Offline Evaluation of a Deep Q-Network Framework for Dynamic University Course Content and Adaptive Learning Paths

University digital learning systems frequently rely on fixed course sequences that do not accommodate differences in learner mastery, pace, persistence, and error patterns. This study evaluated the Dynamic Adaptive Course (DAC) framework, a Deep Q-Network-based mechanism that integrates learner-state diagnosis, content-unit selection, difficulty calibration, and learning-path restructuring within a unified reward-driven process. An offline controlled algorithmic evaluation was conducted using 18,900 learner-course sequences comprising 642 content units, 1,486,320 timestamped events, 214,760 quiz responses, and 96,430 feedback records from Chinese university courses.