Reproducibility

It might well be surprising that the computational sciences require researchers to have basic knowledge of licensing and copyright laws. We say that a result is reproducible if another team of researchers is capable of reproducing the result. As the science community grows, so does the importance of being able to reproduce the work of others, both for verification and for comparison. Read the horror story Empiricism is not a matter of faith.pdf.

Reproducing a result requires access to both the exact data used, and the exact analysis software. In serious cases, this can be verified using checksums; not all data and software packages are amenable to checksumming, though!

Imagine software in which the date and time of install, or a data set that records the time of download. If these differ between two installations, checksums might prove different. Read about all the ways software packages can be non-reproducible in this article on Reproducible builds.