Data and code for "Cross-dataset transportability of kinematics-based lower-limb joint-moment prediction during level walking"

This version 2.0.0 reproducibility package supports strict nested participant-disjoint development and frozen cross-dataset validation of kinematics-based sagittal-plane hip, knee, and ankle OpenSim generalized-moment prediction during normal-speed level walking. Five models were evaluated: a participant-balanced mean waveform, phase-only Ridge regression, kinematics Ridge regression, a pointwise multilayer perceptron, and a temporal convolutional network. Development used 224 Camargo cycles from 22 participants. The confirmatory Scherpereel Tier A cohort comprised 1,127 cycles from 10 participants. The archive includes nested split manifests, frozen predictions and neural weights, model-matched transfer outputs, 5,000-replicate participant-cluster bootstrap results, phase-specific extremum and hierarchical agreement analyses, neural refit and TCN sensitivity analyses, figure source data, publication figures, and source-reconstruction programs. Raw source datasets are not redistributed. Their official repositories, licenses, and required citations are documented in the archive. Version 2.0.0 supersedes the original 1.0.0 analysis files.