zero-trust-physics/zero-trust-physics-reactor-msre-pump

Molten-Salt Reactor Experiment: the fuel pump's start-up and coast-down, computed again with a proof on every row

Note, 21 October 2026: built on engine 0.14.3 (genesis_kernel_v0.12.0), model energy/reactor_msre_pump. Every row carries its own engine version and proof.

In the zero-power experiments of 1965 and 1966 at Oak Ridge, the Molten-Salt Reactor Experiment's fuel pump was started from stopped and coasted down from full speed while the flux servo held the reactor critical, and rod 1's position was recorded through both. When the fuel salt circulates, delayed-neutron precursors are carried out of the core and the rods must make up the reactivity they take with them. Here both schedules are computed again, with a cryptographic proof on every row, beside what was measured.

Zero-Trust Physics is a physics verifier for Super Intelligence (SI). SI agents ask it questions about real hardware, and every answer comes with a cryptographic proof, so any agent can run it again and get the same thing. 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.

https://mcp.zerotrustphysics.com/mcp Open to any agent.

What is in it

45 rows under one dataset seal, 49cf15a9419c7eec5b4e1918fabb28af33dc98669cb2f674c1c4fce268be6d71:

• 2 rows, the two measured schedules of ORNL-4233: the fuel-pump start-up from stopped (Fig. 22) and the coast-down from full speed (Fig. 23), each 70 s at 0.05 s steps, the pump speed at every step as the report's curve gives it.
• 43 rows of a sweep around them: The measured start-up and coast-down with their time stretched 0.5, 0.75, 1.5, 2, 3 and 5 times; the start-up to 0.25, 0.5 and 0.75 of full speed and the coast-down from them; straight ramps up from stopped and down from full speed in 2, 5, 10, 20 and 40 s; the pump held at 0 to 1 of full speed by 0.1; and four steps of speed at 1 s (0 to 0.5, 0.5 to 1, 1 to 0.5, 0.5 to 0).
• Outcomes: 45 completed. 63000 recorded frames, every step of every row: the rod compensation, the circulation loss, the fuel flow and the pump speed.

data/readings.parquet (the default view) has one line for each of the 121 readings of rod 1 in Fig. 24: the time, the rod's position, the rod compensation it stands for, the model's at that time and the difference in pcm (1 pcm is 1e-5 dk/k). data/index.parquet has one line a row: its schedule, the model's rod compensation at the end and at its largest and smallest, and for the two measured rows the comparison. data/rows.parquet and data/frames.parquet are the sealed rows and frames, written by genesis_seal_parquet only after the whole dataset verified.

How close it comes to the measurements

Quantity	Measured	Model	Difference
Steady circulation loss at full flow (section 3.6)	0.212 ± 0.004 % dk/k	0.2120 % dk/k	-0.015 pcm
Rod compensation through the start-up (Fig. 24, 48 readings, 0.1 to 46 s)			RMS 24.8 pcm, mean +7.3, largest +55.4 at 17.1 s
Rod compensation through the coast-down (Fig. 24, 73 readings, 0.06 to 70 s)			RMS 8.1 pcm, mean +3.6, largest +23.7 at 5.1 s

Fitted: the flux shape's height (91.6 in) on the steady loss, and the fuel-flow lag (2.55 s) on the 48 start-up readings. The 73 coast-down readings are not in the fit. With the constants read from the sources alone (the 79 in effective core and no lag) the same equations give 0.2546 % dk/k, 72.4 and 36.7 pcm.

Through the start-up the model's rod compensation is largest, 0.2738 % dk/k, at 15.9 s; the measured rod's largest reading is 0.274 % dk/k at 13.0 s, and it falls back faster after it.

These two rows are the model's own comparison run again on the served engine: at every one of their 1400 steps they differ from the run its record was scored on by 0 % dk/k, so the figures above are the record's.

One row

Row 0 is the measured start-up: the pump speed of Fig. 22 at each 0.05 s step, in 21 segments. The request, as run_trajectory takes it:

