[afarrell]

> never going to get you to 100% reliability

Both Atheist and Muslim SREs agree: Only god is 100% reliable.

Reasoning in non-absolute magnitudes is more effortful but usually more effective.

If 2 weeks of effort spent on the single layer would get it from 95% to 95.3% reliable, then you're likely better off with another layer. If 2 weeks of effort spent on the single layer would get it from 95% to 99.9% reliable, that seems like a wise choice. However, since your mental process for judging the reliability of a single layer is probably less than 99% reliable, adding another layer helps protect against unknown errors.

> So isn't fixing bugs in existing C codebases just throwing good money after bad?

I'd agree, with this train of thought:

The author says "in C, writing reliable software is somewhere between extremely difficult and impossible." To me, that sounds like writing in C is 2%-10% reliable. If writing in rust would be 40-80% reliable, that is a powerful argument for incrementally porting something to rust if that thing will continue to need to change.