Resonance Gap Curiosity (RGC): A Source-Aware Ask–Verify–Update Curiosity Architecture for Memory-Bearing AI Systems

This record archives Resonance Gap Curiosity (RGC), a source-aware Ask–Verify–Update curiosity architecture for memory-bearing AI systems. RGC addresses a key failure mode in persistent-memory AI: the system may recognise an entity as familiar while failing to detect that the new information contradicts stored memory. The central contribution is the Conflict Curiosity Blindspot, where familiar-but-contradictory memory-update conditions are formalised as a separate bounded curiosity regime distinct from novelty. RGC decomposes curiosity into four bounded components: Novelty Curiosity (Cn), Conflict Curiosity (Cc), Ambiguity Curiosity (Ca), and Reward Curiosity (Cr). It further introduces a CSLV-compatible verification interface through CSLV_Conf and CSLV_Contradiction, enabling source-aware local BM updates instead of direct memory overwriting. The archived package includes the RGC v1.0 / v1.4b preprint, DOCX manuscript, and simulation supplementary materials. Three controlled MiniLM-embedding validation packages are reported: v1.2 core validation (5/5 PASS), v1.3 advanced stress tests (7/7 PASS), and v1.4b focused validation (5/5 PASS), covering Ca forward validation, recognition-guarded conflict triggering, CSLV-compatible update gating, local update isolation, and R-only blindspot routing. RGC is a research-stage architecture. These simulations are controlled mechanistic tests, not deployment benchmarks or certified safety validation. The framework may be useful for future memory-bearing AI agents, enterprise assistants, retrieval-augmented systems, autonomous reasoning layers, and AI trust infrastructure where systems must ask, verify, and update memory without blindly overwriting prior knowledge.