In my opinion, Zenodo is the safest option. But CRAN offers a sort of long-term persistence: the sources of a package remain available even after it has been archived.

You can try with_rng_version from withr.
withr.r-lib.org/reference/wi...

Sorry, that's not glamorous, but start by taking a serious course on linear modeling with applications in #rstats.

Thanks for reporting for evolution. I agree that aggregating by paper is also important to get an idea of the absolute usage.