I think one of the harder things about working on RNG heavy arcade/run based games is letting go of the desire for every run to be consistently-ish difficult. Because these games are experienced over a ton of runs, the inconsistencies will average out
Actually it's not just difficulty, consistent depth is an even better way of putting it. Sometimes encounters will be exercises in frantic, dynamic decision making. Sometimes spamming 1 move will be just what you need. It's ok to let both coexist side by side across diff runs
Though I guess one of the other issues is that if it doesn't average out, that can seriously underesell games on yt replays and stuff which is a factor nowadays. Cause people are only seeing the run where everything went as planned, rather than the ones that devolved into pure chaos
Something like that is quite difficult to do because it's scary to not be able to predict how your own game will play out, even though it's very worthwhile if you nail down the execution of RNG elements.
be like Garegga where mad ball can just decide to never go low enough for a full 900k bomb until like 12 attacks later woohoo