Research Software Systems: Exploration and Infrastructure in Observational Cosmology

Abstract

In the last 20 years software has become an increasingly problematic object in the sciences. While it has become integral to research in many fields, scientific communities have struggled with problems of reproducibility, reuse, and the validity of findings produced by complex software systems. This has been attributed in part to the unplannable and unpredictable nature of scientific work itself, which makes it difficult to test research software or to define its requirements ahead of time. In this dissertation I draw on ethnographic work with a research group in the field of reionization cosmology in order to develop an understanding of how researchers develop novel software instrumentation in these contexts of uncertainty. I elaborate the notion of exploratory programming as a central process in scientific work that is distinct from software production, and I further develop the concept of the research software system as a system of heterogeneous software processes that is organized to reconcile the need for rigidity and flexibility in research infrastructures. This contribution reconceptualizes how software is used in scientific work as well as the process by which software engineering methods might be integrated into that work, and it provides new conceptual handles with which to support both activities.