KDA 50 and 27 Deaths: Two Extremes of the Dota 2 Scoreboard and What KDA Cannot Measure

At two in the morning in my analysis room in Seoul, I pulled up OG's match record from PGL Arlington Major 2026. bzm's "Deaths" column sat at zero for forty-four minutes while "Kills" climbed to 27 and "Assists" stopped at 23. KDA 50. Across the entire body of fully recorded professional Dota 2, that figure appears exactly twice. The other instance came from a match that lasted only 22 minutes: Shirley, Bounty Hunter, 10 kills, 40 assists, zero deaths, SL i-League StarSeries Season 3 China qualifiers, 2026.

At the far end of the same scale, Vol of Blue Pikachu died 27 times across 51 minutes on Centaur Warrunner at the ESL One Katowice 2026 qualifiers. One player never went down. One player went down roughly every two minutes. Both made the record book.

Three records, two extremes, one question: what is the scoreboard actually measuring?

When the numbers do not lie, that is when my heart starts listening. This time the numbers said two contradictory things, and I had to choose a reading before writing a single conclusion.

I have followed professional Dota 2 since 2026, starting as a competitor and tournament organiser before moving fully into data analysis. My working habit is not to rewatch highlight fights but to reopen the post-match summary: KDA, match duration, deaths by minute bracket, gold distribution by role, and kill participation by map zone.

KDA is kills plus assists divided by deaths. A KDA of 50 with zero deaths means a player accumulated fifty kills or assists without going down once in the entire match. At a glance, that is the absolute ceiling of individual performance. The formula carries a structural flaw: it rewards not dying without distinguishing what role the player was holding.

The three records in front of me involve completely different roles. bzm played Ember Spirit in the mid lane, the primary damage dealer who has to dive into the middle of fights to finish kills. Shirley played Bounty Hunter, a support built around vision and gold generation through Track. Vol played Centaur Warrunner in the offlane, the initiator who absorbs damage.

One formula, three roles, three meanings. That is why I refuse to read KDA the conventional way.

There is something I do not have: the patch version of any of the three matches. No public record states the patch number. That is a hole in my own model, and I will return to it at the end.

Tournament tier is the second variable. bzm's record came from PGL Arlington Major 2026, a Major-level event where teams prepare thoroughly and rarely experiment with unusual drafts. The other two came from qualifiers — SL i-League StarSeries Season 3 and ESL One Katowice 2026. Qualifiers are usually played as BO1 or short BO3 series, preparation is lighter, and stronger teams sometimes test line-ups on purpose. The probability of an extreme number appearing in that environment is far higher than on a main stage.

Start with Shirley. Forty assists in 22 minutes works out to 1.82 assists per minute. I have never seen that repeated at professional level. Over 22 minutes the total number of fights is far lower than in a 44-minute match, so staying alive is mathematically easier. Participation density runs the other way: nearly two assists per minute means Shirley was present in almost every kill his team took. Bounty Hunter does not produce large damage. His value sits in Track — bonus gold for teammates and sustained vision on the marked target.

Shirley's record is therefore a record of density, not of endurance.

bzm is the inverse. Twenty-seven kills in 44 minutes is 0.61 kills per minute, plus 23 assists, for 50 total involvements in 44 minutes, or 1.14 per minute. Lower density than Shirley, but twice the duration. And Ember Spirit is one of the riskiest heroes to keep clean: the hero lives by diving in, burning resources, and repositioning out, and every dive carries a probability of being held in place. Zero deaths across 44 minutes on that hero is a record of endurance.

The two KDA 50 records are not the same species. One measures participation density, the other measures sustained risk tolerance over time. Folding them into a single figure is a methodological error.

Then Vol. Twenty-seven deaths in 51 minutes works out to 0.53 per minute, or one death every 1.9 minutes on average. Read conventionally, that is a disaster. But Centaur Warrunner is not picked to avoid dying. He is picked to open fights with Blink Dagger, pull damage onto himself, and free up space for the carry behind him.

I do not count deaths. I count what each death bought. A death that opens a tower, a death that pins three opponents while the team takes Roshan, a death that forces the enemy to spend gold on defensive items instead of damage — those deaths pay. A lone death in the jungle that buys no objective and hands over free gold is leakage.

Vol's record does not tell me which kind those 27 deaths were. It only tells me he died more than anyone ever has. The scoreboard books a death as a loss, when in reality some deaths are an investment. That is the largest blind spot in every current Dota 2 statistic.

One small detail confirms how hard these records are: flyfly once reached KDA 49, one point short of the threshold. Another player getting that close proves the ceiling sits at the absolute edge of possibility rather than being a statistical accident. It also shows how fragile it is: a single death erases the entire record.

The Dota 2 record board holds other names — Puppey with the most professional matches played, SumaiL with the youngest The International title, plus Illidan, Madara and TaiLung in other categories. I keep the three cases above because they form a closed comparison set: same scale, different roles, different durations.

From there I built a tool for the next match instead of only commenting on the last one. I call it the Space Index, SI. The formula has three components: damage taken inside fights, major objectives created during or immediately after the death (towers, Roshan, barracks), and the gold the opponent received from that kill. If the first two clearly outweigh the third, the death paid.

Applied to Vol: 27 deaths in 51 minutes, and if most of them fell inside team fights with an objective attached, his SI comes out strongly positive regardless of the number 27. Applied to bzm: zero deaths means SI cannot be computed this way, so we switch to a different measure — how many times he forced the opponent to spend crowd-control abilities without catching anyone.

Before every match I run a fixed five-item checklist: total sprint count, distance covered after minute 20, timing of strategic adjustments, pressure events in the enemy's third of the map, and accumulated gold at ten-minute marks. That checklist cannot measure whether a death paid, and that is precisely the gap I am trying to close.

This is why I carry a fixed analytical frame into every match I watch: without a frame each match is an isolated story; with a frame each match is another test of the model.

The frame has limits, and I have to state them plainly.

Correlation is not causation. bzm's KDA 50 does not prove bzm is better than Shirley; it proves two players met two different match structures, in two different eras, in two different roles. And Vol's 27 deaths do not prove Vol played badly; they prove he held a role where dying is part of the job.

The real blind spot sits elsewhere. I still have no metric for space — the thing Vol creates and the thing bzm consumes. Every existing statistic measures the final result, not the process that produced it.

And I have to break one of my own assumptions. I keep saying historical data is trustworthy. But Bounty Hunter had Track and Jinada reworked between 2026 and 2026. Shirley's record may be impossible to reproduce under the current patch — meaning it is data from a different game, not this one. Comparing it directly with bzm is comparing two different frames of reference.

In my world, luck is only the residual that has not been explained yet. But a record that cannot be reproduced is not luck — it is an artefact. And artefacts belong in a glass case, not in the toolkit for measuring the present.

That is why I put the environmental variable into every analysis, even when the match in question happened four years ago.

I have counted every gap on the map when the crowd disappeared. The gap Vol leaves behind each time he goes down never appears in any statistical table. So does the gap bzm never leaves at all.

I do not believe in inspiration — I believe in standard error. And standard error says one thing here: the next Dota 2 record will not come from KDA. It will come from a metric nobody has named yet — seconds an opponent is pinned, gold forced into the wrong purchase, objectives traded for a single life.

Whoever builds that ruler first reads the match before the scoreboard has time to appear. Records like KDA 50 and 27 deaths will stay exactly where they belong: in the glass case, as a reminder that we are still measuring in old units.