Longitudinal Evidence of Architectural Sustainability in Adaptive Software Remodularization: A 280-Release Empirical Study

Evaluating software remodularization at a single release snapshot is widely accepted as insufficient. However, still it remains the dominant practice in literature. This paper reports the longitudinal experimental results of AMORE, an Adaptive Multi-Objective REmodularization framework, assessed across 280 consecutive releases drawn from 15 mature open-source Java systems spanning 5 to 10 years of evolution. We measure four purpose-built longitudinal metrics namely Modular Stability Index (MSI), Change Localization Ratio (CLR), Predictive Maintainability Index (PMI) trend, and Architectural Drift Index (ADI). AMORE achieves an average MSI of 0.81 — 16–31% higher than competing baselines — while maintaining a CLR of 0.74, an ADI of 0.19 (up to 39% lower than structural-only methods), and a flat-to-improving PMI trajectory. All gains are statistically significant (p < 0.05) with medium-to-large Cliff's delta effect sizes. Taken together, these findings demonstrate that adaptive, multi-dimensional remodularization produces genuinely durable architectural improvements, not transient snapshot-level boosts.