No Angle Clear Enough: Anatomy of a Decision Made on Zero Signal

No Angle Clear Enough: Anatomy of a Decision Made on Zero Signal

Minute 63. The referee signals the VAR team, then runs to the monitor at the touchline. He puts on the headset, both hands on the earpieces, eyes locked on the wide frame. Behind him, the stand rises. Thirty thousand people already had the answer in their heads twelve seconds earlier, when the collision was still happening at real speed. The referee has ninety seconds, seven camera angles, a few frames that can be cut at two different points, and one question he is not allowed to answer by feel: was his original decision a clear and obvious error.

He turns back to the pitch. No change.

On the scoreboard, nothing changes. In the news feed, the line will read "VAR checked, no intervention". On social media, the line will be an adjective. And in the match record, this incident will exist as a dash.

I have spent most of my working life reading those dashes. Since VAR entered the major tournaments and then spread across almost the whole professional game, most public argument has circled the decisions that were made: right or wrong red card, penalty or no penalty, why a goal was disallowed. Very little argument circles the other category — what I call the null result: a check carried out fully under protocol, ending with nothing changed, and with nobody recording that the system actually looked at the incident.

That is the biggest blind spot in the modern game. And it starts from a very dry paradox: the system for evaluating referees is designed to detect errors, but it is not designed to detect silence.

---

Context: a protocol built on three layers of threshold

To understand how a check can end in nothing, you have to start with the architecture. The VAR protocol, laid down by the International Football Association Board and then specified by each confederation and domestic league, runs on three overlapping thresholds.

The first layer is the reviewable category. Only four groups of incident can be reviewed: goal or no goal, penalty or no penalty, direct red card, and mistaken identity. Everything outside those four — a throw-in awarded at the wrong spot, a challenge that should have been a second yellow, a corner given the wrong way — cannot be corrected, even when the whole stadium saw it clearly. This is a hard boundary, and it explains most of the anger I have witnessed from the stands.

The second layer is the intervention threshold. The VAR can only recommend an on-field review when there is a clear and obvious error, or a serious missed incident. The phrase "clear and obvious" is not descriptive prose; it is a technical condition. It forces the reviewer to reach a level of certainty before overturning a decision already made on the pitch.

The third layer is the principle that the original decision stands. Where there is not enough basis to conclude that the on-field decision was wrong, that decision is preserved. It is not preserved because the referee was right. It is preserved because there is not yet enough ground to say the referee was wrong. Those two sentences sound identical in ordinary language and are completely different in the legal text of a match.

The third layer is where the null result is born.

In seven years of working with reports, I have learned that most viewers, including people who watch a great deal of football, cannot distinguish three different states of a VAR check. State one: the system looks and confirms the decision was correct. State two: the system looks and overturns the decision. State three: the system looks, finds the camera angles do not permit a conclusion, and stops. To a viewer, all three display identically on the scoreboard. To a practitioner, they are three entirely different types of event.

---

The three minimum inputs of a decision

Football has no VAR, only blind spots waiting to be exposed. That line of mine is not meant to diminish the technology; it is meant to describe the nature of the problem exactly: a review system does not manufacture truth, it manufactures observability. And observability has physical limits.

A VAR check needs three minimum inputs before it can reach a conclusion. The first is the number and quality of camera angles. A challenge inside the penalty area needs at least one angle showing the point of contact, one showing the direction of travel of both players before contact, and one showing the position of the ball. Miss any of them and the reviewer is holding one-dimensional data.

The second input is temporal resolution — the ability to separate frames. This is the point I have spent many nights dissecting.

The third input is the chronological sequence of the incident — the ability to establish what happened before what.

Of these three, the second is the most underrated. Most spectators believe a camera records everything. No camera records everything.

For a high-speed passage of play, the gap between two consecutive frames corresponds to a considerable distance on the pitch. If contact occurs exactly in the gap between two frames, the system will not have a single image showing contact. It has one image before contact and one after. The reviewer must reason from those two images. Reasoning is not evidence.

I once spent a great deal of time analysing camera angles, calculating foot speed and contact angle, on a decision a colleague said took ninety seconds. He said I was spending too much time on something meaningless. I disagreed. Those ninety seconds are, I think, the most misunderstood unit of time in modern football. Ninety seconds is not the time to find an answer. Ninety seconds is the time to establish whether an answer exists.

