Abstract
<title>Abstract</title> <p>Software quality management needs evidence about how systems absorb change, not only whether teams follow disciplined processes or whether architecture snapshots contain known weaknesses. This paper introduces Change Localization and Externalization Measurement (CLEM), a behavior-centric framework that makes the structural mechanism of change absorption the unit of quality-relevant observation. CLEM classifies activity into Modification, Extension, Low-code, and Configuration and separates three reporting layers: ModeStrength, a neutral CLEM externalization score (CLEM-ES), and an optional context-weighted CLEM measurement index/profile (CLEM-MI). We operationalize CLEM through transparent path-based heuristics over version-control histories and evaluate it on two public Huawei ModelEngine repositories (472 classifiable commits, 26 monthly windows). We also include an exploratory derived-data check on antisuger, a private healthcare application built with Kotlin, Ktor, and Kotlin Multiplatform (135 classifiable commits, 4 monthly windows), with limited evidential claims. Descriptively, the cases occupy visibly different structural bands, covering modification-dominant exploration, balanced extension, and externalization-dominant platform behavior. In ModelEngine, mixed-mode work accounts for 28--36% of classifiable commits, and the ModeStrength vectors separate a balanced plugin/orchestration framework from an Extension-dominant platform. CLEM-ES is only moderately associated with activity volume and shows no reliable next-window association with the bug-fix proxy in this exploratory sample, supporting complementary structural use rather than defect prediction. A completed 160-commit double-annotation sample produced identical dominant-mode assignments; broader public replications and outcome validation remain future work.</p>