Reproducibility by Construction: Open Algorithms, Public Benchmarks, and Cloud-Native Artifact Pipelines

Abstract

Contemporary computer science research often publishes high-level algorithm descriptions without sufficient implementation details, relies on closed or non-portable code, and uses custom or non-standard benchmarks. These practices undermine reproducibility and prevent fair comparison across tools and studies. This paper proposes a practice-driven model for reproducible research software that focuses on implementable algorithm specifications, open-source licensing, explicit software lineage and citation, and standardized data and benchmark formats to support transparent reuse and replication across computer science domains. The model recommends integrating version control, build tools, and automated testing with cloud- or VM-backed continuous integration systems, ensuring that code pushes automatically trigger dependency pinning and benchmark execution across agreed-upon suites. This produces shareable, executable artifacts instead of static, text-based descriptions. The approach highlights the importance of community-curated public benchmark repositories and artifact evaluation processes to replace ad-hoc comparisons with repeatable, platform-agnostic measurements aligned with software engineering best practices. By treating artifacts as first-class research outputs, the model reduces bit-rot, clarifies implementation choices, and improves portability and verification across systems, programming languages, and heterogeneous hardware. The work concludes with a proposal for a community-endorsed platform that isolates environments, automates builds and tests, and continuously validates results as dependencies evolve, ultimately aligning incentives for sustainable, verifiable computer science research software.