When a Scouting File Contains "88 Amenities": A Classification Failure in Football's Data Layer

Tuesday morning, seven o'clock. At an academy, everyone sees the goals. Few see the Tuesday morning at seven. I open the forty-first file in a batch of sixty that has arrived from a data-aggregation vendor. The classification field is unambiguous: football. Inside are twenty-five information points. Point sixteen states that every bedroom has a corner reserved for a study desk. Point eleven mentions eighty-eight selected amenities and an entire floor given over to a library. Point twenty-two compares an asking price of sixty-nine million Vietnamese dong per square metre against a market level of more than ninety million per square metre in the southern part of the city. There is no club in the file. No player, no coach, no competition, not a single line of football data. I sit still for about forty seconds, long enough to drink a mouthful of coffee that has already gone cold, and then I do the thing I have done for fifteen years: I check the label instead of checking the contents. The feeling in that moment was not anger. It resembled September 2026, when I submitted a twelve-page assessment of a sixteen-year-old and believed I was right. Two stages, one weak link Most football data rooms now run on a two-stage model. Stage one breaks a text into discrete information points: names, numbers, dates, quotations. Stage two reads those points and rebuilds them into analysis. The process arrived in football later than in banking or retail, but since 2026 it has been close to default, because academies lost the ability to observe directly during lockdowns and were forced to buy data from outside. I was a beneficiary of that turning point. In March 2026, when the Premier League was suspended indefinitely, I lost my freelance contract and spent six months in a Manchester flat rebuilding my scoring system. When football returned, clubs short of data from cancelled youth competitions came to me. Huddersfield Town paid fifteen thousand pounds for a report on five Brentford youngsters. A small market, but a real one, and it survives on a single assumption: that the label on every file is correct. The weakest link in the whole chain is not in the analysis layer. It sits in the small line of text that records the subject category. A file mislabelled at stage one passes through stage two unchallenged, and every score, every ranking, every model downstream inherits that error silently. The scoring system I built over six months in 2026 had ten criteria stable across three seasons. If I fed it a residential property brochure, it would still return a number that looked entirely plausible. That is the frightening part. So what was the document in the file? A product introduction for an off-plan apartment project in the southern corridor of Saigon, specifically the Nha Be area. No media outlet is named. The developer is described with an internationalising noun phrase, a Vietnam–Japan–Singapore joint venture, but with no legal entity attached. No independent expert is quoted. The single testimonial comes from a customer identified only by first name, an accountant, with nothing verifiable behind it. There is no publication date, so even timeliness cannot be assessed. A classification error, and that is the whole story of stage one. But the story worth writing sits elsewhere: the structure of that document is uncomfortably similar to what football produces for itself every week. The eight-beat skeleton of an advertorial I read the document a second time, this time not hunting for faults but for a template. It followed exactly eight beats, without deviating once. It opens with a sensory experience at a sales centre, where the physical model apartment has been replaced by a three-dimensional projection theatre. Then comes the completed ecosystem, with schools and amenities already operating. Then future infrastructure: two bridges and two metro lines described with the phrase 'now being implemented'. Then product differentiation through an 'intellectual apartment' positioning. Then the handover specification. Then the price advantage. Then the gentle payment structure. Then one customer testimonial. Finally a closing paragraph asserting appeal to thousands of customers during a period of market filtering. Those eight beats are designed to pre-empt objections, not to test them. And they appear in football in three familiar shapes. The first is the statement-signing feature during a transfer window. It opens with the unveiling and the stadium tunnel. Then the academy that produced the player. Then the new facility that exists on a planning drawing. Then the distinctive coaching model. Then the long-term contract detail. Then the fee described as reasonable. Then the multi-year instalment structure. Then one unnamed source inside the club. Then a conclusion about ambition. The second shape is the stadium or training-centre announcement. The third is the academy partnership press release. All three share the same skeleton; only the vocabulary changes. Why does this matter to someone who works with data? Because when you read a file built on eight beats, you are reading a text engineered to survive scrutiny rather than to invite it. An honest scouting report needs a ninth beat: the beat of what is wrong, of the missing data, of the sentence 'I do not know'. I have never seen an advertorial with a ninth beat. Nor have I seen a club transfer announcement with one. The arithmetic that does not close Information point twenty-three offers a very comfortable figure: twenty-nine million dong per month, no borrowing, no exposure to fluctuating interest rates. Point twenty-two gives an asking price of sixty-nine million dong per square metre, below a market level of more than ninety million in the same area. I put the two numbers side by side. A unit of roughly sixty-five square metres at sixty-nine million per square metre comes to about four point four nine billion dong. Paying twenty-nine million a month, interest-free, takes roughly one hundred and fifty-five months, close to thirteen years. Standard Vietnamese off-plan practice spreads only a minority of the price across construction, with the balance due at handover. The document states no total price, no number of instalments, no handover date, no obligation payable on receipt of the keys, no penalty clause. The arithmetic is correct. The logic does not close, because the denominator is hidden. Thirteen years is a number that holds only under an undisclosed assumption. I see exactly this structure in transfer reporting. A