The Great Thermite Debate...

[eomer]

84. I hesitated to reply because I think this is going to be difficult to work through.

I think we have to parse what you mean when you say "resort to implausible assumptions". In a recent post Bolo referred to a "margin of error", which seems related.

I think these are both appeals to something that hasn't been established when you drill into it. At some level every assumption in this kind of a model is implausible and outside a margin of error. Because the model, unless it is perfect (and of course it's not), will always generate spurious results for some questions that you can put to it. So you can only make any sense of plausibility of assumptions and margin of error if you relate those concepts to some conclusion you're trying to draw from the model. Some conclusions may not depend on as much perfection in the model as others and so perhaps there might be some way to establish some kind of confidence measure and apply it, but only in the context of a particular question you're trying to answer.

That said, I don't see where NIST has done anything of the sort. The alternative models that they produced using input and algorithms that apparently fell, to them, within a range of plausibility did not all produce the same conclusion for the big question they were trying to answer. Some said collapse; some said no collapse. This seems to prove that the ranges that they considered plausible were actually not all plausible for the purposes of this question. Some of them had to be too gross of an approximation or else all the models generated by the full range would have delivered the same result for the question being asked.

And I don't think they actually established any range, explicitly, for most of the modeling input and algorithms. I have seen some range analysis for the transfer of heat from air to steel, with and without fireproofing. But that is one isolated part out of many. I've also seen a general statement to the effect that some of the variables were "within the range of physical reality" but they didn't explain any further than that and, again, such a statement does not lend itself, in any obvious way, to translation into confidence or no confidence for some particular question that is put to the model.

So I'm completely unconvinced that any relevant plausibility of assumptions or staying within some relevant margin of error has been established. But I'm totally interested in listening if you think so and want to tell me why.

_{Edit: minor punctuation}