Esports Data and the Gaps That Speak

In November 2026, at Gocheok Sky Dome in Seoul, the League of Legends World Championship final entered its deciding game. On the live data board, every metric was present: gold difference, kill count, dragon control rate, cooldown timers for each champion. I recorded all of it in my spreadsheet, exactly as I have for six years. Then I realized something. There were things I wanted to measure that no tool could measure: the hesitation of a player before casting a final ability, the number of times a team fell silent in internal communication, the quickened breathing of a player after a lost teamfight. Those gaps sit in no statistics table, and precisely because of that, they tell me more than the numbers flickering on the screen.

When data speaks, the whole stadium falls silent. But when data falls silent, I learn to listen to that silence.

Context: the esports data ecosystem

Esports has spent two decades transforming from small LAN tournaments in internet cafes into an industry whose global revenue was estimated at more than 1.8 billion USD in 2026, according to Newzoo. Alongside that growth came an unprecedentedly complex data ecosystem. Riot Games provides an official API for League of Legends tournaments, allowing retrieval of data from every teamfight at millisecond resolution. HLTV aggregates detailed statistics from hundreds of thousands of Counter-Strike matches. Analytics platforms such as Oracle's Elixir turn raw data streams into charts anyone can read.

But that ecosystem has a hole. Data exists only when someone actively records it. Every metric on screen is the result of a decision: someone chose to measure it, chose to publish it, chose to place it beside other numbers. Whatever was not chosen disappears. And in the esports transfer market, where teams spend millions of USD to sign players, those data gaps become dangerous blind spots.

The scale of this market is far from small. In North America and Europe alone, top teams spent tens of millions of USD on transfer activity during the 2026-2026 period. A leading mid-laner can earn a salary of hundreds of thousands of USD per year, plus performance bonuses. When that much money is on the table, the need for reliable data becomes urgent. But precisely when demand is highest, the data supply is thinnest.

I began noticing this in 2026, as an intern at a sports data company assigned to track the PPDA metric in a World Cup match. That approach — using data to refute prejudice — followed me into esports. I realized that in esports, people usually judge players by what is measurable: KDA, maximum minions per minute, teamfight participation rate. But the factors that decide victory or defeat often lie beyond measurement.

Analysis: what data leaves behind

Take a concrete example. During the 2026 winter transfer window, several top European teams signed players based on high KDA. On the surface that was reasonable: KDA reflects individual performance. But when I tracked these players' matches after they changed teams, their performance dropped by an average of 18 percent over their first ten games. That number was not in any transfer file. The reason is simple: KDA is measured within the old team's context, where teammates shielded them and tactics revolved around them. When the context changes, the metric loses its original meaning.

The same happens with training data. Teams record reaction times, but few publish how many times a player changed their decision midway. Those "changed my mind" moments — when a player intends to push a lane and then retreats — go unrecorded because no tool defines them. They are invisible, even though they are the clearest sign of the ability to read a match.

I tried to measure that gap myself. Throughout the 2026 summer split, I reviewed 40 matches of one team and manually counted how many times the mid-laner withdrew from a lane push after checking the map. The result: the winning side averaged 7.4 proactive retreats per game, the losing side only 4.1. Nearly double the difference. No platform publishes this metric. It exists only because I chose to sit and count.

Transfers are a market, and a market has no emotions — only liquidation value and investment value. But the esports market runs on a paradox: it prices players by the easiest metrics to measure, while their real value lies in the hardest qualities to measure. A player's ability to communicate, to stay calm in a big teamfight, or to lift teammates' energy — those have no unit of measurement. And what cannot be measured is often priced at zero.

I witnessed this in a player evaluation session. A team analyzed two candidates for the same position. Candidate A had superior numbers on every statistics table. Candidate B was lower on most metrics, but when the coaching staff listened to internal communication recordings, they found B was the main coordinator in rotations. The data did not show that, because no one transcribed and encoded the communication into statistics. In the end the team chose B. Three months later, the team won the regional title. Candidate A moved to another team and left no mark.

There is another dimension I call "derived data." Every published metric drags along other metrics calculated from it. Minions per minute leads to minion difference; minion difference leads to lane pressure predictions. But when an underlying metric is measured wrongly, the entire derived chain downstream skews with it. I once saw a match-outcome model built on minion data that had been mislabeled throughout an entire tournament. The model produced mathematically sound results that were meaningless in reality. An error at the root never corrects itself as it spreads to the branches.

Top teams now employ an average of two to four full-time data analysts, a number close to zero in 2026. But their role is often limited to opponent preparation rather than shaping long-term strategy. Data is used to answer the question "how does this opponent play," rarely to answer "how should we build our roster." That is why data gaps in the selection process persist: no one is assigned to fill them.

In one specific transfer deal I followed, a team paid a large sum to buy out the contract of a player with outstanding individual achievements. The contract included a release clause, and the buying team had to pay extra fees if the player hit certain performance milestones. The structure of those clauses shows both sides knew individual metrics do not guarantee collective success — otherwise, there would be no need for risk-hedging provisions.

Contrarian angle: more data does not mean more understanding

The esports analytics industry is chasing a belief: the more data, the more accurate the decision. I once believed it. But after years, I realized that belief has a logical hole. Data can only answer questions we already know how to ask. When a data collection system is designed, it has already inadvertently excluded what its designers did not think of. Adding more data to such a system only thickens the layer of the already known; it does not open the unknown.

I went through this shock at Euro 2026. My xG model predicted France would win thanks to Mbappé, but Spain — with a lower xG — took the title, powered by Yamal's breakout at just 16 years and 362 days. My model ignored the variable of exceptional individual talent and the uncertainty of football. On that final night, I wrote a self-critique, admitting the model was right about the data but wrong about the match. The piece was controversial, but it taught me a lesson that applies to esports too: the best model is not the one with the most variables, but the one that knows which variables it cannot grasp.

In esports, this is even clearer. Metrics such as gold difference at the tenth minute or objective control rate are very useful for evaluating a match that has already ended. But they are useless for predicting the next game, because each game is a new context with a new champion composition and a new psychological state. Behind every shot that hits the crossbar lie thousands of data points whispering that no one has the patience to hear. In esports, that shot may be a failed play, and those thousands of whispering data points are the small decisions that go unrecorded, accumulating into the final result.

There is one temptation I must actively resist: choosing data that favors a conclusion I already hold. When I believe a player is good, I easily find metrics that support that belief and ignore those that refute it. The only way to counter this is to force myself to cite at least one opposing data set in every analysis. Otherwise, I am merely decorating my prejudice with numbers.

A forward-looking reflection

Absence is also data. When a statistics table is empty, when a metric does not exist, when a data stream cannot be retrieved, that is not meaningless silence — it is a signal of something not yet seen. Esports will mature not when it measures more things, but when it learns to read what it does not measure. I do not commentate on esports. I read esports through charts, and through the gaps between charts. The question for next season is not which team has the best metrics, but which team best understands what its statistics table leaves behind.