The Empty Arena Summer of Esports: Valuing Players With Numbers That Refuse to Lie

At the 31st minute of the deciding game in the playoff round, the home team's mid laner misclicked a decisive skill. The arena fell silent for exactly two seconds, then the cheers erupted from the away side. I stayed behind after the match and rewound that footage seventeen times. Not to admire the play, but to count. From the moment the teamfight broke out to the moment the skill was misclicked was 0.42 seconds. I wrote that number in my notebook, then reopened the same player's game from the same stage two years earlier: 0.37 seconds. A difference of 0.05 seconds. In a discipline where teamfights are decided by thousandths of a second, 0.05 seconds is the distance between saving a teammate and losing a turret. And in a transfer window, 0.05 seconds can be the distance between a six-figure contract and a seven-figure contract. I am writing this piece seven weeks after the season's transfer window closed. There are no official matches. There are no spectators. There are only contracts being negotiated behind the scenes, closed tryouts with no cameras, and thousands of hours of recorded scrims that no one is allowed to publish. This is the period when most fans think the discipline is asleep. But for someone in the valuation trade like me, this is when data falls drop by drop. Before getting into the numbers, I need to rebuild the context. For many years, the esports transfer market has run on feeling. A player shines in a short tournament, a highlight play spreads on social media, and immediately their price spikes. Teams sign contracts based on memories of beautiful moments, not on a probability distribution of what that player will do over the next twenty months. I have seen this repeat often enough to treat it as a rule rather than an accident. What makes esports different from football is that it has more data but less reliable data. A football match lasts ninety minutes and generates about a thousand recorded touch events. An esports game lasting thirty minutes can generate tens of thousands of events: every skill cast, every movement, every damage exchange is recorded at the frame level. That sounds like a data person's paradise. But that very abundance creates a trap: we think we have more information, when in reality we have more noise. I learned my first lesson about this when I was a content writer for a sports data company. I was twenty-three, freshly out of the Journalism and Communication program in Berlin. I published an analysis using expected goals to argue against a Bundesliga club sacking its manager. The editorial board called me naive. Five matchdays later, that club earned eleven points and survived relegation. A year later, at a World Cup, I pointed out that a national team's PPDA was at a disastrous level and predicted they would be eliminated in the group stage. The result confirmed it. From then on, I abandoned the match-feel style of writing entirely. Every analysis of mine must be tied to a verifiable metric. When I moved into esports, I carried that discipline with me. And I discovered that it is far harder to apply, for three reasons. First, esports patches constantly. Every update can upend the value of an entire role. A metric built on an old patch can become meaningless after a single patch. In football, the offside rule changes once every few decades. In esports, the rules of play change once every few weeks. Second, small sample sizes. A player only plays a few dozen official games per year. If you filter for statistically meaningful games — balanced games, against opponents of equal caliber, unaffected by an early snowball — the number can drop to just a handful. Building conclusions on a handful of games is building a house on sand. Third, and most seriously, most public esports data measures outcomes, not process. We know how many kills a player got, but we do not know how many correct decisions they made that went unrecorded. We know which team won the teamfight, but we do not know who called the shot, who chose the timing, who actually created the advantage. Because of these three reasons, I built a tool I call the decay coefficient. I developed this concept back when the football season froze due to the pandemic, when I sat down to watch an entire season's matches and realized that without spectators, home win rates dropped sharply. One club famous for its wall of supporters lost more than sixty percent of its points compared to when it had fans. I turned that finding into a forty-page report, and it was that report which moved me from a pure writer to a valuer. The decay coefficient measures how a player's or a roster's form declines over time, like a decaying physical quantity. For esports, I measure three things: reaction speed, per-minute laning efficiency, and early-game teamfight win rate across patches. These three do not decay at the same rate, and that is the crux. Let me start with reaction speed. This is the metric every fan can feel but few can measure. In a game, reaction shows up in the interval between when a threat appears and when the player responds. For a nineteen-year-old player, this number typically sits around two hundred milliseconds. For a twenty-seven-year-old, it typically rises by a few tens of milliseconds. It sounds small. But when you accumulate it across hundreds of teamfights in a season, you find it turns into dozens of lost fights whose cause no one notices. I built a simple model to illustrate this. Suppose a player has a base reaction of two hundred milliseconds at nineteen, and their reaction rises linearly by four milliseconds per year. At twenty-three, they sit at two hundred sixteen milliseconds. At twenty-seven, they sit at two hundred thirty-two milliseconds. An opponent in the same role at twenty-one would have a sixteen-millisecond edge in every direct confrontation. In a game where a decisive skill is cast within roughly four hundred milliseconds, sixteen milliseconds is four percent. Four percent sounds negligible until you remember that the win rates of top teams are often decided by margins under five percent. Of course, reaction is not everything. Experience compensates for speed. An older player reads the game better, positions more wisely, and rarely falls into situations requiring a rushed reaction. This is why the decay coefficient of reaction does not equal the end of a career. But it does mean the player's role must change. They cannot keep being the one who leads the charge into teamfights. They must become the shot-caller, the tempo controller, the one who plays safe and waits. The second metric is per-minute laning efficiency. This is something public data measures fairly well, but it is frequently misinterpreted. A player with more CS than their opponent at ten minutes is not necessarily playing better. They may simply be getting resources handed to them by teammates. I always divide this metric by the amount of resources the player receives, to get a fairer measure. After normalizing, I found that many players praised for high CS were in fact just