Data, Materials, and Reproducibility Files for "Socratic Dialogue and Evidence Verification in AI-Supported Learning: A Multisite Randomized Factorial Study"

This record provides the data, study materials, analysis code, and reproducibility files accompanying the study “Socratic Dialogue and Evidence Verification in AI-Supported Learning: A Multisite Randomized Factorial Study.” The formal-stage dataset contains 2,600 undergraduate participants from 80 classes across 10 study sites. The study used a 2 × 2 randomized factorial design crossing explanation-first versus Socratic-first AI dialogue with summary versus evidence-verification learning activities. The primary outcome was delayed independent transfer without AI, reference materials, or activity scaffolds. A subsequent advice task recorded initial judgments, advice quality, verification behavior, final decisions, and confidence. The repository includes the de-identified analytical SQLite database, data dictionary, study materials, analysis and sensitivity-analysis scripts, saved statistical outputs, allocation-reproduction files, power-validation materials, and documentation for reproducing the reported analyses. The public package excludes direct personal identifiers and internal operational records not required for reproduction. The allocation-reproduction workflow independently replays all 2,600 formal-stage assignments and stratum labels. The package also contains retained outputs and audit scripts for the study’s operating-characteristic simulations Ethics approval covered Pilot A, Pilot B, Pilot C, and the formal stage. Full identifying information is withheld during anonymous peer review.