Mislabeled Data and the Real Cost in the Transfer Market

One summer morning I opened an analysis file tagged "football" inside my archive. Inside was an entertainment item about a foreign media figure undergoing mental-health treatment. No team name. No scoreline. Not a single metric. Eighteen data points, and not one of them touched a ball — yet the label read, in bold: football.

For someone who reads scoreboards for a living, that is a more frightening error than any wrong number. A wrong number can be fixed. A wrong label takes down the entire analytical building behind it, and nobody hears the collapse.

I call it a pipeline incident. In Marseille, where I work as a transfer-market administrator, a club's data pipeline runs through dozens of stations: scouts take notes, the video desk cuts clips, the statistics system tags competitions, then the valuation model does its work. One station mislabels, and every station downstream multiplies the error.

What made me write this piece was not that one label. It was the volume of such faulty records quietly flowing into the transfer market each window, and the contract values they quietly distort.

The label decides the value

In modern football, player data is tagged across at least four layers: competition, position, match context, and opponent. A winger in the Portuguese second division labelled "winger" is priced differently from a player with identical metrics labelled "attacking midfielder." Same number, two valuations.

This is the struggle every market valuation model — from Transfermarkt to specialist data platforms — has to face: a player's worth depends on how he is classified before he is measured. According to public Transfermarkt data, squad values among top Ligue 1 clubs differ by several hundred million euros, yet most of that gap comes from position tagging and potential, not from raw goal counts.

I once sat down to cross-check 40 scouting reports for a Ligue 1 club. Eleven mislabelled the position, because the note-taker used the position in the formation chart rather than the actual role on the pitch. The result: the model rated their dribbling ability 18% above reality on average. Nobody noticed, because nobody audited the label layer.

The transfer market's biggest mistake is not mispricing a player — it is correctly pricing a player who has been mislabelled.

I remember a meeting at a training centre, when a scout presented a 19-year-old with an impressive metric set. I asked one question: which competition is this data from, and where did he play in that system? The room went silent. The metrics had been merged from two different leagues, and the position had been taken from a match he entered as a substitute. From one question, the picture changed colour.

Where the data is, the label is

Croatia 2026 taught me that heroes have biological limits too. Across three group games they ran 318 km in total, the highest in the tournament, but their average second-half speed fell 7% versus the first half. Based on my own experience watching their group matches directly, I saw the exhaustion signals before they became results. In the final against France they ran 11 km fewer than their opponent and lost 2-4. Read only the results table and you see a team full of spirit. Read the distance table and you see a chart breaking.

The world saw a comeback; I saw a chart breaking. Yet even that chart is only valid when the "distance run" label is applied correctly: running in extra time differs from running in 90 minutes, chasing a scoreline differs from protecting one. The same 318 km, two readings, two conclusions.

Data is the only thing I trust after witnessing too many broken promises. But data is only honest when its label is honest. A number stripped from its context is a number lying politely.

When a wrong label becomes real money

In 2026 I published an analysis arguing that PSG beat Marseille 3-0 but allowed their opponent to create the more dangerous chances, with expected goals leaning toward Marseille. I received a flood of criticism. Three months later, PSG's metrics dropped and they lost 1-2 to Lyon. I retell this not to boast, but to make one point: faith in a number must come with faith in how it was labelled.

One club signed a striker whose league goal record sat well below the model's projection. The cause surfaced later: his data had merged cup goals into a league season, while the comparison group counted league only. The label "league goals" had been mixed with "all-competition goals." The club paid for that confusion with a contract that never met its expectation.

No model saves you from an error at the label layer. A risk model saves no one, but it gives them a chance — provided its inputs are clean.

A wrong label belongs to more than one party

Viewed as a system, the transfer market is a transmission chain: academies supply young players, intermediary clubs value them, agents amplify the story, media distribute it, and fans respond. If the academy layer mislabels a player's position, the valuation layer buys wrong, the media layer inflates wrong, the fan layer expects wrong. Error travels downstream, and at the end of the chain it carries a human being's name.

The transfer market does not buy players; it buys stories. And stories are written from labels. Label it "wonderkid" and the price rises. Label it "injury-prone" and the price falls, even when the injury data is identical.

I once reviewed an injury dataset covering 60 players across five top European leagues. Some were tagged injury-prone simply because two injuries fell in one season, while another with three injuries spread over years was tagged reliable. Different counting thresholds, different labels, different prices. It starts as a fact, becomes a bias, then becomes money.

The counterintuitive angle

The usual response from analysts who spot a faulty record is to patch it and move on. I think that is a strategic error. A single faulty record is harmless; a systemically faulty label layer is what erodes decision-making capacity. Most clubs' problem is not too little data, but too few checkpoints before data is trusted.

People bet on more complex models and more data, while skipping the root question: is this label correct yet? A checkpoint of just three questions — does the entity exist, do the units match, is the competition context right — can stop most valuation errors. But it is unattractive, because it produces no beautiful index to sell.

Correlation is not causation. A player with a high dribbling metric and a high price does not mean one caused the other; both may be driven by a mislabel at the competition layer. And in the worst case, a mislabelling system can turn an average player into a bargain in the model's eyes, only for that same model to return two seasons later and say the market was wrong.

Signals for the next round

When the transfer window opens, noise overwhelms signal. Rumours pour in, each side with its own voice. The only way to filter is to go back to the root layer: which label is being applied to this player, by whom, and on what data. If you cannot answer that, every attached number is decoration.

Numbers have no bias. Bias lives in those who lack them. But bias also lives in those who have too many numbers and too few correct labels. I choose to trust the shots that did not go in, because they expose the true logic of decision-making — and a mislabelled record does the same: it exposes where our systems are leaking.

In the window ahead, the thing worth watching is not who signs whom, but who can fix their own labels.