cognitive_offloading_analytics

Cognitive Offloading Analytics is a research prototype that combines behavioral proxies—copy similarity, revision, verification, delayed recall, and confidence change—into an inspectable offloading score. The project is described as "a transparent benchmark for detecting possible cognitive offloading in AI-assisted learning." It is designed to support sensitivity analysis rather than diagnose individual dependence on AI, because "scores are normalized within each cohort and the bands are heuristic." The repository includes a reference pipeline with five stages: interaction traces, behavior features, transparent scoring, sensitivity analysis, and risk-band reporting. Core outputs include copy_similarity, revision_ratio, verification_rate, delayed_recall, confidence_shift, and offloading_score. The project emphasizes transparency and reproducibility: "The scoring pipeline keeps each behavioral proxy visible, including copy similarity, revision, verification, recall, and confidence change." All bundled data and results are synthetic demonstrations, and the project states that "nothing in this repository should be interpreted as evidence about real learners, teachers, or institutions."