Self-admitted scientific debt: Navigating cross-domain challenges in scientific software
Date
2024
Authors
Awon, Ahmed Musa
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Scientific software development faces unique cross-domain challenges, requiring expertise from both scientific and software engineering disciplines. These challenges often manifest as technical debt, specifically in the form of Self-Admitted Technical Debt (SATD). While technical debt is a well-recognized issue in software engineering, its impact within scientific software remains underexplored. In particular, the integration of domain-specific scientific knowledge with robust software engineering practices presents ongoing difficulties. This work investigates these cross-domain challenges in scientific software in various fields—including high-energy physics, astronomy, molecular biology, climate modeling, and applied mathematics—through SATD analysis. We examined 28,680 code comments from nine open-source scientific projects, identifying 11 types of technical debt. Among them, we introduced a novel category termed Scientific Debt, representing the issues that arise when integrating scientific findings with software development. We identified five key indicators of SD: assumptions, missing edge cases, accuracy challenges, translation challenges, and the incorporation of new scientific discoveries. Our findings reveal that Scientific Debt accumulates at a significantly higher rate than it is resolved, with the Missing Edge Cases indicator being the most frequently addressed. To further support the management of this debt, we explore the potential of Large Language Models (LLMs) in identifying and predicting cross-domain challenges. Our preliminary investigation suggests that LLMs could help detect issues requiring both scientific and software expertise, offering a promising direction for future efforts to manage and mitigate Scientific Debt.
Description
Keywords
Self-admitted technical debt, Technical debt, Scientific software, SATD, TD