wow what? ffs, how 100x saving can ever come from the architecture itself? Doesn't make sense. And how much saving comes from testing smaller pieces? And the real question is what integration failures might these smaller tests miss? Very curious here!