---

The offside case: the physical limit of a frame

No category exposes the system's limits more clearly than offside.

For years, offside was handled by drawing lines onto a frame, projected from the position of the last defender and the nearest valid part of the attacking player. It sounds like pure geometry. It is not pure geometry; it is geometry plus an estimate.

Two sources of error always ran in parallel in that method.

The first is timing error. The frame chosen for measurement is not necessarily the frame in which the pass was played. It is the nearest frame the system captured of the passer's foot touching the ball. The gap between the real moment and the nearest frame can equal several tens of centimetres of movement by an attacker at sprint speed.

The second is body-reference error. "Nearest valid part" is a legal definition, but the exact position of that part on a moving body can only be established from images, and images have resolution.

In a situation where the offside margin is smaller than the combined error band, the system technically cannot reach a certain conclusion. In that era, some leagues operated very frankly: where the measurement error exceeded the offside distance, the on-field call stood. Viewers saw a line and believed it was a fact. Practitioners saw a line and knew it was an estimate with a confidence interval.

Every foul is a question about intent; data only gives us answers about consequence. With offside, the consequence is position, and position is measurable. But even the measurable has a threshold.

Bringing semi-automated offside technology into operation in some leading leagues changed the error structure, and I will return to that detail later. What matters here is that the change itself revealed something about the old system: for years, a share of offside decisions were reached not on the basis of complete evidence, but on the principle that the on-field call stands when evidence is insufficient. The public called that injustice. Procedurally, it was a null result recorded correctly.

---

The handball case: the language of the natural silhouette

If offside is a measurement problem, handball is an interpretation problem. And this is where null results occur most frequently.

The handball law has been adjusted several times over the past decade. There was a period when the boundary was pushed up close to the shoulder, turning any contact from shoulder down into an offence. There was a period when the law was loosened, with a provision that attacking handball is only penalised if the contact occurs immediately before scoring or creating a goal. Each adjustment was a re-collection of the interpretive boundaries.

What the adjustments could never resolve is the core question: was the arm in that position a consequence of body movement, or a deliberate movement toward the ball.

The referee has a frame with which to answer that question. The frame shows the final position. It does not show intent. The movement of an arm during a turn, an aerial duel, a jump to shield the ball, follows bodily momentum. Distinguishing momentum from intent toward the ball demands something no camera supplies: motive.

When a VAR looks at a handball and cannot establish whether the arm actively sought the ball or merely happened to be there, he must choose between two unsatisfying options. Either intervene on a guess about motive, or do not intervene and accept that an offence may have existed without being addressed. The protocol, in principle, prefers the second.

This is why I always say VAR does not fix errors, it only changes who carries the blame. Before VAR, the blame sat with the on-field referee, who had one look at real speed. After VAR, it moved to a sealed room with more angles, but with the same limit on motive. The difference lies in who gets criticised, not in whether more information exists.

---

The penalty-area case: intervention threshold and the psychological variable

Inside the penalty area everything gets harder because the consequences are bigger. A penalty decision can change a match result, change a league position, and in some cases change a manager's fate.

A challenge in the box can be read in at least four different ways, depending on where the reviewer places the emphasis. Focus on the defender and the question is whether he made contact with the opponent before the ball. Focus on the attacker and the question is whether he actively sought contact. Focus on the ball and the question is who touched it first. Focus on the intensity of contact and the question is whether the force was enough to cause loss of balance.

Those four questions can produce four different conclusions from the same clip.

In such situations, the intervention threshold becomes the only tool for consistency. A high threshold means more marginal incidents stand as called. A low threshold means more marginal incidents are overturned. Both choices carry a price.

The price of a high threshold is uncorrected errors. The price of a low threshold is correct decisions overturned without solid ground. For a process designer, the second is worse, because it breaks the system's predictability.

There is a psychological variable that very few analyses mention. An on-field referee called to the monitor knows his decision is being examined by a colleague in the VAR room, by tens of thousands of spectators in the stadium, by millions on television, and by a review panel that will meet the following Tuesday. That pressure does not disappear when he looks at the screen. It merely shifts from the pressure of deciding to the pressure of confirming.

