zero-trust-physics/zero-trust-physics-reactor-spert3-ccore

SPERT III C-core: 111 step transients from 1964, computed again with a proof on every row

Note, 20 October 2026: built on engine 0.15.0 (genesis_kernel_v0.13.0), model energy/reactor_spert3_ccore. Every row carries its own engine version and proof. The previous version of this dataset was served under the ID reactor_spert3_ccore_twin, on engine 0.14.3.

In tests reported in 1964, the SPERT III C-core, a stainless-steel-clad plate core of highly enriched fuel in a pressurized-water test reactor in Idaho, was put through 126 step insertions of reactivity at pressures from 0 to 2500 psig and flows up to 20,000 gpm. Here the 111 usable ones 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

196 rows under one dataset seal, 6391afa81c8dd2518756039db6451cda5a0bb822081dd3e42487c85f87a62b4fd45d32e272487b822a72b78ea0a47820:

• 111 rows, one for each usable transient of the report's appendix (IDO-16918, pages 34 to 37): 29 at 0 psig, no flow; 22 at 20 to 110 psig, no flow; 21 at 175 to 505 psig, no flow; 14 at 1000 to 2500 psig, no flow; 10 at 230 psig, 200 to 2040 gpm; 15 at 230 psig, 5000 to 20,000 gpm. Each row's request carries that transient's pressure, initial temperature, flow and reactivity as the model takes them.
• 85 rows of a sweep around them: No flow at 25 degC: 0, 110, 500, 1000 and 2500 psig, 0.40 to 1.40 dollars by 0.10; at 230 psig and 30 degC: 200, 2000, 5000, 10,000 and 20,000 gpm, 0.85 to 1.35 dollars by 0.10.
• Outcomes: 118 ended, 78 completed. 19092 recorded frames (power, energy, net reactivity, the hot-spot plate temperature and the void feedback through each burst, about 100 a run).

data/index.parquet (the default view) has one line a row: the transient, its conditions, the model's peak power, energy to the peak, time of the peak and hot-spot plate rise at the peak and at its first maximum, and for the 111 transients the report's measured values, the report's marks, the difference, and the model's own comparison run beside it. 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

The error is the model over the measurement, less one. The plate figures are rises of the hot-spot plate above the initial temperature. RMS, worst and bias in percent; "within 10%" counts the transients inside the report's power calibration of about 10 percent (IDO-17138 p.39).

Transients	Peak power: n, RMS %, worst %, bias %, within 10%	Energy to peak	Plate rise at peak power: n, RMS %, worst %, bias %	Plate rise at its first maximum
All 111	105: 10.2, 57, -1.4, 74%	105: 16.6, 109, +4.7, 60%	104: 11.9, 60, -1.7	95: 10.6, 56, +0.2
0 psig, no flow (29)	29: 8.7, 25, -1.1, 79%	29: 13.4, 32, +2.3, 66%	28: 8.9, 28, -1.4	28: 7.3, 22, -2.8
20 to 110 psig, no flow (22)	22: 9.4, 20, -1.0, 68%	22: 25.0, 109, +6.3, 73%	22: 6.7, 17, -3.4	22: 4.4, 11, +1.1
175 to 505 psig, no flow (21)	21: 7.9, 19, -0.1, 86%	21: 10.1, 19, +6.6, 48%	21: 12.2, 24, +1.3	21: 11.8, 25, +5.8
1000 to 2500 psig, no flow (14)	14: 8.0, 12, -3.8, 64%	14: 11.7, 21, +6.2, 50%	14: 10.2, 20, -3.5	14: 10.1, 24, -4.0
230 psig, 200 to 2040 gpm (10)	9: 10.1, 17, +6.4, 56%	9: 22.1, 47, +10.3, 56%	9: 17.7, 30, -11.5	3: 13.4, 22, +8.9
230 psig, 5000 to 20,000 gpm (15)	10: 19.3, 57, -9.7, 80%	10: 12.2, 24, -2.8, 60%	10: 20.0, 60, +5.9	7: 23.8, 56, -2.3

At the median transient, peak power is 5.8 percent from the measurement, energy to the peak 7.1, the plate rise at peak power 6.9 and at its first maximum 3.4.

These rows are the model's own comparison run again on the served engine: on every compared quantity they agree with the run its record was scored on within 4e-06 of the value, so the figures above are the record's.

The record against itself: transients repeated at one pressure and flow with periods within 3 percent scatter 8.9, 13.7, 8.5 and 5.5 percent RMS in the four quantities (28 transients).

