Universal Metadata Dictionary: A Platform-Agnostic Metadata Representation Model for Cross-RDBMS Schema Drift Detection

This paper addresses the research objective: to design, specify, and validate a platform-agnostic metadata representation model that normalises database schema metadata from heterogeneous Relational Database Management System (RDBMS) environments into a unified structure sufficient for automated cross-RDBMS structural schema drift detection. Schema drift — the unintended divergence of database schema structures between deployment environments or between the database schema and application source code — is a leading contributor to deployment failures in enterprise information systems. The proposed Universal Metadata Dictionary (UMD) normalises database schema metadata from six widely deployed RDBMS platforms — Oracle, Tibero, SQL Server, PostgreSQL, MySQL, and SQLite — into a unified 29-element relational structure governed by four design principles: platform independence, completeness, normalisation, and hashability. A five-stage normalisation pipeline (extraction, parsing, mapping, enrichment, and storage) and a seven-category data type mapping constitute the core implementation specification, eliminating false drift reports arising from platform-specific type nomenclature. The UMD was validated through: (1) expert review by 13 domain practitioners using a five-point Likert scale, yielding composite mean scores of 4.52–4.70 out of 5 across five criteria; and (2) structural completeness assessment via direct query execution on all six target platforms, confirming that all 29 elements are extractable through JDBC-accessible metadata interfaces and system catalogues. An empirical evaluation protocol comprising 15 structured fault injection test scenarios is defined for subsequent implementation-phase validation. Three specific contributions distinguish the UMD from all existing approaches: a cross-RDBMS metadata normalisation model with no comparable published equivalent covering all six platforms simultaneously, including Tibero; a per-object O(1) hash comparison mechanism enabling O(n) total Phase-1 change detection; and a metadata model directly supporting AI-based Semantic Code-Schema Analysis using transformer embeddings.