An Earthquake Drill Filed Under Football: The Label Error and the Silent Cost of Data

On my desk in Turin, there is a file I have opened and reopened for three weeks. It sits in the football folder. The classification label reads clearly: football. By the third line, I realised I was reading about an earthquake drill in Mexico — five hypothetical seismic scenarios, magnitudes from 5.2 to 7.7, taking place at noon on 19 September 2026.

Not a single player. Not a single club. Not a match, a contract, or a table. Only Mexico's National Civil Protection Coordination, known as CNPC, and a disaster-response training schedule.

I sat still in front of the screen for a long while. On one hand, this is small: one file lost among thousands. On the other, it is precisely these silent errors that frighten me most in any data system — and I have watched them cause real damage in the transfer market for years.

Context: data pipelines do not announce their own errors

Every day, hundreds of millions of football data points move through automated classification pipelines: news, match reports, contracts, player metrics, transfer rankings. No one reads them all by eye. The machine reads first. Humans verify later, and usually only where the machine appears doubtful.

The problem lives between those two steps. A classification error does not trigger an alarm. It is not like a missed penalty that the whole stadium sees. It is like a clumsy goal conceded in the 90th minute: only when the final score appears does anyone understand what happened.

I have known this kind of error for a long time. In 2026, when I was one of the few women with a press-room pass in Serie A, I sat in a room full of men and heard a colleague discuss the transfer of a South American striker. The figure quoted was three million euros off the real contract. No one checked, because the number was repeated throughout the meeting and became true by repetition. A meeting room full of men in 2026 taught me that the market trades in seating posture as well as players.

Since then, I cite sources at the end of every piece. Not to show diligence. So that next time a number goes astray, someone knows where it wandered from.

Where that file wandered from, I do not yet know. But I know how it wandered, and that matters more.

What is actually inside the file

This file describes the Second National Drill of 2026, a multi-region earthquake-response exercise organised by CNPC. Its content comprises five hypothetical scenarios: the Central region, with an epicentre at Tehuacán, Puebla, magnitude 7.7; the Northwest, with an epicentre at Mexicali, Baja California, magnitude 7.1; the Northeast, with an epicentre at Ciudad de Allende, Nuevo León, magnitude 5.2; the Yucatán Peninsula, with an epicentre at Cabo Catoche, Quintana Roo, magnitude 7.0; and the Gulf of Tehuantepec, with an epicentre at Tonalá, Chiapas, magnitude 7.6.

The drill date is fixed: Saturday, 19 September 2026, at noon Central Mexico time. That date is no accident. It commemorates the devastating Mexico City earthquakes of 2026 and 2026.

That is the whole content. There is no football. Yet the label still reads: football.

The mechanism of a silent error

I imagine three possibilities. First, the source's classification system bundles everything under a broad news tag, then mistakenly drops this into the football branch. Second, a regional mapping table inside the pipeline mislabelled it. Third — simpler — an error at the tagging stage of the archive itself, where a file was dragged into the wrong folder.

None of these is a disaster on its own. But they share one worrying trait: they are silent. No red flag. No one woken at midnight by an error email.

This is the point I want to stress, because it explains why football data pipelines are more prone to contamination than people think. A mislabelled file does not break anything immediately. It just sits there. It accumulates. When the system finally needs to reconcile, it surfaces — and by then it is too late.

The real infection vector: false co-occurrence

If this file were loaded into a football knowledge base, it would not produce an early error. It would produce something worse: a false co-occurrence edge between Mexican place names and any club sharing a local name.

Football data archives operate by linking entities. When they meet the word Puebla, they automatically link to Club Puebla. When they meet Nuevo León, they link to CF Monterrey and Tigres UANL. When they meet Quintana Roo, they may link to Cancún FC. When they meet Chiapas, they link to the clubs that once existed there.

But this file says nothing about those clubs. It only speaks of earthquake epicentres. The false link happens quietly, and from then on, a civil-protection exercise becomes raw material for a phantom causal map.

In the transfer market, this kind of contamination is far more dangerous than an obviously false rumour. A false rumour can be refuted with a single tweet. A false data edge buried in an entity graph is invisible until a player-valuation model returns a strange result and no one knows why.

Geographic overlap: coincidence, not evidence

The regions in the drill scenario happen to sit on the map of Mexican football. Puebla has Club Puebla in the top flight. Nuevo León has two major clubs. The Yucatán Peninsula has lower-division football.

I must say plainly: this overlap is coincidence, not evidence of anything. Anyone building a table of which clubs exist in these zones from this file is fooling themselves.

This is the trap I call confusing correlation with causation. A shared place name is not a relationship. Sitting on the same line of a data file does not mean a real connection exists.

The strongest football touchpoint, and why it still is not enough

The strongest possible touchpoint between the file and football is the 2026 World Cup, hosted by Mexico alongside the United States and Canada in June and July 2026. One of the scenario regions, Nuevo León, contains Monterrey — one of the host cities.

But the drill date, 19 September 2026, falls after the tournament ends. There is no operational overlap. The file never mentions the World Cup. And the scenarios are hypotheticals, not forecasts.

Here is what I want readers to remember: once you have seen a false correlation, you become very prone to seeing false causation everywhere. A good analyst must pull themselves back.

