The Hidden Cost of Reproducibility: Commercial Software Dependency in Published Research and the Version Accessibility Gap

Abstract The reproducibility crisis in science has prompted widespread efforts to improve data and code sharing. However, a critical yet underexplored barrier persists: dependency on commercial software whose specific versions are often inaccessible for replication. We conducted a cross-sectional empirical study of 10,000 papers published between 2020 and 2025, sampled from PubMed using stratified random sampling across seven research fields. Our analysis reveals that 18.5% of papers relied on commercial software, with an estimated mean replication cost of $1,900 per paper for those requiring proprietary tools. Only 17.0% of detected software had associated version numbers, and the majority of cited commercial software versions were classified as likely unavailable from vendors. Complementing these empirical findings, we conducted a systematic survey of legacy version access policies across major commercial software vendors, finding that no vendor explicitly offers "reproducibility licenses" for verification purposes. We propose a framework for addressing this "version accessibility gap" through the establishment of reproducibility licenses, mandatory version archiving, and enhanced reporting standards. Our complete dataset and analysis pipeline are publicly available to support further research on computational reproducibility.