One pattern I've noticed: the urge to optimize often spikes right before shipping, not because something is genuinely broken, but because shipping feels scary and tweaking feels safe.
A useful reframe I heard somewhere — "done is a feature." The version that exists and can be iterated beats the perfect version that ships next quarter.
For me the trigger is user feedback lag. If I can't get real signal in the next 48 hours by shipping now, the optimization might be solving a problem that doesn't exist for users. Ship, observe, iterate.