The Misrouted File: A Mexican Electoral Dossier in the Football Data Room

On 24 October 2026, a file arrived in my analysis room carrying exactly one label: football. I opened it after lunch, while my screen still held the PPDA table from the weekend's round. Inside there were no players. No coaches, no clubs, no leagues, not a single shot, not one line of xG, not one contract clause. There was a woman named Guadalupe Taddei, presiding councilor of Mexico's INE — the national electoral authority. There was a supreme audit institution called ASF that had just opened an investigation into contracting irregularities at the INE. There were legislators in the Chamber of Deputies weighing a vote to appoint a new presiding councilor. And there was one decisive sentence: as of Saturday, 24 October, Taddei had not resigned, and the INE had issued no formal announcement that she would leave her post.

Fourteen information points. Not one of them belonged to football.

I sat still for about three minutes. Not out of confusion, but because I recognised the exact disease I had spent a decade writing about — this time not on a pitch, but inside my own profession's data pipeline.

Context: a three-stage pipeline and the one stage nobody audits

The sports data analysis I practise in Lyon runs on three stages. Stage one collects and deconstructs: an article, a press release, a stretch of interview gets torn into discrete information points, each with a source. Stage two analyses: a nine-dimension framework is applied to those points — tactics, club finance, results cycles, league landscape, rules compliance, dressing room, risk profile, media narrative, industry transmission. Stage three is the product: a report, a model, a recruitment recommendation.

Between stage one and stage two there is a door. That door is called domain classification. And in almost every system I have ever worked with, that door is the least audited component in the entire chain. Nobody measures its error rate. Nobody scores it. It simply assigns a label and pushes the goods to the right.

I have spent most of my career doubting the numbers at stages two and three. I had never spent enough time doubting the label at stage one. That was my blind spot, and the file of 24 October exposed it.

The cost of this error is not small. A wrong scouting report can cost a club fifteen million euros. But a scouting report built on material from the wrong domain is worse: it is not wrong in its conclusions, it is wrong in its standpoint. And anything wrong in its standpoint always presents itself in the most credible tone, because nothing inside it contradicts itself.

Core: the anatomy of a file with no football in it

Fourteen information points, zero entities

When I went back through all fourteen points, the entity list came out very clean. Guadalupe Taddei, a person, presiding councilor of the INE, the subject of scrutiny and the one denying a conflict of interest. The INE, the electoral authority, stating it will cooperate with the ASF. The ASF, the supreme audit body, which opened an own-initiative investigation. Party representatives and parliamentary coordinators, who publicly demanded she step aside. The Chamber of Deputies, where a vote could be called. And supplier companies, unnamed, sitting inside allegations of contracting irregularities.

Six entity groups. None of them can score a goal.

A file labelled football that cannot contain a single club leaves exactly one thing that can be analysed honestly: the label itself. Every tactical conclusion drawn from here would be organised fabrication. And organised fabrication, in my trade, is a sellable product — which is precisely what makes it dangerous.

Nine framework dimensions and the ritual of returning null

I tried applying the nine-dimension framework to this file mechanically, as if I were a system with no right of refusal.

Tactics and technique: no system, no formation, no style, no match data. Sophistication, execution, personnel fit — all three cells empty. Club finance and transfer market: no broadcasting revenue, no commercial revenue, no wage bill, no net debt. The only financial content here is public procurement compliance, governed by Mexican administrative audit law, entirely unrelated to UEFA's financial fair play or the Premier League's profit and sustainability rules.

Results cycle and public pressure: no table, no form, no fixture list. But one point is worth recording — a genuine pressure cycle exists, it is simply political. Party representatives publicly demanding Taddei step aside. Media reports of a possible replacement. An active ASF audit. Three pressure sources pushing the same way. Yet the same information points also confirm there is no resignation letter and no formal announcement.

League landscape: no league, no tier, no position in the food chain. The power structure here comprises an audit body, an electoral authority, a legislature and parties — mapping it onto a football pyramid is an invalid transfer of framework, not a finding.

