The Empty Dataset: When Football Writing Invents Its Own Foundation

On the evening of 21 August 2026 I sat in row seven of the press area at the Etihad Stadium. Manchester was drizzling, the kind of rain that makes you pull your collar up. On the pitch, Pep Guardiola's Manchester City were running, for the first time that season, the shape England would later call the diamond: a back three, two holding midfielders beneath the ball, four attackers spread across the width, and at the very top a player who was not really a centre-forward. The match finished 1-1. Wayne Rooney opened the scoring for Everton on 35 minutes, Raheem Sterling equalised on 82.

That night I stayed in the newsroom in central Manchester until nearly two in the morning, opened the personal spreadsheet I have kept since 2026, and logged 4,312 Twitter comments plus three Manchester fan forums within six hours of the final whistle. Most of them argued against abandoning a traditional striker. But what kept me there longer was not the noise. I read seven pieces published that night about the same match, and three of them cited metrics I could not trace in any data source I had access to. They were not technically false. They were placed next to each other in a way that made them look meaningful, while in reality they were filling a gap.

I keep the rhythm for a team by writing down even the things nobody wants to read. That night, the thing nobody wanted to read was this: the foundation that a great deal of modern football analysis stands on can be empty, and that emptiness is rarely admitted.

Ten years of datafication and the price of a template

From the 2026-15 season onwards, English football went through a quiet but total shift. Event-data providers such as Opta, and later StatsBomb, Wyscout and Twenty3, expanded their metric sets from a few dozen to several hundred variables per match. Positional tracking data began appearing in some competitions. The Premier League opened APIs to media and betting partners. By 2026-20, a reporter in the press box could open a laptop and reach a library of numbers that the previous generation needed three days to assemble by hand.

This shift has two faces. The first is real: judging a match by process rather than scoreline alone saved writers from naive conclusions. A team can win 3-0 and play badly. A team can lose 0-1 and function correctly. Before 2026, saying that required describing many specific passages of play, and descriptions can always be contradicted by someone else's eyes. After 2026, a writer could quote expected goals and end the argument.

The second face is the one I care about. When a tool becomes easy to use, it becomes a template. And when a template exists, people tend to fill it even when there is nothing to fill. My trade taught me that a football article does not need data to exist. But an article filed under "analysis" does. That pressure is real, and it produces a very specific behaviour: the writer does not invent the match. The writer invents the foundation.

I have seen this mechanism many times in its crudest form. Across roughly seven years working with newsrooms in England, I watched at least a dozen occasions when a data-extraction system returned an empty result: a blank table, no information, no entities, only a domain label assigned automatically by a classifier. Technically, the cause was usually a paywall or bot wall, a JavaScript-rendered page the scraper could not execute, an encoding failure, or an HTML selector that missed the article container. The cause was almost always in the pipeline, not in the article.

And here is the part worth noting: on nearly every one of those occasions, the workflow downstream did not stop. The blank table travelled. It reached an editor, then a writer, and the writer knew the deadline was six in the morning. What emerged was not a wrong article. What emerged was an article that was correct in form, complete in structure and entirely empty in substance, with every data-shaped hole filled by language that sounded like expertise.

That is the first mechanism. It explains why an empty foundation is more dangerous than a wrong number.

Metrics without provenance

Take expected goals, xG, as the cleanest example.

xG is the probability that a given shot becomes a goal, calculated from a model based on shot location, angle, shot type, the move leading to the shot, and the number of players between ball and goal. There is no single xG. Opta has Opta's model. StatsBomb has StatsBomb's model. Understat has Understat's model, trained on a different dataset with different weights on different variables. The same shot can produce three values across three providers that differ by three or four tenths of a goal.

In 2026-23, Erling Haaland scored 36 Premier League goals for Manchester City. His total xG that season was lower, and the gap varies by roughly four to nearly seven goals depending on which provider you choose. Around the same period, the xG figures for Bukayo Saka at Arsenal and Jude Bellingham at Borussia Dortmund also differed between sources by two or three tenths of a goal per season, a small gap that is nonetheless enough to flip the conclusion of an article about finishing efficiency.

That means a sentence like "the home side had 2.7 xG but scored only once" carries no information unless it names the model. In practice, most such sentences do not. I spent nearly a week during the 2026-22 season cross-referencing xG for the same twelve matches across four providers. The average spread between the highest and lowest source per match was around 0.4 goals. For a match that finishes 1-0, that spread is enough to reverse the entire conclusion of the analysis.

PPDA, the number of passes an opponent is allowed before each defensive action, has a similar problem in a different form. A low PPDA means intense pressing. But PPDA is dominated by game state: a side two goals up will press less, and its figure rises automatically with no relation to pressing quality. It is also dominated by the opponent: a long-ball team makes your PPDA look artificially low because they do not pass much in the measured zone. And it is dominated by how the provider defines a "defensive action".

