Null Results in Esports Analysis: Why 'Esports' Is Not a Data Point

In March 2026, a forty-two-page dossier arrived in my inbox. The cover page carried the tournament name, the team name, the player names, the game version number, the match date and the data sources. I spent two days reading it and four days writing, then returned a report with three scenarios: optimistic, baseline, pessimistic.

Four months later, another dossier arrived at the same address. The cover carried a single word: esports.

Inside, blank.

No game title. No tournament. No team. No player. No dates. No win rate, no pick-ban rate, no match duration, no patch number. Attached was one line of instruction: "Assess the potential of this market for me."

I did not assess it. I returned a document containing exactly one conclusion: not assessable. In the business of pricing things that have not happened yet, that is the hardest answer to write, the hardest to sell, and usually the only correct one.

What made me sit down and write this was not the empty dossier itself. It was the reply I got when I sent it back: "Just estimate something, who's going to check?"

Someone will. I will. And that is precisely the problem.

A two-stage pipeline and one rule with no exceptions

The way my data unit works — and the way most analytics rooms in this industry work — has two stages. Stage one deconstructs: read the source document, extract "information points." An information point is an atomic unit of fact: it has a subject, a timestamp, it can be verified, and it can be cited without fear of error. Stage two analyses: it takes that set of information points as its substrate, builds a model, and draws conclusions.

The underlying rule is simple and admits no exceptions: every stage-two conclusion must be traceable to the information point that produced it. No information points, no conclusions. No win rates, no assessment of squad strength. No patch number, no meta analysis.

The July dossier was empty at stage one. So stage two had nothing to build.

That should be the whole story, and it is not complicated. But there is a detail worth more attention: that dossier still carried a valid classification stamp at the domain field. The label "esports" was there, correctly spelled, correctly categorised. Which means the classifier ran. Only the extractor failed to run, or ran and returned nothing.

Two components executed out of sync. One said "this is an esports document." The other said "I found nothing." The system merged both into a file that looked legitimate.

That failure mode is more dangerous than a hard failure. A hard failure gets noticed and fixed. A silent failure does not, until someone reads the output and assumes "no risks detected" means "no risks exist." Those are very different sentences. One is a conclusion. The other is an admission that nobody checked anything.

A label is not a variable

This is the point anyone doing esports analysis must internalise, and the point automated models violate most often.

Esports is not a sport. It is a category.

Beneath that label sit competitive structures that are not interchangeable. A MOBA match is decided by lane tempo, gold difference at fifteen minutes, objective control rate, and the ability to convert an early lead into pressure on the enemy base. A tactical shooter match is decided by round economy, pistol-round win rate, first-kill conversion, and the in-game leader's call quality. A battle royale match is decided by placement points, zone distribution, and the ability to rotate a squad when the map closes in.

The same words — win rate, form, injury — mean different things across those three families.

The Esports World Cup held in Riyadh in 2026 is the clearest illustration of the category's scale: more than twenty titles under one framework, a prize pool above 60 million US dollars, and a club championship that aggregates points across entirely different games. Looking at that, it is easy to believe you are looking at one sport with many disciplines. In reality each discipline has its own ecosystem, its own patch cadence, its own tournament structure and its own metric set.

An analyst handed a file that says only "esports" is not short of data about a team. They are short of a game. And without a game, there is no unit of measurement.

What is the PPDA of esports

I came out of football, so my first question is always: what is your unit of measurement.

In football, the answer lives in PPDA — passes allowed per defensive action — and in high-speed running distance. When those two shift, tactics shift. At the 2026 World Cup, Germany's PPDA sat at 8.7, meaning they let opponents circulate the ball far too comfortably in midfield, and the group-stage exit reflected that number exactly.

In esports the units exist, but they must be chosen per title. In MOBA titles: gold difference at fifteen minutes, major-objective control rate, damage per minute, vision score per minute. In tactical shooters: win rate after winning the pistol round, survival rate in one-versus-two situations, and comeback rounds won from a deficit. In battle royale: average placement points, average finishing position, top-four rate.

Until you know which family you are in, the question of the unit has no answer. And without a unit, every claim about form is just a feeling dressed in a confident tone.

