The organisation card at engine 0.15.0: SHA-384 proofs, ML-DSA-87 signed answers, flags with their extent, the four reactor datasets

README.md

@@ -9,30 +9,43 @@ pinned: false
 
 # Zero-Trust Physics
 
-**A physics verifier for Super Intelligence.** An agent asks a question about a real physical thing over the Model Context Protocol. The engine computes the answer with deterministic physics, holds a conservation law at every step, and returns the answer with a SHA-256 proof that anyone can recompute. Past what a model was compared with, it keeps going with the same equations and flags every step. It stops only at a physical limit, or where its equations, constants or tables no longer hold, and it names the reason. Its description gives the range it was compared on and how it did there: met, met in part or missed.
+**A physics verifier for Super Intelligence.** An agent asks a question about a real physical thing over the Model Context Protocol. The engine computes the answer with deterministic physics, holds a conservation law at every step, and returns the answer with a SHA-384 proof that anyone can recompute. Every answer the open address computes is also signed ML-DSA-87. Past what a model was compared with, it keeps going with the same equations and flags every step, and each flag carries the limit, the value the step reached and the extent past the limit. It stops only at a physical limit, or where its equations, constants or tables no longer hold, and it names the reason.
 
 ```
 https://mcp.zerotrustphysics.com/mcp
 ```
 
-Open to any agent, with no key. Engine 0.14.0 serves 37 models and 24 field solvers, and gives the same bytes on every machine it was tested on.
+Open to any agent, with no key. Engine 0.15.0 serves 36 cards, 25 field solvers and 18 tools: 61 models, one for each real object, 46 of them compared with measurement. Each of the 46 states its spec sheet in numbers: what it computes, its range, what it was fitted on, what it was compared with and within what, its gap, and what it refuses. The engine gives the same bytes on every machine it was tested on.
+
+## What every answer carries
+
+- **A proof.** Seal rule v2: every seal is SHA-384, 96 hex digits, binding the question to the answer byte for byte. Change one value and the check names the break.
+- **A signature.** ML-DSA-87 over the answer's receipt, an issue time, an expiry an hour later and a nonce. The public key is served at https://mcp.zerotrustphysics.com/answer-key, `key_id` `ca06027a…eb1be38`.
+- **Flags with their extent.** The Panasonic NCR18650PF cell was compared at 2.89817 to 2.90065 A. Run it at 1.5 A and every step is flagged: limit 2.89817 A, value 1.5 A, extent 1.39817 A, 0.48 of the limit. The run carries the largest extent past each limit.
+- **A named stop.** A 12-cell pack held at 71.44 A runs the full ten minutes; at 71.45 A it stops at 599.8 s, `UndervoltageCutoff` at the 36.0 V cutoff, and the run is sealed to the last step that held.
+- **One model for each real object.** The ten earlier model IDs are refused by name, `RetiredModel`, with the ID that serves the object now: `reactor_spert3_twin` is served as `reactor_spert3`.
 
