Use when a question requires causal reasoning — "what will happen if…", "why is X happening", "what are the knock-on effects", predicting how a system reacts to a change, or any situation where intuition would be unreliable and hidden variables matter. Builds an explicit causal world model — facts vs assumptions, interventions, second-order effects, calibrated confidence — instead of pattern-matching an answer. Invoke proactively for any non-trivial "why" or "what if" question.
