When the Machine Labelled a Dating Show as Football

Three in the morning in Liverpool. Rain taps the window of the small flat like studs on an empty stand. My screen lights up: a new content packet has dropped into the queue, labelled football, high priority. I reach for the pronunciation notebook beside the desk — a habit from Moscow. Whenever a new source arrives, I read the names first, then analyse. That is my ritual for entering the temple. I open the packet. There is no football. Inside is a reality dating show. A host returning after four years away. A US streaming platform. Ten contestants. Six episodes. Premiere on the eleventh of November, finale on the ninth of December. No club. No player. No scoreline. No stadium. No league. No governing body. I sit still. Then I do what twenty-three years in the trade have taught me: I read it again from the top, slower, as if a club name had been misspelled and I had to find where. This time no name was misspelled. Only a label was applied wrongly. And that label said football. From the balcony of an old newsroom, I hear the derby echoing through layer after layer of memory. Tonight, the echo comes from a room with no match in it. ———— What was in the packet A reality dating show. That is the entire subject. It has a familiar host returning after four years off air. It has a streaming platform as distributor. It has a premiere date, a finale date, an episode count, a contestant count, a list of producers. It has a dozen real names, real jobs, real companies. I read it all. Then I read it again. Twenty-nine information points, if counted properly. I count. Twenty-nine. Among those twenty-nine, not one touches football. No club. No player. No coach. No match. No transfer. No table. No federation. No disciplinary panel. The closest thing to sport sits in the corporate-ownership section: the parent group of that streaming platform also runs a few sports properties. That is a shareholder relation, not a tactical one. It tells me nothing about a back three, about a holding midfielder, about a corner routine rehearsed in training. If I used that shareholder link as raw material for a football analysis, I would have to invent a causal chain the data does not contain. I do not do that. The football label at the top of the packet is the only football attribute present. And nothing in the content backs it up. It is an orphan label. ———— Anatomy of a labelling error In a modern content system, every packet passes through several layers. The collection layer. The classification layer. The labelling layer. The routing layer. The distribution layer. At each layer, a small decision is made. The football label is born at the classification layer, and from there it travels with the packet like a passport. Nobody at the routing layer re-reads the content. They trust the passport. When the passport is wrong, the whole journey goes wrong with it. What stands out here is not that someone labelled it wrongly. Machines label things wrongly every day. What stands out is that no layer stopped it before it reached my hands. A packet about a dating show passed through an entire pipeline without anyone asking the simplest question: does this content contain a single football entity? That question is cheap. It is fast. It can be answered in five seconds. But it was not asked. Here, I think of old newsrooms. There, before a number went to print, someone sat and cross-checked. Before a name was printed, someone read it aloud. That layer — the human layer — was not the slow layer. It was the blocking layer. It was the defence of the truth. We removed that defence to run faster. And now we concede from shots nobody sees. The goal does not come from a beautiful combination. It comes from a gap nobody is standing in. A gap at the blocking layer. I have said before that people call it tactics; I call it the breathing of the team on the pitch. A data pipeline breathes too. And its breathing is failing. ———— Four empty columns When I try to apply my standard analytical frame to this packet — the frame I use for every deep analysis — I get four empty columns. Column one, tactics. No formation. No shape. No pressing scheme. No build-up pattern. No set-piece design. No in-game adjustment. Nothing to compare with anything in a football match. This column is not empty because I lack data. It is empty because the subject does not exist. Column two, finance. No transfer fee. No wages. No amortisation. No wage bill. No financial-fair-play exposure. The production roles listed in the packet — executive producer, casting director, director — are entertainment-industry functions. They have no football equivalent. If I mapped them onto sporting-director or technical-director roles, I would be wrong about the nature of the job, and I would drag a chain of wrong conclusions behind me. Column three, results and the public-opinion cycle. No table. No form curve. No fixture list. The only pressure present in the packet is a personal reputational pressure: a host involved in a 2026 controversy, off air, then returning in 2026. That is a celebrity-opinion cycle. If I mapped it onto manager-sack pressure or dressing-room morale, I would commit a category error. And a category error, in my trade, is the worst kind — it is not wrong about numbers, it is wrong about the world. Column four, rules and governance. No federation. No committee. No transfer regulation. No disciplinary sanction. A controversy over comments was handled as a labour-and-reputation matter inside a private television brand, not under any football rulebook. Four empty