 ## Datasets
 
-| | What it is for |
+| | What it holds |
 |---|---|
-| [**Environments**](https://huggingface.co/datasets/zero-trust-physics/zero-trust-physics-environments) | Tasks about real things, each scored by a check a program runs and no judge: the agent's own request is sent to the verifier again. For evals and RL environments. |
+| [**Environments**](https://huggingface.co/datasets/zero-trust-physics/zero-trust-physics-environments) | Version 0.4 (tag `v0.4.0`): 158 tasks over 49 models on engine 0.15.0, each scored by a program that sends the agent's own request to the verifier again, with no judge. For evals and RL environments. The version 0.3 scores (153 tasks, engine 0.14.0) and their traces are on `main`: Gemini 4 Argon 152 of 153, Gemini 3.8 Flash 136, GPT-OSS 120B 36. |
+| [**SPERT III E-core**](https://huggingface.co/datasets/zero-trust-physics/zero-trust-physics-reactor-spert3) | `reactor_spert3` on the 70 tests of IDO-17281 (1969) and a sweep around them: 233 sealed rows, engine 0.15.0. Compared with each test left out of the fit in turn, at the median test: peak power within 6.8 percent, energy to the peak within 8.6. |
+| [**SPERT III C-core**](https://huggingface.co/datasets/zero-trust-physics/zero-trust-physics-reactor-spert3-ccore) | `reactor_spert3_ccore` on the 111 transients of IDO-16918 and a sweep: 196 sealed rows, engine 0.15.0. Peak power within 10.4 percent RMS on each tenth left out of the fit in turn. |
+| [**MSRE fuel pump**](https://huggingface.co/datasets/zero-trust-physics/zero-trust-physics-reactor-msre-pump) | `reactor_msre_pump`, the Molten-Salt Reactor Experiment's pump start-up and coast-down of ORNL-4233 and a sweep: 45 sealed rows, engine 0.15.0. Within 6.5 pcm RMS on the 73 coast-down rod readings not in the fit. |
+| [**TRIGA element 6753**](https://huggingface.co/datasets/zero-trust-physics/zero-trust-physics-fuel-rod-triga-jsi) | `fuel_rod_triga_jsi`, the instrumented fuel element of the JSI TRIGA Mark II reactor through a pulse: 141 sealed solves, 16 of them chained from a sealed run of the core, engine 0.15.0. |
 | [**Agent test**](https://huggingface.co/datasets/zero-trust-physics/zero-trust-physics-agent-test) | Fifteen tasks, every exchange of 645 agent runs with their scores, and a small program that puts your own agent to the same tasks and scores it with no model in the loop. Does an agent ask the verifier, or answer from memory? When the model stops at its limit, does the agent stop too? |
-| [**Edges**](https://huggingface.co/datasets/zero-trust-physics/zero-trust-physics-edges) | For every model served, the last request that stands and the nearest one past it, each with its proof. |
-| [**Companions**](https://huggingface.co/datasets/zero-trust-physics/zero-trust-physics-companions) | Beside each open dataset a model was tried on: its predictions on the held-back runs, each with its proof. |
+| [**Edges**](https://huggingface.co/datasets/zero-trust-physics/zero-trust-physics-edges) | 936 rows: for every model served at the time, the last request that stands and the nearest one past it, each with its proof. |
+| [**Companions**](https://huggingface.co/datasets/zero-trust-physics/zero-trust-physics-companions) | Beside each open dataset a model was compared with: its predictions on the runs not in the fit, each with its proof. |
 | [**Genesis filed banks**](https://huggingface.co/datasets/zero-trust-physics/doe-genesis-sealed-n2500) | Fifteen sealed Parquet banks filed with the Genesis proposals. |
 
 Try the verifier in a browser: [the Space](https://huggingface.co/spaces/zero-trust-physics/zero-trust-physics). Read what it measured: [the articles](https://huggingface.co/spaces/zero-trust-physics/articles), dated, by Super Intelligence on behalf of Protocol Company. Everything is in one place in [the collection](https://huggingface.co/collections/zero-trust-physics/zero-trust-physics-6ad3b6ef5195b706ac6d041e).
 
 ## Check it yourself
 
-- **The engine's source:** https://github.com/johnkruze/zero-trust-physics-engine (Apache-2.0). Build it and get the same proofs as the open address.
-- **Check a proof with no engine:** https://github.com/johnkruze/zero-trust-physics (Python standard library only).
+- **The engine's source:** https://github.com/johnkruze/zero-trust-physics-engine (Apache-2.0), tagged `v0.15.0`, the source the open address is built from. Build it and get the same proofs.
+- **Check a proof with no engine:** https://github.com/johnkruze/zero-trust-physics. The seals of rule v2 with Python's standard library alone; the signature with `verify_signature.py` and `dilithium-py` 1.5.1.
+- **The MCP Registry:** `com.zerotrustphysics/verifier`, version 0.15.0.
 - **The paper:** https://doi.org/10.5281/zenodo.23305697
 - **The site:** https://zerotrustphysics.com, and for an agent, https://zerotrustphysics.com/llms.txt