Sometimes especially when test sample is not that huge a few very engaged players can alter the picture entirely. Is it possible to filter 2-5% most engaged players to get more median results?