The Empty Record and the Data Crisis in Esports Analysis

In June 2026, in a studio in Moscow, I sat in front of a screen as a rules-checking expert for a Russian television channel during the World Cup. France versus Australia, minute 55, the referee awarded a penalty after consulting VAR. I said openly on air that the "Referee Review Area" procedure under the 2026 IFAB standard had not been fully followed. Colleagues called me rigid. But the question I carried away from that broadcast was not whether I was right or wrong. The question was: if I had no footage, what would I have said?

That question returned to me years later, when I held in my hands an esports analysis report with nine dimensions, running to thousands of words, presented in the language of a serious professional document. Tournament name: absent. Team name: absent. Player name: absent. Patch version: absent. Publication date: absent. In their place were the repeated lines "insufficient information, cannot assess" at every position where data should have been, and in three fields, the verbatim system instruction had been inserted where an extracted value should have gone.

That was the moment I realised: the disease of esports analysis does not sit on the pitch. It sits in the pipeline.

I read the match record before I read the news, because the record does not lie. But an empty record — a sheet with a heading, a frame, a signature, and no content — is more dangerous than a wrong record. A wrong record can be checked. An empty record presented as a complete report will be believed.

Context: what esports analysis needs in order to exist

To analyse a football match, I need only two things: a piece of footage and a book of rules. That is how my career in Marseille began. In 2026, when I was an assistant data editor at a football site in Marseille, I reviewed the footage of Olympique Marseille versus AS Monaco in Ligue 1 and found that the referee had ignored Law 11.3: defender Kamil Glik had deliberately touched the ball before Dimitri Payet shot. My 1,200-word analysis with three situational diagrams drew 40,000 reads in 24 hours, five times the site's usual daily figure.

Esports does not give me that condition. A match in League of Legends, Dota 2, CS2 or Valorant cannot be read from a single piece of footage. It requires at least nine layers of information stacked on top of one another, and each layer has its own grammar.

The first layer is patch and meta. An update can move a champion's win rate from 52% to 46% on the strength of a single stat edit. I have followed many seasons and seen it clearly: the patch is an invisible referee with the power to decide a championship, and adaptability to the meta is routinely mistaken for raw strength. A team that wins because of a favourable patch will be re-evaluated the moment the patch changes.

The second layer is tournament structure. Single elimination, Swiss format, round-robin points, or best-of-one, best-of-three, best-of-five series — each choice produces a different upset probability. A strong team can be stable in a best-of-five and fragile in a best-of-one. Without format information, any judgement about strength is empty talk.

The third layer is roster and players. Not just names, but positions, contracts, ages, injuries, and above all the roster phase — stable, adjusting, or rebuilding. Here lie the two most valuable early-warning tools I have learned over the years: the aging player's cliff and the new roster's honeymoon. Both are useless without a name.

The fourth layer is the regional picture. The same country can hold an entirely different status depending on the title. China is strong in League of Legends but relatively weaker in Dota 2 and almost absent in CS2. Without knowing the region and the title, I cannot even select the correct vocabulary of metrics.

The fifth layer is club finance. Sponsorship revenue, publisher distributions, salary expense, capital injection. A salary-to-revenue ratio above 80% is an industry characteristic, but assessing a specific club requires specific numbers.

The sixth layer is rules and governance. Competitive integrity, transfer rules, contract compliance, protection of minors. This is where I work most, and where I see the structural paradox most clearly: the publisher is simultaneously the rule-maker and a commercially interested party, with no independent third-party arbitration body.

The seventh layer is the risk profile. Competitive, financial, personnel, rules, public opinion, systemic. A six-row matrix.

The eighth layer is public narrative and expectation. The heat of opinion, the emotional cycle from budding to climax to backlash, and the gap between market expectation and objective reality.

The ninth layer is industry transmission. From the publisher upstream, through clubs and streaming platforms midstream, to sponsorship and derivative markets downstream.

Nine layers. Each requires at least one verifiable piece of data. And the report I read that day had all nine layers — but not a single piece of data.

