The Empty Report: When Sports Analysis Deceives Itself With Data That Never Existed

On a July afternoon in 2026, in a second-floor café north of Umeda Station in Osaka, a 42-page scouting report was placed in front of me. The subject: a 21-year-old Brazilian striker playing in the São Paulo state second division. The cover was immaculate. A heat map of activity zones spanned page 12. Expected goals appeared on page 18. Chances created sat on page 23. Skimmed, any sporting director would want to sign immediately. Then I traced it back to the raw data. The sample was 14 matches. Of those, only 9 had complete video footage. The event-coding column — the one that should list every action with a timecode — was empty. Not a single situation was coded. Not a single coordinate was checked against footage. Those 42 pages did not analyse data. They analysed expectation, and expectation was the only thing in the document that could not be verified. Over 29 years in this trade I have lost count of how often I have seen this story. The frequency only rises, because the data infrastructure gets prettier while the discipline of cross-checking it gets worse. The transfer window is the phase in which the sports market runs on belief more than on evidence. Every rumour is an arrow fired at a wall, and people measure the quality of the shot by how many turn to look, not by whether it hit the target. Inside that machine, the scouting report becomes a currency. The thicker it is, the more charts it holds, the higher it is priced. The trouble is that a thick document does not mean a thick source. In 2026, working for a major betting exchange in Osaka, I watched this machine run from the inside. New sports platforms were racing to publish gut-feel analysis, slapping the label of data onto claims with no denominator. I published a study comparing PPDA — passes allowed per defensive action — across 18 J-League clubs. The result: Shimizu S-Pulse finished a season with 11.3 fewer actual goals than expected goals. The media called it bad luck. The table said otherwise: it was a defensive structure with a hole in central midfield, where opponents kept generating quality chances that the goalkeeper masked with saves beyond expectation. My prediction then: Shimizu would finish 14th. The media predicted 8th. They finished 14th. Numbers never lie; the liars are those who choose how to read them. But there is a deeper layer it took me years to see clearly: for numbers to lie, there must first be numbers. The most dangerous thing in sports analysis today is not misreading data. It is data that never existed being presented as if it did. So I put that 42-page report back on the transparency scale and peeled it layer by layer. Layer one was the denominator. The report claimed 14 matches analysed. Only 9 had footage. The other five were reconstructed from the notes of a local scout who never published his coding method. The real denominator was not 14. It was 9, plus 5 in hearsay form. In statistics, hearsay carries a weight close to zero. Layer two was the unit of measure. The report placed xG on the same axis as actual goals and concluded the striker had outperformed expectation. But xG in the São Paulo second division was computed by a model trained on European data. Apply a European model to a league with different defensive quality, pitches and fixture density and you are no longer measuring a player. You are measuring the distance between two football cultures. That error was noted on no page. Layer three was double counting. The same action was coded three times under three labels: once as a quality chance, once as an unlocking pass, once as a breakthrough action. Three labels, one action. Combined, the figure inflated by roughly 30 percent. This is the most common technical error in commercialised scouting reports, and it always happens to favour the subject being sold. Layer four was small sample size. A 21-year-old with 9 fully filmed matches can swing by tens of percent between two draws of the sample. Any conclusion about his trend sits inside the noise. In other words, most of the report was describing noise and calling it signal. Layer five was sourcing. No match had a date, opponent, scoreline or round context. A report without dates is a report that cannot be verified. And what cannot be verified has no professional value, however many pages it fills. Those five layers together produce what I call a data loan. The writer borrows expectation from the future and pays interest in the reader's silence. During the transfer window the loan is rolled over continuously: a club buys a player on the strength of a report, the media reports the deal on the strength of the transfer, and a fresh report is written on the strength of the media. The loop closes without ever touching real grass. What people call transfer data is often just the surface paint of a deeper order, where an agent's incentives, a club's PR needs and a newsroom's revenue pressure sit stacked on top of one another. When everyone looks one way, I start examining the empty space behind their backs. In the case of the 42-page report, that empty space was the blank event-coding column. I remember June 2026, invited as a data commentator for the trial version on DAZN Japan during Japan versus Colombia in World Cup qualifying. In the first half I mispronounced the name of midfielder Hotaru Yamaguchi three times. Viewers noticed at once. What kept me awake was not those three mispronunciations. It was the goal conceded in the 39th minute. Tracking data showed Japan's defensive line stretched to an average of 42 metres, breaking the pressing structure the coach had built through the group stage. Mispronouncing a name is not the error; the failure is not seeing the outline of a system. I spent the following month reviewing every group-stage recording, not to fix my pronunciation but to learn to see that 42-metre gap faster. That experience taught me a rule I apply to every report that passes through my hands: when the source data does not exist, the only correct answer is to say that it does not exist. You must not fill the gap with decorative guesswork. You must not turn the absence of evidence into evidence of the presence of talent. This is where I want to pause a little longer, because it runs against most of the market's intuition. Among analysts there is a professional reflex that makes people fear blank space. Faced with a spreadsheet with empty cells, the first instinct is to fill it — with interpolation, with experience, with feeling. That instinct was nurtured from the moment content-production speed was paid better than content accuracy. The result is a generation of reports with no room for the words I do not know. In statistics, however, blank space has its own value. It is information. An empty column in an event-coding sheet tells me where the collection process failed, not that the player had no actions. Those two readings lead to entirely different transfer decisions, and only one of them is right. I also have to speak plainly about a symmetrical temptation. Once you are used to tracing empty reports, you drift toward the illusion that every public number hides a deeper order. That is another trap. Some phenomena are simply random, and the surface explanation is enough. Occam's razor must still be applied before declaring the discovery of an underlying system. If a player performs poorly because of a muscle injury, do not rush to build a three-layer psych-tactical model to explain it. Recovery is never a miracle; it is merely what you saw in the data three months earlier. And sometimes, three months earlier, the data really was just a muscle injury. The balance lies here: disciplined about blank space, but not paranoid about it. Every odds move is a pulse; I can only hear it with my ear on the data ground. But I also have to accept that some pulses carry no message beyond themselves. So what does a filter tight enough for the transfer window look like? It begins with four mandatory questions for any player file. First: how many matches have verifiable footage, and how many are cited in total? The wider the gap, the lower the reliability. Second: on what data was the metric model trained, and does the player's league sit inside that base? Third: are actions double-counted under multiple labels? Fourth: does every match carry a date, opponent and scoreline? These four questions require no high technology. They require something harder: the willingness to say I do not have enough data to conclude. In a market that pays for decisiveness, that sentence is a countercultural act. I track the matches of the young South American players pushed to Europe each transfer window, and the pattern repeats almost by rule. Players with clean data files — full denominators, clear dates, transparent video cross-checks — tend to start at lower transfer fees but succeed more often at the destination league. Conversely, the group packaged in thick reports with thin sources tends to generate high fees and low adaptation rates. I have not yet published a full quantitative study of this correlation, so I state it clearly here: this is an unverified observation, not a conclusion. But it is enough to have changed how I track. Instead of reading the conclusion first, I start with the appendix. If a report has no data appendix, I read its conclusion with a weight of zero. If the appendix carries dates and timecodes, only then does the conclusion earn the right to be argued with. An era does not begin with technology; it begins with a question sharp enough to cut through the well-worn path. The sharp question for the summer 2026 transfer window is simple: strip every chart out of the report and does what remains stand up against the footage? If the answer is no, then what is being sold is not a player. It is a presentation. And as a man sitting on the other side of the table, what I practise every day is not reading the presentation faster, but recognising it faster.