Abstract
Software is a fundamental pillar of modern scientific research across all fields and disciplines. However, adequate means to cite and reference software remain elusive due to complexities in authorship, varying development roles, and the dynamic nature of code. This paper contributes to the development of proper guidelines for software citation, drawing on the extensive experience of Inria, the French research institute for digital sciences. We propose a richer taxonomy for software contributions, advocate for human-centered evaluation, and emphasize the critical distinction between citation and reference.
Introduction
In the modern scientific landscape, software is as fundamental as the papers that describe it. However, the academic "credit economy" has been slow to adapt. Researchers often struggle to receive proper recognition for software development because current systems treat code like static papers. This paper addresses the gap between how we use software and how we attribute and archive it, proposing a framework that prioritizes both academic credit and technical reproducibility.
The Complexity of Software Attribution
Attributing credit for software is significantly more complex than for traditional publications. Authorship can range from a single developer to a large, evolving team over decades. Furthermore, software development involves diverse roles—architects, coders, debuggers, testers, and managers—all of whom contribute to the intellectual weight of the project.
Automated metrics, such as line counts or commit frequency, often fail to capture the true value of these contributions. A ten-line bug fix may represent more intellectual effort than a thousand lines of boilerplate code. Consequently, the authors argue that only a "human-in-the-loop" moderation process can accurately assess the quality and impact of research software, moving away from algorithmic evaluation toward expert review.
Infrastructure for Archival and Preservation
To address these challenges, Inria has developed a robust internal infrastructure. Central to this is the BIL Database (Base d’Information des Logiciels), which tracks hundreds of projects, their maturity, and contributor roles.
A key innovation is the integration between HAL (the French national open access portal) and Software Heritage. This workflow allows researchers to deposit code into HAL, which then triggers a moderated archival process into Software Heritage, ensuring long-term preservation and accessibility.
Citation vs. Reference: Solving the Reproducibility Gap
A core contribution of this work is the formal distinction between citation and reference:
- Citation is a human-readable pointer intended to provide academic credit to authors (e.g., "I used the Coq proof assistant").
- Reference is a machine-actionable pointer to a specific version or even a specific line of code (using intrinsic identifiers like SWHIDs) to ensure an experiment can be perfectly replicated.
This distinction is vital because of "dependency hell." As illustrated by the dependency graph of common libraries like matplotlib, a single package can rely on a massive, hidden web of OS-level dependencies. Referencing the exact environment is a monumental task that requires specific, versioned identifiers rather than general citations.
Conclusion and Recommendations
Inria’s model demonstrates that software self-assessment can be successfully integrated into career promotion and team evaluation cycles. By adopting a qualitative taxonomy for contributions and distinguishing between credit-based citation and technical reference, the research community can better support the software that powers modern science. While challenges remain—including the manual effort required for moderation and the need for global standards like CodeMeta—these practices provide a blueprint for institutional and national research agencies to ensure their code is both recognized and preserved.
Summary updated