AI-assisted code generation introduces a structural gap between what an AI system is authorized to modify and what it actually modifies. Testing validates outcome correctness but not scope authorization. Code review evaluates quality but not structural boundary compliance. Monitoring tracks operational metrics but not decision attribution. None of these mechanisms answer a prior question: did the generation stay within its authorized scope? This paper introduces Decision Analysis, an effect-oriented analytical framework that determines whether observable effects of AI-assisted code generation fall within structurally authorized boundaries. The framework inherits its structural vocabulary from three prior works: Anchor Architecture provides the coordinate system (A,R)(A,R), Viewpoint-Structured Specification provides decision carriers as specification elements (V,E)(V,E), and Boundary provides the execution-time admissibility resolution (V′,E′)(V′,E′) and B(t)B(t). Decision Analysis introduces two constructs not present in any upstream work. First, Decision Effect, defined as an observable state deviation Et=Δ(St0,St)Et=Δ(St0,St), serves as the analytical entry point. Second, three Anchor-based Scopes — Visible Scope SvSv, Artifact Scope SaSa, and Outcome Scope SoSo — decompose the execution lifecycle into pre-authorized context, intended modification target, and actual modification result. Five mutually exclusive analytical states — DA-N (Normal), DA-O (Over Outcome), DA-I (Indeterminate), DA-V (Decision Vacancy), DA-U (Unobservable) — are defined as set-relational conditions over SvSv, SaSa, and SoSo. These states support a pre-merge structural scope audit: an automated determination of whether each generation event remained within its authorized boundary, without requiring semantic understanding of the generated code. The framework provides review triage (directing human attention to scope-exceeding changes), cumulative scope tracking (detecting cross-session boundary erosion), and constraint quality feedback (using state distributions as a basis for governance adjustment). Risk interpretation is explicitly excluded; governance implications are addressed in a companion Decision Risk paper.
Spark Tsai (2026) studied this question.