Yes, that appears to be a mistake. In the 36.8% of cases where the best candidate shows up in that first 36.8%, it would seem the algorithm must keep running until it is forced to select the final candidate, which is essentially a random choice out of all but the very best (assuming ordering has no relation to goodness) -- very different from being "stuck with the second best." However, the rest of the quote -- "but still, if you like favorable odds, this is the best way to go." -- might very well be true.
Presumably the optimal strategy would be to gradually lower your standards the further you go. It's easily proven that the optimal strategy for the second-to-last candidate is to select the candidate if the candidate is above average. So the algorithm as described in the article seems to be missing something.