The referee is the only person on the pitch not permitted to be led by emotion. In a match with forty thousand people screaming, the one person who must maintain the separation between emotion and judgement is the one holding the whistle. That is an occupational requirement, not a moral virtue. And it is the hardest requirement to meet inside ninety seconds.

---

The ninety-second sequence: step by step through a check

I have watched enough matches, and logged enough moments of referees running to the monitor, to describe the sequence that unfolds in that short window. The sequence is not officially published as a diagram, but it can be reconstructed from released audio and post-match reports.

The first step is establishing whether the incident falls within a reviewable category. The VAR must answer this before considering any detail. A missed second-yellow challenge is not reviewable, and the check can stop here, before any angle is opened.

The second step is establishing whether the incident sits within the attacking sequence leading to a goal. This is the step that expands the review scope, and it explains why a goal check can run several minutes: the system must rewind the chain of events back to the start of the move.

The third step is assessing the level of certainty. Here the VAR poses the central question: does a clear and obvious error exist, or a serious missed incident. This must be answered before the on-field referee is called to the monitor.

The fourth step is deciding whether to call the referee to the monitor. Summoning a referee to the screen is a weighty act, because it signals to the whole stadium that something is in question. In some cases the summoning itself generates pressure toward overturning, even when the evidence is not as strong as the signal suggests.

The fifth step is the on-field referee reviewing and making the final call. He is not bound to follow the VAR's recommendation. In practice, the share of on-field referees who keep their original decision after a review is not small.

The sixth step is recording the outcome. And this is the step I care about most.

In a great many reports, the sixth step exists only as one line: checked, no intervention. That line does not distinguish between three different situations. It does not say whether the system examined the incident closely and confirmed the decision, or examined it and found the evidence insufficient, or examined an incident outside the reviewable categories. Those three situations mean entirely different things for assessing refereeing quality, and they are merged into one.

---

The null result: when a check ends with nothing to say

A null result is a valid output. I want to state that plainly before analysing it, because most public argument runs the other way.

When an evaluation system is designed to avoid type-one errors — concluding guilt where there was none — it will tend to produce many null results. That is the mathematical consequence of a threshold, not a sign of weakness.

The problem lies in the fact that the null result is treated as a silence rather than as a datum.

A referee who correctly confirmed a difficult call and a referee who could not conclude because the angles were insufficient generate the same line in the report. But their competence is not the same. The first completed a judgement. The second completed a process and stopped in the right place.

For years I sent analyses to editors and they were routinely cut down. Once I wrote three thousand words on a penalty shootout in a final, citing data from the fourteen most recent shootouts, analysing the distance from the spot to the goal, ball flight time, and how a goalkeeper chose the right direction for most kicks by studying opponents' habits. The draft was cut to five hundred words for being too technical.

I understand the reason. But it does not change the fact that the part cut was the part carrying information.

---

Silent failure: the matches the system never triggered

There is an error category more serious than a wrong overturn. It is the error nobody notices.

In any review system, most mistakes do not lie in the contested decisions. They lie in the decisions nobody watched. A challenge that should have been a penalty but was not given, after which play continued with no check triggered at all, will not appear in the match statistics. It only appears if an independent post-hoc review process exists.

In some leading leagues, an independent panel meets after each round to examine key incidents. The panel comprises members independent of the competition organiser, including former players and former coaches, alongside representatives of the refereeing body. Its job is to answer two separate questions for each incident: was the on-field decision correct, and was the VAR intervention or non-intervention correct.

Separating those two questions is the most important element of the entire evaluation architecture. A decision can be correct while the intervention was wrong. A decision can be wrong while the non-intervention was right. The two questions measure different things: the quality of judgement and the quality of process.

In the panel reports I have read, most incidents are assessed as a correct decision and a correct non-intervention. This is the category the public calls "nothing to discuss". But in the data, it is the largest category by far. The system works correctly most of the time, and most of that time generates no news.

A smaller group is the set of incidents where the panel judges the on-field decision wrong but the non-intervention correct. This is the group that confuses viewers most, because it sounds like a contradiction. It is not a contradiction. It is the direct consequence of the principle that the original call stands when evidence is not strong enough.