Three distortions live inside one metric. I once saw PPDA used to conclude that a team had "lost its press" after the winter break, when the only thing that had changed was a fixture list containing three long-ball opponents. Kevin De Bruyne shows the inverse on the attacking side: in his peak seasons his open-play chance creation was far above the rest of the league, but that figure only means something alongside Manchester City controlling more of the final third than anyone else. Drop it into a mid-table side and it becomes an empty compliment.

Pass completion is the most abused and, in isolation, the most meaningless metric of all. A centre-back completing 92 per cent of sideways passes is not better than an attacking midfielder completing 78 per cent with six line-breaking balls. Yet in English coverage pass completion still appears as a standalone quality marker, because it is easy to pull and easy to understand for readers who did not watch the match.

The point I want readers to keep: when a metric has no provenance, no definition and no tactical context, what remains is an integer placed in the right slot of a sentence. It performs its role perfectly, because it makes the reader believe verification happened somewhere.

The transfer market: people buy hope and sell memory

If match data can be filled with metrics of unknown provenance, the transfer market is filled with something even harder to verify: structured rumour. In the summer of 2026 I followed a deal reported by four English outlets within two weeks at four different fees, spread over nearly eighteen million pounds. None of them was wrong. Each figure matched a different structure of the same contract: fixed fee, fixed fee plus add-ons, maximum achievable add-ons, and the whole package including wages.

This ambiguity is not an accident. It is a product of how deals are designed.

The most troubling contract type in the current cycle is the loan with an obligation to buy. Formally, it lets a small club acquire a player without paying a fee immediately. In substance, it transfers the entire risk to the receiving side and concentrates that risk at a single future moment the small club does not control, because the obligation triggers on conditions set by the lending club. Those conditions are usually appearances, survival in the division, or league position.

The result is a cash-flow structure a small club only half knows. They know the money is coming. They do not know exactly when, and in many cases they do not know how much financial room they will have when it arrives, because the Premier League's profit and sustainability rules operate on a three-year cycle and recognise costs through amortisation. A thirty-million-pound purchase signed in June is spread across the contract years, but it still occupies space inside the permitted loss threshold.

In England this mechanism has produced concrete consequences. On 17 November 2026, Everton were deducted ten points for breaching profit and sustainability rules; on 26 February 2026 the deduction was reduced to six on appeal; on 8 April 2026 the club received a further two-point deduction in a separate case. On 18 March 2026, Nottingham Forest were deducted four points. Those sanctions are the product of financial decisions taken two to three years earlier, including sums triggered by obligations the club did not fully control.

A small club cannot simultaneously serve as a finishing school for bigger clubs, retain autonomy over squad structure, and comply with a loss threshold designed for clubs earning ten times its revenue. Those three things do not coexist. When forced to choose, a club usually chooses whatever keeps it in the division for the next twelve months, and the bill arrives later.

The transfer market taught me this: people buy hope and sell memory. And most transfer rumours exist to price a negotiation, not to inform supporters.

Nine months for a ligament, and the rest of a career

I wrote about injuries for fifteen years before I started reading sports-medicine literature properly. There is one thing I want to state clearly for Vietnamese football readers, because I see it misunderstood fairly often.

When a player ruptures an anterior cruciate ligament, the figure most often quoted is nine months. That figure is correct as a statistical average and wrong as a career plan. It is the window for returning to full training, usually falling in the eighth or ninth month after surgery. It has never been the window for returning to peak performance.

Rodri of Manchester City ruptured the ACL in his right knee against Arsenal on 22 September 2026, and returned in the following season. The early months after a return are what I call the second phase, and it is the most ignored phase in all reporting. A returning player can run at full speed. He cannot turn quickly enough. He cannot decide to commit to a collision within a fraction of a second. And above all, he cannot stop thinking about his knee.

Psychological fear is harder to repair than the body. For central midfielders and defenders operating at high intensity, the first thing lost is not straight-line speed but reaction time before a potential collision. That hesitation is too brief to appear on video, yet it is enough to tilt a duel and enough for a coach to feel.

So the thing I always record when tracking a player returning from injury is not minutes played. I record how often he engages in a fifty-fifty duel in the first half. If that rate rises week by week, the second phase is going well. If it is flat for six matchweeks, there is a problem no medical department will mention in a press conference. Rushing that phase to serve one big match is the fastest way to ruin the rest of a career.

Cup shocks and the trap of the word miracle

I do not use the word miracle. Shocks in domestic cup competitions are almost always the product of two predictable variables: rotation by the strong side and high pressing by the weak side.

