Replication package: causal machine learning analysis of wearable physiological stress and sports injury data

Analysis code, derived feature data, and pre-computed statistical results for a causal machine learning study of wearable physiological stress responses and sports injury risk. The analyses cover three independent, publicly available datasets: a controlled within-subject wearable stress/exercise protocol (Hongn dataset), the WESAD wearable stress dataset, and a longitudinal athlete-monitoring sports injury dataset. Scripts use fixed random seeds and relative paths; every number in the pre-computed JSON result files can be regenerated from the included data. Supplementary experiments include decomposition of injury-prediction performance into game-day exposure versus within-game-day discrimination, per-subject leave-one-subject-out cross-validation case analysis, class-imbalance robustness (ROC-AUC, PR-AUC, Brier score), and paired effect sizes.