Review of: "Simulating Lay Health-Seeking Behavior with LLM Personas and Illness Vignettes: Reproducibility, Prompt Sensitivity, and Slice Dependence"

This manuscript offers a timely and valuable methodological audit of LLM-based behavioral simulations.The slice-aware analysis-particularly the focus on red-ag versus non-red-ag scenarios -is a notable strength.However, the current experimental design makes it dif cult to fully support the paper's main conclusions.The lack of strict control over the generation con guration-speci cally stochastic settings (e.g., temperature, top-p) and platform-level behaviors like context caching-makes it impossible to con dently isolate the true source of the observed variability.Moreover, three runs per condition are insuf cient to robustly estimate stability.While the authors raise intriguing hypotheses, establishing them requires tighter experimental controls and a deeper engagement with the existing literature.