Nine Empty Sections: The Line Between Sports Analysis and Fabrication

Three in the morning in Boston, January, snow pressed in white streaks against the window frame. I opened a fourteen-page PDF that a young editor had sent with a curt message: “Can you check whether this deep-dive analysis is publishable?”

The file had all nine sections. It had tables. It had a risk column, a probability column, an impact column, a mitigation column. It even had a three-tier transmission diagram running from publisher down to clubs and then to derivative markets. Formally, it was more disciplined than most of what I have read in eighteen years in this trade. Substantively, nearly every cell across the nine sections said the same thing: none.

No game title. No patch number. No team. No player. No tournament. Not a single financial line. And on the last page, the author scored the information value of all four criteria: one star out of five, every one.

What kept me at the desk until four in the morning was not the emptiness. It was the fact that the young writer had refused to fill the gaps. In my profession that is the hardest skill there is — harder than building a model, harder than reading a heat map.

The process that produced that file is a two-stage pipeline. Stage one deconstructs: it reads the source, extracts discrete information points, assigns a domain label, identifies entities, assesses time sensitivity and source quality. Stage two takes stage one's output and does the expert work: positioning the event on the competitive pyramid, weighing rosters, decoding finances, screening governance risk. The binding rule is simple — every judgment at stage two must anchor to a specific information point returned by stage one.

Stage one returned a structurally valid, substantively empty payload. The domain label was populated — “esports” — but nothing underneath it was. That is the most dangerous kind of failure in any data system: the system does not crash, it quietly returns zero, and if nobody checks, that zero flows downstream disguised as a report that looks real.

I recognised the situation because I have stood on the opposite side of it.

June 2026, at Foxborough, New England Revolution hosted Toronto FC. Toronto held seventy-two percent of the ball, fired twenty-one shots, and finished with an expected-goals total of 2.3. They lost 0-1 to a single Diego Fagundez goal. I was an intern writing match reports, and my editor asked me to celebrate the home side's “inspired” defending. I pulled the data from StatsBomb, wrote the opposite argument, and the piece drew fifty thousand reads in twenty-four hours and forced a correction. Results are the lie that time has memorised; xG is the confession.

That lesson still runs through everything I write. But it has a reverse side it took me years to face. When data becomes your brand, you start fearing empty cells. And the person who fears empty cells is the person most likely to fill them with the most attractive thing within reach: confidence.

A year later, at the 2026 World Cup, I built a PPDA table for all thirty-two teams before the quarter-finals. Croatia sat at 8.9 — they allowed opponents an average of 8.9 passes per defensive action, the lowest of the eight remaining sides. I wrote about Marcelo Brozović: 13.8 kilometres run, nine ball recoveries against Argentina. When Croatia reached the final, a Championship club hired me as a part-time data consultant.

In 2026, the pandemic turned the world into an experiment nobody had applied for. I wrote a report on 372 Bundesliga matches before and during the behind-closed-doors period: home win rate fell from 45 percent to 31 percent, penalties dropped 28 percent. Huddersfield Town brought me in for the last eight rounds of the Championship. I proposed a rotation model built on sprint distance above 6m/s — anyone below 80 percent of threshold for two consecutive matches sat out. They took 14 of 24 points and survived by exactly one point.

All three of those were cases where the data arrived complete. And in all three I could conclude quickly, because I had the right to conclude. The three-in-the-morning PDF was the same test in reverse: when you have no right to conclude, do you have the discipline to stay silent?

Start with the narrowest point. To say anything at all about a patch, the first move is choosing a patch-cadence model. Riot ships every two weeks, Valve drip-feeds around majors, Tencent follows the season. Those three rhythms produce three kinds of roster, three kinds of bench, three kinds of coaching contract. Without a title, none of those models can be selected, and every sentence about “the meta” collapses. Here the gap is worse still: the competitive logic of MOBA, FPS and battle royale is non-transferable, so a missing game title means a missing frame of reference for everything else.

I once made this mistake in a milder form. Years ago I carried the esports telemetry toolkit into football and was proud of how precise it looked. Then I realised I was not measuring pressure at all — I was measuring touches in the opposition half. A proxy is only worth something when it is compatible with the thing it claims to represent. Without a title and a patch, you cannot even begin choosing the instrument.

The second section is tournament format. A BO1 event and a BO5 event produce upset probabilities so far apart that they cannot share a model. My World Cup work shows the same team can survive a group stage on three single matches and then collapse in a knockout series simply because the sample size changed. Morocco in 2026 is the counter-example: they were not riding luck, they had a defensive structure that survived a rising sample. But to say that, I need a named tournament, a format, a bracket, a calendar. Without them, every claim about a “shock” or a “collapse” is storytelling.

The third section is the one I know best: teams and players. A name opens a whole chain of questions — form curve, age curve, injury history, contract status, bench depth, role fit. No name means that chain is empty. More importantly, no name means you cannot separate competitive value from commercial value, which is the split I run all transfer window: a player can sell shirts and still be a loss on the pitch.

