Reproducible distributed research in practice

Have you ever tried version-controlling a data set with git and then using code from the same git repository to track it? Or tried to track different versions of outputs from one analysis as you use them in another? Or shared and collaborated on those analyses outputs as you might share code, docker containers or virtual machines? If so, you have probably run into limitations of our current tooling, just like all the other researchers who hit these problems every day. In this talk we show how a multilingual distributed analysis approach we developed enables researchers to stop thinking about the details and facilitates collaboration and reproducibility. Our approach ("orderly") tries to address shortcomings in existing data-versioning and workflow tools to enable researchers to automatically track provenance in data and analysis products as they move from output of one analysis to inputs in another. Our tool is neither a workflow or data-versioning tool that forces data to behave like code. Instead, it leans on metaphors from containerisation by allowing researchers to compose together analyses from building blocks of previous analyses which expose a simple interface describing their inputs and outputs. Our system is designed to be progressive, scaling from a single user on a laptop to a distributed collection of researchers sharing "packets" of data via a web application. The world probably does not need more workflow tools, but our experience indicates there are important researcher needs in data management that are still poorly met. A recording of this session is available on YouTube: https://youtu.be/lkPgihFQbrk