Overall(last I checked), the testing is roughly equal between the stable distro's(Debian/Ubuntu/etc) and NixOS. The difference is stable distro's back-port bug-fixes. NixOS rarely does, since their release cycle is only 6 months long.
> * Alienating you from upstream: why contribute upstream if you'll only benefit from them in 2 years.
I contribute upstream, regardless of if I'm running NixOS, Debian Stable or Windows. It makes no difference to me which OS I'm running when a bug shows up. If I find a bug in X package, I go fix X package. Sure I also fix it locally in my running instance(s), but that's my problem, regardless of which OS I'm running.
> To mitigate breakages we should be aiming for better test coverage, at various build levels: class, package, program, system.
Yes, yes we should. Most software has a terrible testing story. There are very few pieces of software with robust testing. SQLite is one such. One could probably name a handful of others, but after that the list gets really hard to add to.