From Signal Availability to Incremental Utility: Reproducibility Package for a Controlled Study of Feature, Node-Identity, and Fusion Effects in Heterophilic Graph Learning
收藏资源简介：
This record contains the reproducibility materials and source data supporting the study “From Signal Availability to Incremental Utility: A Controlled Study of Feature, Node-Identity, and Fusion Effects in Heterophilic Graph Learning”, prepared for submission to Neurocomputing. The archive contains 2,895 audited primary result files: 750 real-data component cells, 450 same-stack external-reference cells, 750 synthetic-development cells, and 945 prospectively frozen seed-held-out follow-up cells. It also includes source code, frozen protocols, analysis and audit scripts, tests, environment records, source data for the manuscript figures and tables, reproduction instructions, and file-level SHA-256 manifests. The materials support component-isolated evaluation and the distinction between predictive signal availability and residual incremental utility under the tested protocols. They do not establish a universal failure of nonlinear fusion, mixture-of-experts modelling, graph routing, or auxiliary graph branches. Authors and affiliations:Lijun Tang — Ningxia UniversityXu Chen — North Minzu University Corresponding author:Xu ChenNorth Minzu UniversityEmail: chenxu@nmu.edu.cn Zenodo DOI: 10.5281/zenodo.21420749Version: 1.0.0