columns. One conclusion can be drawn — and it is a conclusion about the data pipeline, not about football. I have sat in newsrooms that no longer exist, to write about matches that never grow old. Today, I sit in a newsroom that still exists — this very room — to write about a labelling error that should never have existed. ———— An industry built on labels Every modern information system runs on labels. A label decides where a packet goes, who reads it, how it is monetised. A piece about football labelled football will reach football fans. A piece about dating labelled football will reach... football fans too. The difference is not where it arrives. It is whether the recipient still trusts the label. When wrong labels repeat often enough, readers start doubting every label. They open a genuine football piece and ask whether it is genuine football. They open a derby report and ask whether it is a dating show in disguise. Trust in a label is the easiest kind of trust to lose and the hardest to recover — like trust in a defence after a season of conceding. Once the back line has lost each other once, every ball that follows makes you hold your breath. And when trust in the label goes, what goes with it is not just one section. It is an entire industry's ability to tell the public what a thing is. Football lives on stories. But a story only means something when the teller is believed. A storyteller caught mispronouncing a name once can still save the telling. A system caught mislabelling every day turns every telling into a suspicion. ———— From advertisement to front page One more thing caught my eye. Almost every information point in the packet carried no source. Only three were attributed — the programme description supplied by the producer and the platform's brand-positioning statement. That is promotional-derived reporting. Its factual core does not come from an independent investigation. It comes from material issued by the producer itself. I have no objection to using press materials. I object to using them without labelling, for the reader, what they are reading. When a report takes its core from a press release and does not say so, its reliability is discounted. Not because the release is false. Because the release only tells the story of the one who issued it. In football, we know this too well. A club statement about full support for the manager is not a fact. It is a signal. And the reader who takes it as fact is usually the one most surprised when the chair falls. We have seen it often enough to stop pretending otherwise. In my trade, a good reporter distinguishes three things: fact, signal, and advertisement disguised as fact. Readers are finding it harder to distinguish the three, and that is our fault, not theirs. ———— The heat cycle of a story Every media story has a heat cycle. It rises, accelerates, peaks, then recedes. With this packet, that cycle is pinned to two dates: the premiere on the eleventh of November and the finale on the ninth of December. What do those two dates tell me? They tell me the platform is targeting the US holiday viewing window. That is a scheduling inference. It says nothing about football. But it tells me one thing about the content industry at large: we live in an economy of short heat cycles. A story is born, pushed, consumed, forgotten. Then the next. Football does not escape that wheel. A transfer has a heat cycle of days. A derby has a heat cycle of hours. A sacked manager has a heat cycle of weeks. And precisely because heat cycles are short, people tend to fill them with anything. Including a dating show labelled football. There is a paradox of expectation here. With this packet, the only expectation question worth asking is not whether a team will take points, but whether a six-episode run with ten contestants can hold attention on a weekly release schedule. That is a television-analytics question. Not mine. ———— The blind spot: when football thins out This is where I start asking the serious question. If a machine cannot tell a dating show from football, there are two possibilities. Either the machine is broken. Or — and this is the possibility that keeps me awake — football, in the shape the machine was taught to recognise, has grown so thin that a dating show slipped in and nobody noticed the mismatch. A silent summer does not announce collapse; it only clears the road for collapse. A football culture that thins out does not collapse in one night. It collapses across seasons. Through each lazy article. Through each careless label. Through each story about a celebrity pushed into a sports section because it drives clicks. Through each headline that talks about a player more than about football. When that culture is thin enough, the machine stops recognising the border. And the border disappears. I once followed a semi-professional club in the Merseyside area through the months when football stopped. Empty ground. No crowd. No shouting. I wrote a series about football in the silence. And I learned that what keeps football alive is not noise. It is attention. When attention withdraws, football is still there — only nobody sees it any more. A wrong label is the sign that someone stopped seeing it. I record this detail not to compare a lower-league club with a dating show. I record it because the same mechanism operates in both places: when a thing loses attention, it loses its existence in the eyes of the system. The semi-pro club loses its crowd, then loses its place in the bulletin. A football section loses its border, then loses its place in the data queue. Nobody announces