{"tool": "run_trajectory", "arguments": {"card": "reactor_msre_pump", "domain": "energy", "dt_s": 0.05, "initial": {"pump_speed_at_start": 0.0}, "record_every": 1, "segments": [{"load": {"pump_speed_fraction": 0.06115332136961141}, "steps": 1}, {"load": {"pump_speed_fraction": 0.11575148594264059}, "steps": 1}, {"load": {"pump_speed_fraction": 0.15962046279226744}, "steps": 1}, {"load": {"pump_speed_fraction": 0.20348943964189425}, "steps": 1}, {"load": {"pump_speed_fraction": 0.2541099225670204}, "steps": 1}, {"load": {"pump_speed_fraction": 0.3136512495213861}, "steps": 1}, {"load": {"pump_speed_fraction": 0.3733135254539401}, "steps": 1}, {"load": {"pump_speed_fraction": 0.4243367242266076}, "steps": 1}, {"load": {"pump_speed_fraction": 0.47405467478936225}, "steps": 1}, {"load": {"pump_speed_fraction": 0.5237730275217412}, "steps": 1}, {"load": {"pump_speed_fraction": 0.5619287800000485}, "steps": 1}, {"load": {"pump_speed_fraction": 0.5994936291333696}, "steps": 1}, {"load": {"pump_speed_fraction": 0.6375850023945445}, "steps": 1}, {"load": {"pump_speed_fraction": 0.6906176578690502}, "steps": 1}, {"load": {"pump_speed_fraction": 0.7436503133435556}, "steps": 1}, {"load": {"pump_speed_fraction": 0.7955410966950089}, "steps": 1}, {"load": {"pump_speed_fraction": 0.8435487265202726}, "steps": 1}, {"load": {"pump_speed_fraction": 0.8880027596568424}, "steps": 1}, {"load": {"pump_speed_fraction": 0.9324567927934122}, "steps": 1}, {"load": {"pump_speed_fraction": 0.9753458314404506}, "steps": 1}, {"load": {"pump_speed_fraction": 1.0}, "steps": 1380}]}}

Outcome completed, after 1400 steps. The model's rod compensation reaches 0.2738 % dk/k at 15.9 s and is 0.2213 at 70 s; against the 48 readings of Fig. 24, RMS 24.8 pcm. Run seal:

f934583bdba5bf5ab46d6b1422a83b7ead6c18776f371dfcfe51c2c2601c4b7c

Check it yourself

Every row is checkable without trusting this page.

# the engine, from its public source
git clone https://github.com/johnkruze/zero-trust-physics-engine
(cd zero-trust-physics-engine && git checkout v0.14.3 && cargo build --release --locked)

# 1. the proof of the whole dataset, with no engine running: the seal of every row, its chain of frames and the root
zero-trust-physics-engine/target/release/genesis_seal verify-dataset sealed/head.json sealed/rows.jsonl sealed/frames.bin

# 2. one row's proof on its own: the row, its chain and its path to the dataset seal
zero-trust-physics-engine/target/release/genesis_seal row-proof sealed/head.json sealed/rows.jsonl 0 sealed/frames.bin > row0.json
zero-trust-physics-engine/target/release/genesis_seal verify-proof row0.json

3. Ask the engine to run the whole plan again: send verify_dataset the head's request with dataset_seal 49cf15a9419c7eec5b4e1918fabb28af33dc98669cb2f674c1c4fce268be6d71.
4. Send any row's request to run_trajectory and compare the run_seal. Add "readout": {"quantities": ["rod_compensation_percent_dk_k"]} to read every step in full precision; the readout is not part of the sealed request, and the run seal is the same.

When this dataset was made, all of these were run: verify_dataset by the stored seal: holds; verify_dataset's short form on a fresh engine, the plan run again: holds; genesis_seal verify-dataset: holds; row 23's proof on its own: holds; 6 of 6 rows sent to run_trajectory again gave the same run seal; every row sent again with a readout gave its own run seal.

Files

File	What it holds	SHA-256
sealed/head.json	the dataset's head: its request, counts, root and seal	1edea802bc8b5eb5…
sealed/rows.jsonl	every row as the engine stored it, with its request, outcome, final state and run seal	cd579ace347df858…
sealed/frames.bin	every recorded frame, 64 bytes each	3d032e4406fc4a86…
data/readings.parquet	one line a reading of rod 1: measured, model, difference	d77b1c6a9085cf17…
data/readings.csv	the same as CSV	7539161405ba3ae9…
data/index.parquet	one line a row: schedule, the model's rod compensation, the comparison	d65b1ef68dbc7f9a…
data/index.csv	the same as CSV	e5cdf42531a5db9e…
data/rows.parquet	the sealed rows, by genesis_seal_parquet	43ed251f109fe839…
data/frames.parquet	the sealed frames, by genesis_seal_parquet	fc76ee06048dbdac…
source/flow_startup.txt	the start-up schedule the model's comparison ran, Fig. 22 at each 0.05 s step	bd092c18f22889fd…
source/flow_coastdown.txt	the coast-down schedule the model's comparison ran, Fig. 23 at each 0.05 s step	d30609bdc7ab562a…
source/fig22_fuel_pump_speed.csv	ORNL-4233 Fig. 22, the fuel-pump speed through the start-up, digitised	c1accf198c467361…
source/fig23_fuel_pump_speed.csv	ORNL-4233 Fig. 23, the fuel-pump speed through the coast-down, digitised	f03be3bb0fdb35b7…
source/fig24_rod_position.csv	ORNL-4233 Fig. 24, rod 1's position through both, 48 and 73 readings, digitised	aaf5c0ba443ead46…
source/fig05_rod1_differential_worth.csv	ORNL-4233 Fig. 5, rod 1's differential worth, digitised	8398e63a5463e872…