And a very small group, with the largest influence on public opinion, is the set where both questions receive a negative answer.

---

The table you should be reading, and the table you are reading

In my own notes I once built a comparison table for a full season, with the aim of classifying every VAR check by its output type.

The first type is confirmation. The majority.

The second type is overturn. The minority.

The third type is stopping because evidence is insufficient. A share I believe is larger than the public imagines, and an output type that is almost never classified separately in public statistics.

The absence of the third type from public data creates a serious cognitive consequence. A reader of the statistics sees the overturn rate, the intervention count, the average check duration. They do not see the number of times the system looked and could not conclude. So they default to believing every check ends with a clear answer. Every later argument is then built on a false premise.

There is a way of testing data validity that I always apply before using any figure. I call it the control question: is this data actually answering the question I am asking. Many numbers in football are attractive as numbers but misaligned as objects. A beautiful figure does not automatically become valid evidence.

Here the control question is this: can the number of VAR interventions answer the question about process quality. It cannot. Intervention count is a frequency metric, not a quality metric. A system that intervenes often is not necessarily a good system. A system that does not intervene is not necessarily a bad one.

---

The inspection record: the value of logging a null result

In daily work I write match-discipline analyses based on referee reports, listing every minute of infringement and cross-checking against the Laws of the Game. That method came out of one act of self-audit.

At twenty-six, at a qualifier in London, I mispronounced a defender's name three times in a row. A colleague laughed. The next day I received a reprimand by email from the editor-in-chief. That evening I opened the recording and watched the passages over and over to understand how a referee determines offside. The initial motive was petty: I wanted reassurance that I was not the only person making mistakes on a pitch. That motive disappeared quickly. What remained was method: log, cross-check, classify.

My writing changed from there. I began using card and serious-foul tables instead of emotional narration. I built a record format for each match in which every incident has a code, a description, a corresponding law article, and a conclusion about whether the process was correctly followed.

In that record, null results are logged as fully as any other result. They have their own code. They have their own description. They have a corresponding law article. Logging a null result takes nearly as long as logging an overturn, and nobody reads it. But without it, the record would produce a false picture of the system.

I believe most of the argument about refereeing in professional football stems from records not being kept in full, not from referees being wrong more often than normal.

---

The disappearance of the record

In any process, the first thing to disappear under pressure is the intermediate record.

An on-field referee stands at the touchline monitor for ninety seconds. During those ninety seconds he watches at least four angles, at multiple speeds, and exchanges words with the VAR room. He reaches a conclusion. The record of that entire ninety seconds exists. Whether to publish it is a policy decision.

When the record is not published, only the final conclusion is known. A final conclusion, standing alone, looks like an imposed verdict. With the record attached, it becomes an inspectable process.

Some leagues have begun releasing audio in periodic review programmes, where the head of the refereeing body analyses key incidents alongside the audio and the angles. For me this is the most significant change in years, more important than any technological improvement. It does not make decisions more correct. It makes them inspectable.

And when a process becomes inspectable, the null result is no longer read as silence. It is read as an output.

I once followed twelve matches purely to log every moment a referee ran to a monitor. That was the first major tournament using VAR, and I was absorbed by the mechanism: who is entitled to request, where the intervention threshold sits, what the sequence is. Afterwards I wrote a five-part series analysing forty-seven VAR decisions, and what I found was not systemic injustice. What I found was inconsistency in the error threshold between matches, between VAR teams, and sometimes between the two halves of the same match.

Viewers see the incident, referees see the moment, I see the whole process. Those three perspectives yield three different conclusions about the same event, and only one of them can be tested against data.

---

When the public demands a verdict

The greatest pressure on a refereeing system does not come from clubs. It comes from the demand for a definitive answer.

After every round, a large volume of content is produced to answer the question of who was right and who was wrong. Most of it operates on the assumption that a single correct answer exists, and that it can be found if someone is clever enough or brave enough to say it. That assumption is structurally false.

In many situations, the procedurally correct answer is that there is not enough basis to conclude. Reaching a conclusion in that situation is not courage; it is fabrication.

