Mersenne: Probabilities in prime95 v17

[Cornelius Caesar]

I just got 3 numbers assigned in the 1.5 M range using the new version 17
and am wondering how the "chance of finding a new Mersenne prime" with
double checking is computed.

For me, it reports "1 in 893774" so obviously it's a slim chance :-),
probably based just on past experiences of error rates for double check
results (I remember George having reported something like 1 in 400 results
being wrong).

Is this true?

Regards,
  Cornelius Caesar