Labelling Failure in Football Data Pipelines: Lessons from a Mexican Film File

A file named "Cocodrilos" sat inside a football data folder. Beside it, the statistical columns lined up neatly: xG, PPDA, passes into the box, aerial duel win rate, progressive carries. Not a single checkpoint in the processing pipeline noticed the file contained no match at all. It contained an exclusive conversation with director J. Xavier Velasco about the film "Cocodrilos" — a work about violence against journalists in Mexico, carrying six Ariel Award nominations and a September 24 theatrical release.

No club. No player. No coach. No contract, no wages, no release clause. One label at the top of the file: football.

I read that report on an evening in Incheon, just after filing my last transfer-window piece of the day. Fourteen information points had been extracted. Not one of them belonged to football. What kept me at my desk was a line near the end, written in dry administrative prose: this item is mislabelled, and eight of nine analytical dimensions cannot be applied at all.

A film. A director. A fictional character named Santiago, a photojournalist. A national film award. A release date. And at the top of the file, one word: football.

The pitch does not lie — only the writer's heart lies to itself. In this file, there was no writer at all. There was only a labelling engine, and a process that trusted it.

A pipeline with no stands

To understand what happened here, you have to look at how football content is produced at industrial scale as of 2026. A football article today rarely begins with a reporter in the stands. It begins with a pipeline.

That pipeline runs through four stations. The first ingests raw material from hundreds of feeds: club releases, federation statements, foreign press, social streams, press-conference transcripts, court filings on transfers. The second labels each item — football, basketball, tennis, film, politics — and that label decides everything downstream. The third decomposes the content into atomic information points: who, what, when, how much, sourced where. The fourth applies a professional analytical framework to those points and publishes.

Get station two wrong and everything after it is worthless. Not worthless in the sense of poor quality — worthless in the sense of a category error. Placing a football analytical framework on a film interview is a logic failure, the equivalent of grading a movie with a league table.

The technical cause is specific. Among the fourteen information points extracted at station three, one contained the word "attack". It sat inside the phrase "attacks on journalists". To a bag-of-words classifier, "attack" is a powerful football signal: pressing, counterattacking, pressure. A naive model reads "attack" and nods. A sophisticated model asks: attacking whom, how, under what framework, and who bears the loss.

The transfer window makes all of this worse. Content volume spikes, an item's shelf life is measured in hours, and update pressure makes any gate look like an obstacle. When a newsroom races on volume, station two is the first budget line cut. The label becomes a formality, and nobody reads a formality carefully.

The consequence: an article about violence against the Mexican press sits in the same database as scouting reports. The film "Cocodrilos" by director J. Xavier Velasco, published as an exclusive by the outlet CONTRA, with six nominations for the Ariel Award presented by the Mexican Academy of Cinematographic Arts and Sciences (AMACC), shares a folder with data on the players I am tracking for my transfer bulletin.

That is the whole story at surface level. The rest is what matters.

A label is a routing decision

A label sounds harmless. In practice it is a routing decision carrying an entire analytical paradigm with it.

Once an item is labelled football, it is pushed onto a track of pre-set questions: which formation, what shape, what expected goals, what pressing intensity, what wage structure, what release clause. Those questions did not invent themselves. They exist because someone designed them for a particular type of event, and they work extremely well for that event.

For a film interview, the system's nine professional dimensions become nine empty rooms. The tactical and technical dimension has nothing to measure. Club finance and transfer market has nothing to calculate. Results and public-opinion cycles have no table to compare. League landscape has no league. Rules and governance has no governing body. Management and dressing room has no dressing room. Industry transmission has no football chain.

Eight of nine rooms empty. The notable part is how the report handles that emptiness.

The most expensive silence in the document

In every empty room, the report does not write a paragraph of speculation. It writes one sentence: insufficient information, cannot assess.

That sentence repeats eight times. And across the whole document, it is the most valuable thing there.

I know why it is valuable, because I once did the opposite. After South Korea beat Germany 2-0 in Kazan on June 27, 2026 — a win that was not enough to advance — I wrote 2,500 words and received 47 comments, nearly half of them calling me sentimental and saying I did not understand football. I spent three weeks rewatching the tapes, checking every touch against the emotions I had described. I still trusted my instinct. But I learned something else: without anchoring to a concrete detail, every beautiful sentence becomes a debt.