I am often asked why I never declare a referee right or wrong in my writing. The answer is simple: I describe the mechanism. I do not sit in the referee's chair for him. I have no access to the touchline monitor, no headset, and no knowledge of what the VAR room said to the referee in those ninety seconds. Concluding that a decision was right or wrong without those three things is a guess presented as a verdict.

There is a strong professional temptation in writing about referees: the temptation to deliver a verdict. It generates engagement. It generates argument. It makes the piece look weighty. And it destroys the piece's value.

What I try to do instead is classify error. An individual error is a wrong decision under a correctly functioning process. A process error is a wrong decision under a process that itself created the conditions for it. These two need different remedies. The first needs training. The second needs system redesign. Confusing them is the most common mistake in every refereeing argument I have followed.

---

The counter-view: a null result is not cowardice

There is a popular reading of decisions that do not change after review: reading them as the system's cowardice.

That reading has strong emotional appeal. It is simple. It has a villain. It produces a story that can be told in thirty seconds.

It is also wrong on one basic technical point: it assumes the null result was chosen when the system had enough data to reach a different conclusion. In most of the cases I have analysed, the null result was chosen because the system did not have enough data. That is a conclusion about data, not a conclusion about will.

Distinguishing those two is the hardest thing in writing about referees, because the crowd's emotion does not distinguish them. To a viewer, an uncorrected error and an uncorrectable error produce the same feeling. To the system, they are two different problems.

The counter-view I consider correct lies elsewhere. The real risk to the modern refereeing system is not that it ignores too much. The real risk is that it can produce a null result while presenting it as a complete result. When that happens, the public believes the system checked and confirmed, when in fact the system checked and could not conclude. That confusion accumulates over time and creates a false belief about the system's accuracy.

That is a more serious error class than any individual mistake, because it cannot be detected by rewatching the footage.

---

On avoiding emotion, and on emotion as a kind of data

I am often described as a dry writer. The description is not wrong.

There is a technical reason for the dryness. When writing about a decision system, emotional language cannot carry the structure of that system. An emotionally powerful sentence tends to compress a multi-step process into a single act. It turns a six-step sequence into one sentence: the referee got it wrong.

That is a very large information loss.

There is one thing, though, that I have had to concede after years in the job: the emotions of the participants are not noise. They are data.

A referee walking into a derby knows that a wrong decision in the eighth minute will follow him through the match and for weeks after. A young player stepping up to the spot in the hundred-and-twentieth minute knows the distance is eleven metres and the average ball flight time is about three tenths of a second. A manager on the touchline knows a decision in the ninetieth minute can decide his future.

Emotion in those three situations is not what blurs judgement. It is part of the conditions of decision-making, and the conditions of decision-making are part of the process.

What I remove from my writing is not emotion. What I remove is the use of emotion as a substitute for analysis. Emotion can be described in neutral language: pressure, timing, consequence, threshold. When those words appear, the emotion has been converted into data and can be used for analysis.

The referee is the only person on the pitch not permitted to be led by emotion. That does not mean referees have no emotions. It means their emotions must be converted into another form before they influence a decision. In writing, I try to do the same thing.

---

Anxiety and retreating into numbers

There is a behavioural pattern of mine that I am fully aware of and do not try to hide: when the environment becomes unstable, I retreat into data.

When the pandemic suspended the leagues, I had no matches to report on. Anxiety rose. I went back and watched an entire season of recorded matches, focusing on the tactical fouling patterns of one team under one manager. I calculated that the team committed roughly a little over ten fouls per match, mostly in midfield, for the purpose of stopping counter-attacks. I wrote an analysis and sent it to an editor. He rejected it on the grounds that nobody reads about football while there is no football.

I still wrote twenty pages of notes. I was absorbed because I had spotted a pattern: how a manager uses tactical fouling to break an opponent's rhythm. Fouls were not an accident in his system. They were a tool in his system.

Later, those notes became the foundation of an e-book on law and tactics that I published two years afterwards. What I learned from that experience was not the value of persistence. It was the value of keeping records when nobody is reading. A record only proves its worth when somebody needs it, and nobody knows in advance when that will be.

The law never stands outside the match; it is a second match running alongside. In that second match, intermediate records matter as much as final conclusions.

---

Which threshold is right, and whether that question is answerable

Having laid out the whole architecture of the process, the natural question is where the intervention threshold should sit.

