Love the rewrite, @iitalics . I've only just now managed to read it fully, learning more Racket on the way (parameters and parameterize).

There is an issue with random/io where it will usually give the same value repeatedly. This is because it recreates each time and Racket seeds it with a function of the current time (down to milliseconds). Usually, two consecutive draws are at the same time and therefore the same seed. When I run this code, I tend to get runs of four or five identical integers:

(main (do
        {random/io >>= {print . ((flip ++) "\n") . show . (+ 0)}}
        {random/io >>= {print . ((flip ++) "\n") . show . (+ 0)}}
        {random/io >>= {print . ((flip ++) "\n") . show . (+ 0)}}
        {random/io >>= {print . ((flip ++) "\n") . show . (+ 0)}}
        {random/io >>= {print . ((flip ++) "\n") . show . (+ 0)}}
        {random/io >>= {print . ((flip ++) "\n") . show . (+ 0)}}
        {random/io >>= {print . ((flip ++) "\n") . show . (+ 0)}}
        {random/io >>= {print . ((flip ++) "\n") . show . (+ 0)}}
        {random/io >>= {print . ((flip ++) "\n") . show . (+ 0)}}))

(Edited) Now I think I realise what you are doing with the RandomT code at the end that you have commented out. It's about chaining the computations together, and this would be relevant to starting with one called to make-io-prng and then feeding that into a stream of computations.)

I agree that it probably makes sense to provide an interface to create a fresh PRNG. It may also make sense to provide a random/io function, but make it use Racket’s global/default PRNG. There’s no need to have a pure, self-contained generator if you can only use it in IO, anyway, since it’s random, and other modifications to the generator state are not meaningfully observable.