Nine Categories Returned N/A: Notes from an Esports Data Analysis That Failed at the Input Stage

2:17 a.m., Busan. Outside the window, a thin drizzle and sodium streetlights. I reopened the report that had just come back from the data-processing stage and counted: nine categories, forty-two cells, and every one of them returned the same string — N/A.

The first cell was the article title. Blank. The second was the source. Blank. The third was the article type. Unclassified. The fourth was the information points. Zero items. The fifth was the entity list, covering game title, teams, players, coaches, tournaments. Not identifiable.

The report did not try to fill the gaps. It stopped, stamped itself HALT, and wrote a sentence I read three times: the first prerequisite of esports analysis — identifying the specific game title — cannot be met.

I once sat in a WK League stand with 347 spectators, counting every high press on the left flank that the single fixed camera missed. That year I rigged a second, low-angle camera myself, and Lee Min-a's 23rd-minute opener came through clearly. A secondary camera is not a lower starting point — it is the angle the stand has never seen.

But tonight was different. The camera was there, the tripod was there, the operator was there. The tape was just blank.

It is worth spelling out the system that produced this document, because it is not an article. It is the second stage of a two-stage pipeline my data team and I use to read esports.

Stage one performs deconstruction: title, source, date, article type, discrete information points, viewpoint summary, author stance, article purpose, entity list, time sensitivity, source quality. Stage two takes that output and applies nine professional dimensions: patch and meta, tournament system and format, teams and players, regional landscape, club finance and business, rules and governance, risk profile, public narrative and expectation, and finally industry transmission.

Those nine dimensions do not apply to a blank page.

The reason is that esports is not one sport. It is a cluster of sports sharing a stage shell. KDA in a MOBA, HLTV Rating in a tactical shooter, gold-to-damage conversion in an arena title — those three figures do not convert into each other, do not share a unit of meaning. Patch cadence is the same: one publisher ships a build every two weeks, another goes months without touching the game, another runs on seasons. Each cadence produces a different kind of meta, and a different kind of analysis.

Even role semantics do not translate. A carry in a MOBA and an AWP in a tactical shooter are both damage dealers, but the way contribution is measured, risk is priced, and career length is calculated differs entirely. Without a game title, there is no dictionary.

Then the tournament system. Single elimination, double elimination, Swiss, or league points — four formats, four upset probabilities. BO1, BO3, BO5 — three numbers encoding three levels of stability for a strong team. Without a title there is no tournament, no format, and every sentence about upset potential is a guess.

The regional map is title-dependent too. A country's standing in one title can be entirely different from its standing in another. Drawing a regional tier map without knowing the title is drawing a map that does not exist.

I learned this through a specific scar. In 2026, reviewing the England–Japan tape from the Tokyo Olympics group stage, I counted 17 fast counterattacks by England in the second half. The official stat sheet recorded 3. Since that day I do not trust any table I have not cross-checked myself. I do not trust emotion; I trust data. Emotion can lie. A table of numbers cannot.

But an empty table says nothing at all. And that was exactly the problem tonight.

One detail in the report matters more than the rest, and it sits in the terminology notes: N/A — insufficient information is a deliberate null marker, distinct from none found. It signals that the analytical question was never testable with the available input, and must never be read as a negative finding.

This is where most operational dashboards go wrong. People are trained to read a blank cell as no problem. But no violation signal found and cannot test whether a violation exists are two fundamentally different statements. No data and data showing no risk are two different statements, and the spreadsheet cannot tell them apart on its own.

The report makes this explicit in the governance section. The checklist covers competitive integrity, transfer and registration rules, contract compliance, minor protection, and publisher governance controversies. All five cells are N/A. And the report says it plainly: the absence of a compliance signal here is an artifact of empty input, not a finding of compliance. It even flags itself — under the risk-first principle, this gap must not be read as reassurance.

I consider that the most professionally correct sentence in the entire document. Because out there, machines are running on the opposite logic: a blank cell becomes a green row, a green row becomes a no-anomalies-detected report, and that report becomes a decision.

Another detail deserves a pause: the patch check. The row for patch claims lacking data support is ticked, but annotated vacuously triggered — it lit up not because a patch claim was caught, but because no patch claim existed to catch. A green tick generated from nothing.

Read only the executive summary and you see a row of processed cells. Read closely and you see nothing was processed at all.

The detail that stopped me longest was the risk profile. Seven risk categories — competitive, financial, personnel, rules, public opinion, systemic, and a seventh the report calls process risk — all N/A. The last one was not. It was rated High, probability High, impact High, described as an empty stage-one payload being mistaken for no risk found. Mitigation: re-run extraction, do not act on this report until populated.

That is a craftsman's sentence. In a table of zeros, only one line carried a real number, and that line was about the table itself.

