Which explanation do you trust… the elegant one or the useful one? Clean stories are comforting. They feel rational. But cognition doesn’t care about comfort. It cares about performance. Sometimes the model that sounds ridiculous explains reality better than the one that sounds tidy. That’s a warning sign worth respecting. If your explanation needs endless patches to survive new evidence, it’s not robust… it’s brittle. Good models earn their keep by making predictions, not by sounding reasonable at dinner parties. The goal isn’t realism; it’s leverage. A model is a tool. If it helps you see what would break, confuse, or generalize, it’s doing its job… even if the metaphor makes you cringe. Don’t ask, “Is this literally true?” Ask, “Does this help me reason forward?” Try this: when evaluating an idea, list what it predicts will fail. Models that can’t fail can’t teach you.
