But some of those steps will go under the median as well. The problem is the distribution is not normal -- if the range of possible outcomes below the median n

But some of those steps will go under the median as well. The problem is the distribution is not normal -- if the range of possible outcomes below the median n is [0,n], the range of possible outcomes above the median is far wider than [n,2n]. Going by the chart the article offered, I suspect a sort of mixture distribution: the better understood tasks have normal distributions, and the poorly understood ones have bullshit estimate distributions.