Jim Kingdon wrote

In the 1990's they increased shuttle reliability (on paper) from 98.9%
to 99.7% (or whatever the exact numbers are, I'm referring to some
studies at least one of which was done by SAIC and showed one in 248
probability of an accident).

Checking previous references to this report, the number was originally
given as 1 in ~140, with 1 in 230 and 1 in 76 being the 5% confidence
bounds. (I think there were later reports that did move the central
figure up by a factor or two or so as a consequence of Shuttle
improvements.)

It's interesting that the as-flown Pa is now running not much outside
the lower end of the range.

One illustration of the difficulties is that that both shuttle
accidents to date were caused not by a random, anticipated event, but
by something which was known, just not fully appreciated.

Yes, a feature and limitation, though not a fault if properly
understood by the users, of the PRA method is that it depends on
identifying relevant fault trees and estimating the probabilities of
the events along each tree.