Comment by sandbar

Take the time to speed up my iteration cycles has always been incredibly valuable. It can be really painful because its not directly contributing to determining/fixing the bug (which could be exacerbated if there is external pressure), but its always been worth it. Of course, this only applies to instances where it takes ~4+ minutes to run a single 'experiment' (test, startup etc). I find when I do just try to push through with long running tests I'll often forget the exact variable I tweaked during the course of the run. Further, these tweaks can be very nuanced and require you to maintain a lot of the larger system in your head.