One in 700 000 years (posted the link before, so it is sourced as per thread rules). What sort of culture one must have to declare such a number...
Let's say, I am 'fearful' enough to think that there is >=1/10 possibility there is yet unknown 1/10 000 years event or mechanism. That gives risk of 1/100 000 years or worse.
Let's suppose that I am 'paranoid' enough to think there is a >=1/100 probability of management failure / corruption, leading to neglect of 1 in 100 years event (what seem to have happened in Japan). That is 1/10 000 years or worse. It seems that not even very slight doubt in the completeness of our knowledge, or a slight mistrust, is compatible with such figure as one in 700 000 years.