Each bubble is a faction: how much of the field it makes up (across) against how often it wins (up). Bubble size is the number of games behind the number, so a spike off a handful of games stays small. The dashed lines mark an even 50% win rate and the average play rate, splitting the plot into four quadrants.
Row vs column: each cell is the row faction's win rate into the column faction. Deeper green or red means further from even; washed-out cells have fewer than 10 games, so read them loosely. Hover any cell for the full record and confidence interval.
A 62% win rate matters less than a 54% one if only a handful of people play the faction. These three measures weight performance by how much of the meta a faction actually occupies.
Decided games × (win rate − field average) - wins taken off the rest of the field beyond what an average faction would have taken from the same number of games. Bars are red or blue when the swing is larger than sampling noise would explain, and grey when it isn't.
Share of players who actually had to play the faction at least once. The paler bar is what you'd
expect if pairings were random, 1 − (1 − p)R; the gap between
the two is Swiss pairing at work.
Share of top finishes divided by share of the field. 2.0 means the faction takes twice as many top finishes as its size warrants. Catches factions that look ordinary on average win rate but decide who lifts the trophy.