An Empty Dataset Is Also a Decision: Lessons from a Deleted VAR Model

Minute 67, Seoul World Cup Stadium, Round 29 of the 2026 K League Classic. Lee Dong-gook receives a lofted ball from the left, controls it with his chest, turns inside the FC Seoul penalty area and strikes with the outside of his boot. The net ripples at the far post. The eastern stand erupts into a single block of sound, and inside my headset all that remains is the commentator's voice breaking into pieces. In the VAR room a few kilometres from the ground, I was rewinding the camera angle behind the goal — the angle that showed the Jeonbuk Hyundai Motors striker's run most clearly. I paused on frame 41 and measured the gap between Lee Dong-gook's shoulder and the shoulder of FC Seoul's last defender: 0.3 metres. I sent the alert. The secondary monitor clock read 14.2 seconds. The FIFA benchmark at the time was seven. The referee could not intervene. The goal stood. The executive director reprimanded me in front of the entire editorial room, and I did not resent him — in a system that runs on a timeline, a signal arriving seven seconds late is equivalent to a signal that never existed. For three nights afterwards I kept rewinding that footage, not to find my own mistake, but to ask a different question: why I had believed that more time would automatically produce the truth. That was the first time I understood something sixteen years in this industry had never made me question: every VAR error is a crack in the mirror that reflects the laws of the game. The 2026 K League Classic was the first season South Korea ran VAR across the entire fixture list, following a trial phase in the national cup. The organisers built three control rooms in Seoul, each staffed with a lead VAR referee, an assistant and a replay technician. The protocol followed the FIFA model: the assistant could only send an alert when a clear error appeared in four categories — goals, penalties, direct red cards and mistaken identity. The signal had to arrive before the ball was placed for the restart. The seven-second figure is not written into the Laws of the Game. It is an operational convention, born from a simple calculation: if the signal arrives later than that threshold, the referee has already issued a final decision and reversing it would create a precedent that cannot be explained to the crowd. In other words, the seven-second threshold exists to protect the referee's authority, not to protect the truth. I logged every incident in an automated journal, four fields per line: detection time, camera angle used, response time, final outcome. By Round 29 the journal had 118 lines. My average response time was 9.4 seconds. That sat inside the range the management considered acceptable, but it also said something nobody wanted to hear: in nearly a third of cases I sent the signal between 11 and 14 seconds, a zone in which my decision had almost no enforcement value. That is why I started caring about a narrower, more uncomfortable question: what happens to the data the system never manages to record. In 2026 I was sent to Russia as a VAR analysis assistant for a Korean broadcaster. My job was not to call the match but to reconstruct controversial incidents within twelve hours of the final whistle. IFAB entered the 2026 World Cup with a rewritten handball law, in which the concept of the "natural position of the arm" became the centre of every argument. I collected 27 handball incidents across the tournament and classified them on four criteria: the position of the arm relative to the body, the direction of the ball's travel, the distance from the point of contact to the torso, and whether the player actively expanded their body. The result: only 31 percent of incidents were handled consistently under the same reading of the law. The rest saw referees in different matches reaching different conclusions on incidents with near-identical geometry. I wrote a 40-page report for the editorial desk. They published a small chart, uncaptioned, next to the group standings. The trap of 2026 was not in the hand. It was in the belief in a definition that does not exist. IFAB wrote that an arm in a natural position is not handball, but IFAB never defined precisely what natural means. No coordinates, no angles, no thresholds. When a concept enters the law without a measurement standard attached, referees are forced to fill the gap with their own instinct — and instinct is not distributed evenly across 36 referees at a tournament. From then on I changed how I wrote. I abandoned the incident-report style and moved to investigation form: state the numbers, quote the relevant clause verbatim, then leave one open question at the end. I started a personal blog to publish all 27 incidents with frames, without asking the desk for permission. The post drew roughly 50,000 reads, mostly from referees, sports lawyers and supporter groups. In March 2026 global football stopped. The broadcaster cut my contract on budget grounds. Rather than look for work immediately, I locked myself at home for six months, downloaded every publicly available VAR dataset from five European leagues and built a sample of 1,247 decisions. The variable I cared about was crowd presence. The results forced me to rewrite my entire hypothesis. Without spectators, referee consultation time fell by 22 percent. But the rate at which referees upheld their original decision rose by 15 percent. Referees reviewed less, yet when they did review, they changed their minds less often. The simplest reading: with no crowd pressure from the stands, referees no longer needed to buy time to placate an audience. But there is a second, less comfortable reading. Without noise, referees lose a social signal they had been using as a form of confirmation. Crowd noise is not written into the Laws of the Game, but it carries legal weight. I wrote a 60-page report and published it to an academic repository. A member of a referees' committee at an Asian federation got in touch and invited me onto a data analysis panel. I accepted, and over the following two years I learned something no degree programme teaches: how to present a hypothesis, a method, and — most importantly — the limitations of the very dataset you are using. In 2026, as a mid-level staffer at a consultancy, I built a model to evaluate defenders from VAR and event data. It returned a verdict on Kim Min-jae, then at Napoli in Serie A: 0.73 dangerous errors per match, placing him in the high card-risk bracket relative to centre-backs across Europe's top five leagues. I recommended the firm leave him off its transfer shortlist. Napoli signed him anyway. In 2026-23 Kim played 35 Serie A matches, ranked among the league's most productive tacklers, and won the scudetto with Napoli after 33 years. When I tore the model apart, three flaws surfaced, each clearer than the last. First, I was counting an individual's errors inside an organised defensive system. Napoli under Luciano Spalletti played a high line with a holding midfielder dropping deep as cover. A centre-back in that system is permitted to step up aggressively because someone is behind him. The same action, placed in a system without cover, is a mistake; placed in a system with cover, it is part of the design. I measured behaviour and never measured structure. Second, I