Wrong Labels on the Data Map: Lessons from a Misclassified Scouting File

In August 2026, I sat in the seventh row of a fourth-tier German stadium, holding a notebook with 47 columns. A Chinese U-20 select side had just lost its third consecutive Oberliga match. On the pitch, almost nobody ran. In my notebook, everybody ran. I measured 20-metre sprint times, receptions between the lines, and the share of passes into the final third for 23 players. Six weeks later, I recorded that Yan Dinghao had improved his ball-processing time by 0.4 seconds. That was the only finding worth keeping from eight matches played by a team that won two.

There was one line in that dataset I had to delete. Not because the player was poor. Because that line did not belong to football at all.

Years later, working with scouting data systems across Asia, I ran into exactly the same type of error, only at a different scale. A record tagged "football" whose content was medical news. A match dated 11 September 2026 — a date that has not arrived. A young player valued on the basis of a relative's account rather than an official document. Each error sounds small on its own. Placed side by side, they form a problem far larger than any tactical mistake on the pitch.

Vietnamese football sits exactly at the intersection of this problem. V.League has 14 clubs, each playing 26 rounds a season. Academies such as HAGL, PVF, Viettel and Nutifood produce hundreds of youth players every year. International data platforms have been fully present here since around 2026. A Vietnamese club can now look up every goal in a national U-19 finals, every minute played by a 16-year-old in the third tier. Data volume is growing faster than verification capacity. That is when wrong labels breed.

I once received a scouting file containing more than 4,000 records of Southeast Asian players. Sixty-two records had been automatically sorted into the "central midfielder" group. Opening each record, I found that 11 of them were reserve goalkeepers in youth teams, four were futsal players, and two were handball athletes. The software was not wrong. The software simply did what it was taught: guess a label from the input text. A data-entry clerk wrote "good at catching the ball, fast reflexes" for a goalkeeper; the algorithm read that as a defensive midfielder because the phrase "catching the ball" appeared so often. Nobody read it back.

A mislabeled record does not stay where it is. It moves.

This is the propagation mechanism that very few people in scouting pay attention to. The mislabeled record enters the entity-extraction model and distorts the position-by-position player list. The distorted list enters the valuation model and pushes the average value of a position group up or down. The distorted value enters the report sent to the coaching staff, causing a club to look for the wrong kind of player in the transfer window. Eleven wrong records out of 4,000 is an error rate of 0.275 percent. That sounds harmless. But those 11 records happened to fall into the exact position group the club was shortest in, and so an entire winter recruitment plan was pulled off course.

I followed one specific case across two seasons. A V.League club needed a central midfielder who could turn the ball over quickly. Their system returned a list of five names, including a 19-year-old tagged "playmaking midfielder" with the highest final-third pass figure in the U-21 league. When I went back to the video, that player was operating as an advanced full-back. The passes counted as "final third" were in fact crosses from wide areas — a completely different action tactically, but swallowed by the same data label.

The club signed him. He played seven matches, scored no goals, provided no assists. In the eighth match he was moved to full-back and played well. Nobody went back to check the label in the system. The record still sits there, still marked "playmaking midfielder", and the following season it appeared again in another club's recommendation list.

This is where I recall the line I keep writing in my professional notebook: the value of a map lies in the lines left blank, not the lines that are drawn. When a data system leaves a cell blank, the reader knows something must be filled in. When the system fills that cell incorrectly, the reader knows nothing at all. A wrong label is more dangerous than a missing one, because it creates a feeling of completeness.

The second error is the time error. In the file I received, one medical event was dated 11 September 2026. I traced the entire data supply chain and found the cause: an optical character recognition step had read the digit 4 as 6. One character. One character sent an event two years into the future.

For a news article, this error is merely annoying. For a player injury-prediction model, it destroys the entire time axis. Imagine a 21-year-old defender with a history of anterior cruciate ligament rupture recorded in 2026. The model will treat that player as a "future injury" — that is, not yet occurred, that is, zero risk. A club could buy him at the price of a fully healthy player, and then lose him in round four.

In youth football data, three date fields decide everything: date of birth, date of first contract, date of injury. Get one field wrong and the development curve is redrawn. I checked 300 records of young Vietnamese players on public platforms and found nine cases where the date of birth was off by one day compared with the birth certificate. It sounds small. But in youth football, one day can push a player from the eligible age group to the ineligible age group in a given tournament, and from there change his entire two-year competitive pathway.

Modern football does not lack spectators; it lacks people who read footprints in melted snow.

The third error is the source error. This is the hardest to fix, because it does not live in the computer but in professional habit. In the file I analysed, most of the important facts came from second-hand accounts by relatives and from an unidentified source. Only a small portion came from official authorities. Once entered into the system, they all became lines of text that looked identical, the same font size, the same format, with no ranking.

