Domain-Calibrated Trust in Stateful AI Systems: Implementing Continuity, Causality, and Dispositional Scaffolding

Research Context: This work is a core component of the Presence Engine™ Living Thesis (DOI: 10.5281/zenodo.17280692). This technical note presents an architecture for achieving dynamic, domain-calibrated trust in stateful AI systems. Current AI systems lack persistent context across sessions, preventing longitudinal trust calibration. Kneer et al. (2025) demonstrated that only 50% of users achieve appropriately calibrated trust in AI, with significant variation across domains (healthcare, finance, military, search and rescue, social networks). I address this gap through three integrated components: (1) Cache-to-Cache (C2C) state persistence with cryptographic integrity verification, enabling seamless context preservation across sessions; (2) causal reasoning via Directed Acyclic Graphs for transparent, mechanistic intervention selection; (3) dispositional metrics tracking four dimensions of critical thinking development longitudinally. The proposed architecture operationalizes domain-specific trust calibration as a continuous, measurable property. Reference implementations with functional pseudocode are provided for independent verification. Empirical validation through multi-domain user testing (120-day roadmap) will follow, with results and datasets released to support reproducibility. This work complements Anthropic's Constitutional AI by extending safety principles from AI system behavior to human decision-making development. Full technical specifications are documented in the accompanying living thesis (v4). Keywords: trust calibration, stateful AI, causal reasoning, domain-specific AI, dispositional metrics, human-AI interaction, Cache-to-Cache, longitudinal validation