Proceedings of the Annual Meeting of the Cognitive Science Society

Recent Large Reasoning Models trained via reinforcement learning exhibit a "natural" alignment with human cognitive costs. However, we show that reasoning distillation via Supervised Fine-Tuning (SFT) fails to transmit this cognitive logic, leading to a "Cargo Cult" where students only mimic length. Testing the Hán Dān Xué Bú (Superficial Mimicry) hypothesis across 14 models, we identify a "Functional Alignment Collapse": while teacher models mirror human difficulty scaling (r = 0.64), distilled students significantly degrade this alignment (r = 0.34). Crucially, they exhibit "Negative Transfer," dropping below their own pre-distillation baselines. Our analysis reveals a "Linear Inflation Law" where students apply a constant verbosity multiplier (≈ 2.44) regardless of complexity. Consequently, distillation decouples computational cost from cognitive demand, revealing that human-like cognition is an emergent property of active reinforcement, not passive imitation.