LYON swept G2 Esports 3-0 at MSI 2026. A week earlier, that same G2 roster had eliminated T1. Both results were reported as upsets, and only one of them should have been. Anyone reading the volatility numbers instead of the win rates saw the second one coming.
This is the gap between watching esports and modeling it. A win rate compresses a whole split into one figure and assumes a team performs the same way in game one as in game five of a bracket series. Teams do not work like that. Some are dull and stable and win what they are supposed to win. Others swing between demolition and collapse, and the swing itself is the signal. Below are the three measurements analysts actually run.
1. Variance in gold difference at 15 minutes
Take a team's GD15 across a split and calculate the standard deviation, not just the mean. A roster averaging +800 with a deviation of 400 is a different object from one averaging +800 with a deviation of 1,600. The second team wins some laning phases by 2,500 and loses others by 900, and the season summary shows them as identical.
G2 in the 2026 Summer Split were the second profile. Their early leads oscillated harder than any other LEC playoff team, which is the pattern that produces a T1 upset one week and a 0-3 the next. High mean, high variance, and the variance is what kills you in a Bo5 against a team with film. The compilers behind the odds and match previews at
source:https://win.gg/esports-betting/ treat this spread as a primary input, because it tells you how much of a favorite's record is repeatable.
2. Draft dependency
Measure how much of a team's win equity is generated before the game starts. Sheep's coverage of the
G2 win over T1 traced how much of it ran through comps G2 had prepared for that specific matchup. Preparation is a genuine edge. It is also a perishable one, because it decays the moment the next opponent has tape. LYON had tape.
The practical test: compare a team's results in games one and two of a series against games four and five. A prep-dependent roster front-loads. A fundamentally stronger one does not.
3. Closing rate from behind
Count the games a team won after trailing at 15 minutes. Teams that only win from ahead are volatile by construction: their result is decided in the first quarter of the game, and they have no second gear when the plan fails.
Hanwha Life Esports beat both LYON and BLG on the way to the MSI title, winning multiple games from a deficit. That is a low-variance signature, and it is why their
Worlds qualification path is safer than a raw record suggests.
Running the model on the Esports World Cup
The EWC group stage begins July 15 in Paris, and the format punishes high-variance rosters. Best-of-ones remove the adjustment window that lets a volatile team recover after dropping game one. G2 open against FURIA. BLG, straight off the MSI final, are drawn with LYON, JD Gaming and LOS.
Volatility does not tell you who wins. It tells you how much confidence the prediction deserves, which is the more useful output. On paper, G2 are the stronger side in most of their Bo1s. On the numbers, they are also the roster most likely to hand one away.
