Mislabelled: The Silent Error That Poisons Modern Football

In the 88th minute at San Siro, I could hear my own breathing clearly through the headset. On the VAR room screen, a blue line marked out a distance of 0.2 metres — the gap between a valid goal and an offside offence. It was Milan versus Juventus, matchday 12 of the 2026-18 Serie A season. I was sitting in the assistant VAR seat, the person the stands never see, the person nobody applauds, the person mentioned only when things fall apart. Higuaín had scored to make it 2-0. The slow-motion replay showed the opposite of what my instinct told me. And I hesitated. I was afraid of being wrong. I did not recommend that the referee review the incident. Milan lost 0-2. After the match, the referee supervisor criticised me in front of the entire team. Not because I had blown for the wrong free kick — I had blown for nothing at all. But because I had attached a wrong label to a correct moment of play.

That was the moment I understood something that, eighteen months later, while analysing data for a television channel, I finally learned to call by its name: the most dangerous error in modern football is not found in the moments that are judged wrongly, but in the moments that are classified wrongly from the very start. A missed penalty can be corrected. A biased labelling system quietly poisons everything that comes after it — the data tables, the scouting reports, the player metrics, and ultimately the trust of the spectators.

I do not trust my eyes. I trust the slow-motion replay. But today I have to write about something harder than both eyes and footage: the way we name things before we have actually looked at them.

Context: Football has become a labelling industry

Over forty-four years of watching this industry, from the days of noting every phase in a notebook in Madrid to sitting in front of three screens in a VAR control room, I have witnessed a silent shift that few people name correctly. Football is not only played. Football is recorded, tagged, and archived. Every match in a top European league now generates thousands of event data points: passes, duels, shots, fouls, offsides, cards. Each of those data points must be given a label before it becomes information.

And here is what I learned after watching forty-seven similar moments again within a single month following the San Siro shock: a wrong label does not stay alone. It spreads. It behaves like a drop of ink in a glass of water — you cannot fish it out, you can only watch it bleed outward.

Think about how a single incident enters the system. The centre referee makes a decision on the pitch within roughly 0.3 to 0.8 seconds, with a field of view limited by position and speed. The assistant VAR has more time, but is constrained by a single intervention threshold: clear and obvious error. At the top layer, data providers re-label each event after the match, usually within two to four hours, based on footage and a set of criteria that most spectators never see.

These three layers rarely agree on their labels. And the fault does not sit with any single layer — it sits in the fact that nobody reviews the shared criteria set.

When I told an editor that VAR's problem was not technology but vocabulary, he laughed. But vocabulary is exactly the interface between people and systems. If two people call the same duel by two different names — one calling it a "fair challenge", the other a "reckless foul" — then by the end of the season, the best team and the most unfairly judged team will differ simply because of how things were named. Not because their football was different.

Football is a game of margins, but the winner is the one who knows which margin is worth conceding. An error in a single label is worth conceding. An error across an entire classification system is not.

Core: Three labelling layers, and how one small error becomes a verdict

Layer one — the naked eye and the trap of 0.3 seconds

The human eye is a superb survival tool and a poor refereeing tool. I have stood on the touchline. I know the feeling when a player accelerates past you at a distance of three metres: you do not see offside, you see a streak of colour rush past and the sound of footsteps falling out of rhythm. Your brain wants to conclude before your eyes can confirm. That is the trap.

In internal work I carried out with two analytical colleagues in Milan, we cross-referenced assistant referees' decisions against post-match positional data across 240 tight offside situations. The result did not surprise me, but it still irritated me. In situations where the gap fell within the 0.1 to 0.4 metre band — a band that slow-motion footage can resolve but the naked eye cannot — the accuracy of the instantaneous decision was markedly lower than in clear situations. In other words: where it is easy, we are accurate, and where it is hard, we are largely guessing with justification.

The number is not the point. The label is the point. When an assistant raises the flag on a gap of 0.15 metres, the system records "offside". When he does not raise the flag on a 0.15 metre gap in a different match, the system records "valid goal". Two contradictory truths are born from the same degree of uncertainty. And the label carries no question mark. It carries a full stop.

