Closing the Loop in AI-Driven Athletic Training Prescription

Training planning and load management in competitive athletics remain largely unsystematic, with most coaches and athletes relying on static tools that generate plans without any mechanism for ongoing adaptation. This open-loop limitation fails to account for the athlete's continuously evolving physiological state, increasing the risk of overtraining and injury, particularly among athletes without access to expensive professional infrastructure. This paper proposes a closed-loop decision-support architecture that integrates annual training plan generation with daily adaptation using consumer-grade wearable data and subjective athlete input. The proposed system could combine knowledge graph plan generation, Bayesian adaptation, longitudinal individual learning, and generative adversarial refinement into unified architecture. Using Design Science Research as the overarching methodological strategy, complemented by a PRISMA-ScR scoping review as its empirical foundation, the research aims to evaluate whether such a system could produce prescription quality comparable to that of experienced human coaches.