A strong side often plays seven to nine matches in thirty days around the fourth round. Rotating four to six positions is compulsory, and it breaks positional relationships built over weeks. A back four containing three substitutes cannot hold its line or its distances between units. That is condition one.

Condition two is the weaker side's motivation. For them, this is the biggest match of the season. Lower-division English sides typically enter such games with a PPDA about ten units lower than their own average, meaning markedly more intense pressing, because they calculate that a patched-together midfield will lose the ball in dangerous areas. When both conditions appear together, the probability of an upset compounds.

On 4 February 2026, Middlesbrough knocked Manchester United out of the FA Cup fourth round at Old Trafford after a 1-1 draw and a penalty shootout. Nothing miraculous happened that night. It was a match in which a lower-division side prepared for exactly one scenario while a top side had to field a lineup chosen by the fixture list.

I write this without any intention of diminishing Middlesbrough supporters' feelings. I write it because the word miracle is a lazy explanation, and it stops readers from seeing what actually happened on the pitch. What happens on the pitch is always more specific, more verifiable and more interesting.

The transmission chain and the cost of a one-day delay

A wrong metric does not stop at one article. It travels a chain. From the article into a stats site's aggregate table. From there onto a television programme. From there into a social media post. From there into a fanbase's expectations. And finally into pressure on a manager. That chain takes eighteen to thirty-six hours, far less than the time needed to verify anything.

Upstream, academies and recruitment departments moved to data years ago, so they are usually less affected. Downstream, the media market and derivative products are hit hardest, because speed is their entire value. My trade sits in the middle, and if I want to keep the rhythm, I have to choose slow.

Slow has a price. During the 2026-24 season I lost at least four breaking stories by waiting for a second source. Each time, a younger colleague published three to five hours ahead of me. But of those four, two initial stories turned out to be entirely wrong, and mine was the only one still standing once things settled. I do not tell this to praise myself. I tell it because it is a simple calculation many newsrooms no longer make: the value of one correct article over six days can exceed the value of one article available for three hours.

The counter-intuitive view

The counter-intuitive point is this: supporters do not really buy data. They buy memory, and memory does not need accuracy to have value.

I discovered this by accident. After England beat Colombia 4-3 on penalties on 3 July 2026 at the Otkritie Arena in Moscow, I sat among roughly two thousand England fans and received forty-seven videos of crying and laughing from viewing points across Manchester within thirty minutes of the final whistle. Eric Dier took the decisive penalty. Carlos Bacca and Mateus Uribe missed for Colombia. In those forty-seven videos, not one mentioned a single metric. Nobody mentioned xG, nobody mentioned possession. People mentioned the name of the street they stood on, the name of the pub, and the name of the person next to them.

That is why an article empty of substance can still succeed. It supplies a framework for an emotion that already existed. What readers need is not a correct conclusion but a name and a route to attach their feelings to.

But the same blind spot exists on the opposite side, and few people discuss it. The analytics community, with entirely good intentions, routinely commits a technical error with philosophical consequences: treating missing data as zero.

In a dataset, an empty cell and a cell containing zero look identical once charted. An empty cell means we do not know. A zero means we know there was nothing. The two are entirely different, and treating them as the same produces unfalsifiable conclusions: a player attempted no passes into the box either because he did not try or because the collection system missed him. If both possibilities are encoded as the same value, everything downstream loses its meaning.

I have seen this at team scale. One data source omitted an entire match for one club. In the season aggregate, the match disappeared, and the club's averages were skewed for twenty matchweeks afterwards. Nobody in the newsroom noticed. I noticed because I have a habit of counting matches.

The third blind spot sits with the most engaged readers themselves. The people who follow metrics most closely are often the ones who verify provenance least, because they have learned to trust numbers as something objective. That objectivity is a belief, not a property. A metric is only objective within the model that produced it and the data that trained that model.

So the correct response to an empty dataset is neither to fill it nor to ignore it. The correct response is to state that it is empty, and to treat that as a newsworthy event.

Internal signals to track

The pandemic took away the crowds, but it gave me back the truest sound of all: silence. An empty dataset works the same way. It is not a defect to hide. It is a signal to read.

Over the next twelve months I will track four specific things. First, whether the major data providers begin publishing model version numbers alongside every metric they output, since that is the minimum condition for a metric to become verifiable. Second, whether Premier League clubs publish phased injury-recovery roadmaps rather than a single return date. Third, how the independent commission handles the next profit and sustainability cases, because each sanction reshapes how mid-tier clubs structure loan deals. Fourth, and most important for my trade, whether the next match report I read cites an observable event, or only a metric of unknown provenance.

When the stadium echoes with absence, I understand that football is a conversation between people. That conversation does not need more numbers. It needs more people willing to sit until two in the morning counting how much they actually know.