Spec-Delta Driven Governance Improves Lakehouse Data Platform Changes.

The concept of Specification Driven Development (SDD) posits that specifications, rather than code, should be the primary artifacts guiding AI-assisted work. While this principle has been formalized in software development and extended to data contracts for schema and quality enforcement, its application to data platforms, particularly the "spec-delta" as the fundamental unit of change, remains largely unexplored empirically. This research formalizes the spec-delta concept, which emphasizes that every change should result in a reviewable increment of requirements. It also proposes a taxonomy for classifying data platform changes based on their suitability for incremental specification. Many data platform modifications, such as new datasets, service-level agreements, metric semantics, and access policies, are contractual in nature rather than purely code-based, making them ideal candidates for a spec-delta approach. The paper outlines a controlled experiment designed to compare a spec-delta-driven workflow against a conventional code pull-request workflow. The study aims to measure key performance indicators including discovery-to-deployment time, the density of defects reaching downstream data layers (Silver and Gold), cross-tool metric divergence, and reviewer cognitive load, using metrics like NASA TLX. The contribution is not a new tool, but rather reproducible evidence and an applicability guide to help organizations avoid the pitfalls of upfront over-specification.