Football Data Never Lies by Itself, but Labels Do

In September 2026, a film review of the psychological thriller 'Verity' entered a football data analytics pipeline carrying a single label stuck on top: football. The entire text contained no club, no player, no ligament. Only Anne Hathaway, Dakota Johnson, director Michael Showalter, and a Tomatometer sliding from 54% to 52% to 50%.

I have spent 52 years reading players' medical files. Football analytics pipelines do not collapse from a shortage of data. They collapse because wrong data is believed to be right.

In football, labels are everything. A player labelled "recovered" takes the pitch. An injury labelled "mild sprain" gets ice and three days off. A medical check labelled "clean" gets signed, then becomes a contract, then becomes a season.

In July 2026, Incheon United signed Brazilian striker Lucas Oliveira from the Portuguese third tier. I was the club's doctor-liaison reporter at the time, and I was shown the medical check. The cartilage in his right knee had already been operated on. Not a single line in the file recorded it.

I warned the coaching staff. They signed him anyway. He played 9 matches, 676 minutes, scored 2 goals, then re-injured and retired early. I spent a full month re-watching 47 of his old matches to chart the correlation between running intensity and knee pain. Not to prove I was right. To understand the root mechanism.

That was the first time I understood that a labelling error is not a technical error. It is an ethical error, hidden behind a folder cover.

In 2026, when European leagues paused, I dug through injury data from the top five European leagues covering 2026 to 2026. I built by hand a model of 2,318 injury cases, then compared it with recurrence rates after the shutdown. In November 2026 I published the result: anterior cruciate ligament rupture rates rose 23.4% at clubs with more than 90 days of rest, concentrated most clearly among players over 28.

The first reaction was not to challenge the data. The first reaction was to challenge my credentials: I am not a doctor. Three months later, UEFA published an independent study arriving at 21.7%. The 1.7 percentage-point gap sits inside the error margin of two different sampling methods.

The notable part is not those two rates. It is that both studies rest on the same kind of input — club injury logs — and both must trust that the label on every injury case is correct.

If 5% of those 2,318 cases were mislabelled — a muscle injury logged as a joint injury, a recurrence logged as a new case — then that 23.4% rise could become 18% or 29% with nobody able to verify it.

That is the lesson from September 2026. A film review slipped into a football data pipeline. Nobody died. No player took the pitch with an unhealed ligament. But if that pipeline was computing load indices for a club, the bad label had quietly ruined the entire sample.

Data is not wrong because it is poor. Data is wrong because nobody verified its label.

In June 2026, at the Kazan training ground, I stood about twenty metres from the touchline and watched Son Heung-min limp after a challenge from a Swedish defender. The South Korea team doctor diagnosed a mild sprain. I rewound the footage and measured the ankle inversion angle: 38 degrees. The usual safety threshold sits below 30.

I wrote an internal analysis predicting Son would still start against Germany, because his calf muscle structure could compensate for the damaged ankle ligaments. He started. He scored the goal that sealed a 2-0 win, and Germany left the World Cup.

Son Heung-min's right ankle beat Germany before the ball was kicked.

I do not tell this story to praise myself. The label "mild sprain" and the label "38-degree inversion" describe the same ankle, and they lead to two entirely different decisions. Which label is correct matters less than who verified it, and by what method.

The most comfortable thing about working with data is getting to blame the algorithm. The model is wrong. The classifier is broken. The pipeline failed. It sounds technical, modern, and utterly irresponsible.

Every football data pipeline shares the same weak point: the input gate. Either it exists or it does not. A real gate only needs to read the first three lines and ask: is there a club in here, is there a player, is there a match date.

But nobody installs that gate, because gates generate no market value. A club will happily pay 5 million euros for a striker and refuse to pay 5,000 euros to verify the label on his medical check. I saw that at Incheon in 2026, and I saw it again in a data pipeline in 2026.

In November 2026, before the Uruguay match, midfielder Lee Kang-in was suffering lumbar periostitis. The team doctor proposed a cortisone injection. I objected, citing my own database: a 41% recurrence rate within six weeks of injection. I sent a memo to the federation. He was injected anyway, played three group matches, scored once. After the tournament he missed 14 matches for Mallorca through recurrence, and the following season lost 187 days in total.

Many in the industry called me mechanical. They went quiet when the next page of the file was turned.

A medical file is the only thing at the negotiating table that cannot be bargained down.

The problem with modern football is not a shortage of data. Clubs collect more than at any point in history: GPS, accelerometers, load logs, MRI scans. The problem is that nobody pays for the person who sits and checks the label before the data enters the model.

A film review slipping into a football pipeline is a small thing. An ACL rupture labelled "muscle strain" is not small at all. Both start in the same place: a label nobody reads twice.

Age 68 taught me this: every player is healthy until the team doctor turns the next page.

And when that page is turned, will anyone in the data room be patient enough to read the first line?