UNAM, Pumas UNAM and a Labelling Error: Excavating a Data Fault Line Inside Football

My spreadsheet in Berlin has a column labelled "provenance". That column exists because I once paid the price for not having it. Every record entering the system must declare where it came from, who wrote it, when it was written, and what share of its content has been independently verified.

On the last Friday of September, a record arrived tagged "Football".

I opened it.

Twenty-four information points. Not one club. Not one match. Not one player, coach, formation, contract or wage bill. There was a march. There were families of 43 missing students. There were meetings with human rights bodies and judicial institutions. There was a government report scheduled for Monday, 28 September 2026.

The tag still read: football.

I did not delete the record. I moved it to a quarantine folder and opened a new file, naming it err-09-26. A data discrepancy, like a disturbed soil layer, is not waste. It is evidence. You only need to know which direction to read it.

People see talent. I see sediment.

Context: the machine that reads names

In Berlin I work as a youth-talent scout. The job is not watching three-minute highlight reels and nodding. The job is reconstructing a player's data biography before the media paints over it. To do that I run a pipeline: collect reports, extract points, apply tags, cross-check, and only then analyse the human being.

That pipeline has a fatal weakness nobody wants to mention at sports-analytics conferences: entity tagging is the weakest link and the least audited one.

A machine reads text, finds capitalised strings, matches them against a proper-noun dictionary, and assigns a topic label. See "Real Madrid", tag sport. See "Barcelona", tag sport. See "UNAM", tag sport.

UNAM stands for Universidad Nacional Autónoma de México, the country's largest public university, founded in 2026. But UNAM is also the short name of Club Universidad Nacional, the professional club known as Pumas UNAM, founded in 2026 and playing in Liga MX.

Two different entities. One university and one football club. They share the same four letters.

Crucially, this was not a machine confusing a university with an unrelated club. Pumas UNAM is legally the university's team. Its players have been students. Its home ground, the Estadio Olímpico Universitario, opened in 2026 and hosted athletics at the 2026 Mexico City Olympics. Its cantera is one of Latin America's most respected academies.

The error was not mistaking a university for a football club. It was mistaking a university for that same university in its football-club edition. The boundary is thin enough that a shallow tagging rule cannot see it.

Core: anatomy of a mislabel

I spent two days taking the record apart. Not out of any claim to authority on the underlying human story — I have none, and I will not offer any. I took it apart for a purely professional reason: this record is a perfect biopsy of a disease the football data industry carries and conceals.

First, source density. Of twenty-four information points, exactly three carried any attribution. Three out of twenty-four. Twelve point five per cent. The remaining twenty-one were bare assertions with no responsible party, no named body, no confirmed date.

In Berlin, any scouting report with attribution density under forty per cent is flagged red and cannot be used for a recommendation. A report claiming a seventeen-year-old has exceptional acceleration, with no confirmation of distance covered, competition, or date, does not technically exist. It exists emotionally.

Second, topic structure. Twelve years. One question. No partial wins to de-escalate, no interim targets, no recorded compromise. That structure resembles something I know well from my own trade: transfer sagas that hang for years. A player pursued from seventeen to twenty-three, across four windows, two contract extensions, three injuries. Every summer the story is republished with a new headline. Every summer nothing resolves.

A rejection is a footnote. The contract behind it has not been written yet.

Third, the gap between announcement and delivery. A government announced a report on investigation lines. Simultaneously, families said they lacked access to information they considered relevant. Announcement activity and disclosure activity are diverging. In football I have seen this hundreds of times: a club announces a comprehensive review after a bad season. Four months later the review is unpublished, but three executives have been replaced.

Fourth, calendar anchoring. This record is pegged to a fixed anniversary, repeated annually. Calendar anchoring is powerful: organisers control the calendar, so the story cannot be forgotten. It also produces a predictable editorial cycle — a peak on the day, and silence around it.

The same disease in scouting

Every football data system has its own UNAM: a pair of identically named entities nobody bothers to separate.

Search any player in public databases and you will find at least four versions of the same person: the parent club, the loan club, the national team, the youth team. Four records. Four fitness curves. Four injury datasets. They do not match. They never have.

One footballing family illustrates both the succession question and the identity problem: Marcos Alonso Imaz, known as Marquitos, a Real Madrid defender of the 1950s; his son Marcos Alonso Peña, who played for Real Madrid and Barcelona in the 1980s; and his grandson Marcos Alonso Mendoza, born in 2026, who also wore Real Madrid's shirt late in his career. Three people, three eras, three entirely different career curves — routinely merged at some layer of the data.

