Camera Angle Error and the Limits of VAR: Revisiting 147 Controversial Incidents in the 2026 CSL

My office in Chengdu looks down on a narrow street, and that night it was raining. I switched on the monitors at eleven o'clock, a screen split into four frames, three of them showing pictures, one left blank. The analysis file in front of me was blank in a different way: no original headline, no source, no argument, no facts. Nine analytical sections, from tactics to club finance, each ending with the same line — insufficient information.

Reading files like that is my job. In 2026, fresh out of the Journalism Academy and starting at Bao Bong Da while also filing for Bao The Thao The Gioi from Madrid, I learned one rule with no exceptions: where there is no evidence, leave the gap open. Young writers are afraid of gaps. I was too. But football taught me that most analytical errors do not come from misreading data. They come from reading too much of what is not there.

I found the mistake not in the centre of the pitch, but at the edge of the frame.

That night, between an empty file and thousands of hours of old tape, I understood that I was facing VAR's single biggest problem. It is not a technology problem. It is the question of how much empty information a human being can tolerate before starting to invent a plausible answer to fill it.

147 incidents and a database nobody asked for

In 2026 I was thirty-six, working as a VAR data analyst for a sports television channel in Chengdu. The Chinese Super League season had twenty-eight rounds. I reviewed one hundred and forty-seven controversial refereeing incidents, watching each an average of eleven times, and one of them thirty-seven times.

Among those one hundred and forty-seven, I found twelve incorrect offside decisions directly linked to camera placement. Twelve out of one hundred and forty-seven. That ratio sounds small, but each occurrence is a goal disallowed or a goal wrongly awarded. In a league where the gap between champions and third place is often a few points, a wrong call in round twenty-five can change an entire season.

The case I remember best is Guangzhou Evergrande against Shanghai SIPG in round twenty-five. Wu Lei's goal was disallowed for an offside margin of fifteen centimetres. At a striker's sprinting speed, that is a distance the human eye cannot resolve even standing level with the play, let alone looking through an offset lens. The problem was that no camera in the broadcast system that night sat on the lateral plane of the last defender. Referee and VAR team had to infer from an oblique angle, and when you infer a fifteen-centimetre margin from an oblique angle, you are no longer measuring. You are guessing.

I started building my own database, which I called viewing angle error. Each incident is a row, and each row holds camera coordinates, distance to the incident, deviation from the offside plane, number of available frames, and estimated error percentage. Nobody asked me to build it. No editor paid for it. But it completely changed how I wrote about refereeing, and it changed how I looked at every other sport I covered afterwards — eight Olympic Games, eight World Cups, several editions of the Giro d'Italia and the Tour de France.

In July 2026 I took that database into the World Cup final between France and Croatia. In the thirty-fifth minute, referee Nestor Pitana used VAR and awarded a penalty to France after the ball struck Ivan Perisic's hand. I measured from six main broadcast cameras. Only one angle showed Perisic's arm extended in a way described as unnatural. And that angle — precisely that one — was the only angle the VAR team in Moscow showed the referee, for one minute and forty-seven seconds.

I wrote a two-thousand-four-hundred-word piece and published it on a blog specialising in the laws of the game. Forty-eight hours later it had more than fifty thousand shares.

A line that exists nowhere

Viewers believe the offside line is a line. On television, the offside line is a line drawn by a person, overlaid on a perspective image, extrapolated from a reference point that does not sit on the plane being measured.

Picture a camera on the touchline, five degrees off the lateral plane of the last defender, twenty metres from the incident. To the naked eye, five degrees is negligible. To geometry, five degrees at twenty metres produces a lateral error of roughly one metre seventy-five — taller than a defender.

Most of that error can be corrected by projection, but projection is only valid when the three-dimensional camera coordinates, the focal length and the lens distortion are known exactly. In most domestic league matches, those three parameters are not recalibrated after every camera move. Operators do not re-measure focal length at half-time.

At a margin of fifteen centimetres, every error is larger than the conclusion.

This is the first principle of VAR analysis: when the conclusion is smaller than the instrument error, the conclusion is not a conclusion. It is a choice. And choices always have a chooser.

Fifty frames and sixteen centimetres

There is a technical problem very few viewers know about, and it has cost me many nights of sleep.

Standard football broadcasting runs at fifty frames per second. On NTSC, the figure is fifty-nine point nine four. Each frame is twenty milliseconds apart. A player sprinting at thirty kilometres per hour, roughly eight point three metres per second, covers about sixteen point six centimetres between two consecutive frames.

Sixteen point six centimetres. The offside margin in Guangzhou versus Shanghai was fifteen centimetres.

In other words, the gap between frames is larger than the gap the referee is being asked to judge. When the VAR team picks the frame for the moment the ball is played, they are choosing between two images in which the player has already moved more than fifteen centimetres. Where in that interval the ball actually left the foot, nobody knows. The system only records the frame before and the frame after.

