If You Did Not See It, You Cannot Call It: A Lesson on Evidence in Modern Football Analysis

If You Did Not See It, You Cannot Call It: A Lesson on Evidence in Modern Football Analysis

Three camera angles. Not one of them shows the ball hitting a hand. The assistant referee is eleven metres from the incident, his line of sight blocked by two shirt backs. The stadium howls for a penalty, players from both sides surround the referee, and in his earpiece the VAR room delivers the shortest sentence in its entire vocabulary: no conclusive evidence. The on-field decision stands. The ball is placed on the line, the game moves on, and twelve seconds later nobody remembers what happened.

I lived inside moments like that for ten years. But this week, for the first time, I encountered the exact same moment off the pitch.

It arrived as a nine-dimension analysis report sent to me for review. It had every heading, every table, every assessment frame, a full conclusion, a risk section, recommendations. Structurally, it was flawless. In substance, every cell from the first to the last carried the same phrase: insufficient information to assess.

No teams. No players. No competition. No scoreline. Not a single figure to cross-reference. Not a single name to verify. A document thousands of words long, correctly formatted, with emptiness inside.

The person who wrote it did the hardest thing in our profession: he refused to invent what he did not have. He had room for a very good story. He chose not to write one.

Context: when the frame looks better than its contents

Modern football analysis runs on data pipelines. A match passes through four layers: collection, extraction, classification, interpretation. Each layer has automated checks to ensure output matches the format the next layer expects. Title has a field. Source has a field. Article type has a field. Information points have a field. Tables have data types. If every field exists and every type matches, the system flashes green.

The problem is that an empty field is still a valid field.

This is what engineers call a silent failure. No red alert. No exception thrown. No error log. The system runs through, prints a document that looks highly professional, and passes it downstream. The next layer trusts the format rather than the content, and begins working with zero.

Football has a mechanism against exactly this kind of failure, and it existed long before any algorithm.

Law 5 of the Laws of the Game grants the referee full authority over all matters relating to the match, and the referee's decision is final. But that authority comes with an unwritten condition every former referee knows by heart: you can only call what you saw.

When VAR arrived, that principle was not replaced. It was written into text. The IFAB VAR protocol permits intervention in only four categories: goal or no goal, penalty or no penalty, direct red card, mistaken identity. The threshold is a clear and obvious error, or a serious missed incident. And there is one line in the protocol that few spectators notice, even though they hear it every week: if the review process does not produce conclusive evidence, the original on-field decision stands.

That is a null-handling rule. Failing to find evidence does not mean the event did not happen. It means that in this instance we have no basis to overturn. A referee is not permitted to convert ignorance into a verdict. He is only permitted to preserve what already exists.

A system is trustworthy only when it can distinguish between "there is nothing" and "we cannot see".

The nine-dimension report managed that distinction. It marked every cell as insufficient rather than filling it with guesswork. But to understand why that is so difficult, we need to go back to a June night in England.

17 June 2026

On 17 June 2026, the Premier League returned after a three-month pandemic shutdown. Aston Villa against Sheffield United at Villa Park finished goalless. It did not finish goalless in an ordinary way.

In the second half, Sheffield United won a free-kick. The ball was delivered into the box, the Aston Villa goalkeeper gathered it in his hands, and as he fell backwards to secure it, his whole body and the ball crossed the line. The ball was in the goal. Sheffield United players raised their arms. The referee showed no reaction.

On his wrist was the watch linked to goal-line technology. That system comprised seven cameras, and on this occasion all seven were obstructed at the decisive moment: one by the goalkeeper, one by the referee standing very close, the rest by players inside the box. The system did not see the ball. Because it did not see the ball, it sent no vibration signal. The watch stayed silent.

The referee had no evidence on which to award a goal. He followed the protocol correctly: no signal, no goal. Play continued. Afterwards, when television replays circulated, the whole country saw the ball inside the goal. The technology provider issued an apology. But the score stayed 0-0, and it was never corrected.

I tell this story not to attack technology. I tell it because it is the perfect illustration of something football's data industry still refuses to learn: the most dangerous failure is not the failure that reports an error. The most dangerous failure is the one that reports nothing at all.

Goal-line technology failed silently. It did not break. It did not crash. It simply had no data, and it passed that emptiness down to the final decision-maker without any accompanying warning. The referee received the signal "no goal" and trusted it, because for years that signal had never been wrong.

This is exactly what happened with the nine-dimension report. A data pipeline stopped working somewhere between collection and extraction. The output still conformed to format. Nothing raised an alarm. The final recipient trusted the structure, and unless that person happens to be the suspicious type, an entire chain of decisions downstream gets built on nothing.

VAR does not correct a match's mistakes — it exposes how we define mistakes. The error at Villa Park was not a bad judgement. It was a correct judgement built on false information. In football, as in data analysis, these two types of error need different names, because the way you fix them is entirely different.

