I just found out that the name "Secretary problem" is given to two different problems. The first one talks about a secretary who mixes letters and envelopes, and ask for the probability that no letter will be put into the right envelope: this is an application of derangements, and the limit value for the probability is $1/e$.
However, Wikipedia defines the Secretary problem as the task to choose the best candidate, if you see them one at a time and cannot keep anybody on hold. Curiously (at least for me), the best algorithm has the identical chance to find the best candidate, that is $1/e$. Is it really a casual correlation - after all, $e$ pops out everywhere - or derangements are somewhat involved?
In the comments at this answer in SO the same question was made, but nobody answered.