UnrealJon/DTDR: Semantic Confidentiality Paper – v1.1

This paper introduces a transform-domain semantic storage architecture in which high-dimensional embeddings are stored as quantised coefficients of a global approximately orthogonal transform. We formalise the concepts of representation-level confidentiality and completeness-gated exfiltration resistance, analysing how semantic utility depends on coefficient completeness. The approach complements encryption by reducing interpretability under partial exposure while preserving similarity search efficiency.