It took thirty-seven replays of one incident in round nineteen that season before I understood that the human eye is not a measuring device. The eye is a very good interpolation device. It automatically fills the gap between two frames, automatically reconstructs the trajectory, automatically picks a moment that feels right. That feeling of rightness is not in the data. It is in the viewer's head.

Clear and obvious is a vague clause

The international VAR protocol rests on one phrase: clear and obvious error. Only mistakes meeting that threshold may be intervened upon. The phrase sounds rigorous, and it is entirely not rigorous.

The problem is that the clear and obvious threshold has never been defined in units. It is not fifteen centimetres. It is not one frame. It is not a percentage. It is a perceptual threshold, calibrated by experience, by competition, by round, and by how many times the referee has already blown his whistle that day.

In my database I tried to attach a number to it. I went back through incidents and asked: if the same incident happened in round one and in round twenty-eight, would the intervention rate be the same? The answer is no. Incidents similar in geometry, in speed and in degree of contact produced markedly different intervention rates between the early season and the late season. Nobody ordered referees to change the threshold. The threshold drifted on its own.

That is why I say the space for subjective judgement in VAR is larger than people think. People assume VAR narrows that space. It only moves it from the middle of the pitch into a closed room where fewer people can see it.

The first camera angle anchors perception

There is a psychological effect inside the VAR room that I have spent years trying to quantify, and I am still not finished.

When a VAR team receives an incident, the operator picks an angle to open with. That angle is usually wide, shot from the main stand, because it is the angle most familiar to the human eye. Within a few seconds the VAR team has formed a hypothesis. From the tenth second onwards, every additional camera angle is read through that hypothesis.

Psychologists call this anchoring. People evaluate new evidence against an initial anchor rather than evaluating the evidence from scratch. In a room of three people, the first to voice a judgement often anchors the other two, and speaking order is not specified in any document.

In the 2026 final, the VAR team opened with an oblique angle from behind and to the left of Perisic. From that angle his arm was clearly extended. But clearly extended from an oblique angle is not the same as clearly extended from a straight angle. I measured from six different angles. Only one showed an extension beyond the threshold I estimated as necessary to describe it as unnatural. Four showed the arm in a neutral position. One lacked the resolution to conclude.

One out of six.

Semi-automated offside and its limits

Semi-automated offside technology, rolled out in major competitions from 2026, is a genuine advance. I want to be explicit about that, because my trade is often read as anti-technology. It is not. The system tracks twenty-nine points on a player's body at fifty times per second, plus a sensor in the ball to determine the moment of the kick. Geometrically, it solves most of the error I was measuring in 2026.

But it does not solve everything, and what remains is the interesting part.

The ball sensor determines the moment of contact with higher precision than broadcast frames, but it still has latency and still has a force threshold. A softly struck pass may not trigger the sensor at the required level. And the question of when the ball was played still requires human confirmation in marginal cases.

Twenty-nine points on the body are still not the whole body. The arm is tracked, but the shoulder has to be interpolated. A defender rotating at high speed can produce a shoulder point the system only measures indirectly.

Technology does not remove judgement. It moves judgement from looking to calibrating.

The decision chain: four variables I always record

When I analyse a VAR incident I record four variables, and I believe anyone who wants to understand a refereeing decision should record all four.

The first variable is time: from the ball going dead to the referee's final decision. The average in my database is sixty-eight seconds. The longest incident I recorded was three minutes and twenty-two seconds, in a round twenty-two match. The longer it runs, the greater the pressure on the VAR team, and decision quality does not rise to match it.

The second variable is camera angle: how many angles exist, how many are viewed, and in what order. Order matters more than quantity. Three angles viewed in a good sequence can be more useful than eight viewed chaotically.

The third variable is movement data: the positions and speeds of the players involved across two seconds either side of the incident. This is the part that tells me which trajectories were physically possible and which were not.

The fourth variable, and the most neglected, is the on-field referee's cognitive limit. Where was he standing, what could he see, and which player blocked his line of sight. A wrong decision made from an obstructed position is an entirely different error from a wrong decision caused by misreading the ball's flight. The present system treats both the same way.

Empty stadiums in 2026 and the data gift nobody wanted

In 2026 major competitions were played in empty stadiums. To me that was a laboratory that cannot be recreated.

In football with crowds, referees work inside a constant field of acoustic pressure. A roar in the fifth minute differs from a roar in the eighty-fifth. Part of what people call game management is the management of that acoustic field.

In 2026 the field vanished. What I observed was that card counts fell in several leagues during the early empty-stadium period, then rose again when crowds returned, with different magnitudes across leagues. Home advantage is not only a psychological edge for players. It is a variable inside the referee's threshold of judgement.

