Analysis code, pre-registered protocols and aggregated result tables for "A Paired Dual-Label Benchmark Shows That Label Provenance Cannot Explain the Cross-Dataset PPG Heart-Rate Gap"

Published: 12 November 2026| Version 1 | DOI: 10.17632/fhp8brnycd.1
Contributors:
Songfeng Wang, Jinling Zhang, Yuhua Wang

Description

Companion research data for a paired dual-label benchmark of cross-dataset wrist-PPG heart-rate estimation. Contains two pre-registered frozen-encoder protocols, the analysis and independent gate scripts, and aggregated result tables and machine-readable verdicts. No raw physiological recordings, no window-level feature tables and no third-party dataset copies are included; the four source corpora must be obtained from their original repositories (DOIs cited in the manuscript). Scripts use a <PROJECT_ROOT> placeholder for the local project directory.

Files

Steps to reproduce

Unzip the companion archive and consult README.md and MANIFEST.sha256 for the file inventory and checksums. Download the four source corpora from their original repositories (DOIs cited in the manuscript's Data availability statement); no raw recordings, window-level feature tables or third-party dataset copies are included. Replace the <PROJECT_ROOT> placeholder in the scripts with your local project directory. Run the independent gate scripts (gate_dual_label.py, gate_step12.py, gate_step13.py) before any analysis, then run the analysis scripts listed in the README and verify every reported number with closed_loop_verify.py.

Categories

Computer Science, Medicine, Biomedical Engineering

Licence