The 2 A.M. Offside: VAR's Real Error Begins at the Labelling Stage

Two in the morning. The analysis room had gone dark long ago, with only the blue glow of a monitor thrown across the wall. I rewound the clip for the eleventh time. Minute 78, Hai Phong FC against Ceres-Negros in the 2026 AFC Cup. Fidelis Ikiri ran onto a through ball and side-footed it into the far corner. The stands erupted; the scoreboard clicked to 2-2. In my frame, Ikiri's left foot was offside by roughly 0.3 metres. I sent the signal up to the officiating team. The goal was disallowed, the match ended 2-1 to Hai Phong. I found that offside at 2 a.m., after everyone else had gone home. The home coaching staff never knew what I did that night. Perhaps they never needed to. But years later, writing about VAR's problems, I realised one thing: the biggest mistake in an analysis room has never been a missed incident. It is a label applied wrongly from the very start. Modern football runs on data. A match in the V.League or the Premier League is no longer recorded by instinct; it is broken into thousands of discrete events. Every pass, every duel, every shot is tagged by a team sitting thousands of kilometres from the pitch. Big matches can generate more than three thousand data points. Opta, StatsBomb and Wyscout turn a football match into a database, and from that database people build xG, PPDA, progressive passes and defensive actions. All of it depends on something very small that no one sees on television: the label taxonomy. When VAR arrived, public debate centred on the whistle. Fans rewatch the slow motion, point at the screen and argue about a decision. That is the surface layer. The submerged layer sits elsewhere: before any event enters a model, it has to be classified. At the surface layer, referees have a standard to hold on to: clear and obvious error. At the submerged layer, the labeller has no equivalent standard. There is no VAR for data. Nobody calls them to a monitor to ask why they logged a touch as a tackle rather than an interception. They work alone, and their errors are never replayed in front of a crowd. Is a defender's touch a tackle or an interception? Is a ball into the box a cross or a key pass? Is a blocked long-range effort a blocked shot or a shot off target? Each labelling choice produces a different statistic, and sometimes an entirely opposite tactical conclusion. Two data providers can watch the same match and produce two sets of figures that do not match. Neither is lying. They simply label according to their own understanding. Inside a VAR room, we call this the classification problem. It is far more serious than a wrong decision on the pitch, because it is silent and systemic. A wrong whistle ruins one match. A wrong label, repeated often enough, ruins a whole season, a whole model, a whole way of seeing a player. I learned this from my own error. In 2026, at the World Cup round of 16 in Russia, I was one of three analysts assisting the referee for Spain against Russia. In the 42nd minute, the ball struck Gerard Piqué's hand inside the box. I did not catch it. After review, the referee awarded Russia a penalty. The match finished 1-1, and Russia won the shootout. I blamed myself for three weeks, quietly rewatching all 64 matches of the tournament and taking notes on every VAR incident. What I realised did not lie in that incident. My error did not lie in a single moment. It was a habit. I had labelled arm close to the body for hundreds of similar situations before, and that habit made my eye skip the rare variation. The old label pulled me toward the old decision. When you must judge within two seconds, what guides you is the label taxonomy you carry in your head. That is why I began writing the series on hidden corners. A clip can be slowed down; the label in an analyst's head cannot. If you follow Vietnamese football, you will see this problem surfacing in the league table without anyone naming it. Take PPDA. Over the last three matches, one team's figure has dropped sharply. The media say that team is pressing higher. But if midfield blocks are tagged as active defensive actions when they were really aimless clearances, the metric improves artificially. The team did not change its tactics. The labeller changed. A more visible example is xG. The model depends on shot location, angle, shot type and the situation leading to it. If a cross is labelled as a shot, a player's xG rises. If a long-range effort that should have been excluded from the sample is counted, the whole team is overrated. In many reports I have read, the gap between two data providers for the same match is enough to flip a conclusion about whether a striker is in good form or simply riding luck. The worst case is mislabelling at the macro level. In tactical reports, teams are usually assigned one of a few postures: low block, mid block, high press. That label is often applied before enough data is reviewed, based on reputation. A team famous for defending deep is automatically tagged as a low block, even when in that specific match they pressed in a mid block. When the macro label is wrong, every metric beneath it is misinterpreted. The analyst is no longer analysing the match; they are confirming their own bias. I once saw a news item that had been labelled completely wrongly. Its real