SPIRALbase: Structured Potential-Indexed Representation in Associative Landscapes

SPIRALbase is a working paper on context-gated associative memory as a modular machine-learning component. The paper studies AsoMemm, the first SPIRALbase instantiation, and shows how masked storage, replay, pseudo-likelihood learning, and structural partitioning affect recall, interference, and capacity in a shared substrate. It also evaluates a text-facing and language-model interface, arguing that the same memory core can support bounded persistent retrieval without collapsing into exact database lookup.