For VAR, the empty-stadium period delivered something else. Match audio — the sound of the ball, of boots, of players talking to each other — became far easier to capture. In many contact incidents I could hear the moment of contact and compare it with the frames. Broadcast football with crowds never gave me that data.

Empty stadiums in 2026 showed me this: VAR does not save football, it exposes football. When the shouting no longer fills the gap, everything we call the feel of a match is stripped down into individual, countable decisions, and many of them cannot survive a single slow replay.

We thought we were seeking justice, and found we were seeking a better camera angle

I wrote that line in 2026 and was criticised heavily for it. I stand by it, but I want to state my meaning more clearly, because it is usually read as technological cynicism.

What I mean is this: the public's demand for justice is real, but the way that demand is satisfied does not pass through justice. It passes through imagery. A decision is accepted not because it is correct, but because it has supporting pictures. And where there are no supporting pictures, the same correct decision will be rejected.

That means the acceptance standard for a refereeing decision has shifted from legally correct to convincing on screen. This is a systemic change, and it has a consequence: it rewards decisions with beautiful footage rather than decisions that were difficult.

There is a case in my database, a handball incident where the main camera was shooting into the sun. The image was hazy and unconvincing. The VAR team did not intervene. Had the same incident been shot the other way — sun behind the referee instead of in front of him — I believe the intervention probability would have been different.

Same law. Same incident. Different sun.

Wasted running and beautiful numbers

In seventeen years of working with data I have learned one thing about metrics. They are not wrong. They simply measure things other than what they are said to measure.

Distance covered and sprint count are packaged as measures of effort. Both are physically accurate. The issue is that they are used to answer a question they cannot answer.

A midfielder who covers twelve kilometres in a match may have covered most of it in situations of no value — chasing a ball already out of play, running back into position after being beaten, running to fill a gap he himself had just created. The same twelve kilometres, allocated by zone and by situation, can describe a far better match or a far worse one.

This is why I never read a metrics table without the position map beside it. Wasted running produces beautiful numbers. Beautiful numbers produce flattering reports. Flattering reports produce transfer value.

Transfer windows: noise drowning signal

In a transfer window this problem worsens exponentially.

I have covered football for twenty-nine years, roughly fifteen of them working with transfer data for several outlets. The pattern barely changes. The volume of rumours in one window is always about ten times the number of completed deals. The volume of rumours sourced directly from a club or an agent is far smaller.

My filter is simple, and I always tell readers it is a filter, not a promise.

Tier one covers things that can be priced: release clause structure, remaining contract length, available wage headroom, and performance payments already written into the existing contract. These are read directly from documents or financial statements, and they determine whether a deal is accounting-viable at all.

Tier two covers behavioural signals: a club selling a cornerstone and immediately signing two players in another position, a player omitted from a matchday squad with no published medical reason, an agent appearing in a city where his client's club is not playing.

Tier three is a single story from a single source with no cross-confirmation. This tier makes up most of the news volume and has almost no predictive value.

What I call real transfer news almost always sits in tier one. It is less exciting. It talks about release clause structure and wage bills. And that is why most fans do not read it.

Modern football does not lack data. It lacks the tolerance for data's boredom. The transfer window is when a correct number is flattened by an entertaining rumour, and nobody feels it is unfair, because the rumour is more fun.

Three trends and one proposal nobody will adopt

Over the next decade VAR will move in three directions.

First, automation will increase. Offside will become almost fully automatic, and offside controversy will fall sharply. That is good. And it will push all the pressure onto the harder part: contact incidents, handballs, challenges inside the box. In those, geometry cannot help, because there is no plane to measure.

Second, the protocol will tighten control over camera angles. This is unavoidable, and it will be resisted for broadcast reasons. A system that only permits conclusions from calibrated angles will generate fewer clips, less controversy, and fewer views.

Third, and this is what interests me most: audio will become a formal data source. Pitch-side microphones, recorded in sync with pictures, can help distinguish light contact from heavy contact in many incidents where frames are not enough. It is far cheaper than high-speed cameras, and it has sat inside broadcast data for years without anyone exploiting it.

My proposal is simple and I know it will not happen: every VAR decision should be published with an error figure. Not a video, a number. If the offside plane has been extrapolated to a margin of plus or minus twenty centimetres, then that twenty should appear on the big screen inside the stadium, beside the line.

Nobody wants that, because it turns a decision that looks definitive into a decision that looks uncertain. And every system of authority needs a look of certainty.

Based on my experience following matches across roughly twelve thousand games in twenty-nine years, most of the damage in football does not come from referees deciding wrongly. It comes from us building a system that does not let referees say they do not know.

An empty analysis file may be the most disappointing thing a reader can be handed. To me it is comfortable. It is the correct shape of most marginal situations in football: six camera angles, three frames, two hypotheses, and a gap no technology can fill.

Next time you see a controversial decision on your screen, set aside the question of whether the referee was right. Ask something else first: where was the camera, and has anyone checked its margin of error.