One point eight eight millimetres

On 1 December 2026, at the World Cup in Qatar, Japan played Spain. In the 51st minute, Kaoru Mitoma chased a ball that appeared to have gone out of play. From the stands, from the sofa, from every human vantage point, the ball was out. Mitoma cut it back, Ao Tanaka finished, and the goal was awarded after semi-automated technology confirmed the ball still had a fraction inside the field of play.

The published margin was 1.88 millimetres.

At that scale, the question is no longer right or wrong. The question is: what are we measuring, and who defined the boundary of that measurement.

The semi-automated offside system uses twelve dedicated cameras around the pitch, tracking twenty-nine body points on each player, plus an inertial sensor inside the ball operating at 500 hertz. It is an impressive machine. And it still answers only one question: at the chosen moment of the pass, where was this player relative to that player.

It cannot answer who chose that moment. It cannot answer why we measure the toe, the armpit, or the knee. It cannot answer why offside is an offence at all, while a foul in midfield — clearer, more dangerous — is worth only a free-kick.

Across several seasons, major leagues have repeatedly moved the lines. At one stage the lines were drawn thicker to give attackers the benefit of the doubt. At another, the debate turned to camera frame rates, because a camera recording at a lower frame rate can miss the exact instant of contact. Every time such an adjustment is made, organisers say they want greater fairness. But fairness is not a physical constant. It is a choice.

This is where modern football analysis usually goes wrong. We tend to believe that as resolution increases, controversy disappears. The opposite happens. When a measurement becomes accurate to the millimetre, it does not remove argument; it moves argument to another level — from observation to definition.

And at the level of definition, there is one thing every measurement system misses.

There is something the offside trap can never catch: a player's intention. A striker running behind a defensive line may be trying to break through, or he may be dragging a defender out of position so a teammate can exploit the space. Identical actions on video. Completely different purposes. No camera records purpose, and no algorithm infers it from coordinates.

As someone who once stood inside the pitch, I can say that most of the hardest decisions in a match sit in the gap between behaviour and intent. That is why I do not believe in a future where referees are fully replaced by machines — not because machines are not good enough, but because the question we need answered is not the question they were designed to answer.

Pedri and everything outside the spreadsheet

At Euro 2026, the semi-final between Italy and Spain was played on 6 July 2026 at Wembley. Italy won 4-2 on penalties after a 1-1 draw.

Before that match I was assigned to track Spain. I spent ten hours reviewing their earlier games. In one of them I wrote two numbers in my notebook about an eighteen-year-old named Pedri: touches, by my own count, above 90, and a pass completion rate that was nearly absolute.

Those numbers were not what caught my attention. His positions were.

Pedri did not run toward the ball. He ran toward where the ball would arrive, and most of the time that place was fifteen to twenty metres from where he currently stood, at an angle nobody was watching. When the pass came, he was already there. When it did not come, he had opened a corridor for a teammate.

No statistical table records that run. Tables record passes, not the creation of space. They record shots, not the creation of a chance for someone else to shoot, ten minutes later, in a different phase.

Pedri does not run after the ball, Pedri runs toward where the ball will arrive — and that is the entire difference. It is also the entire problem with the xG model.

Expected goals estimate the probability that a given shot becomes a goal, based on location, angle, the type of pass received, defensive pressure and other factors. It is a good tool. But it can only begin calculating from the instant the ball leaves a player's foot. The entire preceding process — off-ball movement, the decision to pass rather than shoot, the act of pulling a defender out of position — sits entirely outside the model.

Put differently, xG measures the quality of a chance. It does not measure the quality of the process that created the chance. A team can post high xG across several consecutive matches and still lose, and the media will call it bad luck. But if you follow that team across a full season, you see a different pattern: they generate high xG because they face defences willing to let them hold the ball in harmless areas, then collapse on them when they push up.

This is why I say gegenpressing, in its extreme form, has been decoded. Not because its principles are wrong, but because in the hands of mid-table sides with good stamina and limited technique, it degenerates into running. Football becomes athletics. And athletics can be measured with a stopwatch, while decisions cannot.

Not zero, but blank

There is one systemic error in football data analysis I encounter more than any other, and it is almost never named: confusing a blank value with a zero value.

Picture a scouting system. It builds shortlists from a data provider's metrics. For each player it records successful duels, key passes, ball recoveries.

Now suppose the provider does not collect defensive metrics for lower divisions. For a defender in the third tier, the successful-duels field is empty.

A good model reads that blank as unknown. A poor model reads it as zero. And so a defender with good physicality and game-reading ability gets filed as weak defensively — not because he is weak, but because the system cannot see him.

Among the clubs I have spoken with, this is the error that makes them miss precisely the players they most need. Big clubs can buy anyone their data already covers. Small clubs are forced to buy the people their data has never covered. Which means that for most of the time, the competitive advantage of a mid-table side lies in reading the blanks better than everyone else. Not in having more data.

This is where the first thing refereeing taught me about data comes in, after my mistake in Russia.

