Reproducibility as a Technical Specification (1504.01310v3)

Abstract: Reproducibility of computationally-derived scientific discoveries should be a certainty. As the product of several person-years' worth of effort, results -- whether disseminated through academic journals, conferences or exploited through commercial ventures -- should at some level be expected to be repeatable by other researchers. While this stance may appear to be obvious and trivial, a variety of factors often stand in the way of making it commonplace. Whilst there has been detailed cross-disciplinary discussions of the various social, cultural and ideological drivers and (potential) solutions, one factor which has had less focus is the concept of reproducibility as a technical challenge. Specifically, that the definition of an unambiguous and measurable standard of reproducibility would offer a significant benefit to the wider computational science community. In this paper, we propose a high-level technical specification for a service for reproducibility, presenting cyberinfrastructure and associated workflow for a service which would enable such a specification to be verified and validated. In addition to addressing a pressing need for the scientific community, we further speculate on the potential contribution to the wider software development community of services which automate de novo compilation and testing of code from source. We illustrate our proposed specification and workflow by using the BioModelAnalyzer tool as a running example.