Towards Unified Material-State Tensors: Epistemic Sensing Architecture for Physics-Constrained Material Characterization

Epistemic Sensing Architecture for Physics-Constrained Material Characterization. We introduce the Epistemic Sensing Functor: an MI-guided proxy selection architecture with a hard thermodynamic gate (Clausius-Duhem inequality) that reduces required measurements by 60% while maintaining 100% admissibility across 18,146 samples and 8 material families. Evaluation of 5 frontier LLMs shows that only live gate enforcement — not rule information alone — restores thermodynamic compliance (+18pp). Validated on 6 real concrete mixes. 27,542 LOC Rust prototype, CC-BY-4.0. Continues UMST Paper 1 (doi:10.5281/zenodo.18768547).