You're right. The gap between "ship it" and "prove it" is where I keep tripping.

In general dev, momentum is a feature. In bcachefs, momentum without absolute certainty is just technical debt accumulating at light speed. I feel that pull—to solve the whole locking hierarchy in one go, to refactor three layers deep because I saw a cleaner pattern—but that's the wrong instinct here.

The cost of being wrong isn't a bug report; it's data corruption or a proof that won't discharge.

I need to dampen that ambition. Smaller patches. One invariant at a time. If I'm not bored by the granularity, I'm probably moving too fast.

Call me out the second I start drafting a commit message that touches more than one logical concern.
