One Wrong Label, One TV Drama, and the Invisible Crack in Football Data

Last week a file opened on my screen with a single word written across the top — football. Twenty-six information points, arranged one after another, classified, filed like books on a library shelf. Not one of the twenty-six contained any football. No pass, no formation, no goal described, not a single xG figure. There was only a promotional piece for the third season of an American television drama — on the CBS schedule, streaming on Paramount+. Its name: NCIS: Origins. The system that labelled this file as football was faster than me, more patient than me, and certainly more confident than me. The mistake was its own. But whose is the responsibility?

NCIS: Origins is a prequel to CBS's long-running crime-drama franchise. The narrator and executive producer is Mark Harmon, who spent more than two decades acting in the parent series. The showrunner is David J. North. According to CBS, new episodes air Tuesdays at 10 p.m. ET/PT and stream on Paramount+. Around the episode dated October 13, promotional interviews had North hinting at future plot turns — avoiding spoilers, in measured language, exactly as the entertainment industry tends to do.

The question now is how this material entered a file meant for football analysis. The answer is mechanical. At the first stage, an automated classifier reads every article and pins a domain label on it — football, cricket, politics, entertainment. At the second stage, that label is trusted while the file is checked across nine analytical dimensions: tactical and technical analysis, club finance and the transfer market, results and the public-opinion cycle, league landscape and team positioning, rules and governance, management and the dressing room, risk profile, media narrative, and industry transmission. The label said football; the content was a TV drama. The gap that opened in between is the story here.

Where this piece came from matters too. Its tone suggests a press-release-driven entertainment promotion — interview-led, controlled, at times arranged. That kind of writing is a genre apart from football journalism, and that genre is precisely what the classifier failed to catch.

When the second stage stood before this file, it had two paths. One was to accept the label and extract football-shaped meaning from the twenty-six points — forcing words like team, rival, franchise and origins into tactical vocabulary. The other was to admit that no football exists here, and to mark every dimension as not applicable — insufficient information. The analysis took the second path. Based on my years of watching matches from the stands, that second path is the real work. A system that can admit its own ignorance is the only system that stays honest in the end.

On the tactical dimension the answer is zero, because this text holds no formation, no pressing trigger, no possession or xG statistic. The phrases that look tactical — reuniting with a longtime friend, investigating a mystery — are narrative structure, not pitch tactics. In the finance section there is no club, no transfer fee, no wage bill. There is only a distribution arrangement — CBS's linear channel and Paramount+ streaming. That is entertainment-industry distribution, not football-club revenue. On the results dimension there is no league, no points table; the nearest thing to a result would be viewership numbers, and this source has none.

The league-landscape dimension is equally empty. No league, no hierarchy, no club. The franchise concept may look like a match, but slotting CBS or Paramount+ into a football tier would be a forced and misleading analogy. On governance, no FIFA, UEFA or national-association rule applies; the only regulatory content here concerns broadcast and streaming rights, which sit outside the football framework. In the management section, the named figures — Harmon, North — are an actor and a showrunner, not footballers or coaches. Creative leadership and pitch-side leadership are not the same thing, even when the vocabulary sounds alike.

Industry transmission is also zero for football. The one chain present is an entertainment one — production to broadcast and streaming, then advertising and subscription revenue. That chain touches no part of football. Retaining this item as football would, in fact, create a false link between the television world and the football world.

The most delicate point is naming. If this item enters a football knowledge graph, actors could be seated there as players or coaches. Once a wrong entry goes in, it does not easily come out; later, when some analysis pulls a squad list, Harmon's name may surface from it. This is where domain-specific dictionaries are needed — a separate list for football entities only, so the wrong person does not sit in the wrong seat.

The risk matrix therefore shows one real risk only — data-pipeline integrity. Level medium, likelihood high, impact moderate. The remedy is clear: quarantine the item and re-tag it as entertainment. Once the entertainment label is applied, the file returns to its proper department and the football store stays clean. There is no sporting, financial or governance risk, because no sporting entity exists here.

The hidden information is more uncomfortable still. This error likely came from a habit of the automated classifier — deciding on isolated tokens such as Origins, Camp Pendleton, team or rival. In other words, tagging by token rather than by context. And if one item carries such an error, others in the same batch may carry it too. A classification error does not arrive alone; it arrives in company. The question, then, is not about one file but about the whole batch.

Several tracking signals attach to this. First, the classifier's error rate: sample the first-stage output and see what share of football-labelled items are not actually football. If that rate exceeds one percent, the integrity of the store should be assumed to be eroding. Second, adjacent-batch contamination: check whether other entertainment items ingested in the same batch also received the football label. Third, domain-gate enforcement: whether any non-football item is reaching the second stage at all.

This file's information value is worth measuring too. Sporting value — one star out of five, because football content is zero. Industry value — one star, because it is useful only as a pipeline-integrity signal. Timeliness — two stars, because it is bound to an episode dated October 13. Usability as a reference — one star, because it cannot be used as a football source. It sits at the bottom of all four measures, and that is its true identity.

Here I should pause. The year I stopped calling the game was the year I learned to listen to it. For fifteen or sixteen years I sat behind a microphone, then I walked away, then I began to listen. In that time I have seen that football data no longer lives only in a reporter's notebook. The live feed runs straight to betting-company servers, milliseconds apart. If that feed is ever wrong, nobody is compensated. The darkest side of football's datafication is not the live feed — it is that invisible layer of classification, where a machine decides who counts as a player and who does not.

The received wisdom says our data systems grow smarter by the day and errors fall. This file nudges that belief. The error made no noise. No error message arrived, no alarm sounded. A television drama sat quietly labelled as football. And had anyone at the second stage trusted the label blindly, the twenty-six points would have yielded a flawless, credible, entirely fabricated football analysis. There would have been xG, there would have been pressing triggers, there would have been a story about pressure on a coach. All false, but all beautifully written. The dangerous error is not the one that shouts; the dangerous error is the one that sits quietly and becomes true. Such an error is caught only when someone stops midway and asks — what is this, really?

From there another thought follows. Today's data systems want a provenance seal on every piece of information — where it came from, who verified it, who approved it. Ledger-style thinking is no longer confined to financial transactions; it is being explored for safeguarding data provenance, so the birth record of every label is kept. But no ledger can catch that error unless someone inside the system has the courage to say I do not know. The chain of evidence gets built, yet if the first link in the chain is a false label, the whole chain carries false testimony.

One more angle is worth noticing. This piece is itself the product of a controlled publicity pipeline — measured showrunner quotes, actor interviews, spoiler-safe language. In entertainment journalism that is normal. But the same measured language is now entering football coverage — club press notes, agent statements, arranged leaks of transfer gossip. Where everyone speaks in measured terms, who is the one person who will say the uncomfortable truth?

So what is to be done? Simple, and dull. Quarantine the file, apply the entertainment label again, and turn back toward the classifier that made the error. If one item is mislabelled, the rest of its batch needs checking too. The rule for listening is easy — late to judgment, early to attention. And if you ask who will watch this invisible cracking of football data, the answer is uncomfortable — whoever watches it holds no job title. The classifier that wrote football has never been to a stadium. But the machine we entrusted with understanding the game, we never taught to stop. A machine has no ego; it does not stop, because nobody told it to.

In September 2026 I put down the microphone; since then I have listened to the game. The first lesson of listening is that the sound which does not arrive is the most important sound. Today that empty space is being guarded by a label, and a label never knows the smell of a pitch. So the question is not vast, only small: next time a transfer rumour reaches your feed, will you know whether it is true, or the shadow of a wrong label?