For a low dimensional space, I think their diagrams make sense. Like, when working with large industrial machines factors that effect safety are probably how close you are to the machine and how fast everything is going and with what urgency.

Even here they have a section on how the safety performance boundary is fuzzy and dynamic.

I wonder though what things look like with super high dimensions. When there are a 100 different things that go into whether or not you're being safe. That boundary's fuzzy and dynamic nature might extend clear across the entire space. And the fact that failures happen due to rare occurrences suggests that we're not starting at a point of safety but actually starting in a danger zone that we've just been lucky enough not to encounter failures for.

100% unit test coverage comes to mind (even for simple getters). Where some might see a slide towards danger as the coverage goes down, another sees more time to verify the properties that really matter. And I don't see why we can't get into the scenario where both are right and wrong in incomparable ways.