How do you decide between a quick fix that improves performance by 30% but adds technical debt and a refactor that yields 80% improvement but takes longer?