The Decision Architecture of Supply Chain Management: Separating Decisions, Solutions, Evidence, and Context

Abstract

Supply-chain management (SCM) research is analytically rich but representationally fragmented: studies are often identified by technology, method, industry, objective, or risk domain even when these descriptors refer to different aspects of the underlying managerial decision. This paper develops and evaluates the Supply Chain Decision Architecture (SCDA), which separates a four-dimensional Decision Kernel from Solution, Evidence, and Context layers. A reproducible review pipeline identified 705 unique candidate records; 659 advanced to Stage-2 screening and 518 were retained. A prespecified deep-review corpus contained 125 studies, including 121 primary analytical studies yielding 124 source-supported claims. Independent second-coder review showed heterogeneous agreement across the nine dimensions (Cohen’s κ = 0.036–0.712). Coder-sensitivity analysis nevertheless showed a consistent K4→KS7→KSV8→SCDA9 reduction in collision under both independent codings, although absolute collision levels differed substantially. Repeated 80/20 record-level holdouts further showed that preferred depth was coder- and sample-sensitive rather than universally fixed. In the frozen 124-claim analysis, collision decreased from 52.42% for K4 to 19.35% for KS7, 17.74% for KSV8, and 12.90% for SCDA9. The primary τ = 0.02 specification selected SCDA9, whereas a looser τ = 0.05 selected KSV8 and the original 24-study publication holdout selected KS7. A permutation diagnostic further showed that arbitrary independently permuted fields can reduce collision mechanically; collision is therefore interpreted only as within-corpus discriminability, not ontology validity. The evidence supports a layered decision-and-solution representation while treating outer-layer necessity as contingent on coding, sample composition, and analytical purpose.