used Korean referees' definition of a foul to evaluate a player operating in Italy. On the same physical challenge, Serie A and K League referees apply different probabilities of a whistle, and the difference is not in the law but in how the law is read. My dataset had folded two reference frames into one column. Third, and most seriously, I had no data on game reading — the thing that appears in no event table. A good centre-back is not merely one who commits few fouls; he is one who stands in the right place before the situation becomes dangerous. No column records the tackle a player never had to make. That December I wrote a ten-page self-assessment and removed the model from the system. Since then, every report I write carries a mandatory closing section titled "limitations of the data". It is not a ritual of humility. It is a technical barrier, the thing that stops me turning an incomplete sample into a complete conclusion. And this is the point I want to linger on. In this profession there is a document type I encounter more and more often: a report with the full skeleton built out — objectives, method, taxonomy, risk matrix, scoring scale — where the entire body of content reads a single line: insufficient information. Thirteen pages, seven sections, dozens of table cells, and not a single fact. An outsider would call it worthless. I do not. An empty dataset is not a neutral dataset. It is a decision, and it is the hardest of all decisions: the decision not to judge. In sixteen years I have watched people more talented than me collapse not because they lacked data, but because they could not tolerate the silence of it. When there are no numbers, they write from instinct. When there is no instinct, they write from authority. When authority runs out, they write in the declarative voice — and the declarative voice is always available. Elsewhere in the industry I see the mirror image. After a wrong decision, governing bodies stay silent for 48 to 72 hours and call it an internal review process. But a wrong decision does not destroy a match; the silence after it destroys trust. The same information void, handled two opposite ways: in analysis the void is acknowledged as a mark of discipline; in match management it is covered up and becomes a mark of power. VAR was born from the fear of error, yet it nurtures the fear of late truth. There is a very common argument I have heard throughout my career: better no conclusion than a wrong one. It sounds reasonable, and most of the time it is. Push it to its limit, though, and it destroys itself. Because in football, delay always has consequences. Once the referee restarts play, once a goal has been awarded, once a team has benefited, the decision not to decide has produced a specific and irreversible outcome. An analyst who refuses to conclude has also produced an outcome: a decision-making group left without data will fall back on worse data — crowd feeling, public pressure, memory of the previous match. What people call "insufficient basis" is often a politer way of saying "I do not want to own this answer". Here I need to separate two things that are usually merged. There is a difference between an analyst who says the data is not yet enough to conclude, and an analyst who will not go and find the data. The first has done the work and reported honestly on the limits of the material. The second has simply not started. I sat in the second category for years. I believed rigour meant adding data. Only when the Kim Min-jae model collapsed did I understand that rigour lies elsewhere: knowing when the dataset you hold, however complete it looks, is in fact still empty. From the 2026 season I added a column to my personal VAR journal titled "what I could not see". It records what never appears on the monitor: the referee's position relative to the point of contact, whether his line of sight was blocked, his distance from the incident, and whether the signal from the VAR room arrived in time or after he had already locked into a judgement. None of those columns feed into any referee performance index. Yet they determine the majority of the controversial decisions that supporters rewatch thousands of times online. Fans call that a serious error. I call it a limit of the observation tool. Between those two descriptions lies a wide gap, and that gap is where a referee can be unfairly dismissed, or an analyst can be trusted wrongly. What I am tracking in the current domestic season is not whether penalties are awarded correctly. It is speed. In recent seasons, average review time in Asian leagues has trended down while the number of incidents reviewed has not. Referees are deciding faster on the same volume of disputes. That acceleration can be a sign of an improved process, or a sign of something more uncomfortable: referees have learned that reviewing for longer does not help, so they review for form's sake. If it is the second, leagues are preparing to accumulate a generation of decisions that even the VAR room will lack the data to defend. From where I sit, I am not proposing a change to the laws. The handball law has been amended three times in six years, and each amendment produced a new set of arguments. What I propose is much smaller: leagues should publish the VAR room's actual response times round by round, as a public operating metric. Alongside it, publish the rate at which original decisions are upheld after review — the figure I once measured in European leagues during the empty-stadium period. Neither metric requires changing the laws, adding cameras or opening control rooms. They require one administrative decision: accepting that data about the referees is also data belonging to the league. I know the counter-argument. Publishing numbers will expose referees to more scrutiny, and more scrutiny will push them towards defensive, safe decisions rather than correct ones. That concern is grounded, and I have seen it happen at match level when a referee knows he is being scored. But the opposite risk is worse. A defensive referee produces safe decisions, and safe decisions are predictable. A system without public operating data produces something far harder to repair: disputes that never end, because there is no point at which to stop. We are not searching the pitch for justice. We are searching for a reason to stop arguing. My VAR journal now runs to 2,084 lines, spanning 2026 to the present, four numeric columns and one written in words. The written column is the one I read most. It contains no conclusions, no team names, no scorelines. It records only what I could not see in the moment of decision. Ten years ago I believed a good analyst was the one with the most data. Now I believe something else: a good analyst is one who knows exactly which part of their conclusion has no data behind it, and says so before anyone asks. The VAR debate in Asian leagues over the next few seasons will not turn on whether the technology is right or wrong. It will turn on a drier question: whether a league dares to publish numbers about itself. Whichever competition answers that first will gain something broadcast rights cannot buy — the right to be believed. As for me, the last column in the journal keeps growing every round. And that is the only signal telling me I am still working, rather than merely taking notes.