That's not what it's called.

Imagine you write software for rabbit-mq. But then you get a requirements change to use zero-mq for whatever reason. Is that deviance? Have you normalized it? No, and no.

Deviance is when you're dropping messages in production when you shouldn't. Normalizing is when you ignore it because "it's usually fine".

You're probably thinking that my analogy doesn't work because changing software libraries is different than accepting an out of spec part. But why? If it's out of spec, it's a different part. You can make a spec around that part and design to it all the same. Just because it wasn't your first choice doesn't mean it's dangerous.

Point is that it is arbitrarily hard to tell whether a response is adaptive. Maladaptive responses are much cheaper than (unmentioned) alternatives, so are systematically more attractive to management.

So nuclear engineers just aren't allowed to do their jobs, and the safety requirements were never good enough to begin with.

They certainly were not good enough at Fukushima, in a half-dozen particulars.

The next pile of reactors down the coast did not melt down, solely on the strength of a single engineer who succeeded, with enormous difficulty, in upholding standards management objected to, but that he knew were essential. Without him, there would now be two major uninhabitable areas there. He must have known Fukushima would melt down, and must have wanted it held to the same standard as his, but lacked authority to extend his standard so far.

> They certainly were not good enough at Fukushima

Then why would remaking the part be any better than redesigning the system?

Either course would produce uncompetitive power. The best would be stop pouring money into the rathole.

Ok, so the safety argument doesn't even matter, so why debate it as a proxy for what you really mean?

The safety argument is one of several good reasons to cancel. They do not contradict, but add.