MRCE: An Epigenetic Cognitive Architecture for Self-Evolving Verifiable Reasoning(Dataset)

The development of sophisticated Cognitive Architectures requires solutions for verifiable logical consistency and resource parsimony. Current Large Multimodal Models (LMMs) exhibit inherent limitations in these areas, particularly regarding compositional generalization (CGSOOD) in visual domains. This paper presents the realized Meta-Recursive Cognitive Engine Plus (MRCE+), a novel, fully implemented epigenetic cognitive architecture structured around a Triple-Loop Hierarchy: Execution (L1), Reflection (L2), and Autonomous Architectural Self-Evolution (AASE) (L3). Key innovations include: (1) a fully specified Hybrid Reasoning Core (HRC) Protocol bridging probabilistic LMM output with deterministic SMT solvers for formal verification, (2) a Neuro-Quantum Unit (NQCU) for highdimensional feature development, and (3) an L2 \u2192 L3 Feedback Loop that translates Reflection consistency scores (SMASR) into fitness objectives for the AASE core. Empirical results confirm the system\u2019s operational superiority: MRCE+ achieves a robust 0.89 CGSOOD and demonstrates a critical 60% reduction in resource cost variance (\u03c3 2 (AC) = 0.11), validating the framework for Dynamic Resource Allocation and Computational Parsimony capable of continuous, self-supervised architectural development.