Automated Capture of Experiment Context for Easier Reproducibility

The manuscript was edited before publication, and this version does not correspond exactly to the final published version. Reproducibility is part of the definition of the scientific method. In computational science, reproducibility has practical benefits in allowing code reuse and hence enabling more rapid progress and the ability to develop more complex models and analysis methods. However, it is widely recognized that published research in diverse scientific domains that rely on numerical computation is too infrequently reproducible, with some commentators speaking of a “credibility crisis ” [4, 11]. In this article, I examine the reasons why reproducibility is difficult to achieve, argue that for computational research to become consistently and reliably reproducible requires that reproducibility become easy to achieve, as part of day-to-day research and not just for that subset of research that is published, and suggest a combination of best practices and automated tools that can make reproducible research easier. 1 The value of reproducibility The term “reproducible research ” in computational science covers a spectrum of activities [3]