I always assumed that chess engines are dumbed down by occasionally inserting random moves but now it occurred to me that a way to get more realistic mistakes might be feed them a distorted version of the game state.