playing within a system that favored them. When that system changes, they collapse. The third metric is early-game teamfight win rate across patches. This is the most sensitive metric and the most easily misunderstood. Early-game teamfight win rate depends on whether your composition is strong early, not just on individual skill. A team that picks an early-game composition will have a high early teamfight win rate, regardless of how good its players are. If you cannot separate these two factors, you will draw the wrong conclusion about the quality of an entire roster. So how do you separate them? I use a method I call situational decomposition. Instead of measuring all early teamfights, I only measure fights that occur in neutral conditions: both teams with equal gold, equal turrets, equal kills, and equal levels. In those conditions, composition strength is neutralized as much as possible, and what remains reflects skill. When I applied this filter to a season's data, the team rankings shifted considerably. Several teams praised for an aggressive style were in fact just exploiting composition advantages. When the neutral-conditions filter was applied, they dropped to mid-table. This is where I need to talk about what I call the empty dataset. During the valuation of a recent deal, I built three data files. The first contained all of the target's official games in the past season. The second contained all the scrim games the team allowed me to access. The third contained data on direct opponents in the same role. The first two files were full of data. The third returned empty, because the target's same-role opponents had not played enough games in equivalent conditions to compare. The temptation then was to ignore the empty file and draw a conclusion from the two remaining files. I almost did. But then I remembered a principle I set for myself years ago: numbers never lie, only the reader's heart turns them into lies. If I ignored the fact that the third file returned empty, I would be turning an honest number into a lie with my own hands. So I stated clearly in the report: insufficient sample to compare same-role opponents, conclusions are for reference only, confidence interval wide. My client was unhappy. They wanted a decisive number. They wanted me to say buy or do not buy. But an honest data person does not give a decisive number when the data is insufficient. I told them: I can give three scenarios — optimistic, base, pessimistic — with a probability for each. I cannot give a single number. In the end they accepted the three scenarios. And that deal, as of when I write this, is unfolding along the base scenario. This story brings me to a concept I consider the most important in the valuation trade: a transfer is not buying a person, it is buying a probability distribution. When you sign a player, you are not buying their present self. You are buying the entire set of what they might become over the next twenty months, along with the corresponding probability of each possibility. Some teams buy the peak of the distribution. Some buy the tail. The best teams buy the expectation. And expectation, in esports, is governed by the three variables I have laid out: the decay coefficient of reaction, normalized laning efficiency, and teamfight win rate in neutral conditions. Add a fourth variable that public data barely measures: the ability to adapt to a new patch. This fourth variable is why so many deals that look sensible on paper fail on stage. A player can have excellent reaction, impressive laning efficiency, and a high teamfight win rate in the current patch. But if the next patch changes how their role operates, and they cannot adapt, the entire probability distribution collapses. I have seen this repeat many times: a player valued highly after one breakout season, then gone after two patches. This brings me to the counterintuitive angle I want to spend the last part of this piece making clear. In this industry, there is a widespread belief that young players are always better than old ones, because their reactions are faster. This belief rests on a real correlation: age correlates negatively with reaction speed. But correlation is not causation. Slower reactions do not cause the decline in form. The decline in form comes from a player failing to adjust their playstyle to match their changed reaction. A thirty-year-old who understands their limits and plays within them can be more effective than a twenty-year-old who wastes their reaction on reckless plays. I verified this on my own data. When I split older players into two groups — those who have adjusted their playstyle and those who have not — I found the adjusted group had a significantly lower form decay coefficient, even lower than the young group. In other words, age is not a sentence. How you respond to age is the sentence. This has direct implications for the transfer market. Teams are mispricing both sides. They pay too much for youth and too little for adaptability. In an empty-arena summer, when there are no matches to judge by, these pricing errors become clearer than ever, because the light of the arena is no longer there to hide them. In the empty-arena summer, I hear data falling drop by drop. Every closed contract is a drop. Every cameraless tryout is a drop. Every coaching change is a drop. The biggest signals of a season are often emitted from an arena with no spectators, and those who know how to listen will hear them before they become headlines. There is one thing I always remind myself when sitting in front of a new dataset: I do not believe in intuition, I believe in the decay coefficient of intuition. Intuition is a model that has not been verified. When you verify it, it becomes data. When you do not verify it, it becomes prejudice. And prejudice, in the valuation trade, is the most expensive thing a team can buy. Every crisis is unlabeled data. A player in decline is an unlabeled dataset. A losing team is an unlabeled dataset. A chaotic transfer window is an unlabeled dataset. The task of the data person is not to deny the chaos, but to label it. And sometimes, the most honest label is an empty one. I return to the 0.42 seconds at the start of this piece. It is not a sentence for that player. It is a signal. A signal that their role needs to change, that their team needs to adjust its expectations, that the probability distribution of their future needs to be redrawn. In this empty-arena summer, there are hundreds of such signals falling, and most of them will go unheard until the new season begins and everything is already too late. The question I leave for those in my trade, and for the fans waiting for the new season: when your dataset returns empty, do you have the courage to say it is empty? Or will you fill the gap with a plausible-sounding story, then call that story analysis? In an industry where the light of a short tournament can inflate anyone's value, the only person who can protect you from yourself is your discipline with numbers. And that discipline, unlike reaction, does not decay over time — if you are willing to forge it every day.