And the truth is most systems of any kind seem to fail from simple things that everyone could have prevented but overlooked. Wasted engineering. Best to think like the Arpanet was designed. Have a way to heal once things are normal, not work around the bizarre failures.
I've spent a bunch of time in light manufacturing. In testing assemblies I've come to the conclusion that for ordinary stuff you just want to reduce the rate of 'bad product' below the rate of installer fuckups. If 9 times out of 10 the reason is failed it because someone fucked it up and 1 out of 10 is because you shipped a defective unit, no one cares.
You might go, well what about safety critical things. Well it's still true. Just at a much lower level of failure.
