Fix the problem by excluding the distribution from the qa checks. The problem
why the test fails is a low statistic in the distribution such that the mean
and the errors fluctuate strongly.