Number 17 never disappears. He is only deleted from the standings.

Concatenate three generations of injury data and you will find a pattern that does not exist. You will conclude a family has a knee-injury predisposition when in fact you have summed the injuries of three men in three positions across three decades. A mis-assigned entity produces a wrong analysis, which produces a wrong transfer decision, which costs a club several million euros and a person several years of a career.

Numbers that look like effort

The modern analytics industry suffers a subtler disease than tagging error: indicators that look like effort.

Distance covered and sprint counts are packaged and sold as proof of commitment. But ineffective running also produces beautiful numbers. A midfielder repeatedly dragged out of position, chasing the ball in zigzags across four different zones, will finish with one of the highest distance totals in the league. The number says he ran a lot. It does not say he ran in the right places. It may say the opposite.

Old footage does not lie. Only hurried viewers mishear it.

A mislabelled record is exactly like a mislabelled effort metric: both hand you a number that appears correct, tidy, quotable, ready for a slide deck and a decision. Both fail at the same point: they refuse to describe what is actually happening.

I once wrote a workload warning about a young Spanish midfielder in the appendix of an analysis, because I was too focused on proving my own framework right. The warning was about match volume. It sat at the end. Few read that far. The lesson was not whether I was right. It was that I placed correct data in the wrong position. Position is part of the data. Structure is part of the truth.

Contrarian: the problem is not dirty data

The reflex when a labelling error surfaces is to blame the algorithm. The second reflex is to blame dirty data. The third is to add another automated check.

All three miss the root.

The algorithm followed the rule a human wrote: the string U-N-A-M belongs to football. The rule is not absurd — UNAM genuinely is part of Mexican football, just not in this context. The data is not dirty in the ordinary sense either: it describes a real event with internally consistent points.

The error lives one layer deeper: the belief that a label is a fact.

A label is a classification decision made by a person or a machine at a moment in time, under a specific ruleset, for a specific purpose. It carries administrative authority, not epistemic authority.

Football forgot this long ago. We read the label "attacking midfielder" and believe it describes a person. We read "winger" and believe the player plays on the wing. We read "promising talent" and believe we are discussing the future. In reality we are discussing a cell in a spreadsheet.

When that cell was generated by a tagging process whose error rate has never been published, what we trust is not data. It is a chain of unexamined assumptions.

Three out of twenty-four

Imagine a report on a seventeen-year-old: twelve claims. Three specify minute, match and date. Nine say things like "can play under pressure", "reads the game well", "physical base beyond his age". Those nine are not wrong. They are unverifiable. And an unverifiable claim is not a claim; it is an impression formatted as one.

When a club decides on twelve such points, it decides on three data points and nine points of belief. If the player succeeds, the scouting department is praised. If he fails, the player is blamed. Nobody returns to audit the three and the nine.

So I keep one hard rule: every claim about a player must carry a video timestamp. No timestamp, no claim. The rule makes me slower and less prolific than most. It also makes what I write falsifiable — and a falsifiable claim is a claim with value.

Cycles and the trap of recurrence

Football has the same anniversary cycles under different names. Transfer deadline day: three weeks of unsourced headlines and unnamed accounts inventing fee figures, then two months of silence. The final matchday: a pre-written story for every position in the table — the champion's character, the survivor's will, the relegated side's collapse — all drafted before kick-off.

The shared trait of these cycles: intensity is guaranteed, depth is not.

A calendar-anchored cycle creates production pressure. Production pressure creates demand to fill space with whatever is available. The cheapest thing available is a pre-labelled record, unchecked, retitled and shipped.

That is how a wrong-topic record survives inside a football data system. Nobody deliberately inserts it. It enters because it has a label. It keeps the label because nobody has an incentive to re-check a label that already exists.

I audited my own archive. Seventeen other records carried a football tag while containing no football entity against my dictionary — seventeen out of nearly three thousand, under one per cent. Without the provenance column I would never have found them.

The takeaway

Everything that becomes a superstar was once an unread question mark in an archive. But a mislabelled question mark is never read again. It stays in the system forever, under a name that is not its own, in a category that is not its own.

The question I kept for that afternoon was not how to fix the tagging rule. It was this: if your system cannot tell a university from a football club, how far do you trust it when it tells you a seventeen-year-old will become a star?

I know my answer. I am waiting to see whether the industry finds its own.