Replication Package: Auditing the Survey Response Data of Large Language Models (CDI and EIS Indices)

Data and code for a three-experiment psychometric audit of 12 large language models as survey respondents, introducing the Cross-Domain Differentiation Index (CDI) and the Explicit Instruction Sensitivity index (EIS). Contains: raw API response logs (10,936 collected records, 10,506 fully valid: Experiment 1 = 1,500; Experiment 2 = 4,636; Experiment 3 = 3,600 collected / 3,170 valid; AIA-CN replication = 1,200), processed scale scores and indices, the complete analysis pipeline (Python), a validation simulation, and human baseline summary statistics (N = 606; n = 480 for the AIA item pool). Human data are provided as anonymized aggregate statistics to protect participant privacy; de-identified item-level extracts are available from the authors upon reasonable request. No credentials are included.