Hullman et al. 2026

This human study did not involve human subjects: validating LLM simulations. Hullman, J., Broska, D., Sun, H., and Shaw, A. (2026). arXiv:2602.15785.

Relationship to this project — Statistical calibration

Distinguishes heuristic interchangeability from statistically calibrated use of synthetic responses. Complementary: calibration presupposes knowing which response surface must be validated — the object this paper measures directly.