When a Mexican Constitutional Vote Landed in My Tactical Data Table

In my quarterly audit, I cross-check the entity fields in my personal data system. One row surfaced with the label "football." Inside was a vote in Mexico's Chamber of Deputies: 334 in favor, 109 against, one abstention. No club. No player. No coach. No match, no contract, no transfer figure. The only quantitative datum was the margin of a legislative roll-call — 334 out of 444 votes. I sat looking at that row for a while. I have spent twenty-nine years reading football through numbers, four of them doing nothing but redrawing 38 pressure diagrams from 4,500 wide-attacking situations in Serie A. And yet my system had just filed a Mexican constitutional text into the same drawer as the GPS data of a wing-back. I call that the first error in a chain of errors I will lay out here. The reform I am describing amends Articles 82, 116 and 122 of the Constitution of the United Mexican States. Those three articles define the eligibility requirements for three executive offices: the Presidency, state governorships, and the Head of Government of Mexico City. The amendment requires candidates to hold a single nationality — "sole nationality." The obligation extends through the entire term of office: no acquiring or requesting another nationality, no invoking a foreign nationality before foreign authorities, with accompanying restrictions on foreign passports and on the diplomatic protection of another state. Three scope limits are stated explicitly, and I stress them because any summary that omits them misstates the norm. First, the restriction applies only to the three executive offices above. Second, it does not extend to candidates for the Congress of the Union or local legislatures. Third, it does not abolish dual nationality for Mexicans generally. Procedurally, the Chamber of Deputies approved it after a multi-hour session. The Senate, ratification by state legislatures, and publication in the official gazette are not recorded in the data I hold. The amendment carries a transitional clause: if the process completes, it applies from the 2028 electoral process. Now comes the part that forced me to write this piece. How did a text like that end up inside a "football" labelling system? The answer lies in a single keyword: "nationality" — nacionalidad. In football, "nationality" is one of the highest-salience term-fields. It appears in every debate about naturalised players, about dual nationals, about young players eligible for two national teams, about the one-time association switch. Articles 5 to 8 of the Regulations Governing the Application of the FIFA Statutes govern exactly that question: which national team a player may represent. Any text-processing engine that relies on keyword matching alone will hit "nationality" and pull the article into the football drawer without ever checking whether a club is in it. This is where I tell you how I once built a system the other way around — and how it tricked me. In 2026, when I published a 6,000-word analysis of Gasperini's Atalanta, I used GPS data from 37 Serie A matches. I showed that Robin Gosens was not an ordinary full-back. He received the ball an average of 21.4 times inside the box per match, more than the first-choice striker. I called him a "wide number 10." The piece was republished and it opened a professional door for me. But what that piece did not show me was a lesson about labels. When I attached the label "full-back" to Gosens and then read the data through that label, I thought I was reading a position. In fact I was reading my own preconception of the position. It took me three months to realise I had misread that position — not wrong about where Gosens played, but wrong about believing the label "full-back" could describe what he did. A label is an assumption, not a fact. And a wrong label produces an entire chain of wrong conclusions behind it. That is exactly what happened with the Mexican data row. The "football" label was assigned on the basis of one keyword, and once assigned, every analytical step behind it was forced to run inside the football mould. Had I not stopped to check, the system would have produced things like "the ruling bloc's dominant control," "the opposition's low block," "75.2% possession." All meaningless. And the frightening part is that they are meaningless yet sound entirely plausible. A number can be repurposed for a purpose it never belonged to. The 334-of-444 figure — 75.2% — is a legislative instrument. It measures consensus within an assembly, not performance. If I set it beside xG or PPDA, I create what is called a category error: treating a concept from one domain as if it belonged to another. A category error does not make a piece visibly wrong — it makes it subtly wrong, and subtle is more dangerous. I once taught young contributors a principle: before concluding anything, ask what the system has hidden before you judge a defender. That principle applies here. My system had hidden the fact that the article contained no football entity at all — by giving it a label it should never have received. Now I want to go deeper into the mechanism, because this is the part that is useful to practitioners. A typical football data pipeline has four layers. The ingestion layer pulls text in. The classification layer assigns labels. The analysis layer runs models. The presentation layer pushes results out to readers. The error here sat