@yes @fristi Unit tests suck, you get better mileage the more you try to maximize coverage from functional/integration tests, especially if you do fuzzing (never do it if you can make a robot do it) and replays of production traffic.

This pisses off the unit-testing cult, because any process cult gets furious when you point out that their process is a net drag. But there's no getting around it: if something breaks a unit test without changing the output of the system when the entire stack is tested, then the unit test didn't matter. Conversely, replays of production traffic will show you really quickly if something that you do has broken something that is actually in use, and you will find out in a hurry if something you've done has broken something important or slowed it down significantly.