Analysis: dissecting a failed pipeline

When I held that report, my first reflex as someone trained in law was to check the validity of the source. Not the content. The source first. Because content can be right or wrong, but the source has to exist.

And the source did not exist.

The original article title read "not applicable". The article source read "not applicable". The article type was classified as "unclassified". The information points were empty. The core viewpoints were empty in every sub-field. The entities involved contained the verbatim internal instruction — "identify from the information points above" — that is, an order addressed to the extraction model, not a value returned by the model. Time sensitivity and source quality likewise contained verbatim template instructions.

In my profession, this is an unmistakable signature. When a referee's record has a signature section whose content is the phrase "write the referee's name here", nobody believes that record was actually produced. The same applies here. The system did not fail to find information. The system never ran the extraction step at all. The schema was printed, but it was never filled.

I once trained fifteen regional commentators in the season after the 2026 World Cup. The first thing I taught them was not Law 11, nor the VAR procedure. The first thing I taught them was: before arguing about a decision, confirm that you are looking at the same match. It sounds absurd. But in practice, many furious television arguments originate from two people talking about two different matches, or two different versions of the laws.

That report violated precisely this first principle. It could not confirm which match it was looking at, because there was simply no match to look at.

The input gate: the step that was skipped

In every process I have ever built, including the 38-criteria checklist for referees during the 2026 pandemic, there is always an input gate. That 38-criteria checklist was one I led the drafting of for matches without spectators in the Marseille region, applied to 23 friendly matches. The result: disputes over referee decisions fell 18% compared with the previous season. That checklist did not save the season — the pandemic raged on — but it saved the reputation of the person holding the whistle.

What made that checklist effective was not the 38 criteria. It was the precondition at the top of the list: if the pitch did not meet minimum conditions, the match was not played. A hard gate.

That report lacked precisely this gate. There was no precondition requiring a title, a patch version, or at least one information point. The result was that an empty input travelled straight into the deep analysis layer without being stopped.

This is the point I want to state clearly, because it is the biggest lesson of the whole story. In esports analysis, the most dangerous error is not analysing wrongly. It is analysing something that does not exist. A disallowed penalty can be corrected; a gap in the law cannot. And a data gap filled with fluent prose is harder to detect than either.

Nine dimensions, nine silences

As I went through each dimension of the report, what stood out was not that they were empty. What stood out was the systematic way in which they were empty.

The patch and meta dimension returned "insufficient information, cannot assess" in all four boxes: meta direction, beneficiaries, losers, key data. No win rates, no pick-ban rates, no average match duration. Nothing to compare, and therefore nothing to downgrade. My rule says that a claim about a patch without win-rate or pick-ban data must have its confidence lowered. Here there was not even a claim to lower.

The tournament structure dimension could not locate a tier. There was no knowing whether the subject was a world championship, a mid-season event, a regional league, or a tier-two cup. The relationship between format and upset rate — the single most important variable governing the stability of strong teams — could not be modelled at all.

The roster and player dimension had no subject. No player, no coach, no position, no contract, no age, no injury. The two most valuable analytical lenses of this dimension — the aging cliff and the roster honeymoon — had no input to operate on. Critical personnel risk categories such as carpal tunnel syndrome, tenosynovitis, burnout, dependence on a single carry, and the final-contract-year effect were all impossible to screen.

The regional landscape dimension could not position any region. The framework's explicit warning — that the same region may hold entirely different status depending on the title — could not even be applied, because neither region nor title was known. Import flows, import-policy constraints, and the health of the academy pipeline were all void.

The club finance dimension could not decompose revenue structure. Revenue concentration, dependence on publisher subsidies, the salary-to-revenue ratio — nothing measurable. And most importantly: the financial risk screen returned a null result because data was missing, not because risk was absent. These two things are entirely different, and I will return to this point later.

The rules and governance dimension could not identify an applicable rules system. Publisher rules from Riot, Valve, Tencent, Blizzard, league rules, third-party organiser rules, or national regulatory policy — none could be selected. No competitive-integrity risk could be screened, because match-fixing, account boosting, cheating, and the joint liability of coaching staff all require a specific allegation or event.