Rules compliance: the only dimension with a structural analogue, and therefore the one most easily force-fitted. An external audit of an institution's procurement functions like a financial compliance review. But the rulebooks, the sanction menus and the appeal routes are entirely different. And most importantly: nowhere in the text has any sanction been imposed or even proposed.

Management and dressing room: no coach, no squad, no playing contract, no age curve. Applying "contract-year effect" or "new-manager bounce" logic to an appointed official is a category error, not insight.

Risk profile: sporting risk zero, club financial risk zero, football personnel risk zero. The only real risk sits inside the pipeline itself.

Media narrative: the actual narrative is "institutional pressure on an electoral authority during an audit". No transfer narrative, no player rumour, no hype cycle to compare against.

Industry transmission: six segments — academy chain, agent ecosystem, broadcasting and commercial, capital networks, derivative markets, national-team ecosystem — none of them has a node connecting to the INE, the ASF, or Mexican legislative procedure. The transmission diagram has nothing to connect.

Nine dimensions. Nine nulls.

And this is where I want to slow down more than anywhere else in this piece.

Why returning null is the hardest skill in this trade

A system that is not designed to say "I don't know" will always find a way to say something. It fills the gap with inference, inference with assumption, assumption with tone. Three layers later, the output is a paragraph of perfect confidence about something that never existed.

I used to be that machine. In 2026 I predicted France would beat Croatia three-one, using a cumulative xG model. The final ended four-two, with two goals coming from individual errors my algorithm had no variable to describe. I was mocked live on French television. Three weeks later I rebuilt the model with ball-stoppage timing and referee error variables, and I added a mandatory section to every report: the limits of this metric.

That section has saved me more often than any metric I ever built. Because it forces me to name what I don't know before somebody else finds it.

Data does not lie; the person reading the data is the one who deceives. And the most sophisticated deceiver is not the one who invents a number — it is the one who invents a context for a number that is real.

Which keywords fooled the door

I have no access to the source code of the router that labelled this file. But I have fourteen information points, and I can read them the way I read match footage to find the decisive phase.

Three suspicious clusters appear in the text. First, "presiding" and "council" — two words that appear densely in every report about club governance structures. Second, "replacement" and "the presidency" — the language of the managerial market. Third, "coordinators" and "the league of coordinators" — in football, the word "league" almost always travels with a sports organisation.

Three clusters. None of them is evidence. But for a keyword-frequency classifier, three clusters is enough to cross the threshold.

This is what I want people who build systems to remember: a keyword classifier does not understand meaning, it understands probability. And lexical probability cannot distinguish "presiding councilor of an electoral authority" from "club president". It only sees the word "president" four times.

The router is not wrong. It did exactly what it was taught. What went wrong is that nobody stood behind the door to ask one question: does this file contain at least one club, one player, one coach, one league, or one competition?

The answer was no. And that question takes four seconds to ask.

An investigation is not a verdict — this file's biggest trap

If there is one place in this file that made me slow down and write carefully, it is this one.

The ASF has opened an investigation. The word "investigation" carries a semantic weight in Vietnamese, English, French and Spanish that the word itself does not deserve. A skimming reader files it away as: there is an investigation, therefore there is a problem, therefore someone is guilty.

Three jumps. Not one of them supported by the text.

Across all fourteen information points there is no ASF finding, no sanction, no ruling. There is an investigation opened on the auditor's own initiative — meaning the audit body raised it itself rather than responding to a complaint. Procedurally, own-initiative audits often indicate the auditor identified the issue independently and holds firmer procedural ground. But that is a statement about process, not about outcome.

The only two "compliance posture" signals in the file are Taddei's denial of a conflict of interest and the INE's statement of cooperation. Both come from self-interested parties. They must be recorded as claims, not as findings.

I learned this lesson a more expensive way. In 2026, when Lyon's stadiums stood empty because of the pandemic, I took a contract with a German technology firm and studied twenty-four Bundesliga matches played without crowds. The result: home teams lost 0.23 expected goals. I wrote a sharp piece concluding that home advantage was a psychological myth. A group of Lyon supporters boycotted me for two months.

