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

This study evaluates a computational memory architecture in which prior experience persists as changes in anonymous numerical state and recursive relational composition rather than as an explicit event record retrieved from a fixed event-level address. The frozen primary battery used 24 deterministic stimulus families to test repeated-exposure reuse, interference from intervening experience, and topology-sensitive selectivity. Using the predeclared structural-novelty score N, the first target presentation yielded N = 18 in every family, whereas presentations 2 through 16 produced N = 0. After 0, 1, 3, or 10 intervening experiences, median N on re-exposure was 0, 0, 1, and 6. A separately frozen near-zero-decay sensitivity control, added after the baseline interference result, yielded medians 0, 0, 18, and 18. Exact, near-value, and isomorphic probes reused learned structure, whereas same-value non-isomorphic probes produced fresh-like novelty. A separately frozen persistent-state intervention then edited stored composition incidence while preserving record counts and numerical coordinates; externally recomputed persistent growth increased from G = 0 to G = 22 in all 24 families, matching fresh controls. A predeclared coordinate-reassignment intervention did not increase G, so that prediction was not supported. Because the incidence-edited state retained relation-derived coordinates and was not shown to be reachable through ordinary learning dynamics, the intervention is interpreted as counterfactual dependence on stored incidence rather than as isolated identification of topology independent of coordinate-incidence consistency. The results establish auditable and internally consistent experience-dependent structural reuse within the tested deterministic battery, not biological mechanism or generalization to arbitrary input distributions.