I'd just about gotten around to acknowledging that nuclear power seems like the least-bad option for energy generation, given that all other techniques either are massive CO2 emitters or don't work at scale. Prominent environmentalists like Stewart Brand agreed. But if Japan can't manage to build nuclear reactors that can withstand perfectly predictable natural disasters like earthquakes and tsunamis, I may have to change my mind again. I mean, it's one thing for a plant like Chernobyl to blow up, given that it was poorly built and operated, but Japan is supposed to a wealthy, technically competent, and not-very-corrupt nation, and one with a good historical memory and functioning social system to boot.
The answer apparently is that the techniques used to design for risk are just completely broken; and do not account for the fact that a single event can cause multiple problems; they assumed independence of risks that were in fact highly correlated. This is amazing to me, but oddly the exact same mistake seems to have been made by financial risk modelers.
I don't know why I'm surprised at this kind of stunning mistake. I guess at some level you assume that people capable of constructing and running a nuclear plant would also be capable of basic probabilistic reasoning; but; but it is never safe to assume competence, especially in the face of economic pressures.
[update: hm, here's a very detailed post by someone who sounds like he knows what he's talking about that says that basically things worked according to plan; despite some failures of some stages of containment and some safety procedures, there was enough backup to ensure that minimal radiation was released. OK.]