club announces a fee, and the fee is always the easiest part of the deal. Performance add-ons, weekly wages, agent commission, sell-on percentage, release clauses, signing bonuses paid to a young player's family: all of it sits outside the published number. A deal described as ten million euros can be twenty-five million by the time the conditions mature. When a newspaper writes 'a nominal fee', I always read on to find the part left blank, and the part left blank is usually the majority. A wrong report is like a shard of broken pottery: handled carelessly, it cuts the hand of the person who wrote it. The writer here is not the developer. The writer is anyone who copies a number without asking for the denominator. Borrowed authority Information point ten does something clever: it fuses Vietnam's tradition of studiousness with two of Asia's leading education systems to certify a property development. The developer descriptor does the same, a three-country joint venture with no legal name. Someone else's reputation is being used as building material. Football does this every day. 'Trained at the academy of a European club' is a sentence containing no data. The right question is: how many competitive minutes, at what level, under how many coaches, across how many seasons. The badge does the work of the minutes, and a badge has no legs. The experience centre gives it away The detail that stopped me longest was the replacement of a physical model apartment with a projection theatre. Technically, that is a sensible solution for a building that does not yet exist. Informationally, it is a confession: the real product is not available for inspection, so experience technology is carrying the load that concrete should carry. In scouting, this symptom has a name. When a player file contains only edited clips, with no full match footage, no opponent context, no notes on behaviour away from the ball, I mark it red. Not because the player is poor. Because whoever sent the file chose to show me the best part rather than the whole. I have watched enough under-18 matches at the Manchester training complex to know the distance between a fine piece of skill and a good performance. A player can produce three excellent actions in ninety minutes and still be responsible for two conceded goals. An edited reel will never show you those two goals. The 2026 lesson and the two-way notebook In September 2026 I was an assistant analyst at a Manchester academy, assigned to track Phil Foden, then sixteen, in an under-19 training match. I wrote twelve pages concluding that the boy lacked the pace and the frame for elite football. Three months later Foden was promoted to the first team and scored in his first Champions League season. I was wrong because I read only physical data and not his reading of the game. But the structural error is the memorable part: I wrote a one-directional report in which every observation served a conclusion I already held. Since then I have kept a two-way notebook: current data placed alongside development potential, with every line answerable to a single question. What makes me believe this: the number or the prejudice? Before I write the name of a star, I have to strip away a thick layer of soil called hype. But before stripping anyone else's hype, I have to strip my own first. Patching the hole at stage one Three fixes, and none of them requires artificial intelligence. Add a domain-confidence check between the two stages. If a file carries a football label but contains no identifiable football entity, it is held for human confirmation. That step costs about forty seconds per file. Add a mandatory field for article type, kept entirely separate from the subject field. A product introduction can live in the archive, but it must be excluded from every analytical score and every reputational judgement. Promotional text is not wrong to exist; it is only wrong when it is scored as a source. Add a batch-level monitoring threshold. If promotional content exceeds ten per cent of a batch, the whole batch is manually reviewed before it enters any model. My job is rereading. Before writing about the future, read today one more time. The contrarian angle: the stranger gets caught, the familiar does not That property document was caught because its vocabulary was in the wrong room. It talked about amenities and study desks inside a data room that talks about xG and minutes played. Its strangeness saved us. The automated classifier almost certainly tripped on the very phrases we use daily: education ecosystem, development pathway, international standard. That is the vocabulary of academy files. Our own language is the camouflage. A football advertorial is never flagged, because it wears the right clothes. It has a club name, a player name, a coach, a competition. It passes every entity filter perfectly, and it is just as short of a ninth beat as that apartment brochure. The same three unverifiable claims recur in both worlds. 'A reasonable price' in property is 'a bargain in the transfer market' in football, and neither carries an outside source. 'No borrowing, no interest-rate risk' is 'a friendly add-on structure', and both hide the denominator. 'Infrastructure now being implemented' is 'a new training ground according to plan', and both are drawings, not concrete. I do not need a perfect player. I need a player who knows he is not perfect. And I do not need a perfect report. I need a report that knows what it is missing, states the gap, and can survive being reread three seasons later. The worry is not that one advertorial slipped into the data room. The worry is that hundreds of others have been in there for years, carrying the right label, the right proper nouns, the right terminology, and nobody bothers to check them because they look too much like the truth. What comes next Among those twenty-five information points there is one detail I kept as a personal note: every bedroom has a corner reserved for a study desk. A small detail, placed exactly where it will make the reader think about their child's future. That is good writing, and it is also writing that needs to be recognised. Our football is full of study-desk corners: a young player with 'a superb attitude', an academy with 'the right culture', a contract with 'a smart structure'. Those sentences are not false. They are simply untested. The job of the person who rereads is to keep that testing alive, even when nobody asks for it, even when the label already looks too right. Forty seconds per file is the cheapest line item in the entire system, and the only thing keeping the rest of it standing.