Computational Notebooks and Workflow Transparency in Applied Physics Projects

Abstract

Computational notebooks have become a central medium for data analysis, modeling, and simulation in applied physics, providing a flexible environment that combines executable code, natural language explanations, and scientific visualizations. Despite their popularity, the non-linear execution models and exploratory development patterns of computational notebooks often compromise workflow transparency, raising significant concerns regarding research reproducibility and peer validation. This paper presents a comprehensive framework for assessing workflow transparency through multi-level difference analysis of computational notebooks in applied physics projects. By analyzing structural modifications, execution states, and code-to-narrative dynamics across sequential versions, we establish quantitative metrics to evaluate scientific workflow clarity. We construct a dataset of open-source physics projects, employing abstract syntax tree parsing and semantic change tracking to identify structural deviations. Our empirical evaluation reveals that a substantial portion of notebooks contain execution sequence anomalies, undocumented parameter variations, and uncommitted runtime state modifications that obscure scientific pathways. The proposed difference analysis framework demonstrates high efficacy in identifying potential reproducibility failures, offering a systematic tool to audit, improve, and communicate computational workflows in the physical sciences.