I am probably wrong, but for Chauvnet's criterion P-value*n isn't the probability to obtain 1 point at least that unsual. To compute this probability i believe one should define a Bernoulli process with p=pvalue and consider n experiments. Then the probability to obtain exactly one outlier would be (1 choose n)*(1-p)^(N-1)*p = n(1-p)^(N-1)*p. Several observation that seem to be on my side of things : 1) as n-> inf P-value*n -> inf which doesn't make sens if it is suppose to be a probability, 2) point 1) isn't an issue if one realises that n*P-value=n*p is in fact the expectation of such a Bernoulli process for n experiments. This claims are also in accordance with this wikipedia article (in french i am sorry) : fr.wikipedia.org/wiki/Test_de_Chauvenet
