Nice work, but one nitpick.
The underestimate of the Castle Bravo yield was not due to a math error. As you noted in the next sentence, it was due to a failure to calculate possible fusion reactions involving 7Li (or, to be more precise, reactions with what results when 7Li is in a large flux of neutrons from the other thermonuclear reactions taking place). From an engineering point of view, if they had allowed a typical safety factor of 3 above their original yield estimate, they would have had a more appropriate exclusion zone.
In contrast, the initial belief that N might undergo a self-sustained thermonuclear reaction resulted from a major math error by Teller when examining that process. The histories indicate he was prone to that sort of problem. (For example, his belief in a particular design for the Super was based on flawed calculations, and he only came up with the Ulam-Teller design after being convinced of the error.) Seemed to go along with his creativity.
You will notice that very large safety factors were built into the LA-602 calculation, ones that went well beyond the known uncertainties of the poorly known reaction cross sections. Similar things were done in the LHC study, but the problem is really one of statistics: these reactions take place all the time in the atmosphere, but the flux is low compared to the LHC. That is why they have to look beyond the earth for their extreme test cases.