When Data Gets Mislabeled: Lessons From a File That Doesn't Belong to Football

There are numbers that don't appear on the stat sheet, they exist between two touches of the ball.

And sometimes, they exist somewhere no one would ever expect.

Last week, I received a data set from the sports news aggregation system I use to track major leagues. In that batch, there was a file labeled "football." The content inside was entirely different.

It was an explainer document about property tax policy in Mexico City — specifically predial tax discounts, water fees, and INVI housing credit programs. No clubs. No players. No tactics. No scores. But at the top of the file, the words "Domain: Football" were still lit up.

I stared at the screen for a while. Not because I didn't know what to do. But because I asked myself: if a system can mislabel a document this obviously, how many other errors are lurking that I haven't seen yet?

That's why this article exists — not to analyze a team's tactics, but to analyze a flaw in how we receive and process football data.

This incident is not isolated. Over five years of working as a data consultant for several clubs and sports information platforms in Southeast Asia, I've witnessed automated text classification systems malfunction more than once. The causes typically come from three directions.

First, the classifier algorithm is usually trained on high-frequency keywords from the training dataset. If a source document about Mexican tax policy coincidentally contains words like "club" (taxpayer association), "league" (tax federation), or "transfer" (money transfer), the algorithm can confuse it with football.

Second, ingestion errors can occur when the pipeline pulls the wrong file from an unchecked folder. In this case, the Mexican tax document clearly came from a source unrelated to football.

Third, and most concerning, is human subjectivity. An editor or data technician may have mislabeled it manually, and the automated system then inherited that error.

I don't have enough data to determine which scenario this specific file falls into. But what I can state with certainty is: any football analysis model consuming data from this file will produce completely noisy results. Not numerically wrong results, but meaningless ones.

Imagine an xG algorithm being fed data about a 30% predial tax reduction. It would try to find correlations between the number 30% and a team's expected goals. The result? Meaningless numbers. Unfounded charts. Misleading conclusions presented as deep analysis.

This isn't just a technical problem. It's a problem of trust.

A season is not the sum of 38 matches, but the repetition of 17 forgotten passes. And a data system is not the sum of millions of records, but the accuracy of each individual record. Just one wrong file in the right place can affect the entire analysis chain downstream.

In my work, I constantly cross-check data from multiple sources. That habit formed in 2026, when I worked as a part-time statistics assistant for a football website in Singapore during the World Cup in Russia. My task was to code every touch of the Spain 3–3 Portugal match. I discovered that Cristiano Ronaldo reached a maximum speed of only 9.8 km/h in that game — below Portugal's team average of 11.2 km/h. But all five of his shots on target came from close-range situations.

My analysis of the "unusually narrow pitch" received over 200,000 views. But what I learned from that experience wasn't how to write a viral piece. It was how to verify a number before believing it.

If I hadn't coded every touch myself, I wouldn't have discovered that Ronaldo moved less than his teammates but more effectively. If I had relied solely on average speed data from a single source, I would have missed the real story.

Now, let's return to the mislabeled Mexican tax file.

What's noteworthy is that the document's content — structurally speaking — wasn't bad at all. It clearly explained who qualifies for tax reductions, who doesn't, by how much, and by when. Numbers like 30% predial tax reduction, 68 pesos bi-monthly, 50% water fee reduction, and INVI credit support levels of 15%/25%/20% were presented coherently.

But there was one point I discovered upon close reading: several figures lacked clear sourcing. Specifically, the 30% and 50% reductions weren't tied to any official document. Meanwhile, other numbers like the cadastral value ceiling of 2,808,466 pesos were sourced from the Secretaría de Administración y Finanzas.

This inconsistency isn't unique to the Mexican tax document. It's a common problem in many sports reports I've read. Numbers cited without sources. Statistics presented without context. And conclusions drawn from unreliable data.

Clubs dissolve, football stops. But data never stops telling stories. The question is whether that story is true.

In a corridor, if you only look toward the light, you'll miss what stands in the darkness. And in a large dataset, if you only trust the classification label, you'll miss the errors silently spreading.

I was once called "what does a girl know about football" after writing an analysis of Mesut Özil's 17 key passes in the Premier League. That article sparked controversy on a major football forum because I pointed out that Arsenal's xG ranking declined when Özil didn't start. Instead of deleting the post, I added three more charts and match-by-match data annotations.

From that, I learned one thing: data doesn't defend itself. The data user must do that.

And sometimes, defending data begins with admitting we might be wrong. That our systems might be wrong. That the numbers we thought were most certain might be in the wrong place.

The Mexican tax file labeled "football" is a reminder. Not about Mexico. Not about taxes. But about how we build trust in data.

If a system can confuse tax policy with football analysis, then more complex models — the ones we use to evaluate players, predict results, value transfers — might be making similar errors. We just haven't discovered them yet.

The question isn't whether your system has errors. The question is whether you'll discover them before they affect your analysis results.

I heard a goalkeeper talk about how she reads the shooter's belly button, something that doesn't appear in a data export file. And I wondered: how many other signals lie beyond our vision, simply because we trust labels too much?

This season, as you follow every match and try to find the tactical currents beneath the league table, take a moment to check your own data sources. Not to doubt everything. But to know that sometimes, the biggest error isn't in the number, but in where we place it.

Because in football, as in data, truth often lies in the least expected places. And sometimes, it lies right inside a mislabeled file.