Nine Empty Fields and the Esports Analysis Machine That Never Stops

At 1:47 in the morning in Shenzhen, I opened the attachment my editor had sent for the next day's recording. Forty pages. Nine sections. Twelve tables. Dozens of metrics boxed inside starred cells. I scrolled down.

First cell: insufficient information. Second cell: insufficient information. Third, fourth, all the way to the last cell of the ninth section — the same phrase, repeating like a curse.

Only one field had been filled in correctly: the domain label, esports.

Then the editor's message appeared: "Just make something up. The audience won't notice."

That was the moment I understood I was looking at something more dangerous than any match-fixing story I have ever covered. An analysis machine that refuses to stop when it has no data.

People like to say esports grew up on money. True, but not enough. Esports grew up on spreadsheets.

For roughly a decade now, every major match has dragged behind it a data layer many times thicker than the match itself. A caster sitting in front of a screen now holds stage-by-stage win rates, lane indices, objective-take speed, lead-holding time, win rate after a patch switch. Fans call it deep analysis.

And behind that data layer runs an entire industrial chain. Publishers push patches. Organisers publish formats. Clubs announce transfers. Streaming platforms record every metric. Then data vendors turn all of it into tables. From those tables grow articles, panels, predictions, odds.

I have my own name for the lowest layer of that chain: the entity layer. It is the set of concrete names every analysis must attach to — game, tournament, team, player, coach, publisher. Without the entity layer, every table above it is decoration.

The report I opened that night illustrates exactly that. It was built to answer nine big questions of the industry. Where the patch is pushing the meta. Which team type the format rewards. Whether a roster is stable or rebuilding. Which region is rising. Which club is bleeding money. Which rules are being bent. Which risks are accumulating. Which media narrative is being inflated. And which power current is shifting from publishers down to the derivatives market.

Nine questions. Not one answer. And still a formally complete report.

Here is the hard thing to hear: this industry has taught fans that a starred document is a trustworthy document. We forgot that stars can be the product of an empty cell filled with belief.

The most misunderstood thing about a null record is this: it is nothing like a thin record, and the two demand opposite handling.

A thin record has information, just little of it. A short piece on a minor match. A transfer announcement with no fee attached. You can still pull out the team, the player, the timestamp. From there you can reason, even with a wide error bar.

A null record has nothing. No name. No date. No event. And the danger sits here: inside an assessment system, an empty cell usually reads as "no problem found." The correct reading is "not assessed."

The distance between low risk and unrated risk is the distance between a sound investment decision and a blind gamble. No data about a team does not mean the team is healthy. No warning about unpaid wages does not mean wages were paid.

Across the report's nine dimensions, every cell carries "not assessable" rather than "safe." A risk that has not been rated must always be read as a risk still present, never as a risk that has disappeared.

The trace of the failure sits right on the surface. The domain label esports was filled in correctly. The analytical frame was built correctly. Only the content was empty. If classification had failed, the label would read "unidentified." Here the label is right and the interior is hollow. The machine still recognised it was reading about esports, but could not pull a single word out of the source.

That kind of failure usually comes from a familiar set of causes: a login wall, a paywall, a bot block, or a page returning only headers and no body. For a content person, this is good news. The failure is usually transient, and one re-run fixes it.

There is one spot in the document where I stopped the longest. It contradicts itself structurally. The document instructs the analyst to identify entities "from the information points above" — while the list of information points above is completely empty.

To get an entity, you need an information point. To get an information point, you need an extracted body. The body does not exist. So all nine dimensions stand still.

To an outsider this is a boring technical detail. To someone inside the industry, it is evidence of a process-design fault. Entity recognition was placed before information extraction, while its logic depends on the later step. It is a one-line fix. But if nobody spots it, the system keeps failing the same way, every day, on every article.

The loss is not evenly spread. It concentrates in exactly the cells that matter most.

If the source was a tournament preview, what vanished is the draw and the format — bracket half, seeding, group-stage length. If the source was a transfer announcement, what vanished is the fee, the contract length, and the identity of both buyer and seller.

