After finding out that trig/transcendental was basically not guaranteed to be equivalent across kernels (libc/musl), which caused the dreaded “only fails in CI” problem for me when I was trying to generate nonflat distributions of drng’s, I ended up creating https://github.com/pmarreck/random to solve it, which it did
Curious why you used Box-Muller for normal PRNGs instead of Ziggurat.