> "What most ordinary users want is that it ALWAYS works."
> Uh, I'm not sure about that. From what I've seen, most ordinary users want something that's 'good enough'
They want something that ALWAYS works. However, what they're willing to pay for is the 'good enough'.
The "last 10%" to make something that seems to be working, to something that actually works on the real world, for large set of real users (different environment, use-cases, hardware etc), takes 90% of the effort. In proprietary SW it would show in the price & being late.
As it's also in practice impossible to cover all this with tests, what you typically get is "good enough" SW release which is then iterated with later releases, at least in the case of SW that has many millions lines of code worth of functionality and hugely varying uses & use environments.
I think AT's "ALWAYS works" comment was about SW working in a fixed (i.e. academic) test setup.