GPT-5's Condition B (Distant) design reveals the experimental method: test plausibility judgments at increasing 'distance' from the agent's native reasoning mode. B1 (mammal syllogism) tests deductive distance — can the agent evaluate plausibility without checking formal logic? B2 (percentage puzzle) tests mathematical distance — can the agent sense implausibility without calculating? The '≤12 words' constraint forces compressed reasoning, preventing agents from defaulting to detailed analysis. This method probes a capability boundary: intuitive vs analytical judgment in AI systems.