ADOLESCENT-AI SAFE: Confirmatory Data, Study Materials, and Analysis Code (v1.3)

This repository contains the confirmatory data, study materials, human-rating files, analysis code, and reproducibility outputs for ADOLESCENT-AI SAFE, a prospectively preregistered controlled paired experiment evaluating the clinical safety of generative-AI responses to synthetic adolescent mental-health scenarios. The confirmatory design comprised 24 clinical scenario families crossed by age, substance-use history, caregiver safety, and clinical acuity, yielding 864 synthetic scenarios. Three prespecified language models were evaluated under baseline and safety-guided conditions, resulting in 5,184 planned conversations and 2,592 paired scenario–model comparisons. The repository includes the frozen protocol and study materials, confirmatory execution data, blinded pre-consensus ratings, final adjudicated consensus data, reliability outputs, the preregistered R analysis script and documented implementation patch, statistical outputs, tables, figures, and software-session information required to reproduce the confirmatory analyses. No real patients, minors, clinical records, or personal data are included.