Analysis Code for the Sociodemographic Benchmark Consistency Study of Korean Synthetic Personas

This software archive contains the harmonization and analysis code for a study evaluating the sociodemographic benchmark consistency of the NVIDIA Nemotron-Personas-Korea synthetic persona dataset against the Korea Community Health Survey (CHS) and selected Korean Statistical Information Service (KOSIS) municipal indicators. The archive includes code for variable harmonization, regional crosswalks, marginal and joint distribution comparisons, total variation distance and Jensen-Shannon divergence, stratified cluster bootstrap intervals, CHS split-half references, independence baselines, sparse-cell and temporal sensitivity analyses, KOSIS linkage, regional downstream compositional sensitivity analyses, and an exhaustive audit of the living-alone classification rule. Source datasets are not redistributed; access information and execution instructions are provided in the README. The code supports reproducibility of the reported statistical analyses and is released under the MIT License.