I took it to be: the original author's point was that nuclear attacks are so rare and our sample size so small that we can't use history to predict their occurr

I took it to be: the original author's point was that nuclear attacks are so rare and our sample size so small that we can't use history to predict their occurrence in the future. This commenter is saying the typical way to handle situations like this is to look for more-common events that are proxies for the less-common ones in some predictable way. To give a different example of my own: on an intersection-by-intersection basis, pedestrians dying from getting hit by a car is relatively rare, so if it's perceived that a particular intersection is unsafe and the design of the intersection is modified with a goal of reducing the chance of death, it can be difficult to know if that intervention worked. Supposing last year there was one death there and this year there weren't any; was that because of the design change, or just chance? But we do know that, in aggregate, some relatively constant percentage of collisions will result in deaths, and collisions overall are much more common than deadly ones, so even if the policy goal is to reduce deaths, measuring all collisions is a very useful, less-noisy proxy for determining if an intersection has been made less deadly. So: original author is arguing that we can't meaningfully reason about these things because they're rare, and this commenter is saying "yes we can, if we find a more-common, less-noisy thing that's a good proxy for the rarer thing," and proposes that nuclear close calls are such a proxy.