Heuristic advocated by Stephen Margheim and Maciej Rząsa. Rather than diving into a full fix, take 10–30 minutes to explore a performance problem and estimate the leverage: 'could I spend 2 hours and save a second here, or 5 days to save 100ms?'. Then optimize the 3 highest-leverage places. Maciej's variant: publicly declare a time box (half a day / a day), ask the team to yell at you when time is up, or iterate one hypothesis per week/sprint to keep the work bounded and endless optimization at bay.