The NRC's Abuse of PRA

I have received considerable pushback on the PRA piece, from some of the insiders in the choir. This is deserved since I did not make myself clear. Lou Lanese correctly points out that the goal of reliability analysis is to identify high probability paths to a casualty. He absolutely right, and it's laudable goal. I have used reliability analysis and event trees throughout my engineering career. I have balanced submarine engine rooms using these techniques. I argued for twin screw tankers using these techniques. Used with judgement in situations where you have decent component failure data, they are valuable tools.

But that's not what the NRC does with PRA. What the NRC does is claim they have computed the probability of a release by using PRA. That's a lie.

At best they have identified a few of the higher probability paths to a release. (And that assumes all of the probabilities in every event chain that they examined are realistic.) But there are literally an infinite number of paths to a release. No matter how many paths you identify, you can always find one more. That's what I meant by a fractal bush.

Even if all these other paths actually are much lower probability than the paths in the tree, there are so many of them, that it's far more likely that the next release will be from one of those paths, and not from one of the paths in the tree. That's why all the releases to date have resulted from a chain of events that was not in the PRA tree. That's why no sane person will take me up on the $10,000 bet.

Lou’s right. The problem is not reliability analysis. The problem is the NRC's abuse of it.