Early detection of tipping points in Software Engineering - A Cross-Domain Perspective on Software System Instability using Early Warning Signs

Abstract

This thesis investigates tipping points in Software Engineering (SE) and explores whether concepts and methods from other complex systems domains can be applied within a SE context. Software projects can experience sudden transitions from stable development into states characterized by instability, declining activity, coordination problems, or project abandonment. While tipping point theory and early warning signs (EWS) have been extensively studied in domains such as ecology, climate science, and finance, they have received limited attention within SE. A mixed methods approach combining a systematic literature review with empirical analysis of SE datasets was applied. The literature review identified tipping point characteristics and tipping point types associated with SE activities. To evaluate the applicability of existing tipping point detection methods, three open source software datasets were analyzed using the ewstools framework. The results show that tipping point behavior in SE can be characterized using a set of EWS, where variance, autocorrelation, and skewness appeared most frequently. Their distribution closely aligned with findings from domains such as ecology and climate science, suggesting that software projects may exhibit similar dynamical behaviors to other complex systems. Tipping point characteristics were identified across multiple SE activities, particularly within Management and Requirements related processes, highlighting the importance of socio-technical and organizational dynamics in software project instability. The empirical analysis demonstrated that existing tipping point detection methods can capture meaningful instability patterns within SE datasets. However, irregular and highly variable repository activity affected the consistency of several indicators across project populations, suggesting that existing methods may require adaptation before they can be reliably applied within SE contexts. This thesis contributes a foundation for understanding tipping points in SE and provides a basis for future research on instability detection and EWS tools within software projects.