That’s actually a really interesting idea. Particularly if you take care to add docstrings and comments to the tests. Thanks!

Given the "ai Mario just pauses the game and calls it a win"

I'd expect that the ai written code doesn't do the right thing outside of the given test cases

We are writting ton of such tests, on daily basis haha.

That is infinite process, we're all stuck within :-)

Watch out, we'll end up writing unit tests 40 hrs a week