Speaker
Description
Research software underpins an increasing share of modern research, yet expectations for such software remain implicit, inconsistent, and difficult to articulate. From exploratory notebooks to reusable tools, public-facing systems, and decision-support applications, what constitutes “good enough” depends strongly on how the software is used. This becomes particularly challenging as software moves from use within academia toward reuse, sharing, and real-world deployment. The growing use of AI-generated code further amplifies these challenges, as software may be functional but lacks clear specification, validation, and ownership.
We introduce an approach to making expectations for research software explicit and actionable. The central idea is to evaluate software not against a single set of criteria, but in relation to different evaluation scenarios, each reflecting a particular way in which the software is used. These scenarios enable the same software system to be examined from multiple perspectives and provide a structured way to reason about transitions from exploratory artefacts to software intended for wider use or deployment.
Our approach focuses on observable aspects of software and development practice, framing evaluation as a comparison between what is present in the software and what is expected for a given scenario. Beyond assessment, it supports guidance by making explicit what is required for software to become fit for a particular use.
The talk will also present RSCompass, a web-based tool designed to support this process in practice. RSCompass helps users make expectations explicit, identify gaps, and explore development pathways for their software. More broadly, the talk will argue for a shift from implicit and generic notions of software quality toward explicit, context-dependent expectations as a foundation for research software engineering.