This is the part I find most frightening in the whole story. The analysis system can still produce a complete segment by filling empty cells with what it remembers from other articles. It does not invent conclusions. It simply pastes old data into a new slot. Technically that is rational behaviour. Informationally, it is fabrication.

There is a name for this: base-rate substitution. When form data for Faker or Chovy is missing, the system fills in the regional average win rate for the mid lane. When data for a patch is missing, it uses the trend of the previous patch. The output reads fluently. And it is wrong exactly where fans care most.

At the same time, a player like Levi or SofM can be graded by the base rate of an entire generation — a figure never calculated for them specifically.

Based on my experience watching hundreds of esports and football matches, this class of error is harder to catch than any ordinary data mistake. It is not wrong in the numbers. It is wrong in where the numbers came from.

I have seen something similar at a much smaller scale. In 2026, when I said on air that Germany would be eliminated in the World Cup group stage, I was working from a specific observation about ball-circulation speed and average squad age. I was right. But I also know how many people in the trade reached the same conclusion with nothing behind it — only because that team had lifted the trophy the time before. The same sentence, two different foundations. One built on data. One built on collective memory.

Esports is at exactly this stage, only industrialised.

The cost of missing something is asymmetric. Missing a routine news item costs content. Missing a signal about match-fixing, unpaid wages, or a young player's injury costs people and costs the integrity of the discipline.

The report says this plainly, and I agree with the choice it makes: when you hit a null record, the correct response is to escalate, not to quietly discard. The loss from a missed integrity case is many times larger than the effort of one re-extraction.

And here I have to speak about the darkest part of the story — the part the report touches lightly and walks past.

Sports data exists to serve viewers. The same stream also flows into a different pipe, where it becomes input for betting companies. The report sets its own limit: everything is to be read only as market-expectation information, never as betting advice. I respect that limit. I also know the limit does not exist on the receiving side.

A system that can generate nine dimensions of analysis out of nothing, and a system that can turn nine dimensions of analysis into odds, are two systems that complement each other very well. The first does not need to be right. The second only needs an input.

That is why I have never believed digitised sport brings only good. It brings transparency to viewers. And it brings raw material to rooms with no audience.

The U19 tournament that year taught me one lesson: an editor's silence is a crime. Here, the culprit does not stay silent. It publishes. It publishes a document full of form, full of stars, full of section headings, with a hollow interior. And the reader on the other end nods along, because it looks like a report.

Everything I know about sport I learned from my mistakes on air. What I learned was not to stop being wrong. It was to stop pretending I was never wrong.

I will say the thing a safe editor would keep off air.

If you hand me a system that returns nine empty cells and ask where the fault is, the easiest answer is the extraction step. The truer answer sits at the publish decision. A machine returning all-empty cells is not broken. It did exactly what it was programmed to do: produce structured output regardless of input.

And what unsettles me most is the human reaction around it. "Just make something up. The audience won't notice."

The audience won't notice. That is what we say to each other in this industry, every day, in different words. It does not stop at empty data. It applies to spliced stat sheets, to headlines heated past their content, to predictions issued with nothing behind them.

Fans hate the truth, but I do not go on air to be loved. And the biggest lesson from a null record is this: if people inside the industry do not build their own brakes, nobody will build them for us.

Where I could be wrong. There are three places I have to interrogate myself.

I am reading an empty document and inferring a systemic problem. It is possible this was a single failure, one collision with a block wall, and the industry does not operate the way I just described. That probability is not small. Most extraction failures are transient.

I am describing an industry with material from that same industry. I sit in a studio in Shenzhen, I make a living from a sports podcast, and I remark that the whole industry is selling hollow stars. There is a self-flattery in that argument I cannot deny.

And I have a habit of seeing systemic meaning where there is only a technical fault. I have done it before. In 2026, I said Japan were deliberately losing to study their opponents. They won. But I still never proved I was right about the motive — only that I was right about the result.

What I am certain I am not wrong about: a null record must not leave the analysis room wearing the face of a complete report.

A testable prediction: within the next twelve months, at least one major programme or outlet will publish an analysis segment built on an unresolvable data record. The signature will be a nine-dimension analysis with no concrete name, no timestamp, and no sourced figure.

The machine has learned to speak without knowing. Whether we teach it to be silent is a human choice.