Research dashboard for the Foundations of Constitutional Physics (FCP) programme — provably safe AI for physical-world applications, grounded in the Clausius-Duhem inequality. This 29-page technical report documents five studies (FCP-I through FCP-V) covering architecture, empirical validation, formal verification, and deployment status as of March 2026. Highlights FCP-I Physics-Gated AI 100% thermodynamic admissibility on 16,146 samples · 2.99 MPa MAE · 785× faster than H-PINN Submitted. Preprint: doi:10.5281/zenodo.18768547 FCP-II Epistemic Sensing 60% measurement reduction · 88% timing-error improvement · 18,146 samples × 8 material families Submitted. Preprint: doi:10.5281/zenodo.18894710 FCP-III–V Working drafts — not yet submitted for peer review Verification & Implementation Formal verification — 59 sorry-free Lean 4 theorems · zero Coq Admitteds · complete Agda formalisation · 13 Haskell QuickCheck propertiesgithub.com/tytolabs/umst-formal Rust kernel — 27,542 LOC Rust/WASM physics engine with ROS2 bridgegithub.com/tytolabs/umst-prototype-2a FCP-I and FCP-II are publicly available as Zenodo preprints and have been submitted for peer review. FCP-III through FCP-V are working drafts not yet submitted. This is a living document; all claims are subject to revision.