Validation-Gated Multi-Agent Governance for Online Adaptation of Thermal-Hydraulic Surrogate Models under Operating-Regime Shift

This research addresses the condition-locking problem in thermal-hydraulic surrogate models deployed outside their pretraining envelope. The study develops a guarded continual-adaptation framework using role-separated agents (Monitor, Diagnosis, Adaptation, Safety-Auditor, and Orchestrator) for experimental thermal-hydraulic loop data. The framework enables AI surrogates to diagnose error signatures, prioritize candidate model families, and review promotions dynamically. This approach provides a systematic solution for maintaining predictive accuracy in thermal-hydraulic forecasting under changing operational conditions.