The report on the "Cocodrilos" file chose to repay before borrowing. It refused to write when there was no basis to write. In an industry measured by publishing volume, that decision is an anti-economic act. A document with no professional conclusion is a document that generates no articles. Nobody pays for a page that says "we do not know" eight times.

Which is exactly why it is worth reading. I have spent thirteen years in this trade, across eight Olympic Games, eight World Cups, multiple Giro d'Italia and Tour de France editions. Young journalists are taught they must have an angle, an argument, a conclusion. Few are taught that one of the hardest skills is knowing what you do not know.

A financial category error

One detail deserves a longer pause: the report calls applying a football finance framework to a film a category error.

A film's economy runs on production budget, marketing and distribution spend, screen count, box-office splits. A football club's economy runs on broadcast revenue, commercial revenue, wage bill, net debt. Both systems share one word: money. They do not share a balance sheet, a cycle, or a way of measuring success.

The report refuses to merge them. It records: insufficient information, cannot assess, and adds that any attempt to connect the two systems would be fabrication.

I wish that refusal were more common in my trade. We live in an era where every sport is read through the football transfer framework. An esports coach is judged in the language of the player transfer market. A track athlete is measured by release-clause logic. A regional league is analysed with an index that only means something at continental level. The convenience of a ready-made framework makes people forget it was designed for something else.

In this film file's case, the confusion is harmless because nobody reads the output. The mechanism is not harmless. The same mechanism, applied to a real transfer story, produces a very confident and very wrong conclusion.

Six Ariel nominations and September 24: an unresolved contradiction

This is the detail I want readers to remember, because it is a perfect example of what automation loses.

The article states two hard facts. First, the film has six Ariel Award nominations — Mexico's principal national film prize, presented by the Mexican Academy of Cinematographic Arts and Sciences. Second, the film opens in Mexican cinemas on September 24.

Those two facts sit awkwardly together. Ariel eligibility conventionally requires a prior theatrical exhibition window in Mexico. A film arriving in cinemas on September 24 would ordinarily not yet be eligible for the same awards cycle. Three possible resolutions: the film had an earlier festival or limited run and September 24 is the wider commercial release; the nominations belong to a different cycle or edition; or this is a paraphrase error in the extraction stage.

I do not have enough data to conclude which is correct. Neither does the report, and it says so plainly, with a medium confidence tag.

My point lies elsewhere: a human editor would circle this. A counting system will not. It records "six nominations" as a fact, pushes it into a database, and that fact will live there for a long time, cited again and again, used as evidence in some argument about the film's quality, while the contradiction is left far behind.

This is what years of bulletin work taught me: hard facts travel easily, but the contradiction between hard facts is where the truth lives. And contradiction cannot be automated.

A single source and the "objective" label

In the report's quality checklist, one item stands out: the "objective" label assigned at the extraction stage is judged wrong.

The reason is simple. The article is an exclusive interview with the film's own director. The only named source has a direct stake in how the film is received. When an outlet grants a subject the right to frame their own story, that is promotional conduct, not neutral observation. The report recommends re-tagging it as "promotional / first-party source".

I read that line and thought immediately of a press room at a training centre I once visited. An agent talks about his client in the language of an investment prospectus. A reporter transcribes verbatim. Hours later the quote appears on a front page with the word "revealed" in the headline. Nobody in that chain lied. But the chain together produces something called information, when its nature is advertising.

A good labelling system distinguishes the two. A bad one files both in the same column, and that column becomes the foundation for every later analysis.

A metric measured by volume

One recommendation in the report matters more than all the others: tracking the misclassification rate as an operational metric.

Almost nobody measures this today, because industry incentives point elsewhere. People measure articles published per day, page views, engagement, dwell time. All volume metrics. Nothing measures classification accuracy, because a mislabel does not produce a visible traffic drop this week. It produces a database that slowly contaminates over time.

And when data contaminates, the consequence does not appear in the wrong file. It appears in the right one. If my database holds a film interview labelled football, the model trained on it learns that content about violence and journalism relates to football. Months later, when I ask the system about a violence story connected to a match, it answers with a confidence it has no right to have.

That is why an error at station two is not a small error. That station does not create content. It creates direction.

Two transmission chains, side by side

The report draws two transmission chains and deliberately places them next to each other.

