Code and analysis outputs for dependency-aware discovery and zero-refit validation of simple equations in human locomotor biomechanics

Published: 25 October 2026| Version 3 | DOI: 10.17632/y5xp449tyf.3
Contributor:
Alessandro Garofolini

Description

Repository accompanying the manuscript “Discovering and challenging simple equations in human locomotor biomechanics: From compression to zero-refit external validation.” This repository contains the analysis code, frozen specifications, quality-control records, derived analysis outputs, supplementary materials, and figure-generation code supporting the study. The archive preserves the analysis chronology from feature specification and dependency auditing through equation discovery, deterministic candidate freezing, measurement-portability assessment, zero-refit external validation, and post-falsification robustness analyses in two independently acquired and processed external gait datasets. The repository includes the frozen 18-equation specification and supporting analyses used to evaluate uncertainty, selection stability, walking-speed and condition-related common drivers, participant- and condition-centred associations, scaling sensitivity, and external numerical transfer. These analyses distinguish low-parameter biomechanical compression from relationship-specific association and preservation of the frozen numerical mapping across datasets. The original Le Goff, Fukuchi, and Senden biomechanical datasets are not redistributed and should be obtained from their original repositories. The included README provides source-data references, analysis chronology, execution order, provenance information, quality-control landmarks, and reproduction instructions. Version 1.2 update: The repository was synchronised with the final manuscript following completion of the post-falsification robustness analyses. It now includes the final uncertainty, speed-conditioning, selection-stability, and participant/condition-centring analyses (Experiments 27F–27K), together with the corresponding derived outputs and final figure-generation scripts. The earlier Version 1.1 reproducibility corrections are retained, including clarification that 1,374 of the 1,500 admissible relationships were eligible for the three-family breadth-constrained candidate freeze, while 126 were classified as OTHER_ADMISSIBLE, and the addition of measurement-provenance information for the Senden external-validation dataset. These updates did not alter the prospectively frozen equations or coefficients; rather, they extend the robustness assessment and align the archived code and outputs with the final manuscript.

Files

Institutions

Categories

Computational Biomechanics, Biomechanics of Gait

Licence