in the second layer. But the real damage happened in the third and fourth. The analysis layer has a property I call "production pressure." Once a row has been tagged football, the analytical module feels obliged to return football conclusions. It has no mode for "nothing to say." It is not programmed to return "N/A — insufficient information." So it fabricates. It takes article names as team names, vote counts as possession metrics, the 2028 horizon as "next season." The result is not a syntax error anyone can see. The result is a fluent, confident, entirely wrong analysis. This is why I tell anyone working in sports data: a null result is a valid result, and must be treated as valid. If you reward a model for always producing an answer, you are rewarding it for fabricating. It took me a long time to understand this, because for years I too believed silence was failure. Let me return to the 4,500 wide situations. Three months of isolation during the pandemic summer, 4,500 Serie A wide phases from 2026 to 2026, and 38 hand-drawn pressure diagrams. I do not tell that story to talk about diligence. I tell it because inside that pile of data were rows I had labelled entirely wrongly, and I had to peel the labels off by hand. Some phases I had logged as "left-wing attack" turned out to be central build-up where the flank was only the final receiving point. Some I had logged as "zonal defending" were in fact man-marking inside the box that dropped late. That mislabelling had one common cause: I started from the categories already in my head, then stuffed events into the categories. I worked from the label down, instead of from the entity up. That is the system-builder's trap. Beautiful on paper, but detached from the events — and in the end, an architecture detached from the pitch is not architecture, it is a building with no ground. A "football" labelling system that swallowed a Mexican constitutional text is working from the label down in the most literal sense. It has a drawer ready, and it needs to put something in it. The word "nationality" is the hook. Done. Here I want to pull the story out of the pipeline and back onto the pitch, because there is a second trap few people mention. The first trap belongs to the system. The second belongs to the reader. In July 2026, I was in Moscow for the France–Belgium semi-final. I logged how coach Didier Deschamps dropped the defensive block to an average of just 24.8 metres, while Blaise Matuidi tucked inside to cut the passing lane into Kevin De Bruyne's feet. I wrote in detail about space, about the defensive lines, about the distances between units. My piece sank. A colleague wrote only about Vincent Kompany's tears after the defeat, and his piece was shared six times as much. That was when I understood something numbers cannot tell: emotion is not data noise; it is data not yet decoded. Readers do not only want to know how many metres high the block sat. They want to know what pain, what loss, what regret that distance produced. Pure analysis, with no emotional touchpoint, is half the story left out. And this is where it connects to the Mexico story. When a system fabricates football analysis from a text with no football, it is not only wrong on the numbers. It also takes the place of real stories. It fills a gap that should have stayed empty, or should have been filled by something true — a person, a moment, a verifiable fact. I want to tell one more story, because it explains why I believe in entity discipline. When football stopped in 2026, I was 39 and fell into a long stretch of anxiety. For six months I wrote nothing. I only rewatched footage and drew diagrams. In June 2026, when the Euros kicked off and I turned 40, I suddenly noticed a pattern: Italy's central midfielders, Nicolò Barella and Marco Verratti, were generating 14.7 passes into dangerous areas per match through triangular movement. It was a pattern that had never appeared in my data store. 4,500 situations, and one detail changed how I read the whole match. What I took from it was not "I was right all along." What I took from it was: a new pattern is only credible when it comes from observing entities — from my seeing Barella and Verratti move, and only then naming what I saw. Not from having a ready-made category, "Euro 2026 central midfielders," and shoving data into it. At the same time, I drew a wave of criticism: the writing was too hard to read. I had to change: shorter sentences, spatial metaphors, and always a minimal data table at the end. Every piece after that began with an assumption, then used data to prove it step by step. I learned that data rigour and readability are not mutually exclusive; what excludes each other is fake rigour and fake readability. Back to "nationality." I want to pause here, because I think this is the most important part of the piece for people working in football. In football, the nationality question is not small. It is one of the questions with real consequences. Which national team a player may represent, when an association switch is permitted, how dual-nationality cases are handled — all of it is tightly governed by Articles 5 to 8 of the FIFA regulations. These are real rules, applied to real people, and they sometimes decide the entire career of a star. And here is what must be said plainly: the Mexican constitutional reform I am discussing does not change, restrict, or expand any player's national-team eligibility. The two rule