This can be used to justify any kind of crap being shipped in newly introduced features, merely baring regressions. Therefore I don't value the approach as pre

This can be used to justify any kind of crap being shipped in newly introduced features, merely baring regressions. Therefore I don't value the approach as presented here. But a sane variant could be interesting: anyway I suspect it will need domain specific judgment, preventing from throwing a nice universal rule of thumb to the world.