A Mexican INAPAM Card Ended Up in Football Data: 42 Years in the Trade, and the First Time I Saw an Algorithm Score an Own Goal

In 2026, I opened a football data package in Shenzhen to prepare a transfer-window analysis. Inside was a document of 18 information points. I read the first line: the credential is granted to applicants aged 60 and over. I read the tenth: an applicant must supply a CURP code. By the fifteenth, the document warned citizens not to pay middlemen, because the procedure is free. No players. No clubs. No matches. It was a guide to obtaining the INAPAM card, the senior-citizen credential issued by Mexico's National Institute for Older Adults together with the Ministry of Welfare. A civil-administration document, entirely foreign to football. And yet the domain label attached to the data package read exactly two words: football. I once got one thing right and the rest entirely wrong — this piece is about the part I got right. What I got right here was not a judgement about any team. It was a systemic failure. And to someone who has worked 42 years in this trade, a systemic failure is worth more than any hot take. People call me a controversialist. I treat that as a job description. But there is one argument I never join: an argument with a wrong label. When the label is wrong, every analysis born of it is organised fabrication. Over the past two decades, the sports-data industry has become a vast ecosystem. Every day, millions of records are pushed into automated pipelines: match statistics, player profiles, transfer data, expected-goals models, pressing indices, conversion rates. Companies build models, build indices, build rankings. All of it rests on an unspoken assumption: that the incoming data belongs to the right domain. That assumption has just been shattered by a document about a Mexican senior card. I spent six months during the pandemic learning expected-goals models. I know one thing about data: garbage in, garbage out. But there is a more dangerous variant few mention: garbage in labelled clean means a catastrophe out. A transfer-prediction model trained on data mixed with Mexican administrative procedure will not throw an error. It will return a number. A number that looks highly professional, enough for some pundit to read on air as truth. Viewers will believe it. Because numbers don't lie, right? Wrong. A number only repeats the label someone stuck on it. That is when I realised the most dangerous thing in our industry is not fake news. Fake news still invites suspicion. The most dangerous thing is false data presented through a correct interface. The point most people in the industry miss: this is not the fault of an algorithm. It is the fault of a process. No check layer, at any step, verifies that a record labelled football actually contains a football entity. No gate counts the clubs, players, competitions or coaches in a text before letting it through. If one existed, the INAPAM document — with zero football entities — would have been stopped at the door. The lesson lies in the zero. Clubs in the document: zero. Players: zero. Competitions: zero. Coaches: zero. Four zeros in a row. Four zeros that should have screamed that something was wrong. But the system never heard the scream, because the system was never built to hear it. Pause on the smallest detail, the one a hurried reader skips. The CURP code stands for Clave Unica de Registro de Poblacion — Mexico's unique population-registration code. A citizen identifier. No player on earth carries this figure in a transfer file. But feed it into an unvalidated model and the model will try to give it meaning. It will find a pattern. It will produce a correlation. And that correlation will be garbage — but systematic garbage. This is the most counter-intuitive part, the part where I might be wrong. The problem is not weak technology. The problem is that we trust technology so much we skip the very first check. We build elaborate pipelines to process millions of records a day, then act surprised when a bad record slips through. Meanwhile an editor with ten minutes and one simple question — does this text mention any player — would have caught it. We do not lack tools. We lack a gate. A gate that counts entities, placed before content is labelled, would cost less than any machine-learning model we boast about. I might be wrong about the solution. A manual gate may not scale to millions of records a day. But I am certain about the diagnosis: a process with no domain-verification step is a process that scores own goals. And an own goal needs no clever opponent — only a moment of lost concentration. I am 58. I have seen everything. But I had never seen a football database score an own goal against itself. Football has not run out of ways to surprise me; this time it surprised me where I least expected: the intake stage. Forty-two years in journalism taught me the scariest thing is not an unexpected result on the pitch. The scariest thing is a wrong result entering the system and being repeated as fact. No idea is too wild to be worth testing — the pandemic taught me that, and a Mexican senior card taught me the same. If you work in sports data, this is your homework. Do not just build more models. Build a gate. A gate that counts entities and blocks out-of-domain content before it is labelled. Because every index you are proud of, every model you advertise, every ranking you publish, rests on one assumption: that the input data belongs to football. The day that assumption collapses, everything collapses with it. Morocco reached the semi-finals, I am 58, and football still has not run out of ways to surprise me. This time it surprised me with a senior card. And if you ask what I predict next: I predict more strange documents are sitting in our databases, waiting for someone to count entities and find them. The question is not whether they are there. The question is who will open the first verification gate.