Semantic Memory Hierarchies: Hot Databases, Cold Archives, and Governed Movement Between Them introduces a tiered semantic memory architecture that separates computational memory from durable semantic memory through governed thermal storage tiers. The paper argues that modern software systems frequently flatten fundamentally different memory roles into a single database substrate, causing archival overload, degraded computational locality, operational opacity, and semantic instability over time. In response, the paper proposes Semantic Memory Hierarchies (SMH) — a framework that distinguishes between: HOT computational memory, WARM indexed archival memory, COLD canonical immutable archives, and DEEP compressed preservation layers. The architecture formalizes: bounded hot databases, immutable filesystem-based semantic archives, governed archive hydration and eviction, rebuildable indexing layers, projection-based archive navigation, and explicit thermal movement between semantic memory tiers. A central doctrine of the paper is: The database computes. The archive remembers. The paper explicitly defines the architectural advantage boundary of SMH, arguing that database-centric architectures remain optimal for highly transactional, fully “hot” systems, while SMH becomes superior for systems possessing significant archival mass, semantic lineage, auditability requirements, or long-term institutional memory. The work situates SMH as a general systems architecture applicable to: AI governance systems, semantic archives, legal and medical systems, media libraries, research infrastructures, compliance environments, and long-horizon semantic operating systems. The paper does not claim invention of filesystems, databases, or content-addressed storage individually. Instead, it contributes a unified architectural synthesis centered on thermal semantic memory specialization, canonical archive authority, and governed movement between computational and archival states.
Adam Ableman Mazurk (Mon,) studied this question.