OTOH if one is designing an ISA that will have a bunch of different implementations - and this includes later implementations wanting to be ASM compatible with earlier ones - they tend to eventually become a footgun for the processor designers. (if I remember correctly and didn't completely misunderstand, MIPS' branch delay slots were absolutely a neat optimisation for the early models, but when they went to a deeper pipeline for later chips required a bunch of extra design effort to maintain compatibility with, without being helpful anymore)
(explicit disclaimer that I'm an armchair amateur here, so if you're a fellow non-expert reading this comment before it attracts better informed replies please default to joining me in the assumption that I've made at least one massive error in what I'm saying here)
I'm sure there are some bad decisions in RISC-V, but I've yet to see one that isn't in the process of being remedied. There was a good argument for a lack of POPCOUNT instruction being bad, but I think that's being added soon