I don't know enough here to comment on who's right, but when the magnitude of catastrophic failure is large enough, model error is appreciably important. So in those cases past experience is a very important input into the decision, even if not perfectly correlated.
If anything, it proves that identifying all failure modes is challenging, and theoretically safe is not the same as practically safe.
"So in those cases past experience is a very important input into the decision, even if not perfectly correlated."
Of course it is, which is why modern reactor designs have incorporated safety features based on these accidents and older designs have been retrofitted (with some exceptions that are legitimately concerning). It's entirely true that it is impossible to predict every mode of catastrophic failure, but that does not mean it's impossible to create designs that are resilient to unplanned disasters. No type of power plant can be perfectly safe, but
for assessing practical safety records, in terms of deaths / TWh generated, even estimating conservatively nuclear power is safer than any other source of power (including wind and solar). Some references for this:
Hydroelectic as well technically. A set of hydroelectric dam failures in China in 1975 killed more people directly from the event (171,000) than is estimated for the direct and future deaths caused from Chernobyl and Fukushima combined.
If anything, it proves that identifying all failure modes is challenging, and theoretically safe is not the same as practically safe.