They were not entirely wrong. I was right about the number and wrong about the scope. Twenty-four matches in one anomalous season are not enough to declare a law about human nature. I read a sample and announced a rule.

An empty stadium is not silence; it is a problem without an answer yet. And an open investigation is the same: a question without an answer, not a verdict waiting to be read out.

Source tiering: this file tells you where it is weak

In my trade, every piece of information has a source tier. Tier one is a primary document with a case number. Tier two is an on-the-record quote from a named person. Tier three is reporting from a newsroom of standing. Tier four is "journalistic versions" with no traceable origin.

Where does the file of 24 October sit? No ASF document number. No INE press release cited. No named parliamentary coordinator. The only figure quoted directly is Taddei herself. Everything else rests on "journalistic versions", one unnamed counselor calling the moment "delicate", and an unattributed judgement by the article's author.

That is tier three and a half. Not the worst tier, but nowhere near enough to support a top-level conclusion.

And this is where our transfer trade should look in the mirror. How many transfer rumours this window are built on exactly that structure: an unnamed source, a "according to journalistic versions" clause, an author carefully adding the word "could"? I have read hundreds of them this window. And I have seen dozens of player valuation models built on top of them.

Every player is a distinct data population, and a good analyst is one who can read their scripture. But nobody can read the scripture of a person they only know by name through an unnamed source.

Shelf life: a story that can expire in forty-eight hours

The last point about the material, and the most underrated.

The article's core conclusion is an absence. Taddei has not resigned. The INE has not announced. No decision has been taken as of 24 October.

An absence-shaped conclusion has the shortest shelf life of all. It holds until the moment someone speaks. An INE statement, an ASF announcement, a vote called in the Chamber of Deputies — any one of those three invalidates the entire article within a day.

In my trade we talk about the validity window of a report. An injury story has a window of hours. A team-sheet story has a window of ninety minutes. A completed transfer has a permanent window, because it is an event that has happened. A "no decision yet" story has the shortest window of all, because it is true only until the next press conference.

A system that does not record shelf life will reuse this file next week, next month, next year — and each time it reuses it, it will be a little more confident, because nothing inside the file says it has gone stale.

The contrarian angle: the wrong label is not the biggest problem

The obvious conclusion at this point is: the classifier is broken, fix the router, done.

I don't believe that conclusion.

Fixing the router is the easiest thing in this entire story. Add a gate before stage two — if the entity field contains no club, player, coach, league or competition, halt and reroute — that costs about a day of work. Blacklisting the combination of "presiding, replacement, league of coordinators" costs half a day more. Done. This error will not recur on this file.

But it will recur on another file, because the problem does not live in the router. The problem lives in the confidence layer generated behind it.

Imagine a system with no ritual of returning null. It receives the file of 24 October. It finds no players, but it finds "pressure", "replacement", "the presidency". It writes: a leader under pressure from stakeholders, likely to be replaced, while the institution projects stability. Then it translates into football language: the coach has lost the dressing room, the board is looking for a replacement, the club risks crisis.

That paragraph will read fluently. It will not contradict itself. It will use the correct terminology. And it will be entirely wrong, because it was translated from one language into another without a dictionary.

That is the most dangerous product this trade makes: text that is formally correct and ontologically false.

And this is where I have to put myself in the dock. I have often written about error as a virtue. I have said that error cultivated long enough becomes destiny. But there is a kind of error that never becomes destiny, because it does not live inside the model — it lives at the point where we forget to check what we are measuring.

Lyon 2026 taught me one thing: numbers know how to rebel, if you are willing to listen. But Lyon 2026 did not teach me this: sometimes numbers do not rebel, they simply belong to a different match.

There is one detail in the file I want to leave for people who do this work. The original author added a line of their own: treating the replacement as decided would be premature. That is an act of self-restraint. In a news market where every headline tries to push events one step beyond reality, an author pulling them back one step is a sign of discipline, not indecision.