The mistake in Russia that year did not teach me how to referee correctly — it taught me how to live with the sound of my own whistle. On 14 June 2026, in the opening match between Russia and Saudi Arabia at Luzhniki, I was twenty-six, working my first event of that scale. In the first half I mispronounced the name of the Russian substitute striker three times. Nobody in the stands knew I was wrong. I did.

That night I went back to the hotel and started a notebook. In it I recorded everything I did not know: names I had never heard, rule situations I had never encountered, moments I had looked at without actually seeing. That notebook followed me through my career. It does not contain answers. It contains a list of unresolved questions.

The author of the nine-dimension report did the same thing, just at a larger scale. He built a notebook with nine sections, and was honest enough to write in each one that the section was empty.

The problem is that such a notebook can still be accepted. And if it is accepted, it enters the process.

What happens when an empty document is accepted

Through March and April 2026, when global football stopped, I worked at the editorial desk. I had an unprecedented amount of time and an unprecedented shortage of matches. This is when I began analysing Liverpool against Atletico Madrid at Anfield on 11 March 2026 — the night Liverpool lost 2-3 and went out 2-4 on aggregate after leading the tie.

At the time I thought it was the last match of an old world, and I was right. What I did not anticipate was the feeling of watching football with no crowd. When the Bundesliga returned on 16 May 2026, matches were played in empty stadiums. And in those matches I heard sounds I had never heard in ten years of refereeing.

When the stands are empty, I hear the ball strike the boot clearly — something ten years of refereeing never let me hear. I hear a defender breathing as he is pulled out of position. I hear a coach issuing instructions from the technical area, clearly enough to count how many times he repeats the same word. Technically, those matches were no different from any other. But they revealed more, because one layer of noise had been removed.

I thought about this while reading the nine-dimension report. If an empty document slips past the format check and enters the process, it does not cause a single error. It causes two. The first is a data error: decisions downstream get built on information that does not exist. The second, and more serious, is a cognitive error: from then on, everyone in that chain believes a correctly formatted document is a substantively correct one.

In football we handle this cognitive error with a very simple mechanism: every decision must be recorded with a responsible person attached. A match report does not say "there was a penalty". It says who awarded it, in which minute, for what offence, and in VAR cases, why the on-field decision stood or was overturned.

That report had no such section. It had room for conclusions, risks, and recommendations. It did not have a single line stating: the input source did not exist, and this is the person who confirmed it.

The counter-argument: emptiness is not always failure

So far the story seems to lean one way: empty data is bad, a broken pipeline is bad, a blank document is bad. I want to push it the other way, because I believe our first instinct — to demand fuller, bigger, more granular data — is the dangerous response.

First, a large part of the value of football analysis lies in detecting what has not yet been recorded. If everything already had data attached, analysis would be nothing more than rearranging numbers anyone can access. The edge is not there. The edge is in looking at a blank and understanding that the blank is not nothing, but a question nobody has asked yet.

Throughout my career I have chosen to write about players the media calls invisible talents. Not because I enjoy going against the crowd, but because I believe most of a player's real value lies in actions our current measurement systems cannot capture. If you only write about what already has numbers, you are writing about football's past, not its future.

Second, there is enormous social pressure on decision-makers to decide. In football, a referee who calls wrongly but decisively is usually treated more kindly than one who hesitates. Crowds do not forgive hesitation. The same applies to analysts. A nine-cell table all marked insufficient information irritates readers more than a nine-cell table packed with conclusions — even when all nine conclusions are wrong.

The media understands this. That is why the underdog-upset story always generates traffic. But only by following a weak team for a full year do you understand the price of a miracle. Miracles are not free. They are paid for with a season nobody writes about, with injuries nobody records, with nights players spend twelve hours on a bus to play a match in front of no one.

Third, and this is the point I want to stress most: if we force every data cell to be filled, we do not create knowledge. We create numbers inferred from other numbers. Each cycle of inference blurs the boundary between the known and the unknown, until nobody can tell them apart. And when that boundary disappears, the most complete-looking report is the one furthest from the truth.

The VAR protocol teaches the opposite of our instinct. It says that staying silent and letting the original decision stand is a valid action. That "I did not see it" is a complete answer, as long as it is honest.

A forward-looking thought

What will distinguish a good football analyst over the next ten years is not the speed of their conclusions, nor the volume of data they can access. It is the precision with which they mark the boundary of what they know.

In data processes, this means a mandatory gate: without at least one verified information point, no verdict. In football, that has been law for more than a century. Without conclusive evidence, the on-field decision stands.

And in writing, it means a question I believe all of us will have to answer more and more often: if every line in an analytical report had to carry a source and a confirming person, how many lines would survive?

I left that nine-dimension report on my desk for one more day before passing it on. Nine cells, all empty. And I wondered: if every referee in the world were taught to stop before a blank like that, would we still be arguing about millimetre lines every weekend.