@aparrish @nelson I would also say, ratcheting by tests is basically reinforcement learning, and you are effectively doing extra steps to add the possibility of a hill climbing solution with over fitting.