SERAPH and NOUS: A Geometric Substrate for Structure Mapping and a Self-Modeling Pipeline Without Symbolic Machinery

SERAPH is a deterministic semantic-geometry substrate that realizes Gentner's structure-mapping theory as a computational mechanism, retrieving by relational structure rather than surface features. Eigenframe promotion implements the systematicity principle geometrically: a frame becomes a stable attractor only when sufficient in-degree accumulates around it. Every frame carries a watermark lineage chain, making provenance walk-back a first-class substrate property.

NOUS is a cognitive architecture built on SERAPH, comprising a seven-store topology, a homoiconic pattern system, and an evidence-gated factualization mechanism that promotes parameters from Frozen to Advisory to Live under accumulated outcome evidence. The paper reports ten experiments (E1–E10) that seal a self-management pipeline end-to-end: shape reading, regime detection, cross-store divergence, pattern retrieval, content-side action, outcome evidence, threshold factualization, and modulated future dispatch. Every stage operates through architecturally specified machinery with no human in the loop. The culminating experiment demonstrates a 230-query causal divergence in event timing attributable to a threshold parameter factualizing under evidence accumulation.

The work is positioned against hypergraph-and-rewrite approaches such as OpenCog Hyperon, offering a geometric rather than symbolic answer to the cognitive-architecture question.

Patent status. Architectural mechanisms described herein are covered in part by US provisional patent application filed 4-9-2026, . Publication under CC BY 4.0 applies to this work as scholarly communication; patent rights in the described mechanisms are reserved. This public disclosure is timestamped for prior art purposes.