A 'Football' Record With No Football: When Labelling Breaks Sports Data

Across the 412 matches played without spectators that I logged throughout the pandemic, I once believed the hardest thing to control in football data was crowd noise. Strip it away, and what remains is a pure technical exercise, where every pass and every pressing action surfaces without distortion. That belief collapsed when I opened a record clearly tagged 'domain: football'. Inside there was not one team, not one player, not one minute of the ball rolling. It told the story of an actor, a few film studios and a string of cinema projects. Yet the label still read: football. What is striking is that this record was far from worthless. On the contrary, it was carefully structured, with clear entities, timestamps and causal links. Only one thing was wrong, but it was wrong at the foundational layer: it belonged to an entirely different field. For anyone who aggregates sports data, this is the most frightening class of error, because it does not make the data look sloppy. It makes the data look trustworthy, right up until someone actually reads it. Picture the journey of such a record. It leaves a newsroom, passes through an automated system, receives a domain label, and is then distributed to feeds, aggregation models and topic filters. Nobody along the way reads it with human eyes. The domain label becomes a certificate of trust. And so a piece of entertainment news quietly sits in a stream otherwise filled with articles on tactics, transfers and injuries. A wrong label does damage on three levels. The first is the reader: someone searching for news about their club suddenly reads a story about a superhero role. The second is the model: if a system counts entity frequency to infer trends, it will log names that have nothing to do with football and push them into a signal ranking. The third, and most dangerous, is trust: once dirty data slips through undetected even once, people begin to believe the data is clean. I once reconstructed the France versus Argentina knockout tie at the 2026 World Cup using a single number: 41 presses. I did not need player names to conclude that the opposing midfield would break; I only needed to look at pressure density in the central lane. That is the power of football data: one correct number, placed correctly, tells an entire story. But for that number to be correct, it must sit in the right drawer. A pressing metric placed in the wrong drawer becomes a meaningless number, or worse, a misleading one. What is notable is that the mislabelled record was full of verifiable facts. List what it actually contained: a male actor who played Spider-Man in two films released in 2026 and 2026, returned to the role in a 2026 multiverse crossover, a project steered by director Paul Greengrass, a television series titled Wild Things, a project called Artificial tied to an artificial intelligence company, plus distributors such as Sony Pictures and Focus Features. All of it is verifiable. All of it is accurate within its field. And none of it touches football by even a thread. No club, no league, no coach, no player, no minute of play. In other words, this was not a badly written football article. It was an article that never belonged to football in the first place, forced into a garment that does not fit. And the very moment of mislabelling is the real subject worth discussing. In the sports data industry, a wrong label is not rare. It is daily. It is just that most mislabelling happens silently, unnoticed, until a wrong prediction, a stray news item or an absurd aggregation makes someone stop and ask: where did this data come from? And the answer is rarely about football expertise. It is about operations. This is where I want to say plainly something few in the industry want to hear. We pour enormous resources into analysing the glamorous things: tactics, models, prediction algorithms. We spend almost nothing on the most mundane thing of all: input quality control. And in any system, output quality can never exceed input quality. The most sophisticated model, fed on carelessly labelled data, will produce careless conclusions presented with great sophistication. That is the hardest kind of mistake to catch, because it wears the face of precision. I remember that across 138 matches played in empty stadiums, the home-win rate fell from 45.7% to 31.2%. A figure that forced the whole industry to look again. But to discover it, the precondition was that I could trust every one of those matches to be a real football match, recorded correctly, labelled correctly, classified correctly. If even a small share of the input data were contaminated, that 31.2% would mean nothing. Every conclusion rests on an assumption about the cleanliness of the data. When that assumption collapses, the conclusion collapses with it, however beautifully it is presented. This is why I treat domain labelling as part of the craft, not an administrative chore. Anyone who makes sports news understands: a chaotic feed destroys reader trust faster than a thin feed. Readers can forgive me for not writing about a match. They will not forgive me for putting a Hollywood story into the football section. So what should be done? The answer needs no advanced technology. It needs a minimal barrier: one entity check, verifying whether a record contains at least a club, a player, a league or a match. With such a simple test, that mislabelled record could never have passed. The difficulty is not technical, it is a matter of awareness. We tend to assume the automated system has handled everything, so nobody bothers to add a manual checkpoint. There is a telling paradox here. Precisely because sports data grows richer by the day, people tend to trust it more blindly. A feed of thousands of records a day creates a sense of completeness, leaving no one patient enough to inspect each record. But completeness is not accuracy. A thin feed is visibly thin. A packed feed laced with impurities is far more dangerous, because we do not know what we are reading. I am not telling this story to pin blame on any particular system. It is common enough to be a feature of the era. When speed is placed above accuracy, verification is always the first thing cut. And football, a field where every number can lead to a real decision, becomes the place where that carelessness exacts the heaviest price. A club can buy the wrong player based on a noisy metric. A coach can miss a genuine pattern because it is buried under junk data. But 51 years on the pitch and in the stands taught me one thing: big mistakes rarely begin with big decisions. They begin with small details overlooked, then grow over time. One wrong label today is one noisy signal. A thousand wrong labels in a month is a broken model. And a broken model, presented well enough, can lead an entire industry to skewed conclusions before anyone notices. The pitch never lies. But the label attached to the pitch can absolutely lie, and lie very politely. Whoever reads the next move will hold the game, true enough. But before reading the move, that person must be sure they are sitting at the right board. A record with no football in it, even filed under football, remains forever an empty board. This lesson is not only for news aggregators. It is for anyone who believes quantity of data can substitute for quality of data. On the night I stayed awake to redraw every ball trajectory in a historic match, what woke me was not a beautiful number, but a correct one. And a number is only correct when it sits in the right place. Next match, when I open any feed, I will ask one question before reading the first line: which board does this record really belong to? The question sounds small. But most failures in this trade, in the end, begin with forgetting which board we are sitting at.