I don’t suppose other people will be as excited as me about statistical rigour, but I found it pleasantly surprising that 6.0 and 100 games are nice round numbers that happen to be just sufficient to say that a 6.0 club member’s performance is statistically different from 50-50 odds of winning over the same number of games.
Obviously more games are better, but at least a sample size of 100 games tells us something slightly meaningful. If you want, it would also allow each step of 1.0 above to be a tier level, as 7.0 would also be statistically different from 6.0, etc.
To chip in about better performance measures, I also agree that the last 100 games would be a much more meaningful metric than lifetime performance, since the underlying win rate is more likely to be consistent once a player exits their initial learning period.
Assumption: Normal approximation to binomial distribution is applicable as we have more than 30 observations (N) and p=0.6 is not close to 1 or 0.
Estimated standard error = sqrt [ p * (1 - p) / N ] = sqrt [ 0.6 * (1 - 0.6) / 100 ] = 0.049
95% confidence interval = [0.6 - 1.96 * 0.049, 0.6 + 1.96 * 0.049] = [0.503 , 0.696]
–> excludes 0.5 (aka 50-50) and 0.7