The Denominator of 24: How Football Builds Monuments on Data Too Small to Stand

The Denominator of 24: How Football Builds Monuments on Data Too Small to Stand

02:40 in the morning, Nagoya

Four days before the summer transfer window closed, my third monitor was running the contract tracker when a Spanish-language line scrolled past: Robo con violencia en el Metro cae 50% durante 2026. Robbery with violence on the Mexico City subway down by half in 2026.

I almost scrolled past it. I stopped, not because of Mexico. I stopped because of the frame. This was the frame I had read eleven times that week, with different labels: a club press release about muscle injuries, a league presentation about referee decision accuracy, a line from an agent saying the parties were close to agreement. The same verb, the same denominator hidden away.

What made me sit up was the denominator. Twelve investigation files between 1 January and 7 September 2026. Twenty-four files in the same period of 2026. Half of 24 is 12. Half of a very small thing is still a very small thing. And I make my living reading percentages built exactly this way.

The frame everyone uses

The report from Mexico City described Operativo Quetzalcóatl, a security strategy deployed more than a year earlier across the Sistema de Transporte Colectivo Metro network. 5,800 Metro Police officers were committed. For four consecutive months between March and June 2026, no investigation file was opened for robbery with violence in the system.

On a first read, this is a success story. On a second read, I realised I was reading exactly the kind of document football clubs send me every week: a percentage calculated on a set so small that random variation is noisier than the actual trend.

The arithmetic is not complicated. When the denominator is 24, each file accounts for more than 4% of the entire year. When the denominator is 12, each file accounts for more than 8%. Which means one bad week in late December can turn a 50% decline into a 25% decline, or into growth, without anything changing on the streets. The report is not wrong. The report is simply standing on ground too narrow to carry the weight its headline places on it.

My point is not that the number was fabricated. My point is that the indicator being used does not measure what readers think it measures. The unit here is files opened, not incidents that occurred. Between those two things sits an operational gap: offence classification rules, the threshold for a case to be classed as violent, the willingness of victims to report, and the existence of a simpler intake process downstream. A police operation with heavier presence can reduce real incidents and simultaneously reduce filings along a different curve, if reporting behaviour shifts at the same time.

I have to be fair here, because I do not want to be read as a reflex sceptic. The report's comparison window is legitimate in timing: Operativo Quetzalcóatl began more than a year before the September 2026 cut-off, which means January to June 2026 sits almost entirely before the operation. Someone chose the right window. The problem lies elsewhere: no victimisation survey data, no ridership data, no published classification rules. Those three are the legs of any security conclusion. Without them, the report is a statement about filing behaviour presented as a statement about criminal behaviour.

And I know this feeling precisely, because it is my daily work.

Football runs the same arithmetic, every week

Based on my experience watching matches over sixteen years, I can say that the football industry operates on an ecosystem of percentages built from sets far smaller than fans believe. Take the most famous case I ever followed minute by minute.

On 1 December 2026, at Khalifa International, Japan beat Spain 2-1. The possession split was published everywhere: Spain 82.3%, Japan 17.7%. Within twelve hours, 17.7% became a label for an entire football culture: the overrun side, the side that lives on counters, the side that cannot control a game.

But possession is a percentage of what? It is a share, not a volume. Spain circulated the ball more than a thousand times in that match. Japan moved it roughly two hundred and thirty times. Each Japanese pass carried different information from each Spanish pass, because they sat inside two different match structures. A share metric compresses two different behaviours into a single cell, and then readers assign that cell the meaning of a single behaviour. It is a category error, repeated every weekend on television.

Japan's two goals came inside roughly three minutes, the 48th and the 51st. No metric in the broadcast data package measures those three minutes. The data package measures the rest of the match, and then the headline takes the rest of the match as the story.

Magic does not exist; there are only those who read the rules closely before anyone else can blink.

When the denominator is small, legends grow themselves

Let me move to what I consider the purest example of a small set generating a cultural story. For decades, English football built a psychological icon called "the penalty curse". England lost most of the major shootouts they contested between 2026 and 2026.

If each shootout is treated as a roughly balanced coin toss, the probability of losing at least 6 of the first 7 is about 6.25%. That sounds small, until you multiply it by the number of national football associations in the world. With dozens of national teams, one of them falling into such a losing streak is close to a certainty. Probability did not choose England for its national character. It simply chose England first.

The most interesting moment came in 2026, when England won a shootout. Within twenty-four hours, "the curse" vanished from the papers and was replaced by "forged mentality". The same country. Almost the same group of players. The same culture. The only difference was the outcome of a sequence of events with enormous variance.

This is where I want to place it beside the Metro report. Both are honest documents at the data level, read as narrative documents. The Metro report reads 12 files as a policy victory. Football media reads seven shootouts as a national psychological trait. Both skip the question of whether the denominator is big enough to carry the conclusion.

VAR, and what the broadcast data package does not measure

Every league publishes its referee decision accuracy rate. That rate rises year on year, and each rise becomes a headline about technology working better.

The problem is definitional. The denominator of that rate is not "number of wrong calls on the pitch" but "number of incidents the league classifies as reviewable". When the classification protocol changes by a single line, the accuracy rate changes while nothing else on the grass changes. This is the same mechanism as investigation files: the indicator measures classification behaviour, presented as an indicator of on-pitch behaviour.

In 2026, when every stadium was closed by the pandemic, I sat at home and rewatched forty-seven Kawasaki Frontale matches. With no crowd and no noise, I saw a detail that is normally obscured: during a VAR review, Kawasaki's under-25 shape ran into its defined defensive positions and held the structure for around twenty seconds while waiting for the decision, like a pre-programmed drill.

