If you’re not sure something is getting better, it’s not
Good heuristic, especially in fat-tailed domains where small differences in performance don’t really matter. In such cases, if it’s not obvious you’ve made a step function improvement, you either haven’t improved or haven’t improved sufficiently for it to matter.
References
Transclude of Today,-@Foundersfund-Led...#6fb3f7
The Guerrilla Guide to Interviewing (version 3.0) – Joel on Software
Transclude of The-Guerrilla-Guide-to-Interviewing-(version-3.0)-–-Joel-on-Software#461c9c