The film chain: research and subject matter, production and distribution, awards season, then public debate on press freedom. The football chain: academies and scouting, the agent ecosystem, broadcasting and commercial rights, capital networks, derivative markets, the national-team ecosystem.

These chains do not intersect. The report states plainly: no inference should be drawn about academies, agents, broadcasting, capital networks or betting markets from this item.

I noticed the word "betting" in that sentence. If anything is currently erasing the boundaries between sport's transmission chains, it is the pressure to read every event as a bettable signal. An injury story is no longer read as news about a person's health. It is read as a line movement. A personnel decision is no longer a story about a family relocating. It is a variable in a model.

The report fences off the film file and says: do not cross here. In thirteen years I have not seen many fences. I have seen more bridges.

What the machine learned from us

Here is the most uncomfortable part.

The first reaction to a mislabelling story is to blame the algorithm. Stupid classifier. Outdated bag-of-words model. Pipeline missing a gate. All true. But it misses a simpler fact: the machine learned from us.

For years, football journalism has labelled non-football things as football. A player's tattoo. A coach's divorce. A stadium's new menu. A social post deleted after three minutes. Each time, we taught any system reading us a lesson: content about famous people in football is football content.

The machine did not invent category blur. It industrialised it. The "Cocodrilos" file is a mirror held up to us. If the word "attack" in an article about violence against journalists is enough to trigger a football label, the right question is elsewhere: why, in the training data, is that word so tightly bound to a single meaning?

The scariest scenario is elsewhere

The report rates the highest risk on one item I believe is central: the risk of generating fabricated football output to fill a template.

The scariest scenario is elsewhere. It is not that a wrong file enters the database. It is that the file enters, and then somebody patiently fills the template. The template is always available: projected lineups, pressing intensity, wage structure, sanction scenarios. Once filled, the output stops looking like an error. It looks like a scouting report. It has figures. It has jargon. It has confidence.

And the scariest part of all: nobody would notice. Nobody audits a document that looks normal. People only audit documents that look wrong.

In May 2026 I sat in Jeonju World Cup Stadium for Jeonbuk Hyundai Motors against Suwon Samsung Bluewings, with zero spectators. The K League had returned after a four-month freeze. I could hear the ball against boot cushioning, coaches shouting instructions, players breathing hard on the bench. My piece on that match ran 1,800 words and reached 12,000 reads, thirty times my usual.

What I learned that day was not about the numbers. It was that when noise disappears, people actually see each other. An empty applause is still a piece of music — if you know how to listen.

A data pipeline has no stands. It has no applause to create noise, and no applause to warn it. When it errs, it errs in absolute silence.

The sentence my trade has lost

If I had to take one thing from this document, it would be the sentence that appears eight times inside it.

Insufficient information, cannot assess.

Thirteen years, eight Olympic Games, eight World Cups. In that time this sentence has almost vanished from football journalism. Every match must now have a story. Every transfer must have a reason. Every silence must have a meaning. Every defeat must have a lesson. And whenever we have nothing to say, we say we need more time — and then say something anyway.

Readers dislike emptiness. I understand. But honest emptiness is more useful than fabricated completeness. A bulletin that says there is nothing yet brings the reader back tomorrow. A bulletin that invents a reason makes the reader believe something false for years.

In a transfer window, that difference is worth money. A club builds a season plan on a wrong detail about a contract clause. A supporter buys a shirt with a name that leaves in three weeks. A small outlet republishes an unverified figure, and that figure becomes truth in the mouth of an entire community.

That document was the first thing I read this year in which the word "unknown" appeared more often than the word "likely". It is not appealing. It has no conclusion. It cannot be published as news.

It is right.

What remains

The pipeline will be fixed. Someone will open a ticket, add a gate, adjust a threshold, re-tag a file. "Cocodrilos" will be routed to the correct desk, and J. Xavier Velasco's film will be discussed by people who genuinely care about cinema, about the story of violence against the Mexican press, about what it means to retell stories others want buried.

But one question followed me after I closed the file, and it has nothing to do with algorithms.

If the system were never fixed, and the machine kept producing confident tactical reports about films, how long would it take us to notice?

I ask that while my trade races toward automation, and the answer I fear most is: longer than we think.

We watch sport not to escape life, but to understand it better. A data pipeline built to understand football should be built the same way: not to fill every gap, but to know which gaps should be left open.

The match ends, but the bulletin of memory never runs out of time.