What the file gets right

Not everything in the file is suspect. One point deserves praise, because it is rare.

The file states clearly that the five scenarios are hypotheticals, not earthquake forecasts, and that there is no prediction of five earthquakes occurring. This is a proactive anti-misreading clause. In football media, I wish injury reports carried such clauses too.

The dates are also recorded accurately and consistently. 19 September 2026 is indeed a Saturday, matching the figure in the article.

A mislabelled file whose content is more disciplined than many correctly labelled football articles. That is a paradox worth pondering.

The risk table: all empty but one cell

When I ran the file through the risk framework I normally use for deals, I got a strange result: nearly every cell was empty.

Sporting risk undefined, because no sporting subject exists. Financial risk undefined, because not a single monetary figure appears. Personnel risk undefined, because no one is named. Rules risk undefined, because no football legal system appears.

The only lit cell is the systemic one: the risk of a non-football file contaminating a football corpus. High severity, high likelihood, medium-to-high impact.

I have used this framework for years. Never before have I seen a file whose biggest risk cell sits not in its content, but in the label attached to it.

From eight World Cups: data always precedes the story

I have reported on eight Olympic Games, eight World Cups, and many editions of the major cycling tours. Across all those events, one rule repeats: the story always arrives after the number.

People tell the tale of a national team through inspiration, through a last-minute goal, through a moment of transcendence. But before the inspiration, there is always a dataset: kilometres run, sprints made, passes completed. Football is a sport whose emotion is built on a foundation of numbers, even if the audience never sees that foundation.

So when the foundation is contaminated, the story built on it drifts too. A mislabelled file kills no one. But it contributes to a transfer decision being made on unstable ground.

I work in transfer-market administration. My job is to value people with numbers. If the input number is wrong, everything downstream is wrong, including the conclusions that sound most reasonable.

Agent noise, algorithm noise

The player agent is the largest hidden cost in the transfer market. I have said this for years. They generate noise: rumours, leaks, inflated figures to push prices. On one hand, that noise distorts the market. On the other, it is easy to spot, because all noise has a visible source.

But there is another kind of noise that is harder: algorithmic noise. It has no clear source. There is no face to suspect. It is just a file in the right folder, in the right format, in the right interface, but with the wrong content.

Between the two, the second is more dangerous, because it wears the clothes of reliability. Everyone knows to suspect an agent's rumour. Few question a data line generated by a system.

This is why I track lost files the way I track dubious contracts. They belong to the same family: they look fine until you open the detail.

The contrarian angle: loud errors beat silent ones

There is a natural reflex on finding an error: find the cause, then fix it. But with data, the right question is not where this error came from, but why it did not announce itself.

A wrong penalty makes the whole stadium howl. A wrong figure in a label never howls. That is why large data organisations deliberately create noise: red flags, review queues, error-rate thresholds. They understand that silent errors cost more than loud ones.

This is where I think of an old image. The empty stadium of 2026 was not a silence. It was a warning sign few read in time. Mislabelled files are the same: they are politely empty, and that very politeness makes us overlook them.

In football, people love stories. But data does not tell stories. It just sits there, right or wrong. No one calls Croatia a miracle when they have each run 400km on Russian soil. The number needs no narrator. The number only needs the right label.

What to watch next

I will track four signals, and I suggest football data people do the same.

First, whether the label is corrected in the next pipeline run. If corrected, the error is file-level. If not, it may be systemic.

Second, sibling files in the batch. If the same supplier has other non-football files also tagged as football, it is a cluster-level error needing model retraining.

Third, false-co-occurrence thresholds. Any pipeline linking Puebla to Club Puebla without textual evidence should be blocked.

Fourth, CNPC's official documentation. If they publish the scenario annexes, we can independently verify the 5.2-to-7.7 figures.

Why a lost file is worth writing about

I know someone will ask: what is there to say about one lost file?

My answer lies in the number. In a pipeline ingesting hundreds of millions of data points daily, a mislabelling rate of even 0.1 percent is enough to create hundreds of thousands of false edges a day. Multiplied over a year, that is a layer of noise nobody cleans.

And that layer does not sit still. It flows into valuation models, into rumour rankings, into the numbers transfer administrators use to decide on spending tens of millions of euros. I have been in this profession long enough to know that the costliest mistakes rarely begin with a wrong decision. They begin with wrong data that no one checked.

The player agent is the largest hidden cost in the transfer market, I once said. But there is a larger hidden cost: dirty data. It sends no invoice. It just quietly leads everyone to make wrong decisions.

Closing

Three weeks ago, I opened a file in the football folder and read about earthquake epicentres half a world away. I could have closed it and forgotten it. But I thought of the other files, the ones I have never opened, sitting in the right folder while holding the wrong truth.

In football, people review goals with VAR. Even an obvious goal must pass a few seconds of replay. But data has no VAR yet. No one has built a review room for the numbers running through pipelines at three in the morning.

Perhaps it is time to build one.

And if you work with football data, start with the smallest task: open any file in your football folder and ask yourself whether it truly belongs there. Because an earthquake drill in Mexico just reminded me that, in the world of data, the label is sometimes more important than the content.