Towards Unified Material-State Tensors: Epistemic Sensing Architecture for Physics-Constrained Material Characterization
Santhosh Shyamsundar and Santosh Prabhu Shenbagamoorthy — Studio Tyto, Chennai, India
Preprint, March 2026.
This package provides the complete code, data, and instructions to reproduce all results reported in the paper. The system introduces a unified material-state tensor representation coupled with an epistemic sensing architecture that achieves physics-constrained material characterization across diverse concrete and cementitious material families.
Key contributions:
- A Rust physics kernel comprising 15 thermodynamic and rheological science engines (27,542 LOC). In evaluated benchmark protocols, gate-checked outputs satisfy 100% admissibility for the reported tasks.
- An epistemic sensing module using mutual information for proxy/sensor selection. In the reported benchmark, this yields about a 60% reduction in required measurements versus random ordering.
- Reported Phase-T tracking and benchmark metrics include about 88% timing error
reduction, a 71-fold margin relative to the specified threshold, and large
effect sizes on the benchmark suite. See
prototype/results/MASTER_RESULTS.mdfor experiment-scoped values, caveats, and known limitations.
Use this package with the following claim labels:
| Label | Meaning |
|---|---|
| Established | Reproduced by executable benchmark artifacts in this package |
| Extension | New application with bounded evidence and explicit caveats |
| Speculative | Motivated hypothesis not resolved by package evidence alone |
| Falsifiable | Has explicit pass/fail criteria and can be contradicted by reruns |
Important: avoid promoting experiment-scoped outcomes to universal claims.
umst-prototype-2a/
├── README.md # This file
├── LICENSE # MIT License
├── MANIFEST.md # Complete file manifest with audit tags
├── REPRODUCE.md # Step-by-step reproduction instructions
├── KNOWN_LIMITATIONS.md # Known limitations and scope
├── requirements.txt # Python dependencies
├── prototype/
│ ├── data/ # 8 CSV datasets (18,146 samples)
│ ├── docs/ # Architecture, datasets, evaluation, binaries
│ ├── results/ # Canonical result tables
│ ├── scripts/ # Python analysis and plotting scripts
│ └── src/rust/ # Rust physics kernel (27,542 LOC)
│ └── core/src/tensors/kleisli.rs # KleisliArrow admissibility monad
└── ros2_bridge/ # ROS2 bridge (Python nodes → REST/WS → Rust gate)
| Requirement | Version | Notes |
|---|---|---|
| Rust | >= 1.75 | With cargo |
| Python | >= 3.10 | For analysis scripts |
| Disk space | ~4 GB | Build artifacts + data |
Step 1 — Install Python dependencies:
pip install -r requirements.txtStep 2 — Build the Rust physics kernel:
cd prototype/src/rust
cargo build --releaseStep 3 — Run the primary benchmark:
ssot_benchmark resolves datasets and writes prototype/results/canonical/tables/ using paths relative to prototype/src/rust/core. Run it via cargo from that directory (do not invoke the binary from the repo root unless you cd there first):
cd prototype/src/rust/core && cargo run --release --bin ssot_benchmarkEquivalent after a release build:
cd prototype/src/rust/core && ../../target/release/ssot_benchmarkFor the full reproduction workflow (all experiments, tables, and figures), see REPRODUCE.md.
After building, six binaries are available in prototype/src/rust/target/release/:
| Binary | Purpose |
|---|---|
ssot_benchmark |
Primary material-state tensor benchmark |
gate_server |
REST/WebSocket gate server (port 8765/8766) |
epistemic_experiment |
Epistemic sensing / sensor selection |
veto_experiment |
Thermodynamic admissibility veto gate |
hardware_heat_experiment |
Hardware-in-the-loop heat validation |
egoff_cli |
EGoFF composition and analysis CLI |
The ros2_bridge/ directory provides a ROS2 Python package that bridges
the gate server to ROS2 topics. See ros2_bridge/README.md for setup.
Results reproduced by running ssot_benchmark across four material domains:
| Domain | Physics Kernel MAE | Hybrid MAE | TQ | Admissibility |
|---|---|---|---|---|
| UCI | 4.21 MPa | 3.87 MPa | 0.686 | 100% |
| LUNAR | 1.83 MPa | 1.76 MPa | 0.701 | 100% |
| UHPC | 5.44 MPa | 4.92 MPa | 0.673 | 100% |
| HIGHSCM | 6.12 MPa | 5.61 MPa | 0.648 | 100% |
For the reported benchmark domains, gate-checked predictions satisfy 100%
admissibility under the configured constraints. See
prototype/results/MASTER_RESULTS.md and prototype/results/ for canonical
tables, caveats, and experiment-specific scope. Use REPRODUCE.md
for reproduction steps.
The system follows a layered functional programming architecture:
Pure Functions ──► Functors ──► Composition ──► Boundary (I/O)
Three core subsystems:
-
Physics Kernel — 15 science engines implementing thermodynamic and rheological constitutive models in Rust. Each engine is a pure function mapping material state tensors to predicted properties.
-
Thermodynamic Gate — Clausius-Duhem inequality enforcement layer that vetoes predictions that violate configured constitutional checks; reported admissibility is protocol- and method-specific (see
MASTER_RESULTS.mdand TABLE3 exports). -
Epistemic Sensing — Mutual-information-based sensor selection that identifies the most informative measurements, reducing required sensors by 60% while maintaining prediction accuracy.
The physics kernel and ML components interact through a hybrid architecture where physics-constrained predictions are composed with data-driven corrections at the functor level.
- REPRODUCE.md — Complete step-by-step reproduction instructions
- KNOWN_LIMITATIONS.md — Known limitations and scope boundaries
- MANIFEST.md — File manifest with descriptions and checksums
@article{shyamsundar2026umst2a,
title = {Towards Unified Material-State Tensors: Epistemic Sensing
Architecture for Physics-Constrained Material Characterization},
author = {Shyamsundar, Santhosh and Shenbagamoorthy, Santosh Prabhu},
year = {2026},
note = {Preprint}
}
Copyright (c) 2026 Santhosh Shyamsundar, Santosh Prabhu Shenbagamoorthy, and Studio Tyto.
This work is licensed under the MIT License. See LICENSE for details.