I contacted head coach Toru Oniki over Zoom. The interview ran forty-five minutes. He confirmed: "We turn waiting time into active time."

What is worth noting is the reaction. One analyst called the finding delusional. But the broadcast data package only measures VAR's duration, not the content of that period. Every VAR metric that season was about how long VAR took. No metric was about what those twenty seconds contained. And in the very season when every metric was helpless, Kawasaki won the title.

The most important person in a match does not run on the pitch; they sit quietly in the stands where nobody looks. In this case, they sat in a video analysis room, counting seconds that nobody bothered to count.

The transfer window: an unsolvable denominator problem

Now apply the Metro frame to the thing running on my third monitor at 02:40 in the morning.

Every transfer window, thousands of headlines are produced around phrases like "talks progressing", "personal terms agreed", "one step apart". What is never published is the denominator: the total number of negotiations described with exactly those phrases, and how many of them collapsed.

Nobody tracks that rate, because nobody has an incentive to track that rate. A transfer success rate can only be calculated if agents, clubs and journalists all publish the number of negotiations that died. None of the three wants to publish that number. The result is an indicator that survives forever because it cannot be wrong.

The transfer market never tells the truth; it only whispers what we are desperate to hear.

The structure is identical to the Metro report: measuring what was recorded, not what happened. And identical to every club medical bulletin, where an injury prevention programme is announced as a "50% reduction in muscle injuries" on the basis of six cases falling to three.

Here I should state plainly what sixteen years in this trade taught me. The most objective benchmark for any injury claim in football is the UEFA Elite Club Injury Study, data running continuously since the early 2000s across thousands of players and tens of thousands of match hours. That is a set large enough that random variation cannot swallow the trend. Any club release announcing a percentage without a denominator at that level is playing in a different epistemological league.

The Kubo case, and the uncomfortable truth about small samples

I wrote his name in my notebook before the stage lights came on.

In December 2026, at the EAFF E-1 Football Championship, I was twenty-three and six months into the job. Japan beat China 2-1. While the entire press room talked about Yuma Suzuki's long-range shot, I sat writing down four dribbles by a sixteen-year-old who came on in the 68th minute: two successful take-ons, one chance created. About twenty-two minutes of football.

My article was called delusional. Three years later, when he scored for Real Madrid Castilla, the same article was called a genius's foresight.

But here I have to be honest, and this is where I differ from the version of myself from eight years ago. Those twenty-two minutes proved nothing. A sample of four dribbles has a confidence interval so wide it carries almost no information. My prediction being right does not turn a small sample into a good sample. It only means I asked the question in the right place and the answer arrived later.

The real value of a small sample is not that it gives you an answer. The real value is that it gives you a question nobody else is asking. Confusing those two things is the central error of modern football analytics, and also the central error of the Metro report: taking an interesting question and packaging it as a completed conclusion.

The Salah paradox is the same mechanism in reverse. In the summer of 2026, I brought a data sheet into the newsroom: a tracking sample of 117 minutes of touches inside the box in the Champions League, with a cross completion rate of 12%. I wrote three pieces asking whether Salah was a brand manufactured by Klopp's system or a player independent of it. Egypt lost their opener, and two thousand comments called me an idiot.

But look carefully at what I did that day: I did not deny the player's ability. I pointed to the limits of the system context. That is a narrow, testable claim, and it was rejected because it collided with a monument.

When a person is cast into a statue, they begin to lose themselves on the grass.

Where I could be wrong

Now comes the part where I have to interrogate myself, because if I do not, someone else will, and they will do it less accurately.

First possibility of error: perhaps percentages are not broken at all. Perhaps they are the product. Football is an entertainment industry, and a "down 50%" headline is a feature of that industry, not a scientific claim. When I demand inferential rigour from a press release, I am asking a cinema to publish error bars on its poster. I myself sell contrarian takes for a living. The demand may be hypocrisy dressed in statistical vocabulary.

Second possibility, and more serious: perhaps the decline in Mexico City's Metro is real. The comparison window is valid. 5,800 officers is a genuine resource commitment, not an empty communiqué. I have no counterfactual data. My scepticism may be an occupational reflex rather than a finding. A contrarian journalist always runs the risk of finding scepticism where only a boring truth exists.

Third possibility: the contrarian instinct may be running ahead of the data. My personal brand is tied to provocation, and there is a real pressure to always provoke. I have seen myself do it. I have seen better people than me do it. When a commentator starts being treated as a statue, he starts reading data to protect his own pedestal.

Prejudice has the capacity of a packed stadium, but no exit for anyone inside it.

What I will be watching

I will be watching two things, and both are verifiable.

First, Mexico City Metro data for December 2026 to February 2027, the holiday season, when ridership peaks and crime composition shifts with the flow of people. If the annual total holds around a 50% decline through the first quarter of 2027, the indicator is credible and I was wrong. If a spike appears and drags the annual total below 30%, then the 2026 indicator has the shape of a trough, and its headline was a media product.

Second, and closer to home: from now until the end of the 2026-27 season, for any club announcing a percentage reduction in injuries, I will read to the footnote to find the denominator. If the denominator is under forty cases, I will log it as a hypothesis, not an achievement. If it is above forty, I will call the head of medical and ask about their classification protocol before I ask about their results.

From the grass to the LED screen, the border between the two worlds is as fragile as a horizontal touchline. And the trap is always in the same place: a denominator left in the footnote, while the headline has already flown everywhere.

The question I leave for myself, and for anyone who has read this far, is not whether football lies. It is this: when was the last time you checked the denominator standing behind a percentage you already believed?