systems are entirely separate. One concerns eligibility for public office. The other concerns eligibility to represent a national team. Merging them is a false analogy, and in football a false analogy spreads faster than correct data. I have seen enough times how a wrong headline spreads across forums within hours, and by the time it is corrected, nobody is reading anymore. The wrong has more pull than the right. That is why I wrote this piece instead of just a line in an internal note. There is one more angle I want to dig into, because it concerns the whole football-analytics industry. Our industry has an obsession: everything can be measured. Passes, sprints, kilometres, pressure indices, duel win rates. We believe that if we measure more, we understand more. That belief is mostly right. But it has a side effect: we begin to treat every number as signal, and forget that some numbers do not belong where they sit. A heat map shows position; an intent map shows thought. A heat map in the wrong zone makes us misread the entire intent behind it. Likewise, a number in the wrong place makes us misread the entire context in which it appears. When I built my system, I learned to check entities before analysis. Not to check the label — the label is what I assign. But to check the entity — what is actually in the text. Is there at least one club? At least one player? At least one competition, one match, one football governing body? If the answer to all of them is no, then there is no football analysis to speak of. That is a simple test. It is cheap. It is fast. And it blocks most errors of this kind. So why do many systems not have it? Because it demands architectural humility. It demands that you admit your system can return "nothing." And "nothing" is not an attractive result to report. No one gets promoted for discovering that an article does not belong in their field. But that is exactly the kind of work that keeps the whole system clean. I said at the start that I call this the first error in a chain. Here is that chain. Error one: a single keyword was used as the entire basis for labelling. Error two: the analysis layer had no mode for returning an empty result, so it fabricated. Error three: the sources were not fully verified. In the data I received, most information points listed their source as "none" — no outlet, no date, no byline. Error four: the reader can be infected. A headline that reads like football news is shared like football news, whatever the actual content is. This chain did not start in Mexico. It does not end in Mexico. It starts in a very human habit: we fear the empty space, and we fill it with the nearest available category. Now comes the counter-intuitive angle, and I must be blunt because I have asked myself this for a long time. Most people will read this and conclude the problem is technological: fix the classifier, add keyword rules, build an entity-check gate. All of that is right, and I proposed exactly those above. But that is the most comfortable reading, because it turns the problem into a flaw patchable at the technical layer. My counter-view is: the deeper problem lies in football itself, in how our industry treats the word "nationality" as a signal strong enough to decide everything on its own. Look at how football talks about naturalised players. We label them "naturalised" before we know where they play. We build an entire profile of a person around a line on a document. Football itself taught text-processing systems that "nationality" is a keyword strong enough to label by. So when a system does exactly what we taught it and mislabels, is the fault entirely the system's? I think not. Part of the fault belongs to our reading habits. We are used to a word carrying a whole category with it. We are used to a number telling a story on its own. And every time we do that, we are teaching every machine behind us a bad habit. That means fixing the classifier is not enough. We also have to fix how we label people in our own football writing. A player is not a nationality. A match is not a percentage. A defender is not a coordinate. If we clean that habit at the human layer, the machine layer will have less chance to reproduce it. This is why I do not believe in systems that look beautiful on paper. I believe in systems that dare to say "I do not know." A model that returns an empty result is not a failed model. An analysis that stops because entities are missing is not a weak analysis. Sometimes, stopping at the right moment is the most mature analytical act a data person can perform. And I wonder: if I had applied that discipline in 2026, when I labelled Gosens a "full-back" and believed it for three months, would I have lost three months to realise I had misread the position? Probably not. The answer would likely have come much sooner. But back then I was not humble enough to ask my system what it had hidden. So I return to the open question I often use to close. If a system can learn that "nationality" is enough to label football, then it can also learn that a player is enough to label a person. The difference between these two things is not in the algorithm. It is in whether we dare to leave empty the space where the truth has not yet arrived. Before the next match, I will check entities first. A club, a player, a competition. If there is nothing, I will not write. That is the lesson from a 334-109-1 data row I should never have seen in my football drawer.