In football scouting this happens daily. News that a 17-year-old at club A is about to move to club B spreads from a social media account. Three days later, a sports outlet repeats it. Five days later, an international data platform updates the player's "current club" field. By then, nobody remembers who the original source was. The record has become administrative fact. If the deal collapses, correcting it takes ten times as long as entering it did.

I rank sources in four tiers. Tier one is official competition-organiser records, medically signed health files, notarised contracts. Tier two is direct observation by a journalist present at the scene, with time-stamped notes. Tier three is second-hand interviews cross-checkable against at least two independent sources. Tier four is everything else. In my own work, only tiers one and two are allowed into the conclusion section. Tier three is explicitly marked as tier three. Tier four is excluded from every calculation.

This tiering makes me slow. I accept slow. Over the past ten years I have had to apologise to editors for late filing no fewer than twenty times. But I have never had to publish a correction for a wrong number.

What is worth noting is that all three errors share one root cause, and that root cause is not technology. Clubs, academies and data platforms in Vietnam are currently investing heavily in analysis software. They buy motion-tracking systems, hire event-data services, build visual dashboards. Very few hire a single person whose only job is to read the input data back and ask questions about it.

This is a familiar paradox. A club can spend billions of dong on a video analysis session, yet have nobody who spends two hours a week cross-checking the birth dates of 40 youth players against original documents. The prettier the software, the harder the input errors are to see, because a dashboard always presents everything neatly, in colour, with charts, with an upward trend.

The transfer market is the dust layer; the deep stratum is what determines the age of the talent.

But I do not want to stop at criticism. In my own work as a data archaeologist, I have learned that a sediment layer of errors is also a layer of information. When you find a mislabeled record, you do not merely find an error. You find the entire process that produced it — the data-entry clerk, the tool, the timing, the purpose. Fixing one record is a small job. Understanding why that record exists is a large one.

In Vietnam, I believe this is the biggest missed opportunity in the football data industry. International platforms have better technology but do not understand local context. They do not know that a youth tournament in central Vietnam may change its name three times in five years. They do not know that some player files are recreated when a player moves from one academy to another. It is the Vietnamese football people themselves who can build this verification layer, and they have not yet done so.

Yan Dinghao back then was a small example of the opposite. He was not the most outstanding player on that list of 23. He did not score in eight Oberliga matches. What he had was a measurable improvement curve — 0.4 seconds over six weeks — and that curve only appeared because I measured the same movement, at the same distance, with the same stopwatch, every week. Had I changed the measurement method midway, that finding would have vanished. Good data does not come from expensive tools. It comes from holding one ruler constant throughout the measurement process.

Mbappé taught scouting that the weapon lies beneath the ankle, not in the scoreline.

At the 2026 World Cup in Russia, I tracked 17 sprints by Kylian Mbappé in the France–Argentina match. The gap between two of his sprints was always under 22 seconds. I wrote a piece about that speed structure. On day one it had 30 reads. Three days later, when Mbappé scored his brace, it was shared more than 500 times. The value of the measurement did not change. Only attention changed.

I tell this story not to talk about my own patience. I tell it to make a different point: every measurement is worthless unless somebody reads it correctly, at the right moment, with the right question. A 47-column spreadsheet says nothing on its own. It speaks only when someone sits down, cross-checks, doubts, and records the date of each measurement.

For Vietnamese youth football, I believe there are three things to do immediately, and none of them requires new technology.

First, every academy should have one person responsible for checking input data weekly. Not tactical analysis. Just reading back the newest records and asking three questions: what field does this record belong to, is the date plausible, and what tier is the source.

Second, every date field in the system should have at least two independent cross-reference sources. For a youth player's birth date, the first source is documentation, the second is the competition organiser's registration record. If the two disagree, the record is locked until resolved.

Third, every report sent to coaching staff should state the source tier of each conclusion. A conclusion from tier four must not appear in the transfer recommendation section, however plausible it may sound.

These three things do not cost much money. They cost time, and in football time is the most expensive thing nobody wants to spend.

The Oberliga map is still lying there; it is just that few people have enough patience to dig.

In eight Oberliga matches in 2026, that U-20 select side won only two. Looking at the scorelines, there was nothing to write about. Looking at the 47 columns, there was a player who improved by 0.4 seconds in six weeks. Two ways of looking, two different conclusions, the same fact. The difference lies in whether anyone was willing to sit down and measure.

What worries me most about Vietnamese football is not a shortage of talent. Academies are producing young players at a rate never seen before. What worries me is that we are building a data layer that thickens every season while nobody sweeps the dust. Every mislabeled record left in place is a trap waiting for a wrong transfer decision in the future. That trap does not detonate immediately. It detonates three years later, when a player is signed because of a statistic that was never real.

Every generation of good players begins as a generation of archaeologists who know how to wait. But waiting does not mean sitting still. Waiting means holding the ruler constant, recording the dates, recording the sources, and checking one more time before believing.

The question I leave behind is not for the algorithms. It is for the people sitting in the data rooms of V.League clubs: in the youth player file you are using to plan next season, how many lines have you never opened again to read?