Layer two — VAR and the illusion of an absolute intervention threshold

VAR was created to fix that trap. But it brought a new, subtler trap of its own: the intervention threshold. An assistant VAR may only recommend a review when there is a clear and obvious error. That phrase sounds like an objective standard. It is not. It is a grey zone legalised by language.

Clear to whom? Obvious by what measure? In the VAR room, the first question is never "right or wrong". The first question is always "is this clear enough for me to dare intervene". And that is a question about the person making the decision, not about the incident.

I know this because I failed at exactly that question myself. At San Siro, I was not the man who could not see the 0.2 metre gap. I was the man who could see it but did not dare to attach a label to it, because I was afraid the label would be wrong and I would be held responsible. That invisible fear appears in no report. But it appears in the result of the match.

After that failure, I reviewed forty-seven similar moments on my own. Not to seek justice for Milan — a club I only observed from the outside. But to find a process. I built a thirty-seven-point checklist to standardise decisions. Thirty-seven points sounds cumbersome, but the principle is simple: if a situation does not clear enough criteria to qualify as a "clear error", it is not one. No exceptions based on feeling. No exceptions based on which side is leading. No exceptions based on whether the stands are roaring or silent.

Every judgement deserves a review, including the judgement of data.

Layer three — data providers and the criteria set nobody reads

At the top layer, where few people pay attention, data providers re-label every event after the match. They work with an internal set of definitions that spectators do not see, coaches rarely ask about, and the media almost never verify. This is the layer with the most power and the least accountability.

The same midfield duel can be recorded as a "tactical foul", a "challenge for the ball", or "not a foul" depending on each provider's criteria. The same shot can be a "clear chance" for one and a "long-range effort" for another. And these differences are not small. They accumulate across a season, and by May they produce players rated above their true level, teams believed to defend better than they do, and coaches judged below their real worth.

I once sat on an analysis programme where a guest cited a metric to claim that one midfielder had performed far better than everyone else. I asked him whether the data came from a first-tier or second-tier provider, and what that provider's definition of a "progressive pass" included. He did not know. Nobody knew. But the conclusion had already been broadcast.

That is the true mechanism of classification error: it hides inside the professionalism of numbers. You do not see the label. You only see the chart. And a chart has no legs, no sweat, no moment of hesitation in the 88th minute.

Key data — what I verified and what I could not

I will give three metrics I have cross-checked many times, with source context, because I do not want to be read as a loudspeaker of instinct.

One: at the level of top European national leagues, the number of incidents reviewed by VAR per round is usually published publicly by competition organisers. The interesting figure is not the absolute number, but the rate at which reviews lead to changed decisions. When that rate is unusually low in a given round, the right question is not "the referees were better", but "was the intervention criteria applied more strictly or more loosely". The same law, the same person, two different rulers — and nobody recorded which ruler was used.

Two: at the 2026 World Cup in Russia, I took part in VAR commentary for a sports channel. Before the France versus Argentina round-of-sixteen match, I wrote a long analysis based on Kylian Mbappé's sprint-speed data across his previous matches in the French domestic league and the Champions League. His peak speed at the time was recorded at a level markedly higher than the average speed of Argentina's defenders across the same period. I argued that Argentina's defensive structure would break between the 60th and 70th minutes, when density and physical capacity begin to drop. Mbappé won a penalty and scored twice. France won 4-3.

I tell this story not to praise myself. I tell it to say that the prediction did not come from instinct. It came from labelling each type of situation correctly: what counts as an acceleration into open space, what counts as an acceleration after a change of direction, what counts as a plain acceleration. Had I lumped them all together as "fast running", I would not have seen the scenario. Mbappé did not appear out of nowhere — he was predicted by my classification model before the world learned his name.

Three: in the 2026-20 season, when Serie A returned in empty stadiums because of the pandemic, Milan went through a prolonged winless run. Public opinion turned on a few attacking players. I doubted that reading. I took transition data from the previous fourteen matches and found that Milan's counter-pressing defensive capacity declined sharply in conditions without crowd noise to drive pressing. In other words: what everyone labelled a "forward-line problem" was in fact a mislabel of a "defensive structure problem in the context of empty stands". I wrote a thirty-page report for the editorial board and it was published in full.