The risk profile dimension returned N/A in all six rows: competitive, financial, personnel, rules, public opinion. But there was a sixth row that did not return N/A. That was the systemic risk row: the extraction pipeline at the previous stage produced an empty schema. Level: high. Probability: it happened. Impact: high. This is the only risk confirmed anywhere in the report, and it belongs to no team. It belongs to the process itself.

The public narrative dimension identified no narrative. Heat could not be determined, the emotional cycle could not be determined, the ratio of social heat to fundamentals could not be computed. The overhyping risk — whether media promotion is planting the seeds of a future backlash — could not be assessed.

The industry transmission dimension could not trace a single link. The upstream trigger was unidentified, so the entire chain from publisher to streaming platform to derivatives market was empty.

Nine dimensions. Nine silences. And what stopped me longest was the note at the end of the report: its reference value is not as esports analysis, but as a worked example of null-value discipline in a multi-layer analytical pipeline. In other words, the document declares itself to have no value in content, only in process.

That is an admirable honesty. And it is also a frightening warning.

What almost happened

I want to spend a paragraph on what almost happened, because that is the part worth discussing.

If the deep analysis layer had received that empty input without an input gate, what would it have done? By the instinct of a language model trained to answer, it would have answered. It would have written a fluent, confident, well-structured nine-dimension analysis, full of professional terminology and plausible-looking numbers. It would have spoken about a champion's win rate, a team's lane dominance, a club's contract structure. All of it invented. All of it very convincing to read.

This is the most dangerous failure mode in analytical publishing. Not the obvious failure — the obvious failure gets caught immediately. It is the failure disguised by fluency. VAR is not wrong. The person operating VAR is, after all, only a person. And a language model operating on empty data is also just an entity trying to complete its task, by filling the gap.

The good news is that in this particular case, the gate worked. The report stopped, declared clearly "analysis impossible", and refused to pass judgement on any game, team, player, coach, tournament or publisher. It said plainly: any such judgement would be fabrication.

That is correct behaviour. It is the behaviour of a referee with enough backbone not to blow the whistle when he cannot see the ball.

But the bad news is this: that gate was not designed as part of the pipeline. It was a manual decision at a later stage. Change the operator, change the model, or merely vary the prompt, and the result could have been entirely different. The systemic risk at the input stage remains intact.

The contrarian view: fan emotion, data discipline, and the distance between them

I must say something that the analytical profession rarely agrees to say.

When an empty analysis is pushed out to the public, the strongest reaction usually comes not from practitioners. It comes from fans. And their reaction is often dismissed as emotional, unfounded, to be set aside.

I disagree with that framing.

In seventeen years of watching this industry, I have found that fans' instinct about a fake article is highly sensitive. They cannot point to the logical error on line three, they cannot cite the null-value handling rule, but they can feel that something is off. Something generic. Something formally correct but empty in content. Like looking at a transfer contract with all the signatures and seals but missing the core clause.

That feeling is valid data. I am not saying it is always right. I am saying it is data, not noise. And someone like me, a person whose trade requires distinguishing evidence from conjecture, also needs to admit that there are signals not recorded in the record yet still of value.

In France, where I live and work, fans have a fairly strong culture of dissent. They are ready to take to the streets, to hang banners, to organise protests outside federation headquarters. In China, where I was born, reactions tend to play out on online platforms, fierce in wording but fragmented in organisation. The same refereeing situation, two markets, two ways of reacting. And in both, what fans react to most strongly is always the feeling of being deceived — not deceived about the result, but deceived about transparency.

A wrong refereeing decision can cause controversy. A transparent process that produces a wrong result still retains trust. But an opaque process, even when it produces the right result, destroys trust. This is why I always tell the commentators I train: when you explain a decision, you are not only explaining the result, you are explaining how the result was produced.

The offside line has never been straight; it is only that today I can see that it is curved. The same applies here. The analytical pipeline has never been watertight; it is only that today a leak has appeared before my eyes.

