A Vaccine Inside the Football Folder: A Cold Reading of a Domain Misclassification

I opened the file at 2 a.m. Rain was drumming on the tin roof in Sylhet. The file had landed inside our football analysis pipeline, and its label said, plainly, Football. What was inside was not a formation diagram and not a pressing map. It was a vaccine launch event. A vaccination network in Vietnam, an EV71 hand, foot and mouth disease vaccine, statements from paediatric infectious disease specialists, supply across roughly three hundred centres, and one large number — 96.8 percent efficacy.

I started with a blank pitch and a spreadsheet that refused to lie. I opened the file at 2 a.m.; by 4 a.m. it was clear there was no midfield here, and therefore no midfield confession to extract.

Where was the error? In words. Of the three words hand, foot and mouth, one has a face that resembles football. The disease caused by enterovirus 71 can produce neurological complications, settles in small children, and in clinical language the full name is a single disease entity. It is not the anatomical foot. It is a term hiding inside a clinical name, with no relationship to a goalpost.

The pipeline this file entered runs on words from the start. Which sector a document belongs to is decided by its domain label. A football label sends the document to a tactical analyst; a health label sends it to a public health desk. Over the past decade the volume of sports content has grown so large that reading and classifying everything by hand is impossible. Hence the conveyor belt. And the conveyor belt has an old habit — it does not ask questions, it simply obeys the label.

My own work sits on the opposite side of that conveyor. In 2026, while I spent my days on logistics ledgers at the Sylhet District Football Association, I started a Bengali tactics blog from a two-room flat. The third post ran two and a half thousand words — how Antonio Conte's Chelsea 3-4-3 used Cesc Fabregas's diagonal to overload Tottenham's back line. Forty-seven thousand reads in nine days. After that I stopped writing match reports altogether. One template became final: a shape diagram, a pressing map, and one paragraph on who owned which channel. I still refuse to publish without a drawing.

The lesson of this intrusion needs to be broken into three layers. The first is lexical. The second is sourcing. The third is verification. Together, the three produce something that is simply another name for football journalism's most familiar disease.

The lexical trap is not new. Anyone who works with data knows the type — goal also appears in sustainable development documents, press in press conferences, wing in hospital wards. Our taxonomy stands on words, but the errors stand on intent. Nobody asks at the labelling stage who commissioned the text, and what the commissioner wants.

At the sourcing layer the thing looks even more familiar. In this vaccine document, the subject is also the principal source. The organisation describes its own work, supplies its own distribution numbers, explains the significance of its own initiative. That is not weak journalism; it is a different species of journalism, called promotional writing. A document that is its own source is not news; it is an extended advertisement.

And here the resemblance to football content is uncannily exact. In the transfer window we read such documents every day. A club's own medical bulletin declares its star fit; an agent, citing a reliable source, declares three clubs are chasing his client. The label says Football, the subject says Football, the source is partisan. We pass it off as news.

The third layer holds the number. 96.8 percent efficacy — a big claim, but attributed to published data, with no journal named, no sample size, nothing. I do not fear numbers; I fear nameless numbers. From May to June 2026, across fifty-four days, I watched ninety-two Bundesliga matches behind closed doors and tagged every high-press sequence. The per-match count fell from 12.4 to 9.8, while final-third pass completion rose. Ninety-two empty stadiums taught me that silence still has a shape — but to describe that shape I had to tag it, not guess it.

In football data, a source has indicated sits in exactly that place. At Russia 2026 I logged all 169 goals across 64 matches into a spreadsheet of my own with twelve variables. Forty-three percent of them — seventy-three goals — came from set pieces, penalties or second balls rather than open-play build-up. Before the final I argued Didier Deschamps would keep Blaise Matuidi as a defensive left midfielder instead of starting Ousmane Dembele, to shield the channel behind Lucas Hernandez. France won 4-2, Matuidi started, and an Indian outlet syndicated the piece — my first paid English byline. Since then, a number beside every claim has been my habit.

What cannot be proven with data cannot be dressed up with data either.

There is another way to test a number, one I learned counting goals. The question is: who moved first? With goals we usually count the last touch, because the last touch has a name. But if the spreadsheet holds only assists and goals, half the story is lost. The same question applies here. In this error, who moved first? The answer: not the classifier. The keyword list moved first. The machine that applies the label is only obeying.

One more thing catches the eye. In the launch narrative there was a sentence — first in Southeast Asia. Launch stories always stitch a first onto themselves, and the reader becomes a stakeholder. Football stitches the same: the first club to use this data, the first league with that VAR system. The stitch is not a lie; it is part of the competition. But a stitch in news is information, while a stitch in advertising is a weapon — and only the sourcing can tell the difference.

None of this means the people involved were wrong. In public health this document may have a role — parental anxiety, disease season, access to vaccination are all real. Sitting in Sylhet I will not deny that. The problem is not the content. The problem is the label.

The easy fix occurs to most people now: add an exclusion rule to the keyword list. If foot appears, route it to the health desk; if vaccine appears, drop it. I thought that first. Then I remembered the next error will not be so simple.

The next error will arrive under the headline club financial crisis, containing a sponsorship announcement. The next will arrive under transfer target, containing an agent's press release. The next will arrive as a children's academy report that is really an advertisement for higher enrolment fees. The domain will be correct, the category will be correct, and only the truth will be missing. A rule recognises words; a rule does not recognise motive.

The real danger is not that a vaccine document slipped into the football folder. That is uncomfortable but not harmful — no analyst will make a tactical decision from it, because four lines in they will see there is no match. The real danger is that documents infected by unsourced numbers and partisan description are passing through every filter we own, because their labels are clean.

One thing should be said plainly. The most valuable part of that Stage-2 analysis was not the analysis — it was the confession. A system that opens a document and declares there is nothing football-related here, therefore I will not invent football-related content — that system is the actual asset. Because the opposite road is always open: find a picture, drop in a formation, turn hand, foot and mouth into a metaphor, and write four thousand words. That would not be analysis. That would be forgery.

So a new rule goes into my next window. Before any claim enters the database, three questions — who is the source, who benefits, and can I pronounce the name of the study? If any of the three goes unanswered, the claim is not discarded; it goes into an unverified box and gets no permission to be dressed up. Alongside it, a negative test corpus: ten documents deliberately built to look like football but actually advertising.

And one question that has not left me since 4 a.m. If a vaccine document can walk into the football analysis folder wearing shoes, what else is already sitting inside that folder, behind doors I have not yet opened?