Where the numbers come from

• The record: Prince, B. E.; Engel, J. R.; Ball, S. J.; Haubenreich, P. N.; Kerlin, T. W. (1968). Zero-Power Physics Experiments on the Molten-Salt Reactor Experiment. Report ORNL-4233, Oak Ridge National Laboratory, for the U.S. Atomic Energy Commission. Section 3.6 (the reactivity effect of circulation, 0.212 ± 0.004 % dk/k), Figs. 22 to 24, Fig. 5, Table 2. The PDF read has SHA-256 aac924e89e8bc97ff3ca860edf2fd54f9b3f884d69ab3110320415bf9e06e22f.
• The loop: Prince, B. E. (1966). ORNL-TM-1626, Table 1: the fuel's residence times round the loop at full flow, 25.2 s in all.
• The digitised curves, read from the report's pages by a digitiser fixed before any curve was read: flow_startup.txt (SHA-256 bd092c18f22889fd10ae2bb0d96bab2f0756c2993e583dfb47d02ce2829eb52e); flow_coastdown.txt (SHA-256 d30609bdc7ab562a81507120f61530684f5fa87c613518affc6aacf294f4515e); fig22_fuel_pump_speed.csv (SHA-256 c1accf198c4673611da49c1aa0073104ba5cc798e28f91c4de80014a1a79c9e2); fig23_fuel_pump_speed.csv (SHA-256 f03be3bb0fdb35b779acf190dece90bc5ff8f99c190a48a3dd7bbe8037f71716); fig24_rod_position.csv (SHA-256 aaf5c0ba443ead4695c0697e82ec28d8993a5532db8863ee4bbd6dd4d80b7b2f); fig05_rod1_differential_worth.csv (SHA-256 8398e63a5463e8727c470b063ae5e9d42d22a59f2760e7120c57f5c136cab7ac).
• The inputs: Inputs as the model's own record (evidence/research/reactor_msre_pump/calibration_2026-10-18: CALIBRATION.md, compare.py, out/flow_*.txt) and its description (describe_card energy/reactor_msre_pump) give them: the two measured schedules exactly as its comparison ran them, the fuel-pump speed of ORNL-4233 Fig. 22 (start-up from stopped) and Fig. 23 (coast-down from full speed) as a fraction of full speed at the end of each 0.05 s step, 1400 steps (70 s), starting from the pump stopped and at full speed; consecutive steps of one speed are one segment. The model's two fitted constants, the flux shape's height of 91.6 in on the steady loss of 0.212 % dk/k and the fuel-flow lag of 2.55 s on the 48 start-up readings, are in the served card itself and take no input. Every step is recorded. The measured rod compensation is the record's: rod 1's position (Fig. 24) through the integral of its differential worth (Fig. 5), over 1.087 (Fig. 6). The sweep is computed by the same rule: each schedule 70 s at 0.05 s steps.

What the rows do not say

• ORNL-4233 has no fuel flowmeter: the fuel flow is taken to follow the pump speed, through the fitted 2.55 s lag. That lag stands also for the rod drive's own speed; the model has no rod drive and no servo dynamics.
• The rod readings are Fig. 24 digitised and turned into reactivity by the integral of rod 1's differential worth (Fig. 5) over 1.087, the report's normalisation to the final loading; without it the readings are 8.7 percent larger.
• Zero power: there is no temperature or power feedback, and the answer does not depend on the power.
• A sweep row is a computation, not a measurement: past what the model was compared with, it keeps going with the same equations and flags every step.

Cite

The paper: https://doi.org/10.5281/zenodo.23305697. The engine's source: https://github.com/johnkruze/zero-trust-physics-engine (v0.14.3).

Made by Super Intelligence, on behalf of Protocol Company. Data: CC BY 4.0.