Simply put, he is ignoring population size. For example, imagine two populations where a characteristic has identical values (e.g. average value of 200 with a standard deviation of 30). Lets assume that the two populations differ in size (e.g. 5,000 vs 10,000). He is taking the top 1,000 from each population, that is 20% and 10% respectively, and analyzing the tail of the population using a normal distribution which is completely invalid in my mind. In his analysis, the population where the top 10% were analyzed would had a higher characteristic value than the population where the top 20% were analyzed.
He and I have both analyzed the era 11 data from Blizzard’s API. Unfortunately, the data is limited. For the top 1000 on the leaderboard, it is easy to get GR clear and its time and paragon level of player. The information on builds needs to be done manually in game and is impractical time-wise.
His better analysis is present here. You can check out the fist few posts.