Calibrating Epistemic Delegation in Agentic AI–Mediated Learning: Learner Verification as the Behavioral Validation Layer

This repository contains the de-identified research data and reproducibility materials associated with the study “Calibrating Epistemic Delegation in Agentic AI–Mediated Learning: Learner Verification as the Behavioral Validation Layer.” The study used a three-arm, individually randomised, multisite controlled design with 336 university EFL learners to examine whether verification-calibrated agentic AI supports learner verification, epistemic accountability, calibration, and reduced-support transfer relative to standard agentic AI and learner-controlled AI assistance. The deposited materials comprise the de-identified quantitative datasets used for analysis, the qualitative data/coding record, study assessment materials, variable and derivation documentation, and analysis syntax required to support reproducibility of the reported analyses. Quantitative data are organised at the participant–occasion–episode level, covering baseline, immediate posttest, and eight-week follow-up assessments. The archive is intended to support transparent verification and reproducibility of the findings reported in the associated article. The qualitative materials are maintained separately from the quantitative participant-level records where no documented participant-level linkage is available. To protect participant confidentiality, unrestricted raw interaction logs containing free-text learner writing are not included in the public repository.