A Wrong Label, an Endangered Dataset: Football Analytics Without a Blockchain-Grade Provenance Layer

The label on the record said football. Inside, there was not a single syllable of football. The whole substance was an internal process of Mexico's governing party Morena, a political figure named Andrés Manuel “Andy” López Beltrán, Federal District 6 in Tabasco, and speculation around the 2027 federal election. There is no club, no player, no contract, no transfer, no match. Yet in the dataset the record sits under a football identity, like a spectator who walked into the stand without a ticket and was never questioned. I spent eleven years in marine insurance. I know a clause when it bites. This label is exactly that kind of bite—small, silent, and quite enough to destroy the credibility of an entire file.

In March 2026, at the IFAB Annual General Meeting, the handball clause in Law 12 was rewritten and VAR was formally embedded in the Laws of the Game. I was 51 then, one of two women in a 40-seat press box at Anfield. I wrote that the word “deliberate” had been quietly replaced by “unnatural silhouette”, and that it had happened eighteen months before anyone said it out loud. The piece was read by 900 people, 400 of them referees. I opened the rewrite at 2 a.m. The offside law had moved to a footnote. That number still matters to me, because it proves something: which rule exists, where it is placed, and who holds the authority to apply it, are questions not only for the people on the pitch but for those who read documents, footnotes and databases.

That is precisely where today's discussion begins. The new field of football analysis is no longer the pitch; it is a data pipeline. And at the door of that pipeline sits an automated classifier whose job is to read an article and declare: this is football, or politics, or business. On the record I am writing about, the classifier was wrong—it tagged a Spanish-language election story as football. The question is how that error happened, and why it is not merely a mistake but a jurisdiction problem.

The real cause of a wrong gatekeeper is a collision of words, not of meaning. The Mexican story keeps returning to a set of terms: registration, candidate, process, structure, organisation, coordination. In football, exactly these words recur: player registration, the mechanics of the transfer window, squad structure, territorial coordination. A classifier that does not separately verify entity type—club, player, competition versus party, person, electoral district—decides on word overlap alone. So the word “registration” can mean both a political candidate's enrolment and a footballer's registration, and the pipeline cannot tell the two apart. This is not a failure of technology; it is a failure to know the borders of a jurisdiction.

What is absent from the story is an even louder piece of evidence. Andrés Manuel “Andy” López Beltrán is the son of former Mexican president Andrés Manuel López Obrador—a familial and political fact, not a football one. He is considering leaving a party executive post for a local project—a political transfer of power, not the resignation of a coach or sporting director. The story names Centro, Jalapa, Tacotalpa and Teapa—municipalities, not clubs. And the source field on the record reads “not specified”. The clause was buried on page ninety-four. That is where the match was lost. A document that cannot even state its own birthplace, placed in a football dataset, is not information—it is an assumption.

This is where my experience applies. In Kazan, on 16 June 2026, referee Andrés Cunha first waved away a Griezmann tumble against Australia, then went to the monitor and reversed himself—the first VAR-awarded penalty in World Cup history. France won 2–1. I had forty minutes to file. I ignored the “was it a penalty” debate and wrote 1,100 words on who now carried the burden of proof—because the on-field decision had stopped being a decision and become a hypothesis. The same applies to this pipeline today. The question is not how accurate the classifier is; the question is who holds the authority to decide which record enters under which identity.

When the Bundesliga restarted behind closed doors in May 2026, I stopped watching football as sport and started watching it as a dataset. I logged all 81 matches of the nine-round restart myself: the home-win rate fell from 43 per cent to 33 per cent, and IFAB's temporary five-substitution amendment—introduced to protect players after a ten-week layoff—was being used tactically before half-time in 61 of them. I printed the table with one sentence: “This rule is not temporary.” IFAB made it permanent in 2026. For eighteen years I had been told I did not understand the game; numbers then became a language nobody could accuse me of faking. In the same way, a wrong label is not mere opinion—it can be counted and measured.

In 2026, at 36, I walked away from a stable salary to freelance on football governance. By 2026, fifteen years later, I was still explaining at every door why a rules columnist deserved a credential. That experience taught me something: a system that does not want proof does not merely distrust proof, it distrusts the people who demand it. In the data pipeline, that same attitude has now become institutional.

Now to the heart of the problem. When this error occurs, it does not stop in one place. If an election record enters under a football identity, two dangers emerge downstream. First, a false entity: a model may read “Morena” or “District 6” as a team or organisation, and it may land in a league table. Second, a false signal: the story mentions distributing a party newspaper, which could be read as regional scouting or commercial activation. Neither is true, but both can enter the dataset—and once in, they do not leave.

So what is the fix? The conventional answer is simple: train the model on more data, make the classifier more accurate. I do not agree. The problem is not accuracy; it is proof. Where there is no immutable account of a document's birth, modification and verification, even a flawless model can hide a wrong label. A blockchain-style provenance layer is relevant exactly here: if the answers to who created each record, who verified it, and when a label changed are inscribed in a chain, a wrong label can never pass itself off as true. In football administration, IFAB does precisely this: every amendment and every clarification is documented with a date, so that no one can later claim the rule was different at the time.

But here is my second, more uncomfortable observation. We usually assume the analyst's job is to fill every cell. When an analyst writes “insufficient information” and leaves a cell empty, we treat it as failure. Yet in this record, the analyst honestly wrote across all nine dimensions: “not applicable—insufficient information”. He did not invent football narratives to fill the boxes. To me that is not failure; it is the cleanest specimen of professionalism. An empty cell is evidence of honesty; a fabricated cell is the beginning of danger.

There is one more layer to this error that is easy to miss. The headline itself is a question—will he be a candidate? That interrogative style is click-bait in both politics and sport. When the classifier compares the tone of the headline with the language of the body, the resemblance can mislead it. The error, then, is not only of words but of tone. And tone cannot be verified—only facts, entities and sources can. So the gate must sit before the words, at the level of the entity.

For me, the real lesson is broader. When the transfer window opens, the flow of news multiplies several times over—rumours, speculation and incomplete claims all arrive together. At that very moment the pipeline's front door is at its weakest, because the pressure of speed pushes verification back. A record that is not football yet sits under a football identity is a product of that pressure. The fix is not to close the door but to place at it someone who owns an accountability ledger. A human reviewer, an entity-verification layer, and an immutable provenance chain—only with all three together will an election story never again write its name on Anfield's scoreboard.

So the question now stands before me like this: will we build football data an open ledger, in which every label carries its date of birth—or will we praise ourselves for a faster, larger, more opaque database inside which a wrong address lives forever? IFAB did not hide its rulebook's footnotes. Football's information industry must learn the same honesty—otherwise our analysis will remain noise outside the pitch and never become proof.