The fourth is the regional picture. Even in an empty report, one claim survives because it is structural rather than empirical: the same region holds radically different standing across titles. China is a powerhouse in League of Legends; that status does not migrate wholesale to DOTA2 or CS2. Anyone who says “region X is strong” without naming the title is selling you an illusion of coherence.

The fifth is club finance, and this is where the risk-first principle earns its keep. In an empty payload there is no wage-arrears signal, no dissolution signal, no sale signal. The temptation is to read that absence as an all-clear. The absence of a wage-arrears signal in an empty payload does not mean wages are being paid — it means nobody asked. That is the most expensive error in sports analytics, and it is not a machine-only error.

The sixth is governance. Without a title and a jurisdiction, no rules system can be selected. But one structural feature of esports bears repeating whatever the article is about: the publisher is simultaneously rule-maker, commercial stakeholder, and final court of appeal, with no independent third-party arbitration. That structure does not create disputes, but it amplifies any dispute once one appears. Football has FIFA and the Court of Arbitration for Sport; esports has no equivalent.

The remaining three sections — risk profile, public narrative, industry transmission — are derivative by nature. They only mean something once a concrete event exists to anchor them. Without an event there is no narrative tag to attach: no new king crowned, no dynasty succeeding, no all-domestic roster making history, no veteran's last dance. And there is no ratio to measure between social heat and competitive fundamentals, because both sides of that division are missing.

The only confirmed item in the entire risk profile is the risk of the pipeline itself: empty input, high severity, already occurred, impact measured as the total analytical loss of the article. The remedy sits inside the finding — gate it at the door, return it to extraction, demand a re-run before it moves forward.

And here is the part I want newsrooms to keep, because it is the most commonly skipped. Such a report can only be rebuilt if extraction returns at minimum six things: the game title and patch identifier; three to five discrete information points, each traceable to a source; at least one named entity — team, player, coach or tournament; an event-type classification, meaning transfer, patch change, tournament reform, financial event or governance controversy; a time-sensitivity and source-quality assessment; and the original source with publication date. Miss any one of those six and stage two has no right to speak.

The young writer got half of it right: he refused to fabricate. The other half is pushing the problem upstream instead of waiting. In data work, disciplined silence is an action — but only when it carries a specific request.

Now the counter-intuitive part.

That empty report is more honest than ninety percent of the full reports I read during a transfer window. A full report can be full in two ways: full of evidence, or full of confidence. The second is cheaper, faster, and better rewarded by readers, because it offers certainty — the one thing sport structurally cannot supply.

I have seen the consequences. Before the 2026 World Cup I published a series arguing Morocco do not defend, they operate on data: Yassine Bounou carried a post-shot expected-goals overperformance of plus 4.3, Achraf Hakimi played 6.8 progressive passes per match. When they beat Portugal 1-0, international platforms called me. But if Bounou had saved seven in ten of those shots instead of nine in ten, my argument would still have been right and the result would still have differed. xG judges nobody; it merely exposes the truth that results conceal. Results, meanwhile, always reserve the right to disown you.

In summer 2026 a Saudi investment fund asked me to appraise Cristiano Ronaldo for a contract extension. I produced a forty-page report with an uncomfortable conclusion: his actual expected goals created was 0.55, inflated to 0.82 by set-piece situations, and that inflation would compress as set-piece volume fell. I recommended against further spending. The fund objected. Three months later his market valuation fell fifteen percent.

That story does not prove I was right. It only proves that correlation is not causation, and that a beautiful metric can be built on a supply of situations that will run dry. Football is luck, but luck is not an excuse to stop measuring. I never kicked my data habit; I just changed suppliers — from the numbers everyone sees to the numbers nobody has asked for yet.

Back to the PDF. Had the young writer filled nine sections with plausible-sounding sentences — team X is in form, this patch favours attacking play, region Y is rising — the report would have been published, would have drawn traffic, and would have gone unchecked. That is how most sports analysis is manufactured today: in the prose of certainty, not in evidence.

Absence of evidence is not evidence of absence. But the presence of confident prose is almost always evidence of the absence of evidence.

The signal for the next cycle is not a new metric. It is a gate. Any analysis system operating at newsroom scale — and transfer windows are when that system runs at full load — needs one hard condition: an input with zero information points must not be allowed to generate a report, however complete the template. The cost of such a gate is close to nothing. The cost of skipping it is measured in credibility.

Three signals to watch in the coming weeks. First, extraction success rate: if the zero repeats, it is a system defect, not a single-article defect. Second, raw source accessibility: if the origin is an image, a video, or JavaScript-rendered content, we have found the reason and need a different parsing path. Third, divergence between classifier and extractor: a populated “esports” label with empty content means the two components are reading different inputs.

Transfer data is like a tide: you cannot learn anything from the surface, you have to measure the seabed. But before measuring the seabed, you must first confirm you are standing on a sea and not in a room with a photograph of one on the wall.

One question I leave for those who have worked this trade longer than I have. If a newsroom had to choose between an analysis that might be wrong and an analysis that cannot be checked, which one has our industry been choosing for the last eighteen years?