Research by Vicky Rampin, Librarian for Research Data Management and Reproducibility at NYU's Division of Libraries, revealed that while there is widespread use of version control among academic researchers writing source code, there are limited approaches to its preservation. In response, Rampin, together with Martin Klein at Los Alamos National Laboratory, has developed an ambitious plan for CoSAI, Collaborative Software Archiving for Institutions, a project that will create a decentralized and federated platform that will knit together several existing archiving and software preservation tools. Decentralization means that no one institution can be a bottleneck or failure point for archiving workflows—a thorny problem on other platforms—while federation both shares costs among partners and implements one of the gold standards in archiving: ensuring the robustness of preservation through having multiple copies of files mirrored across independent sites. CoSAI will focus on research software and aims to archive not just the code developed on sites like GitHub, but the (currently) ephemeral record of supplementary material related to the code (e.g., discussion threads, issues, etc.). By leveraging existing open source tools like Memento Tracer and building on workflow engines such as OCCAM, CoSAI will be able to capture web resources from code repositories at high quality and in a reproducible manner.

The review process for software is analogous to but in some ways different from a manuscript review; in addition to assessing the integrity of the methods manifested in the software’s algorithms, reviewers can consider features of the code itself and how well it is “bundled” for use by others. Does it run well under varying conditions? How interoperable is it with other platforms? What is the quality of its documentation? Such review is important for two main reasons. First, software that receives high marks by reputable reviewers lowers barriers to use.  Scientists can trust that well-reviewed code is robust, trustworthy, and easy to implement, even if they did not write the code themselves.  Second, well-regarded software reviews (and citations) can signal value and thus increase the incentives for software engineers and others to develop and maintain research software.  This grant funds a project by ecologist and data scientist Leah Wasser to further advance research software review in Python, arguably the dominant programming language for data science. pyOpenSci will mimic many of the core functions of the rOpenSci ecosystem including a grassroots process to develop common community standards, a transparent review process that leverages critical tooling from the Journal of Open Source Software, and efforts to build a strong, well-connected, diverse network of developers, engineers, and working scientists committed to the project.