Decision Fidelity of Longitudinal Synthetic Clinical Cohorts: Aggregate Results, Protocols, and Reproducibility Code

This repository provides the non-person-level reproducibility materials supporting a two-source longitudinal benchmark of synthetic clinical cohort validation in psychiatric risk stratification. The study evaluated whether conventional statistical resemblance, dependency preservation, symptom-state transition fidelity, predictive calibration, and threshold-specific decision utility were preserved when models trained on synthetic cohorts were applied to untouched real external populations. The package contains frozen protocols and amendments, aggregate analysis outputs underlying the manuscript tables and figures, replicate-level and survey-bootstrap uncertainty summaries, sensitivity-analysis outputs, quality-control records, analysis scripts, figure-generation code, and machine-readable metadata. The benchmark used the Medical Expenditure Panel Survey (MEPS) and the Health