The LCK Stat Sheet Cannot Measure What Decides Matches — and the Transfer Market Is Paying for That Blindness

In June, I sat in an analysis room in Seoul and watched a stat sheet get printed and taped to the wall. On it, the team's jungler ranked second in the league for gold per minute, held a 78% kill participation rate, led the team in wards per minute, and survived more than 60% of teamfights. The sheet's conclusion was that this was a jungler with almost no weaknesses.

Three weeks later, the team lost five of seven matches. The jungler was moved to the bench, and in the press conference the head coach said something I wrote down verbatim: "He has every stat except the stat we needed."

That sentence describes a structural flaw in esports analysis precisely. Official statistics measure the result of an action but not the reason the action existed.

How the stat sheet is manufactured

Most stat sheets viewers see on broadcast come from a single source: event data the publisher supplies through an API. The publisher logs every kill, every tower destroyed, every major objective taken, every death. A third party takes that package, runs it through a fixed set of definitions, and outputs numbers that appear neutral: KDA, gold per minute, damage per minute, kill participation, vision per minute.

The problem is that every metric is a division, and the denominator is never shown. Gold per minute divides by total match time. Kill participation divides by the team's total kills. Vision per minute divides by minutes on the map. Because those denominators are not published alongside the numbers, a reader has no way of knowing what a number is actually measuring.

The consequence is that the entire official statistical system rewards presence, not correctness. A jungler who shows up to every fight will post a very high kill participation rate, even when his presence in the bottom lane cost the team a Herald in the top lane. The stat sheet records that he participated. It does not record the price.

I am not saying this to dismiss official data. I am saying it because for years I began with those numbers, and then discovered they were only the outer paint of a building I had never entered.

Why I don't trust ready-made sheets

In 2026, as a middle school student in Busan, I kept the habit of hand-tallying every pass in domestic football matches. In a match between Busan IPark and Seoul E-Land on July 12, 2026, I counted 412 successful passes for the home side, while the official sheet reported 389. A gap of 23 passes, roughly 6%. I posted the comparison to a small forum and got two kinds of response: people who thought I had miscounted, and people who thought my definition of a successful pass differed from the provider's.

Both were right. And that was the first lesson: a correct number can still be a polite lie if it is separated from how it was produced.

Since then I have kept raw data from nearly 50 matches, not to prove myself right, but to have a second frame of reference. When I moved into esports and started working with replays, I applied the same method: every number on broadcast gets checked against data I pull from the match record myself.

The result repeats almost by rule. The official sheet is always numerically correct. It is only wrong in meaning.

Autopsy of a jungler: four variables nobody publishes

Back to the jungler in that Seoul room. To understand why a near-perfect stat line led to a benching, I had to extract the records of all seven losses and rebuild four other variables.

The first is a pressure index. I define it as the number of times a team proactively creates contact with the opponent, divided by total defensive actions in the same window. This framing resembles football's PPDA, which measures how aggressively a team presses rather than the outcome of pressing. For this jungler, his individual pressure index during the laning phase was 4.1, among the lowest in the league. He did not create pressure. He appeared after someone else had created it.

The second is objective-window distribution. I split a match into 90-second windows around each major objective spawn and calculated the share of objectives a team secured in each window. This team took 68% of dragons but only 29% of Heralds through the mid game. The aggregate broadcast number read "51% objective control" and looked balanced. But 51% is the average of two extremes, and it is the gap between the extremes that opponents exploit.

The third is intra-team resource spread. I measured the gold gap between the richest and poorest player on the same team at minute 15. On winning teams this spread averaged 1,850 gold. On this team across its seven losses it averaged 3,400. The jungler ranked second in the league for gold per minute, but much of that gold came from areas that should have belonged to the mid laner or the marksman. He grew richer while his teammates grew poorer. The stat sheet calls that individual performance.

The fourth is vision latency. Wards per minute only says a player placed wards. It does not say whether they were placed before or after the opponent moved. I timestamped every ward and cross-referenced it with when the opponent appeared in that area. For this jungler, 61% of wards were placed within roughly 12 seconds after the opponent had already left the area. Technically he placed the most wards on the team. Strategically, most of them recorded only the past.

Four variables, and the picture inverts completely. A player the official sheet described as flawless, examined through raw data, was a resource drain: slow to create pressure, misaligned on objectives, and late on vision.

The contrarian angle: correlation is not causation

This is where most esports analysis collapses, and I have to remind myself of it every time I hold a dataset.

A player with high damage per minute is not necessarily a good player. He may simply be playing on a slow team that funnels resources and extends games to optimise teamfights. A team with a high teamfight win rate does not necessarily have good coordination; it may simply be choosing fights only when it already holds an item advantage. A jungler with 78% kill participation is not necessarily a good connector; he may simply be standing near fights other people created.

In statistics this is the confounding-variable problem. In esports it is a system problem. The same metric carries opposite meanings inside two different systems. That is why I refuse to draw conclusions from a single metric, however convincing it looks.

Another example sits in how people are valued. Current transfer models assign very high value to young players with good numbers on weak teams. The logic is easy to follow: strong numbers in a bad environment imply upside. But that model ignores a variable that cannot be quantified: the ability to integrate into an established system. A player accustomed to receiving 30% of his team's resources will hit a structural wall when he joins a team that distributes resources along a different axis. His numbers do not change. His role does. And the market usually recognises that about six months later, after the contract is signed.

The reverse also holds. Players with average numbers but high five-on-five fight win rates are undervalued, because no column on any public sheet reads "made the right decision in 0.4 seconds." That value is very real on the map. It does not exist in any public dataset.

At league level the same problem scales up. Since 2026 major leagues have used franchised models with fixed slots, and the initial fee for a slot has been recorded at eight figures in US dollars. Since 2026 some leagues have applied salary caps with exceptions for long-tenured players. Those mechanisms were designed using data. And the data used to design them is precisely the kind I have just shown cannot measure what needs measuring.

Signals for the next cycle

Recent large international events, with total prize pools exceeding 60 million US dollars, have pushed money into esports at a level the industry has never seen. When money moves faster than data quality, valuation error grows exponentially.

In the coming split I will track two specific signals: the individual pressure index of junglers during the laning phase, and intra-team resource spread at minute 15. If those two diverge from the official sheets by a large enough margin, we will have quantitative evidence that the market is paying for something other than what decides matches.

Every metric leaves an ink trail if you take the trouble to follow it. The trouble is that most sheets are not designed to be followed — they are designed to be skimmed. And when a number is skimmed often enough, it becomes true simply because nobody checks. If your team is valuing a player off a broadcast stat sheet, you are valuing him off something even its own creators would not defend in court.