One wrong label at the source, one wrong conclusion at the end. With a football data vault, the mechanism is identical, only the scale differs. An item labeled "football" while its content is real estate will corrupt every count. If someone tallies "football articles per week" from this vault, the number will be higher than reality. If someone trains a topic-classification model on this vault, the model will learn the error too. If someone compiles "prominent keywords of Vietnamese football," words like "Ring Road 3.5" or "metro line 14" may appear in the ranking — an absurdity only a careful reader would catch. I call this data contamination, and it is dangerous because it is silent. An injured player is visible to all. A sacked coach is read by all. A bad data point is seen by no one, until it has passed through ten processing stages and becomes a conclusion cited back as fact. By then, tracing the origin of the error becomes nearly impossible, because it lies scattered across dozens of intermediate reports. In football, the most common form of contamination is not in movement data but in sourcing. Many times I have opened a transfer aggregator and seen dozens of rows marked "source: unknown." A transfer fee without a source, a wage without a source, a release clause without a source — all are items mislabeled in a subtler way. They are not wrong in topic, but wrong in reliability. And when they enter an aggregate, they carry a weight they do not deserve. I rank information sources into four tiers. Tier one is direct confirmation from the club or player. Tier two is indirect confirmation through a reputable agent. Tier three is a report by a named journalist with an accurate track record. Tier four is an unsourced rumor. Most transfer content fans encounter daily is tier four, yet it is presented in the tone of tier one. The gap between the real tier and the tone is where contamination breeds. A young colleague once asked me why I never cite a single number on its own. I answered that a number standing alone is a meaningless number. It only means something when placed in the context of space, the coach's decision, and the person who executed it. That is why I built a four-layer analytical frame: data, space, decision, person. Drop any layer and the number loses its meaning. I was once "disowned" by data in the literal sense. In 2026, at the World Cup, I mispronounced the name Isco three times on television despite careful preparation of my notes. Viewers complained fiercely. That night I wrote in my diary: "I studied tactics for 20 years, yet I am judged for a single name." After the tournament, I spent a full month rewatching 52 matches, built a pronunciation notebook of 342 player and coach names, and created a three-step check: consult the official source, listen to native commentators, and record my own voice for comparison. I learned that a small error at the data-entry stage can destroy the credibility of an entire long process behind it. One wrong name does not just make one sentence wrong. It makes the listener doubt every remaining sentence. That is why I looked at that real-estate data file with different eyes. It is not merely a single error. It is a sign that the gatekeeper of the data vault is asleep. And if the gatekeeper sleeps on one item, it is very likely sleeping on many others in the same batch. A contaminated data vault does not contaminate item by item. It contaminates by batch, by wave, by the same systemic error. This brings me back to the transfer window. In mid-July, as V.League clubs race to complete their squads, every contract comes with a story. Some stories are confirmed by the club. Some are confirmed only by an agent. Some are confirmed by no one, only repeated by many sites. To fans, those three kinds of story look the same, because they all appear on the same timeline. But to a data worker, they belong to three entirely different reliability tiers. Mixing them into one table without distinguishing reliability is another form of contamination, subtler and more common. Before buying a player, I let him run three matches, and only then do I trust the offer. That principle applies to data too: before trusting a number, I let it pass three checks, and only then does it enter a conclusion. Many people's first reaction will be: one bad item is no big deal, just delete it. I used to think so. But after spending six months in 2026 rewatching footage of 200 European matches from 2026 to 2026 to record every repeating pattern, I understood something else. The problem is not one bad item. The problem is that no one knows how many bad items are sitting in the vault. When the pandemic halted global football, I fell into deep disorientation, and rewatching 200 matches was how I steadied myself. It was in that process that I realized the human eye misses a great deal, but data also misses no small amount — the difference is that data misses silently. The biggest blind spot of football data analysis is not a lack of data. The blind spot is believing that having data means having truth. A contaminated data vault still produces numbers that look highly professional. It still draws charts. It still prints reports. It still makes readers feel at ease because everything is quantified. That very ease is the most dangerous thing, because it makes people stop checking. In the transfer window, this trap runs deeper. Fans are drowning in rumors, and they need a reliability filter. But most of the filters in use are the very sources causing the noise. An account posting unsourced transfer news, an unverified roundup, a fee figure copied across five sites — all form a dirty data vault that readers willingly load into their heads every day. When something is repeated enough, it begins to look like truth. For someone in my trade, this is a stern reminder. I used to place data on the table and let it plead its own case. But I have come to understand that data pleads correctly only when it is clean. A mislabeled item does not plead — it lies on behalf of its creator. And its creator, in many cases, has no idea he is lying. Before trusting any aggregate, I ask myself three questions: who labeled this item, where is its source, and does it belong to its own field. Those three questions do not make me slower. They make me more right. Vietnamese football is entering a period in which data will only grow. The way we guard the data vault today will determine the quality of every conclusion tomorrow. A real-estate article slipping into the football vault is a small thing. But the habit of letting it through is a big one.