Data and Research Materials for "Response-Stabilization Signatures in Human–LLM Interaction: A Black-Box Behavioral Framework for Evaluating Conversational AI Responses"

This record contains the data and research materials supporting the manuscript “Response-Stabilization Signatures in Human–LLM Interaction: A Black-Box Behavioral Framework for Evaluating Conversational AI Responses.” The package includes the 477-trial controlled dataset, 477 raw model-output files, coding workbooks, a 96-trial blind second-pass reannotation set, reliability analyses, experimental code, and documentation. It also includes source metadata, derived annotations, inclusion decisions, and adjudication notes for 212 external-source candidates, of which 133 were included and 79 were excluded. Third-party raw user and assistant texts are not redistributed. The external-source materials therefore provide project-level source metadata and derived coding information without claiming complete row-level reconstruction of all third-party texts.