A Cat Sterilisation Notice in Iztapalapa, and the Labelling Hole in Football's Data Industry

At seven in the morning on September 30, in Iztapalapa — the most populous borough of Mexico City, home to roughly 1.8 million people — a line formed along the pavement outside the intake point. No scarves, no flags, no drums. In their hands were carriers, and inside the carriers were cats.

The entry conditions were strict. Cats had to be between six months and six years old. Owners had to fast them for six hours before arrival. No vaccines within 15 days. Not pregnant, not lactating, not in heat, not a flat-faced breed. A vaccination card was mandatory. The service was entirely free.

A record of that event sits inside a content aggregation system I have access to. It carries the label: football.

I am not joking, and I have no intention of writing an article about cats.

---

I have been in this trade for more than twenty years, eleven of them hosting a late-night football programme, and I spend most of my working hours reading data rather than scorelines. At this stage of my career, the job comes down to one question: how was this record produced. Not what it says, but who typed it in, by what criteria, and which system agreed with it.

With the 2026 World Cup, that question gets much harder.

The tournament opens on 11 June 2026 at Estadio Azteca in Mexico City, according to the schedule FIFA has published. It becomes the first stadium in the world to host matches at three separate World Cups: 2026, 2026 and 2026. It is also the ground where, on 22 June 2026, Diego Maradona scored the goal known as the Hand of God. This year's edition expands to 48 teams, 104 matches, across three host nations and 16 cities. After renovation, the Azteca holds more than 80,000.

For nearly thirty years, the working definition of "football" inside data systems has been narrow. A match has two teams, a referee, a scoreline. A football story has a club, a player, a manager. Anything that did not fit the mould was pushed out.

In 2026 I wrote that Manchester City paying 50 million pounds for Kyle Walker was a tactical mistake, and bet my editor they would lose at least three home games before the turn of the year. They lost one, won the title with 100 points, and Walker contributed six assists. People laughed at me over Walker. Three years later, they were laughing through tears at the price of defenders.

That shock killed my habit of betting on instinct. I started tracing every number back to its source before using it. That principle is what took me to Iztapalapa.

---

A modern football data pipeline runs through three stages. A crawler scrapes content. A classifier assigns a topic label. An index pushes the record into downstream products: rankings, sentiment indices, pricing models, editorial recommendations, and the training sets themselves.

The classifier does not understand football. It counts the co-occurrence of entities. "Mexico City" plus "September" plus "event" plus "registration" plus "free" produces a vector that, in a training corpus shaped by a World Cup year, sits almost permanently beside host-city coverage. For a model fine-tuned on a Mexico City 2026 corpus, that phrase lands in the "football" bucket at a probability far above what it deserves.

What is missing is a gate. Almost no system asks: does this record mention a specific match. They ask which team, which player. When the question is wrong, the answer is wrong too, and it is wrong silently.

What happens next is the part worth discussing. The record enters the index. A sentiment reading on Mexico City ticks up, because one more document has been sorted into a trending topic. A model estimating market liquidity reads the total volume of host-city football text as a demand signal. An editorial recommender proposes promoting the item. Nobody in that chain reads the content.

The error of one record is negligible. The error of a system is not.

The core point is not the cat. It is that football has built an entire quantitative layer on top of pipelines almost nobody audits. Over two decades, clubs have hired hundreds of data scientists. Yet most of the numbers fans argue about every weekend come from third-party aggregators nobody names. We read them, argue about them, rank players by them, and have no idea how many gates they passed through.

Based on my experience watching matches from the stands at the Azteca and at lower-league grounds in northern England, one observation holds. Insiders are excellent filters. They look at a table of numbers and automatically discard the meaningless rows, because they carry fifteen years of context in their heads. Outsiders have no such filter, so they are forced to ask. And it is the naive question — why is this record here — that exposes the fault.

I do not disagree because I want to be different. I disagree because the majority has been wrong about me before.

---

Now the part where I argue against myself.

The likeliest scenario: this is noise. Mislabel rates in large pipelines typically sit between 0.1 and 1 percent. Across a corpus of tens of millions of records, tens of thousands of bad labels are invisible to every statistic. If that is the case, I am inflating a small matter, and this piece is just another sound in the noise.

I accept that risk. But one detail stops me from dismissing it.

World Cup 2026 is the first edition in which peripheral text outweighs core text. Three host nations, 16 cities, 48 teams, 104 matches. Airports, transport, security, public health, municipal services, rents, vaccination drives — all of it is World Cup content in an operational sense. When peripheral text outweighs core text, a classifier calibrated on core text degrades. An error rate measured on old data no longer holds on new data.

The second scenario is harder to stomach. Perhaps the classifier is right, and my definition is the outdated one.

Four decades of hosting World Cups have turned the host city into an operating system. In the week of the opening match, everything in Mexico City sits inside that system: the metro runs longer, hospitals add shifts, schools reshuffle timetables, and public health programmes get pushed forward to ride the traffic. If football in 2026 means that entire machine, then a cat sterilisation notice in Iztapalapa falls inside the coverage — not in essence, but in operation.

I do not like that conclusion. It sounds like an excuse for sloppiness. The pitch and the esports arena share one law: whoever pretends will be exposed. If I call a wrong label a right one to protect a system, I am pretending, and I will be exposed at the next World Cup.

What I know for certain is this. A classifier that cannot tell a cat sterilisation day from a football match cannot tell many other things apart either. The problem is not the Iztapalapa record. The problem is that we have grown used to reading the output of systems nobody understands and calling it data.

---

At 43, I still speak hot, but the fire has learned to wait.

So here is a falsifiable prediction. Before the referee blows the whistle at the Azteca on 11 June 2026, at least one large football data pipeline operator will publish its labelling logs, or publish a quarterly misclassification rate. If nobody does, then by the next World Cup we will still be reading numbers nobody dares to guarantee.

If you want to know how clean your football data is, there is a small exercise. Open the system, find a record about cats, and see where it has been filed.