This is a question with no single technical answer, and it is important to understand why.

The intervention threshold is a choice between two error types. A high threshold reduces the number of correct decisions overturned while increasing the number of wrong decisions allowed to stand. A low threshold does the reverse. No single point on the curve eliminates both error types at once, unless the quality of the input data improves.

This is why technological improvements matter more than arguments about the threshold. Improving the temporal resolution of the capture system reduces error on both sides of the curve. Adding camera angles reduces the number of inconclusive incidents. Deploying semi-automated offside removes one specific error source and turns a reasoning decision into a measurement decision.

After one error source is removed, the arguments do not disappear. They migrate to the next error source. This is a rule I have observed long enough to trust: a decision system never runs out of argument, it only changes the subject of argument.

That sounds pessimistic. I do not think it is. It means the quality of the system is measured not by the volume of argument but by the type of argument. When argument shifts from "did the referee see it" to "is the threshold right", the system has advanced a step, because the second question is one that data can answer.

---

The most worrying thing is not a wrong decision

In discussions about refereeing, all the attention pours onto wrong decisions. That is a reasonable reflex: mistakes cause consequences, and consequences must be handled.

But the error type with the largest long-run consequence is not a specific wrong decision. It is a process that looks complete but is in substance empty.

Such a process has all its data fields populated. It has an incident name, a match code, a classification, a conclusion. The conclusion is a line stating that nothing changed. No field records how many angles the system actually used, at what resolution, for how many seconds, and with what degree of certainty.

A record like that can pass an entire quality-check system undetected. It is not wrong in format. It is only empty in content.

This is the error type I believe is the most common and the least discussed in every expert evaluation system, not only football. It has a dangerous property: it produces no symptom. A large mistake produces a conceded goal, an outcry, an inquiry. An empty record produces nothing at all. It only makes the file look more complete than it is.

The remedies for this error type are very specific and very unattractive in media terms. The record must state the volume of inputs used. Every conclusion must be accompanied by an assessment of the sufficiency of the evidence. A null result must be classified as a null result by confirmation and a null result by insufficient evidence. Those two must sit in different columns in every statistics table.

When those three requirements are met, the refereeing system becomes auditable. And an auditable system is one that can be improved by data rather than by argument.

---

On being allowed to say you do not know

There is one professional competence I consider the most important in any decision system, and it is also the least rewarded: the competence to say you do not know.

In an environment where every question is expected to have an answer, saying you do not know is read as weakness. But in a decision process with consequences, saying you do not know is a valid output, and forbidding that output is a design flaw.

A referee placed in a situation where he must always produce a definitive answer will produce definitive answers in situations where the data does not permit one. That is worse than letting him keep his original decision.

The same holds for writing. For years I received requests to write pieces with clear conclusions about situations where I did not have enough data to conclude. I declined some of those commissions, and I know some people read that as evasion. I accept that reading. Writing a conclusion I have no basis for is an act of professional sabotage, and it sabotages more slowly and more discreetly than an obvious mistake.

Football has no VAR, only blind spots waiting to be exposed. The job of the process person is to record exactly where that blind spot was exposed and to what degree. The job of the writer is to describe that process without adding a conclusion the data cannot support.

---

A forward thought: audit before you judge

If I were allowed to propose a single change to the professional refereeing system in the coming years, I would not propose more technology. I would propose a change to the record.

Every VAR check, including those ending with no change to the decision, should be logged with four pieces of information: the number of camera angles used, the temporal resolution of the key frame, the duration of the check, and a classification of the sufficiency of the evidence. Those four fields resolve no specific argument. They make every argument auditable.

Once the record contains those four fields, a question that currently cannot be answered will have an answer: this season, how many decisions were not changed because the system confirmed them, and how many were not changed because the system lacked sufficient evidence.

The difference between those two numbers is the most precise available measure of the quality of a refereeing system. Today, nobody knows that number. And in an environment where every mistake is logged and every silence is erased, not knowing that number is a larger gap than any single contested decision.

Viewers see the incident, referees see the moment, I see the whole process. And within that process, the thing most deserving of a record is not the moment a decision was made, but the moment a decision could not be made.