A Football Analysis System Returns Empty: A Lesson in Data Integrity

Three in the morning in Shenzhen, the automated analysis pipeline finishes its run. I open the results table. Nine categories. Article title: N/A. Source: N/A. Type: unclassified. List of information points: empty. The system still returns a complete analytical framework — nine full sections, tables and subheadings included — but inside there is not a single line of real data. It does not raise an error. It stays silent and presents emptiness as if it were a conclusion.

I have seen this kind of failure before, though not on a computer. I saw it on grass, on the night of 30 September 2026, in London.

In the 34th minute of Tottenham against Liverpool, Luis Díaz put the ball in the net. The assistant referee raised the flag for offside. The goal was disallowed. In the stands, nobody knew that at that moment the VAR team believed the decision had been a goal. Darren England, the VAR, believed the on-field decision was a valid goal, so when the check was done he said "check complete" — meaning the goal stood. He did not draw the offside line. Nobody drew it. The goal should have stood, but it was cancelled, and the match ended 2-1 to Tottenham after Joel Matip's own goal in the 90th+6th minute. Three days later, the English referees' body (PGMOL) released the audio and called it a significant human error.

What is striking: the technology did not fail. The camera angles were all there. The software was still running. What broke was an assumption — the VAR team assumed the wrong input state, and built its entire conclusion on top of that. That is exactly what I saw in the empty results table at three in the morning: a system proceeding on an input state it had never verified.

Since I became a rules commentator, one question has obsessed me: when the data does not exist, what should the system do? Modern football has become a chain of pipelines. Expected goals (xG) models calculate chance quality. Semi-automated offside technology (SAOT) renders players in three dimensions: 12 cameras under the roof, 29 data points on each player, sampled 50 times per second, plus an inertial sensor in the ball sending data 500 times per second. The system was used at the 2026 World Cup in Qatar and then brought into the Premier League. Transfer-rumour platforms gather thousands of sources a day. Every pipeline is only as good as its input link.

There are two kinds of failure, and they are not equally dangerous. The first: empty input, the system goes silent, nobody notices because there is nothing to read. The second: empty input, the system still emits a conclusion — and dresses that conclusion in the robe of professional authority. The second is far more dangerous, because it does not look like a failure. It looks like an analysis. A nine-part framework full of tables and terminology will make readers believe a serious investigation took place, when in truth it is a hollow shell, carefully decorated.

The correct principle here is null handling. When the list of information points is empty, the only honest output is "insufficient information to assess". Not a conclusion, not a guess wearing a professional label, but a decisive refusal. It sounds simple, but in this industry it is rare. Because systems are built to always answer, and readers are trained to always expect an answer.

I learned this lesson through a specific wound. In 2026, when I was 35 and working as a senior expert for a new sports platform, a colleague asked me to analyse a disallowed goal by the striker Alan in an AFC Champions League semi-final. The referee gave offside. I rewatched the footage 47 times, measured the run angle of the opposing defender, and concluded that Law 11 of the laws of the game had been applied incorrectly. My 6,000-word article with 14 illustrated frames was shared more than 200,000 times. But what I remember most is not that number — it is the gap between two seemingly identical questions: "was the attacking player ahead of the last defender" and "did the attacking player gain an advantage from that position". Law 11 does not fully answer the first question; it demands the second. A system that reads the law correctly is one that knows which question it is answering, and knows when it does not yet have enough data to answer.

Based on my experience watching matches, I have realised that every major refereeing error has the same structure: an input assumption that was never verified. Tottenham against Liverpool in 2026 was that. And the empty results table at three in the morning was that too, except it had no crowd to react.

What worries me about that hollow framework is the process risk. A pipeline that lets an empty input through the first check will let it through the second, and by the third it is ready for publication. The greatest risk is not in football; it is in the joints between processing stages.

In the transfer window, this is even more true. Every day brings hundreds of rumours, and most rumour-aggregation systems have no concept of a null value. They do not say "unverified"; they assign each rumour a confidence level that looks scientific. A confidence level generated from an empty input is the most dangerous thing of all, because it turns fabrication into a number.

Here a counter-intuitive view appears. Our natural reflex is to demand that the system always give an answer. When VAR stays silent too long, the stands boo. When an algorithm gives no prediction, people call it useless. But it is precisely that demand that is the error. A referee under crowd pressure will blow for a foul rather than admit "insufficient information". A system forced to always answer will invent an answer. Empty is not failure. Empty is information. It tells us the input link has broken, and that is what we need to know before trusting any conclusion downstream.

I still hold a strange belief: empty stands are a referee's finest laboratory. When the roar is gone, when the crowd pressure is gone, the match reveals its purest operating rules. An analysis pipeline with an empty input is that kind of laboratory too. It tells us nothing about the match, but it tells us about the system itself: whether it is honest or whether it is acting. And the penalty law, as I often say, is not for the taker but for the one who reads the taker. Likewise, an analysis system does not belong only to the person who builds it, but to the person who reads it. The reader must be able to read both the text and the whitespace.

My mistake on live television is the foundation for a new system. In 2026, at the World Cup in Sochi, I explained Pepe's handball in the Spain-Portugal match and declared "handball is a deliberate offence" — a ruling that was wrong on the law. Social media turned on me at once. That night I downloaded the VAR data of all 12 opening matches, logged every decision into a 2,000-row spreadsheet, and spent a month cross-checking against FIFA's original laws. I was forced to abandon the style of firm assertion and switch to a multi-track structure: if clause X applies, the conclusion is A; if Y applies, the conclusion is B. And if there is not enough data, the conclusion is: cannot yet conclude.

That is the proposal I want to leave behind. Call it the null gate. Every football analysis system — from semi-automated offside software to the data models inside the dressing room — should be designed to have the right, and the duty, to say "insufficient information to assess". Not as an apology, but as a valid output on equal footing with every other conclusion. Because a system is only trustworthy when it knows how to be silent at the right moment. And a spectator only truly understands football when they understand that sometimes the most honest answer is a blank space.