A constraint-based optimization approach to prioritizing agile issue backlogs

Abstract

Prioritizing software artifacts such as bugs is challenging due to multiple, often conflicting criteria. Traditional approaches either rely heavily on stakeholder input or attempt full automation without adequately considering these factors. This thesis investigates the use of Google’s CP-SAT constraint solver combined with simulated user elicitation to generate issue orderings. Using historical Jira data, we evaluate how constraint configurations including priority class, issue type, dependencies, and creation date approximate a gold-standard resolution order, and we assess the impact of varying both the quantity and accuracy of user input. We conduct project-level analyses, apply constraints at the assignee level, carry over issues between sprints, and compare CP-SAT with Multi-Objective Particle Swarm Optimization and Ant Colony Optimization. Our results show increased elicitation generally improves alignment, the system is robust to errors, local prioritization enhances performance, CP-SAT usually outperforms nature-inspired methods, and issue rollover improves performance, though the method can be further refined.