content belonged to another field entirely, yet it was assigned to a sport. If that item enters a sports model, it poisons everything downstream. In the world of data, an item tagged with the wrong subject is more dangerous than a missing item. A missing item knows it is missing. A wrong item thinks it is right. I write for Vietnamese readers, so I will speak plainly about the V.League. There, event data is thinner and less systematic than in Europe. Many matches have a single data source and no cross-checking. That means if that source labels wrongly, there is nothing to compare against. In a system with only one gatekeeper, that gatekeeper's error becomes the truth. Back to that 2026 AFC Cup incident. What I want to tell is not that I caught a 0.3-metre offside. What I want to tell is the pressure behind it. I had only seconds to send the signal. In those seconds I had to choose a camera angle, place a line and decide the final point of contact. Each choice was a label. If I chose the wrong frame, I missed the offside. If I chose the wrong point of contact, I could disallow a legitimate goal. Both would be my error, though only one would enter the record. The crowd sees only the final outcome. They see a goal disallowed or allowed. They do not see the taxonomy behind it. That is why I tell younger colleagues: inside a VAR room, the laws are only part of it. The rest is how you name things. The transfer market is where a wrong label does the most damage. A player tagged as injury-prone will be valued below his true worth, and that label follows him across clubs, countries and years, even after fresh medical data refutes it. A contract is like an offside call: one beat out of step and everything collapses. A club buys the wrong man because of an old label, and the price is not paid in the transfer fee but in a lost season. In esports, the crowd sees the play; I see the mouse click a hundredth of a second before it. That hundredth of a second is usually named wrongly. People call it reflex, when it is really the output of a decision model trained over thousands of hours. When you mislabel the nature of a click, you mislabel an entire method of training. Data analysts are moving into the dressing room, and that is a good thing, on one condition. Their conclusions often sit apart from the real rhythm of a match. A model can say player X should shoot more from outside the box. But the model does not know that player X has gone three matches without rest, that his ankle still hurts, that the opposing defence changed its shape after half-time. Correct data can still produce wrong advice, if the person reading it forgets the human part. Fans, and most of the media, believe VAR's problem lies in technology or in refereeing competence. They demand more cameras, more angles, semi-automated offside. Those things help, but they do not touch the root. The root lies where no camera points: the event-classification stage, the labelling stage, the stage where a person decides which box an incident belongs to in the taxonomy. When everyone blames the 19-year-old, the person in the VAR room must stand up. But when everyone blames the referee, the data labeller stays seated, and no one says their name. That is the biggest blind spot of modern football. We have built a system in which the most visible decision-maker is accountable to the public, while the person producing the data is invisible. Emotion feeds this distortion too. Fans remember a controversial decision longer than they remember a definition. A whistle creates emotion; a line of data does not. So the debate always flows toward the whistle, while the structure behind it fades into silence. We fix the loud thing and abandon the quiet one. When I wrote about Nguyen Van Truong, Hai Phong's 17-year-old during the 2026 shutdown, I deliberately gave him no label at all. Not a young talent, not a nervous kid. I sent a separate report on his strengths to the technical director and asked that he train with the U19s. Six months later he made his debut and scored a goal that helped the club survive relegation. I tell this story not to boast. I tell it to prove that sometimes the correct act in analysis is to refuse to apply a label. The referee is the only person on the pitch who is not allowed to take sides — and I stand behind them. But the data labeller stands behind both the referee and the public, and no one has ever audited them. There are offsides no one sees, but the camera never blinks. The problem is that the camera does not name what it records. People do that, and people err. If football genuinely wants VAR to be trustworthy, the task is not to buy more cameras. It is to build a new role: a data-integrity auditor, independent of both coaching staff and statistics providers, empowered to challenge every label before it enters a model. And above all, it is to teach the next generation of analysts that labelling is an ethical decision, not merely a technical operation. A millimetre changes a club's fate; I learned to live with that. A wrong label is the same, except it leaves no mark on the scoreboard. It leaves marks on how we understand the game, and those marks often take years to surface. For me, the next thing worth doing is to open a notebook, record every label I have ever used, and start questioning them, one by one.