An Experience-Dependent Relational Memory System Based on a Numerical Relational Field and Recursive Relational Structures

This study proposes a memory architecture that differs from storing events at fixed addresses and reading them back later. In the proposed architecture, experience does not remain as a copy of the original event; instead, it modifies anonymous numerical states and recursive relational structures, and subsequent experience interacts with this already modified persistent state. For the architecture to function as memory, prior experience must selectively alter subsequent processing, and that difference must depend on persistent relational state rather than transient runtime state. To test this, we sequentially examined reuse of the same experience, interference from subsequent experiences, relational generalization and discrimination, and direct intervention on persistent state across 24 preregistered stimulus families. Structural novelty for the first target experience was 18 in every family, but remained 0 from the second exposure through the sixteenth. When the target experience was reintroduced after 0, 1, 3, or 10 intervening experiences, the median novelty increased to 0, 0, 1, and 6, respectively; under a control condition in which fieldDecay was minimized to 1×10^-8, the corresponding values were 0, 0, 18, and 18. Exact, near, and isomorphic inputs all yielded novelty 0 in the trained state, whereas an input with the same numerical values but a different topology yielded 18, identical to the fresh control. Finally, when memory size and stored numerical coordinates were held constant while only the composition topology was made non-isomorphic, persistent growth recomputed from the externally serialized state increased from 0 to 22 in 24/24 families, matching the fresh control. By contrast, cyclically permuting the cell assignment of base coordinates while preserving topology left the value at 0. Within the scope of these experiments, the results show that persistent composition topology formed by prior experience is a state variable that determines subsequent structural reuse. The study does not claim identity with biological neural mechanisms; its conclusion is limited to functional similarities in history dependence, sensitivity to interference, and relational generalization and discrimination.