In recent years I have spent most of my time on what I call the decay coefficient — a way of measuring how fast a player or a roster declines, treated as a deterministic physical quantity. Reaction time decays with age; it must be measured in milliseconds and re-measured every season. Champion pool depth contracts as patch cadence accelerates; it must be measured as the number of champions that clear a competitive threshold per version. Lane performance per minute declines as younger opponents enter the pool.

What those series taught me is that the decay coefficient spares no one. Faker is the longest continuous data series this industry has — more than a decade at the top, across champion pool changes, roster changes and entire system changes. Chovy is a different curve: mid-lane stability, low variance, high creep score per minute, muted movement across patches. Two different curves, one ruler, and neither readable unless you know which game and which version you are talking about.

On the tactical shooter side the same rule applies: s1mple and ZywOo are two of the rare long-horizon datasets in that discipline, and their value lies in season-to-season stability rather than in any single event. That is why contracts built on one brilliant week tend to return far less than contracts built on three stable seasons.

A transfer is not the purchase of a person; it is the purchase of a probability distribution. And to buy a probability distribution, you first have to know what you are measuring.

Tournament format is the most neglected variable

There is a layer of data outside the game that directly shapes every conclusion about squad strength: the format.

A team playing single-match series has far higher result variance than a team playing best-of-three or best-of-five. This is not intuition; it is a direct consequence of sample size. The same roster, the same form, can post a materially different apparent win rate in a group stage versus a knockout bracket purely because the format changed.

Swiss systems, double elimination, group-then-crossover — each structure produces a different opponent distribution. A team landing in a bracket half with two strong opponents has a lower advance probability than an equally strong team landing in a lighter half. Ignore this variable and the analyst attributes "knockout mentality" to a team when the actual explanation was the draw.

I do not trust intuition — I trust the decay coefficient of intuition. And that coefficient says most of the "against the odds" stories circulating online can be explained cleanly by three variables: format, bracket half, and number of games.

The layers outside the game

If you read only patch notes and in-game statistics, you will miss two layers of data with far greater destructive power over the value of a team or a player.

The first is club finance. The most common early-warning signal in this industry is unpaid wages. It appears before a roster dissolves, before players speak publicly, and before a governing body issues sanctions. It is an observable variable, provided anyone bothers to observe it. The Overwatch League is the great case study: franchise slots were valued at up to 20 million US dollars each at the peak, and only a few years later the league had to move to an entirely different framework. No model built on in-game statistics could have predicted that collapse.

The second is governance and competitive integrity. This is the layer Vietnamese esports had to confront at an unprecedented scale: in March 2026, the national league organiser, working with the publisher, announced sanctions against a series of players and coaches linked to match-fixing. That event appears in no patch note. It appears in contract files, in transaction histories, and in anomalies in the betting markets.

Long before that, the industry had international precedents: a match-fixing case at a 2026 Counter-Strike event led to lifetime bans for a group of players in early 2026, and match-fixing cases in Korean StarCraft forced an entire ecosystem to rewrite its rulebook.

Every crisis is unlabelled data. The analyst's job is to label it before it becomes a headline.

The missing state in every risk matrix

Back to the empty dossier.

Its biggest problem is not the blank space. The problem is what that blank space will be read as once it passes through someone else's hands.

In most risk matrices I have received, there are only two states: risk present, or risk absent. There is no third state. But reality always needs a third state — the state of "not assessed."

The difference between "no risks detected" and "no data examined" is the difference between a conclusion and a blank. When that blank is pushed through a display layer, it is usually auto-filled with a zero. A zero looks like a measurement. Nobody is warned that it was merely an empty cell.

This is a systemic failure with reach: an empty field propagates carrying the appearance of a value. Risk is transferred from the analyst to the reader, and the reader is never told.

I once hit a manual variant of this failure. In 2026, when European leagues had to play behind closed doors, I sat down and calculated home win rates across an entire Bundesliga season and found they fell from roughly 46 percent to roughly 29 percent. For Union Berlin — a club built on its crowd wall at the old stadium — the points given up reached more than 60 percent compared with matches played in front of fans. But if I had presented a table with only two states, "risk present" and "risk absent," the clubs with the smallest behind-closed-doors sample would have appeared as the safest clubs.

