Oldes R3 has a pretty good idea in testing, of using Rosetta code problems:
I'm doubtful many of the solutions are his, since he speaks about not having spare time. Several seem to come from Red versions of problems people solved, such as:
Rebol3/src/tests/rosetta/Conway's_Game_of_Life.r3 at master · Oldes/Rebol3 · GitHub
But the idea of pulling them in as tests seems to be his (although I can't see where he is running them automatically and verifying the outputs? Maybe he is not, but plans to...)
In Ren-C, I have some "proof of concept" examples that are in the tests. But the only real "puzzles" that I put in running automatically were @iArnold's programs he posted here, because I figured they ran and had repeatable results so why not:
Anyway, this is the sort of task that AI would be very useful at inventorying and working on. Historical Rebol testing was very much isolated bug tests, but whole example continuous integration tests compositional behavior that exposes things those fragments don't.