the death. It simply vanishes from the category. ———— Do not blame the machine The easiest reaction, and the most cowardly, is to blame the machine. Blame artificial intelligence. Blame the labelling algorithm. Blame the automation layer. I do not buy that reading. The labelling machine is built on what humans taught it. It was taught that something looks like sport when it has a famous figure, a controversy, a premiere date, a distribution platform, a heat cycle. It was taught by our own habits. When an entertainment story has a figure, drama, and a schedule, it carries enough signals for a classifier trained on consumer data to place it in whichever box shines brightest. And the brightest box in the attention economy is often the football box — because football is the universal language of attention. If football becomes the box that swallows everything, football also becomes the box that means nothing. Here is the paradox. We take pride that football is the most universal sport on the planet. We boast that it crosses borders, languages, classes. But when it crosses everything, it begins to blur into everything. And a thing blurred into everything is no longer a thing. A pronunciation gatekeeper like me knows this from another angle. When you stop reading a person's name correctly, you have stopped seeing that person. When you can no longer tell a football club from a dating show, you have stopped seeing football. Both are the same illness. The illness of an eye that has stopped looking. And that illness does not spread from machine to human. It spreads the other way. ———— Moscow and humility June 2026. World Cup in Russia. I was thirty-one, by then a senior specialist. In Portugal's opener against Spain, when Cristiano Ronaldo took the free kick in the eighty-eighth minute to make it 3-3, I had called defender Nacho by the wrong name three times in the first half. Three times. I was ashamed. For the following month I rewatched the footage of all sixty-four matches and built my own dataset of player-name pronunciations in the correct native languages. Since then, I never leave home without the pronunciation notebook. Moscow taught me that every name has its own breath; to mispronounce it is to strangle a soul. And if mispronouncing a name strangles a soul, then mislabelling a section strangles a world. Humility is not knowing you are wrong, but learning another person's name correctly. I learned that in Moscow. Now I learn its extended version: humility is reading the content again before trusting the label. The label is an assumption. The content is the fact. A decent professional must trust the fact before the assumption. That is everything I learned in Moscow, and everything I need in Liverpool at three in the morning. ———— Three signals to track I leave three signals for those working in the industry — those behind the data pipeline, those managing the corpus, those building entity graphs. Signal one: the labelling error rate. Sample packets randomly at the first layer and compare label with content. Whenever a non-football packet is labelled football, that is not an isolated incident. It is an indicator of the health of the entire corpus. A corpus infected with wrong labels produces wrong conclusions downstream — in sentiment indices, in entity graphs, in transfer-rumour trackers. Signal two: source coverage. Measure the share of information points with a named source. When source coverage drops, the reliability of the entire corpus must be discounted, not just individual pieces. In the packet in my hand, only three of twenty-nine information points had a source. That is the number I wrote in my notebook and underlined twice. Signal three: the lifespan of an error. When a labelling error is discovered, how long does it take to fix and to prevent recurrence? A pipeline without a correction layer is a pipeline that will repeat its old mistakes. These three signals do not shout. They are quiet. But they are what separates a system that knows what it is doing from a system that is merely running. And football, in the end, is a system — a system of memory. A system that, if you let it be infected with wrong labels, will forget itself. ———— What I want to record I am not writing this to shame a wrong label. I am writing to record a moment. Every dynasty has a minute of leaving the pitch that nobody notices; the observant will record that moment. For me, that moment is three in the morning in Liverpool, when I opened a packet labelled football and found a dating show. In that moment, a border disappeared. That border can be rebuilt. Not by banning the machine. But by putting humans back into the blocking layer — the one who cross-checks before a number goes to print, the one who reads a name aloud before it is printed, the one who asks the cheap, fast question: does this content contain football? That question does not need a doctorate. It needs attention. And attention, in our economy, is the most expensive thing there is. The ball rolls through time, but the memory of a derby night rolls through the heart. I want to keep one small thing from this night without a derby: that reading correctly is still cheaper than reading wrongly. That a correct name still matters more than a click. That football, in order to exist, must be seen — and to be seen, it must be called by its right name. Three in the morning. The rain keeps tapping. I repackage the content, tag it for re-routing, and send it back where it belongs. Then I open the pronunciation notebook and write one line for tomorrow: read the name first. Always read the name first.