> I would probably score about the same, does this prove I also rely on training data memorization rather than genuine programming reasoning?

It doesn't even prove the models do that. The RLVR environments being mostly Python isn't "training data memorization". That's just the kind of dumb thing people say to sound savvy.