Mislabeled: A Mexican Card, One Classification Error, and How Football Fools Itself

On the morning of August 13, 2026, I opened an analysis file and read the first line: domain — football.

The file held fifteen information points. I read all of them. No clubs. No players. No coaches. No matches, no transfers, no financial fair play. The content was about the Instituto Nacional de las Personas Adultas Mayores, Mexico's national institute for older adults, and a discount credential issued to citizens over 60: valid through 2026, older cards not automatically expiring, the procedure issued free of charge, and three replacement cases covering loss, irreparable damage, or the need to correct personal data. The confirming sources were INAPAM itself and the Secretaría de Bienestar.

Nine deep-analysis dimensions ran through the engine. All nine returned the same line: insufficient information to assess.

The engine was not broken. It ran with merciless precision. What was broken was the label at the top of the file. A single wrong label nullifies the entire value of everything beneath it, and nobody in the operating chain notices until the results come back empty.

I have seen that exact mechanism in football. Not once. Every week.

Football databases run on labels. Wyscout, InStat, StatsBomb, Opta, Transfermarkt — they all begin with a tagging act: this player is a full-back, that one is a playmaker, this league is tier three, this event is a shot, this payment is a free transfer.

Labels turn raw data into comparable information. The same labels turn comparison into nonsense when they are wrong.

In Vietnam, I once sat with two scouting departments from two different V.League 1 clubs, watching the same match, using the same data source. They argued for two weeks about a 21-year-old. Their conclusions diverged not because their eyes differed, but because one filtered by the label defensive midfielder and the other by the label central midfielder. One player, two profiles, two valuations forty percent apart.

The crux is not who was right. The crux is that nobody re-checked the label.

A line I keep repeating to young people in this industry: Numbers do not lie, but those who can read numbers always know how to make others believe the opposite.

Four label layers do the heaviest damage in football.

The first is the position label. A player tagged as a right-back but actually playing as a right winger in a back-three system. His defensive numbers sit below full-back benchmarks, so the file reads weak defensively. His attacking numbers run high, so the file reads unusually strong going forward. Both lines are arithmetically correct and professionally wrong. Tag him as a winger and he becomes an average player. Tag him as a full-back and he becomes a phenomenon. One human being, one season, two careers.

The second is the league label. A goal in V.League 1 and a goal in a European second division enter the same expected-goals model when the league coefficient is missing. I once read an internal report comparing the chance-creation index of a Vietnamese central midfielder — the kind of player Nguyễn Hoàng Đức is — directly against a European midfielder, with no adjustment for opponent quality, fixture density, or pitch condition. The report concluded the Vietnamese player was equal. That report was then used in a wage negotiation. It was wrong at the label layer, and entirely correct at the calculation layer.

The third is the event label, the least visible of all. A cross tagged as a pass instead of a shot skews both expected goals and the expected-goals chain of a single move. PPDA is computed on a set of defensive actions that carry labels; mix up the tackle tag and the interception tag and the pressing index of an entire system shifts, and a coach reading the report adjusts his block based on a number that never existed. Based on my experience watching matches in V.League and across the region, this kind of error clusters most heavily in games played in heavy rain or on poor pitches, when the tagger has to judge fast inside a noisy frame.

The fourth is the financial label, and this is the layer I care about most. A deal recorded as a free transfer means the transfer fee is zero. That label says nothing about signing-on fees, agent commissions, image rights, or performance-linked instalments. Real money still moves; it simply moves inside a box that financial fair play never inspects. Mexico's INAPAM card procedure is recorded as free — true on the fee line, while the real cost sits in documents, travel and waiting time. The same kind of label: right in wording, wrong in substance.

One more layer sits deeper, in youth academies. A 16-year-old is tagged as a striker for three straight years because he scores heavily at that age group. At 19, pace is no longer an absolute weapon, and he is released for failing to develop. The file never records that he was never coached in a suitable position. The label outlives the player.

From VCS to the World Cup, I learned one truth only: whoever holds the data holds the whole game. In 2026, when I wrote an analysis of GAM Esports and Đỗ Duy Khánh (Levi) at VCS Summer using European football's gegenpressing model, the esports community pushed back hard. They said I was applying one sport's label to another sport. They were half right. I applied the wrong label to the right data. What was more telling is that none of them re-checked whether the underlying data carried the correct label. They only argued about the conclusion.

If the misclassification rate for playing positions in Europe's top leagues sits below five percent after cross-validation — and the major data providers genuinely run periodic audits — then most of the argument above is surplus worry. I am putting that threshold here to tie my own hands. Conversely, if that rate exceeds fifteen percent, then every player-metric ranking we currently read deserves a warning line attached to it.

My second possible error: label noise is sometimes useful. Some valuation models have performed better when input data is lightly blurred, because it forces the algorithm to lean less on any single variable. Small-scale mislabeling can act as a natural regulariser. At large scale, it stops regularising and starts polluting.

Where I was certainly wrong in the past: in March 2026, when global leagues stopped, I publicly proposed using a weighted-average model to allocate European cup places in La Liga, and I had to take the piece down a week later over image-rights issues. My error then belongs to exactly the category I am dissecting here: I placed my-data as a label on someone else's data. That wrong label was not in the conclusion. It was in the source.

During the winter transfer window of 2027, I will randomly sample two hundred player profiles across major data platforms and cross-check position labels against actual match positions in the 2026-2027 season. If the mismatch rate comes in under five percent, I will publicly retract this entire piece.

When everyone looks at the giants, I saw the Viking quietly smiling — because the small club with no budget for expensive data is often the only one still checking labels by hand. The crowd is data, and I always read it backwards.