AURA: A Dual-Memory Resonance Architecture for Bounded Meaning-State Generation in Memory-Bearing AI Systems

Memory-bearing artificial intelligence systems can store and retrieve prior information, but they often lack an explicit mechanism for generating a bounded, context-sensitive meaning-state from the interaction between present input and accumulated memory. This record archives AURA (Artificial Unified Resonance Architecture), a simulation-stage dual-memory framework based on Temporal Memory (TM), Bold Memory (BM), and TM-BM resonance. The core formulation is E0 = tanh(R(TM,BM) - D + lambda_value), where R(TM,BM) represents resonance between present context and continuity-weighted memory, D is a decay term, and lambda_value is a condition-sensitive stabiliser. Using synthetic four-dimensional simulations, AURA shows that the same present input can generate distinct bounded meaning-states under different BM profiles, separating it from stateless, BM-only, and retrieval-only baselines. In contradiction-shock tests, lambda_value reduces coherence-collapse depth by 25.4% and maximum output jump by 25.3%, with no increase in post-shock variance. A preliminary Trainable R extension improves mean resonance magnitude on relevant queries by 342.3%, while not improving categorical anchor selection and increasing E0 volatility by 8.4%. All results are simulation-stage findings based on synthetic embeddings. This work does not claim machine consciousness, genuine human emotion, clinical safety, safety certification, or production-grade deployment.