Each player is assigned a composite score built from their live ranked stats. ELO is the dominant signal — it captures overall skill — but the model also rewards consistency, not forfeiting, and top speed:
ELO rating
50%
Win rate
20%
Completion rate (non-forfeit)
10%
Best time (normalised)
10%
Average completion time
5%
Low forfeit rate
5%
Players with no ELO data are treated as heavy underdogs with a baseline score of 400, equivalent to a low-coal ranked player.
match outcome
For each match, the win probability is calculated using a logistic (sigmoid) function on the score difference between the two players. The variance setting controls the spread — low variance means the higher-scored player wins almost every time; high variance introduces more upsets by flattening the probability curve.
predictions
The predictions tab runs 10,000 full tournament simulations. In each one, every match is resolved probabilistically — the favourite doesn't always win. After all 10,000 runs, each player's percentage for a given round is simply how often they reached that round across all simulations. This gives a distribution rather than a single outcome, and captures how variance compounds across 10 rounds.
simulate tab
The simulate tab runs a single tournament from start to finish. This is one sample from the distribution — a plausible outcome, not the most likely one. Upsets are highlighted in orange. You can re-run as many times as you like to explore different possible outcomes.