And the distance between fan emotion and data discipline is not an unbridgeable distance. It is a distance that practitioners are responsible for narrowing, by making their processes public — not by teaching fans that they ought to trust.

What remains open: levels of certainty and items awaiting verification

Based on what has been verified, I offer the following provisional conclusions.

First, the failure of source-data integrity in this case is certain to a high degree, because the evidence lies inside the document itself: the entity, time-sensitivity and source-quality fields contain the verbatim instructions of the extraction layer. This is an unmistakable trace of an unfilled output.

Second, the root cause most likely lies in the fetch or parse layer upstream of extraction, rather than in the original article genuinely being empty. This is a low-confidence inference, because a truly empty source article is a less likely scenario but cannot be excluded.

Third, the full analytical value is likely recoverable if the original article still exists. This is a medium-confidence inference.

Fourth, and this is the most important part: a null risk-screen does not mean the subject has no risk. All six risk categories return not-applicable because data is missing, not because they were checked and found clean. In my work, this is the most dangerous interpretive error. A monitor showing no alarm may be safe, or it may simply be switched off.

I do not know who or what the subject of the article was — which team, which title, which tournament. I do not need to know in order to draw a conclusion about the process. But if someone reads this report and understands it as "no risks were found", that is a logically false conclusion, and it will lead to false decisions in practice.

Proposed improvements: one clause at a time, not a revolution

I do not believe in calls for a total overhaul of the system. A system overhauled wholesale without a roadmap usually collapses faster than the old one. Reform must be measurable in cost, risk and roadmap. In that spirit, I propose four specific clauses.

Clause one: hard schema assertions at the extraction-layer boundary. Any output with an empty information-points field, or with an entity field containing a string matching a known template, must be rejected automatically. Not warned. Rejected. Because a warning can be ignored, while a rejection blocks the pipeline.

Clause two: metadata tagging of failed records. Every record with an empty title or source must be tagged "extraction_failed" and excluded from all aggregation sets, all archival indices, and all datasets used for training or evaluation. An orphaned record pollutes the entire archive over time.

Clause three: logging at the fetch layer. The system must be able to distinguish "empty document" from "extraction produced nothing from a non-empty document". These two situations require two different remedies: infrastructure fix versus source-quality downgrade. Without logging, they cannot be distinguished, and without distinguishing them, every attempt at repair is guesswork.

Clause four: a gate based on minimum criteria, not volume. The gate must require a title, a source, and at least one information point. It should not require a high minimum count of information points, because a short official announcement from a tournament is a valid input. The gate must distinguish legitimate brevity from genuine emptiness.

These four clauses require no new technology. They require discipline. And discipline, in my profession, is always harder to build than technology.

A progressive thought

I once thought my job was to catch errors. After years as an assistant data editor, then a rules-checking expert, then a mid-level staffer at a regional federation, I was called a rules robot because I never said "probably right". I thought that was my entire value: seeing the error faster than others.

But the more I work, the more I see that catching errors is the easy part. The hard part is designing a process in which errors cannot pass through silently. In football, I cannot undo a decision that has been issued. I can only ensure that the next decision is issued within a better process.

Esports is at exactly that stage. The industry has passed the era when a good analysis only needed an author with a reputation. Now, reputation must be proven by traceability of sources. An article with no source, no timestamp, no verifiable number will increasingly be treated like an empty record — with a frame, with a signature, but with no legal value.

When I read that nine-dimension report, what I saw was not an embarrassing failure. I saw an opportunity. Because a pipeline that dares to declare "I cannot analyse" when there is no data is more trustworthy than a pipeline that always has an answer. The match does not end with the whistle; it ends when people have finished reading the record. And an honest record, even an empty one, is still better than a record full of words that are all invented.

The question I leave behind is not how to analyse more. It is how this industry can learn to say "I do not yet know" without fearing the loss of reputation. Because in esports, as on the pitch, the one thing that cannot be fabricated is the truth — and the one thing worth building is a system that makes fabricating it structurally impossible.