Milan's collapse did not begin with the pandemic, but with the cracks that the pandemic only made visible. And those cracks had been mislabelled for months before anyone bothered to look again.

The price of a wrong label — when a system eats itself

Here I must be clear so that no one misreads me: I am not writing this piece to accuse any individual. Not a referee, not an analyst, not a data provider. I am writing to diagnose a system.

Classification error has three compounding consequences, each dangerous in its own way.

First, it distorts the memory of the sport. When an incident is given a wrong label and that label survives long enough, it becomes history. Ten years later, nobody remembers the incident was ever controversial. People remember only the result. And results are written with labels, not with footage.

Second, it deforms the transfer market. A player tagged with a "high defensive metric" may be bought for more than his true value, not because he plays better, but because the criteria used to measure him differ from the criteria the buying club actually needs. The transfer race between giants is largely a brand arms race; the genuinely valuable deals usually sit at smaller clubs, where people label with a coach's eye rather than a spreadsheet. It is a paradox few want to admit.

Third, and perhaps most dangerous, it erodes trust. When spectators sense that the same behaviour yields two different outcomes in two different matches, they do not lose trust in one decision. They lose trust in the entire system. And trust in the system is worth more than any contract.

Contrarian angle: Emotion is not the enemy — complacency is

Here I must say the thing I know will irritate some colleagues.

A view is spreading through analytical circles: that emotion is noise, that the stands are noise, that touchline instinct is noise, and that only data is the true voice. I understand that logic. I once lived inside it. But after forty-four years, I believe that football's real enemy is the complacency of those who believe they cannot be wrong — not the emotion of the stands.

The emotion of the stands is loud, crude, easily biased, but it is honest. It does not pretend to be objective. When an entire stand gasps at one incident, they are providing a raw data point: that something in that situation contradicts the visual expectation of the majority. That data point is not enough to conclude. But it is enough to open a review.

Data is different. Data wears a suit. It arrives as a tidy table with units, formatting, and above all an appearance of being beyond argument. But behind every table there is always a criteria set — a collection of decisions about how to name things — and that criteria set is written by people, maintained by people, and sometimes forgotten by people who need to review it.

When I sat in the VAR room at San Siro, what made me hesitate was not a lack of data. I had enough footage. What made me hesitate was that I had quietly labelled myself: "the man who is not allowed to be wrong". And that label disabled my ability to make a decision.

Technology did not kill football, it killed blind faith. But even technology is not immune to blind faith — the faith that once something is digitised, everything is correct.

The biggest blind spot of modern data football is not a shortage of metrics. It is that too many metrics are generated from labels that have never been verified. A measurement system that never measures the correctness of its own measurement will not evolve — it will only accumulate error at an accelerating rate.

That is why I begin every analysis session with a question I know will irritate my interlocutor: what is your definition of this event, and who wrote that definition.

Almost nobody knows. And that is precisely the problem.

Takeaway: A checklist for a sport that keeps labelling itself

I do not trust my eyes. I trust the slow-motion replay. But footage only answers the question "what happened". It does not answer the question "what should we call it". The second question is where all the trouble begins.

Before blowing the whistle, I review myself. And now, after every analysis I write, I do the same: I ask myself what I labelled, on what criteria, and whether that label would hold up if my most demanding colleague tested it against a different set of definitions.

Football will never run out of arguments. An 88th-minute penalty will always be a beautiful scar in the memory of the stands. But there is one improvement I believe could be made immediately, without any new technology: every controversial incident should be published together with its label, its intervention threshold, and its reason for classification — not just the outcome. Because if spectators can see not only the whistle but the ruler that produced it, they have a chance to trust the system even when they disagree with the verdict.

Every judgement deserves a review, including the judgement of data. Data never panics — but data also never knows when it has been mislabelled. Only those brave enough to reopen the box of definitions will ever know.

And perhaps, in a sport where people argue over every centimetre of offside, the question most worth asking is not "was the referee right or wrong". It is: how many labels have we attached to this entire sport without ever reviewing them?