The report also listed what it could not rule out: unpaid wages, match-fixing suspicion, a patch targeting a dominant playstyle, injury to a core player. None confirmed, none denied. Simply outside the testable range. In my trade, an untested risk must be recorded as untested, never as absent.

The public narrative section returned N/A as well, but left a note worth keeping. It states that the traffic-filter test — checking whether a crowd is excited by the competition itself or by a headline — cannot be applied, because there is no headline to test. In my trade that is the most important test and the most frequently skipped.

Industry transmission is the same story. The map from publishers upstream, through clubs and streaming platforms midstream, to sponsorship and derivatives downstream, cannot be drawn because the publisher is unidentified. Riot, Valve, Tencent, Blizzard, Krafton — each name on that list carries a different patch cadence, governance posture, and commercial strategy, and each produces a different transmission effect across the chain. Drawing one map for all of them is wrong from the first stroke.

Then the section that hurt most to read: the information value rating. Four dimensions — competitive value, industry value, timeliness value, reference value — all one star out of five. Not because the subject was dull, but because there was nothing to rate. The report records itself as not citable, retained only as a process-audit artifact documenting an extraction failure.

A document declaring itself unworthy of citation. I have read a great many analysis reports over twelve years, and this was the first I had seen voluntarily rate itself at the floor.

Technically, the report offered three unblocking paths. First, a populated stage-one result: title, source, date, at least three information points, and an entity list with the game title mandatory. Second, the original article text or URL, so deconstruction and analysis can run in a single pass. Third, a structured briefing — game title, entities, and three to five key facts to assess — when the source is unavailable or paywalled.

The minimum viable input for a defensible stage-two report is four things: game title, article date, source outlet, and at least one named entity — a team, player, coach, or tournament.

Four things. Only four. And we had none of them.

I remember 2026, when global competitions paused, sitting down to build a dataset of 214 matches involving the Korean women's national team from 2026 to 2026. The result: only 23.7 percent of goals came from set pieces, well below Japan's 41.2 percent. I sent the report to the head coach and got back an email inviting me to collaborate on opponent analysis during the October camp. 214 matches, 214 problems: the pandemic did not stop football, it only changed how we read a match.

That 23.7 percent carries weight because it stands on 214 matches. A number with no sample behind it is just a style of writing.

And this is why the report chose to stop. Without a game title, every metric selected is invented. Without a publisher, every patch claim is invented. Without a format, every upset projection is invented. Without a title there is no metric system to choose, and any analysis produced afterwards is fiction packaged in professional language.

The danger of fiction packaged in professional language is that it looks exactly like the real thing. Same layout, same words — meta, roster depth, tempo, transfer window. Different in one respect only: there is nothing inside.

Esports is not a young person's sport — it is a sport for those who read the meta before stepping onto the stage. And reading the meta begins with knowing which game's meta you are reading.

There is a layer here I want to address separately, because it is my beat. Women's sport in general, and women's esports in particular, suffer the same reading error with heavier consequences. The data infrastructure is thinner: fewer matches filmed, fewer stat providers, fewer people willing to recount a tape. When infrastructure is thin, blank tables appear more often — and because they appear more often, people default to assuming there was nothing worth counting. Historical neglect does not require anyone to decide anything. It only requires a blank cell to be left alone.

That is why I do not treat rigging a second low-angle camera as a technical matter. It is an editorial act. And it is why I treat refusing to fill a blank cell as an editorial act of the same kind.

There is an inverted reading, and I think it is the correct one.

In an industry where insight is the unit of currency, a report that returns null is a failed product. In an industry where decisions rest on reports, a report that returns null is a successful product, because it blocks a bad decision before that decision is made.

The blind spot of sports journalism is not wrong numbers. Wrong numbers get caught, usually fast, because someone is always cross-checking. The blind spot is missing numbers presented as zero numbers. A metric that was never collected passes through three processing layers and emerges as a green cell. And nobody cross-checks a green cell.

I have written before that possession percentage is the most deceptive metric in football, because many teams grind out 60 percent with meaningless sideways passes. But possession at least deceives with real numbers. Far more dangerous is a metric that deceives with numbers that do not exist, because it leaves no trace to argue against.

That is why I consider the report's final recommendation — a schema-level guard rejecting any stage-one output with zero information points, so that empty payloads fail loudly instead of flowing downstream as no risks identified — the highest-transfer-value passage in the whole document. It is not about the article. It is about how an organisation transmits information.

A good broadcaster is not the one who talks most, but the one who knows when to let the data speak. And sometimes letting the data speak means telling the audience: I have no number here.

At three in the morning I closed the file and wrote one line in my notebook: next time someone hands me an all-green table, the first question is not what does it conclude, but what went in.

A blank tape is still a document. It tells you the camera was running and the lens cap was on. The job of the person in the edit room is not to colour in the frame — it is to go find the cap.

And if the cap cannot be found, the right thing is to say so: I have not seen anything yet.