Post date: Feb 23, 2014 8:24:20 PM
Okay. It turns out that we can pretty easily compute best-possible bounds for the bridge reliability under independence assumptions and p-box (or interval or distributional) inputs.
The Frechet case that makes no assumptions may not even be too hard; there are multivariate versions of the Frechet inequalities, and I think multivariate versions of Williamson and Downs are also possible. There's a paper on the subject (although not under these topics) by some French-speaking Belgian economists at Université catholique de Louvain, if I remember correctly.
Scott