They were not safe. They were merely under-sampled.

An empty stadium summer is when I hear data falling drop by drop. And every drop that lands in an empty cell gets read as a zero.

A dossier that closes its own loop

Inside that empty document was a detail I find more frightening than the emptiness itself.

The field "entities involved" read: identify from the information points above.

The field "source quality" read: assess from the source fields of the information points.

Both fields are dependent fields. Both point at an empty list. The loop closes on itself and the system has no mechanism to detect that it is closing on itself.

This failure mode appears in almost every analysis process I have seen, including human-operated ones. A form asks you to fill one box by referencing another. When the source box is empty, the person filling it has three options: leave it blank, guess, or walk away. Very few walk away. Very many guess.

And when guessing is packaged inside a properly formatted template, it becomes indistinguishable from a properly formatted conclusion.

In Vietnamese esports, where most analysis is written from match feel rather than from a cited dataset, this failure is the default rather than the exception. The writer starts from a conclusion already formed, then selects the metrics that flatter it. I call that white cheating. It breaks no law, violates no regulation, and nobody gets punished. It simply corrupts the entire evidence base an industry depends on.

The counterintuitive point: a null result is the highest-value product

Here is where I break with the habit of an entire industry.

I hold that the most valuable product an analytics unit can generate is not a correct forecast. It is a null result, properly published.

The reasoning is pragmatic. A correct forecast may simply be lucky, and in this industry the luck rate is not small. A null result has no luck propping it up: it appears only when the analyst has checked and concluded there is not enough data to say anything at all. It is the only kind of conclusion that cannot be faked by sampling.

The paradox is that the market does not pay for that kind of conclusion. Clients pay to hear a name. Editors pay to get a headline. Algorithms pay for engagement. A document saying "I cannot assess this because data is missing" produces no name, no headline, no engagement.

The result is that the whole system incentivises fabrication. Not blatant fabrication — a subtler kind: filling an empty cell with a statistically plausible value that has no data foundation, then presenting it in the confident tone of a measurement.

Numbers never lie — only the reader's heart turns them into lies.

One variant of this problem deserves separate mention, because it recurs constantly in transfer discussions: confusing "currently hyped" with "genuinely good."

A player who shines across six matches at an international event is a six-match sample. A player who sustains stable output across three seasons is a three-season sample. Those are not the same grade of evidence, though they are routinely compared as if they were. In the most recent valuation file I worked on, the candidate the media mentioned most was the candidate with the widest variance, and the candidate dismissed as boring was the candidate with the narrowest probability distribution.

Nobody gets excited about a narrow distribution. But a narrow distribution is the only thing that can be priced.

One clarification, to avoid being misread: I am not arguing that every analysis must end in a null result. That rate must stay low. If an analytics unit returns null results for most of the files it receives, the problem is the unit, not the files.

But when a null result is the right answer, it must be delivered. And the person delivering it must be protected. In Germany, sports media operates on a principle I learned in my first days on the job: if you do not have a second independent source, you do not publish. That principle is written nowhere. It exists only because the people before us paid a price for it to exist.

Esports has not paid that price. And it is spending capital it never accumulated.

What to demand from any analysis

If I had to distil a minimum checklist from this whole story, it would have five items. A readable piece of esports analysis must name the game. It must name the patch being played. It must name the tournament and its format. It must name a subject and a specific date. And it must name the source of every statistic it cites.

Missing any one of those five, the remainder can still be read as commentary. But it is no longer analysis. It is an opinion decorated with numerals.

For Vietnamese teams competing internationally — from long-serving figures such as Levi in the jungle to younger players being promoted into starting rosters — this checklist matters more, not less. When a team plays only a handful of international matches each year, every match becomes a precious data point, and the risk of over-reading a single game is at its highest.

Some matches end when the referee blows the whistle — and some only begin when the data speaks. The empty dossier I received in July contained no matches at all. It contained one label, and a label is not enough to begin.

What I want to know next season is not which team will win. It is how many dossiers sent out during this season share the exact structure of that empty file — valid label, valid classification, empty content — without anyone having noticed.