lbsh: Pounding Science into the Command-Line

The definition of science is rooted in the notion that results must be reproducible. However, this must be more than just a presumption. Scientific disciplines must strive to ensure and facilitate that results actually are reproduced and build upon. In order to facilitate this in Computer Science, a detailed step-by-step record of how each result-set was generated is critical. This is called Data Provenance and when done well it enables researchers to understand the precise nature of each other’s work. However, provenance is often only described at a high-level and this makes followon work, collaboration, and healthy scientific scrutiny more difficult. This is often because the complete documentation of all important processing steps is a very difficult task and is not always done. In this work we present a new open source utility called lbsh (pronounced “Pound-Shell”) that provides a simple way for researchers to automatically log their work into virtual lab-books, automatically capture the provenance of their result-sets, query their lab-books for the provenance of individual result-sets, automatically re-run their prior work (with new or old inputs), and share this information with others via a very simple format. lbsh is publicly available, and we present 2 case-studies describing how it has already benefited actual Computer Science researchers.