One row

Row 109 is transient 03614: 230 psig, 5000 gpm, 33 degC at the start, a period of 21.5 ms. The request, as run_trajectory takes it:

{"tool": "run_trajectory", "arguments": {"card": "reactor_spert3_ccore", "domain": "energy", "dt_s": 0.0001, "initial": {"initial_temperature_k": 306.15, "system_pressure_psig": 230.0}, "record_every": 41, "segments": [{"load": {"coolant_flow_m_per_s": 1.3716, "reactivity_inserted_dollar": 1.1310720849681186}, "steps": 4070}]}}

Outcome ended (PastPlateTemperatureMaximum), after 3970 steps. The model's peak power is 172.4 MW at 0.3421 s with 7.139 MW s released to it, and the hot-spot plate rises 108.0 K by the peak and 186.4 K at its first maximum; the report measured 175 MW, 7.2 MW s, 111 K and 174 K. Run seal:

f1016fcc3cc6022aaab31f5381c90bf9730367c6f79e205c895073f8d3e146be0abf992c8dbdf3f05e48d84994db06e4

Verify it yourself

Every row can be verified 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 v0.15.0 && 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 6391afa81c8dd2518756039db6451cda5a0bb822081dd3e42487c85f87a62b4fd45d32e272487b822a72b78ea0a47820.
4. Send any row's request to run_trajectory and compare the run_seal.

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 144's proof on its own: holds; 6 of 6 rows sent to run_trajectory again gave the same run seal.

Files

File	What it holds	SHA-256
sealed/head.json	the dataset's head: its request, counts, root and seal	7b8d7d02aa813a12…
sealed/rows.jsonl	every row as the engine stored it, with its request, outcome, final state and run seal	ca8c10d989623d36…
sealed/frames.bin	every recorded frame, 64 bytes each	47bb832d0f23810f…
data/index.parquet	one line a row: transient, conditions, model figures, measurements, differences	8627a5a7bee2d576…
data/index.csv	the same as CSV	0cb5ac55ff03e55b…
data/rows.parquet	the sealed rows, by genesis_seal_parquet	8ca91c7fdaeb6c74…
data/frames.parquet	the sealed frames, by genesis_seal_parquet	263c7642aec73f91…
source/spert3_c_core_ido16918_appendix.csv	IDO-16918's appendix, the 126 transients as printed	a560939fa3375e9c…
source/split.json	the 15 transients set aside and each row's inputs	455ecd330972242d…
source/plate_dt0.0001.tsv	the model's own comparison run at 0.1 ms: the reactivity of each period, and its figures	f3d3f092069a5c34…

Where the numbers come from

• The record: Toole, C. R. (1964). The Effects of Pressure and Flow on Room Temperature Power Excursions in Spert III. Report IDO-16918, Phillips Petroleum Co., Atomic Energy Division, for the U.S. Atomic Energy Commission. OSTI 4042264, https://www.osti.gov/biblio/4042264. The PDF read has SHA-256 010709246980ca5053acad8a809128b45bd07a375fa58a54efcbf47acc7bc338. Distributed without restriction by the Department of Energy.
• The core constants: Condit, C. M.; Scott, J. F.; Johnson, R. L. (1967). IDO-17138, Table B-1, OSTI 4389198.
• The transcription of the appendix, typed from its page images and verified cell by cell against them: spert3_c_core_ido16918_appendix.csv (SHA-256 a560939fa3375e9c33477e472f84a68c1d2c60fd2c3ef7950173a21b4967f17a); split.json (SHA-256 455ecd330972242d321f497eddd407c109c4f4bb09d0f1a465e029206eb73dfe); plate_dt0.0001.tsv (SHA-256 f3d3f092069a5c34c469deed4d8a794e784301f6bc2241771f22179ce9188910).
• The inputs: Inputs as the model's own record (its calibration, analysis and driver, kept with the transcription of IDO-16918) and its description (describe_card energy/reactor_spert3_ccore, engine 0.15.0) give them: each transient's printed pressure (0 to 2500 psig, as the card takes it) and initial temperature (theta_i + 273.15 K); its printed flow as a mean velocity through the core, gpm x 18/20000 ft/s (IDO-16918 p.1: 20,000 gpm is 18 ft/s) x 0.3048, as split.json carries it; its reactivity as the card's own inhour value of the printed period, read to the last bit from the model's comparison run (source/plate_dt0.0001.tsv, rho_dollar); each run at the 0.1 ms step that comparison used, to the card's declared end at the first maximum of the hot-spot plate after the power peak, or to the first step whose clock, as the card carries it, is at or past 1.88020 s after the peak where that comes first (the comparison time fitted with the constants: the record gives no scram time). The engine runs at most 1,000,000 steps a run, 100 s at 0.1 ms: a transient that needs more takes 0.2 ms, which runs to 200 s as the model's driver did. That is transient 01425 (0.412 dollars, its power peak near 103 s), and the five with no power peak in 200 s (all marked still rising at the scram, none compared). The card's step ceiling is 0.3 ms, so no row takes more than 0.2 ms. The sweep is computed by the same rule.

What the rows do not say

• The report tabulates no time of the peak. The model's, from its start at about 10 watts, is in the rows and is not compared.
• 15 of the 126 transients are set aside: a temperature the report marks as read beside an abnormally high reading on another thermocouple (11), a likely misprint (2), and an initial temperature missing or in doubt (2). A quantity the report marks still rising at the scram is not compared, nor a plate maximum it marks higher at the scram.
• The record gives no scram time. Where the plate has no maximum after the power peak within 1.88020 s, its value then is compared; that time was fitted with the model's constants.
• The plate temperature is one thermocouple, spot-welded to a plate at the flux peak of a central assembly.
• 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 (v0.15.0).

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