A classical mathematical mistake - i.e. to impose a Gaussian distribution on a population sample known to be non-Gaussian - whether it‘s 2% or 6%, you cannot pre-define how many people are underperformers (or overperformers, for that matter).