And here is the paradox: that article is more careful than many football products I read every day this transfer window. It does not invent sources. It does not declare an investigation to be a verdict. It does not turn a possibility into an event. It simply belongs to another domain.

We can learn exactly one thing from it: restraint in language is a technical skill, not a moral quality. And it can be taught.

The second blind spot: the transfer window is where this error breeds

I have to be blunt about this window.

In a transfer window, noise exceeds signal by a ratio none of my models can measure. Thousands of information points pass through pipelines every day: a deleted tweet, a clipped interview, a "source close to" somewhere, an airport photograph. All of them have the form of information.

The labelling mechanism works exactly as it did with the file of 24 October. An article about a player negotiating a personal sponsorship gets pushed into the "transfer news" branch. An article about a club replacing its commercial director gets pushed into the "managerial change" branch. An article about a player opening an academy in his hometown gets pushed into the "negotiating an exit" branch.

Each time, a valuation model absorbs one more assumption that isn't true. Two months later, that model will be more confident about a deal nobody ever mentioned.

I do not believe in miracles on grass. I believe error cultivated long enough becomes destiny. But I also believe most of the "destiny" in a transfer window is not cultivated by error — it is cultivated by the wrong label.

And there is a professional consequence I want scouts to write down. When an internal report says "player X is a target for club Y", your first question should not be "does Y need that position". Your first question should be: what tier is the source point behind this sentence, and does it actually describe a transfer?

Two questions. Four seconds. And it filters more than any xG model I have ever built.

Operational lessons: four things to do before the next data cycle

I am not writing this section as a recommendation to others. I am writing it as a to-do list for myself, before the next data cycle lands in the room.

First, place a domain gate before the analysis stage. Minimum condition: the file must contain at least one recognised football entity — club, player, coach, league, or competition. If not, stop. Do not analyse, do not infer, do not translate.

Second, make returning null the default rather than the exception. A dimension with no data must be recorded as having no data, with the precise location of the gap. This is what I learned after the 2026 World Cup and have applied to every report since. But I had never applied it at the label layer.

Third, attach absolute timestamps to every absence-shaped conclusion. "No decision as of 24 October 2026" is a verifiable sentence. "No decision yet" is a sentence that will quietly become false without anyone noticing.

Fourth, and hardest: log this incident in the error register with a diagnosis. The diagnosis here is compact — stage one's deconstruction worked correctly, the core viewpoints are coherent and internally consistent for a political story, only the domain label is wrong. A clean, diagnosable failure, and therefore a cheap test case for the whole system.

Clean failures are gifts. They tell you exactly which door is broken, whereas dirty failures only tell you something is broken somewhere.

What I take away from this file

I have spent twenty years teaching people that data is a witness, not a judge. That a match is a trial, and the most suspect person in the room is the one reading the spreadsheet.

The file of 24 October forced me to extend that sentence beyond the pitch. Before asking what the witness said, ask whether the witness was at the scene. In this case the witness was in Mexico, and the scene was in Lyon.

Fourteen information points, nine analytical dimensions, not one player. A file perfect in form and empty in substance. It did not deceive me with a wrong number. It nearly deceived me with a label in exactly the wrong place.

I wrote somewhere that a win is only a coordinate in the ocean of data, but people mistake it for the whole ocean. Today I add another coordinate to my map: a file labelled football that contained nothing but elections. That coordinate does not tell me which team wins next round. It tells me where my map is still un-drawn.

And I will track four signals in the coming days, because they decide whether this file lives or dies: an official INE statement on Taddei's status; a case number or finding from the ASF; a vote published on the Chamber of Deputies' agenda; and the appearance of any named official in follow-up coverage. Four signals. Each one that appears lifts the file from tier three-and-a-half to tier two, or removes it from my desk.

As for the door at stage one, I will not let it swing on its own again.

Data can sit quietly for years and then overturn every theory. But before I let it overturn anything, I have to be sure I opened the right door.

The next data cycle arrives tomorrow morning. I will be standing at the door.