Nine Chapters, Four Tables, Zero Data: The Crack in Football's Analysis Pipeline

At 2:47 in the morning, on a night in March, in a fourth-floor flat in the 11th arrondissement of Paris, I opened a PDF sent by a football data analysis firm along with a consultancy contract. The file was 1.8 megabytes. Nine chapters. Four data tables. A seven-row risk matrix. An eight-entry glossary, fully defining xG, PPDA and the “panic premium”. At the end, a disclaimer, correctly worded.

I read the whole thing in nineteen minutes. Not one club name. Not one player name. Not one match date, one league table, one transfer fee, one official seal.

Chapter one, headed “Tactical & Technical Analysis”, recorded its conclusion as: N/A – insufficient information. Chapter two, club finance: N/A. Chapter four, league landscape: N/A. Chapter eight, media narrative and expectation: N/A. Nine chapters, more than two hundred cells like that.

Then I reached chapter seven. That was when I sat up straight.

The risk matrix in chapter seven has seven rows: sporting, financial, personnel, rules, public opinion, systemic, and the last one. The first six are blank. The last one reads: Process — analysis produced on an empty Stage-1 input. Level: High. Likelihood: Certain, already occurred. Impact: High. Mitigation: halt distribution, re-run Stage-1 extraction.

That is the only sentence in the entire 1.8 megabytes carrying a real judgement, and the judgement is about the document that contains it. The only risk that can be evidenced inside an analysis report is the risk of the machine that produced it. Everything else — tactics, transfer fees, dressing-room pressure, the possibility of breaching financial fair play — can be invented, and once invented well enough, no reader can tell the difference.

My trade changed shape at some point without anyone announcing it. In 2026, when I started writing for Báo Bóng đá, a post-match analysis was a notebook, a pen, and forty-five minutes sitting in the press room listening to a coach talk. To say which side pressed higher, I had to count. To say which player ran more, I had to ask the organiser for the stats sheet. No numbers to count meant no article.

Now there is a pipeline. An extraction stage reads the source article, pulls out information points, tags entities, grades the source. An analysis stage takes that output, pushes it through nine frames, and every frame is required to carry a conclusion, a comparison, evidence and a confidence level. Then a publishing stage. The pipeline is fast, it is consistent, and it has one property that anyone who has run a similar system knows well: there is no gate at the exit. It was never designed to say “I don't know”. It was designed to always have an answer.

In the V.League, which I still follow from a distance every season, that pressure arrives earlier and from a different direction. Most clubs have no data department; they have an assistant coach who knows spreadsheets and a laptop. When an article needs numbers, people take them from wherever numbers exist: a foreign aggregator site, a social media group, a stats table with no visible update date. I have no data to prove that ratio, so I record it as a professional observation, not a conclusion. But the gap between a young football nation and an old one is not about which has more numbers. It is about which has someone checking the numbers before they go to print.

I reconstructed the timeline of that file the way I reconstruct a sponsorship dossier. Input first: the extraction stage returned an empty list of information points, a title of N/A, a source of N/A, an article type of “Unclassified”. Then the analysis stage. It did not stop. It ran on, and did exactly what it was programmed to do — fill nine frames.

Read closely, the file is honest in the way a machine can be: every chapter ends with a line reading “Minimum input required to execute”. Chapter one needs team names, a coach, the paper formation and the in-game formation, two or three process metrics. Chapter two needs a club name, a fee, a contract structure, the financial fair play context. Chapter five needs the alleged breach and the governing body with jurisdiction. The system knows precisely what it is missing. It simply has no authority to say so in the first line.

This is where the problem sits, after three layers of cross-checking. Hand that file to an inexperienced editor, or to a language model, or to me on a tired day, and what comes out next is an article. Not an obvious fabrication — a fabrication that reads exactly like the real thing. It will have “likely”, “signs of”, “worth monitoring”. It will have xG. It will have a manager under pressure. It will have a contract running down. And no reader, however demanding, will be able to tell.

I have worked the other direction, and I know what it costs. In 2026, when the desk asked me to review Olympique Lyonnais' financial report, I spent eleven weeks on a single conclusion. The shirt sponsorship with the travel company Mammoth Travel, worth 12 million euros, double the Ligue 1 average. I cross-checked three independent layers: the financial report; the company register, showing 5,000 euros of share capital and exactly three employees; and the payment flow, arriving from an investment fund in the Cayman Islands. Three layers, three sources, none of them aware of the others.

Eleven weeks for one sentence. That machine produced nine chapters in forty seconds. I don't listen to apologies. I read bank statements.

A year later, at the 2026 World Cup in Russia, I worked another way: building a data timeline. The striker Gunnar Helgason, 27, improved his burst speed from 8.1 to 9.4 metres per second in four months. Physiologically, that slope does not exist. I checked FIFA's public testing records: three results from the first half of the year had vanished. The article was held back because the desk feared pressure over broadcast rights. I archived the full dataset and sent it to UEFA's investigators. Two Iceland players were banned for eighteen months.

A fitness file does not tell you about winning; it tells you the price people were willing to pay to win. Nearly fifty years of watching matches has taught me one reflex: an anomalous metric only means something when another metric stands beside it to compare. 8.1 alone is meaningless. 8.1 next to 9.4 within four months is a witness. Numbers never lie; only the people reading them lie to themselves. That nine-chapter file had no numbers to lie with. It had only the form of numbers — chapter titles, table cells, rating scales, stars. And the form of numbers lies very well.

There is a counter-reading, and I want to give it room because it is not unreasonable.

In an industry where “no story” counts as personal failure, a system willing to return N/A across all nine chapters is a rare form of integrity. It invented no manager. It pinned no doping case on anyone. It even wrote, in its own words, that the absence of a stated compliance issue in an article is not evidence of compliance. A sentence like that, written down at all, beats most of what I read on sports sites each week.

The cost sits elsewhere. The system stopped at the ethical line but still stepped over the aesthetic one. It published anyway. It kept chapter headings, kept the seven-row matrix, kept a section called “Highlights & Opportunity Identification” with two bullets and a certainty rating. A skimming reader will catch the word High, the word Certain, a row of empty stars — and the brain files that as a document with weight. The form of truth was preserved. The substance of truth had been hollowed out long before.

That is the blind spot of the entire modern football analysis business, and it belongs to no one in particular. We taught readers to read with their eyes instead of reading with sources. An article with tables is more credible than one without, whatever the tables are made of. An article containing the letters xG is more credible than one containing the words “I watched”. That is a very bad trade, and we signed it long ago.

In football, the most expensive thing is not the player; it is the silence of the witness. That file was a silent witness in the literal sense: it was there, it had the pages, and it said nothing.

I sent the file nowhere. I wrote in my notebook: input empty, extraction stage failed, re-run. Then I sat for another forty minutes thinking about something else — if a system can write nine chapters about nothing and still read as real, what should readers be given?

My answer is narrow: the right to see the input. One line at the foot of every analysis piece, stating how many information points it was built from, from which source, on what date. On the Lyon dossier, that line would read: three document layers, eleven weeks. On the nine-chapter file, it would read: nothing.

A line like that does not make the writing better. It only makes the trade slightly harder to fake. And for an industry that has learned to speak beautifully about things nobody has checked, slightly harder is all that is left to ask for.