zero-trust-physics/zero-trust-physics-reactor-spert3-twin

SPERT III E-core: 70 reactor tests from 1969, computed again with a proof on every row

Superseded, 19 October 2026: engine 0.15.0 serves one model for each real object, and this model is now served as reactor_spert3 (the ID reactor_spert3_twin is refused by name, RetiredModel, with its replacement). The dataset on engine 0.15.0, with seal rule v2 (SHA-384 proofs), is https://huggingface.co/datasets/zero-trust-physics/zero-trust-physics-reactor-spert3. These rows stay as they were made, on engine 0.14.3.

Note, 26 October 2026: that dataset is re-sealed on engine 0.15.2, the engine the open address serves since 22 October 2026: the same requests and run seals, dataset seal 9bcc225fa463abe5…, every row signed by the open address. These rows stay as they were made.

Note, 19 October 2026: built on engine 0.14.3 (genesis_kernel_v0.12.0), model energy/reactor_spert3_twin. Every row carries its own engine version and proof.

In 1969 the SPERT III E-core, a small pressurized-water test reactor in Idaho, was put through 70 tabulated power excursions. Here every one of them is 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

233 rows under one dataset seal, 9dbf53ad44fff545cade33cd14ece2a3586cf50cf951eb6e8277fa929131d060:

• 70 rows, one for each test the report tabulates (IDO-17281, Tables III, V and VII): 30 cold startup; 15 hot startup, 260 degF; 17 hot startup, 500 degF; 5 hot standby; 3 operating power. Each row's request carries that test's conditions as the model takes them.
• 163 rows of a sweep around them: each series' nominal conditions with the reactivity run from below the smallest tested insertion to above the largest.
• Outcomes: 233 completed. 23091 recorded frames (power, energy and reactivity through each burst, about 100 a run).

data/index.parquet (the default view) has one line a row: the test, its conditions, the model's peak power, energy to the peak and time of the peak, and for the 70 tests the report's measured values with their standard deviations and the difference between the two. 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 1969 measurements

At the median test of each series: the size of the difference from the measurement, in percent. Beside it, how many tests are within two of the report's own standard deviations (10 to 17 percent on peak power and energy) and the root mean square difference in those standard deviations. The mean time is signed; negative is early.

Series	Tests compared	Peak power	Energy to peak	Time of peak	Peak power within 2 sd	Energy within 2 sd	Peak power RMS	Time, mean
Cold startup (Table III: 53 to 92 degF, no flow, atmospheric pressure)	30 of 30	4.2%	4.5%	2.7%	30	29	0.47 sd	-1.2%
Hot startup, 260 degF (Table V: 251 to 259 degF inlet, 1500 psig, 2.4 to 22 ft/s)	13 of 15	15%	12%	3.4%	13	13	1.06 sd	+2.2%
Hot startup, 500 degF (Table V: 498 to 501 degF inlet, 1500 psig, 4.8 to 24 ft/s)	15 of 17	4.3%	4.5%	3.7%	15	13	0.50 sd	+3.8%
Hot standby (Table VII: about 505 degF, 14 ft/s, about 1 MW at the start)	5 of 5	15%	13%	4.2%	4	5	1.70 sd	+1.3%
Operating power (Table VII: about 503 degF, 14 ft/s, about 19 MW at the start)	3 of 3	13%	82%	39%	2	0	1.61 sd	-35%
All 70	66 of 70	6.0%	7.4%	3.4%	64	60	0.84 sd	-0.8%

Over the 66 tests compared, at the median test the model's peak power is 6.0% from the measurement, its energy to the peak 7.4% and its time of the peak 3.4%. Peak power is within two of the report's standard deviations on 64 of them, and energy to the peak on 60.

Without the five tests whose clock was started by hand, the time of the peak has a root mean square error of 6.1%.

The model reactor_spert3_twin takes no insertion rate: it moves its own transient rod.

Table VII gives the net energy to the peak, the energy above steady operation. It is compared with the model's energy to the peak less the initial power times the time of the peak.

4 tests ended by a scram before the peak are kept in the rows and left out of the comparison.

One row

Row 64 is test 81 (hot standby, Table VII). The request, as run_trajectory takes it:

{"tool": "run_trajectory", "arguments": {"card": "reactor_spert3_twin", "domain": "energy", "dt_s": 0.00043000000000000004, "initial": {"coolant_flow_m_per_s": 4.2672, "initial_power_mw": 0.9, "inlet_temperature_k": 535.3722222222223, "system_pressure_mpa": 10.44}, "record_every": 24, "segments": [{"load": {"reactivity_inserted_dollar": 1.17}, "steps": 2327}]}}

Outcome completed, after 2327 steps. The model's first maximum of power is 281.1 MW at 0.135 s; the report measured 330 ± 30 MW at 0.135 s. Run seal:

213882bad4d1c3ffa4c71ffd22ba740e3d1c55b7a52cae18f745a22cbfe03dac

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 v0.14.3 && 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 9dbf53ad44fff545cade33cd14ece2a3586cf50cf951eb6e8277fa929131d060.
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 73'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	23fab115f5d96c95…
sealed/rows.jsonl	every row as the engine stored it, with its request, outcome, final state and run seal	522ccaf94b06ec4c…
sealed/frames.bin	every recorded frame, 64 bytes each	fefe24e90db754c3…
data/index.parquet	one line a row: test, conditions, model figures, measurements, differences	c0530e7cc20dda02…
data/index.csv	the same as CSV	05ea8ba83e736c12…
data/rows.parquet	the sealed rows, by genesis_seal_parquet	6d2f56747f6e567d…
data/frames.parquet	the sealed frames, by genesis_seal_parquet	3f19e200484f756b…
source/table_III_cold_startup.csv	the transcription of IDO-17281 Table III	067e646ea090e0b5…
source/table_V_hot_startup.csv	the transcription of IDO-17281 Table V	214341b2b05d75d0…
source/table_VII_hot_standby_operating_power.csv	the transcription of IDO-17281 Table VII	6a73f6c146eadb46…

Where the numbers come from

• The record: McCardell, R. K.; Herborn, D. I.; Houghtaling, J. E. (1969). Reactivity Accident Test Results and Analyses for the SPERT III E-Core: A Small, Oxide-Fueled, Pressurized-Water Reactor. Report IDO-17281, Phillips Petroleum Co., for the U.S. Atomic Energy Commission. OSTI 4792676, https://doi.org/10.2172/4792676. The PDF read has SHA-256 2d6482caa77021a60b0542622df04d13b0e7c0c3da4511f008c15470f3e30349. Its cover reads "Distribution of this document is unlimited."
• The transcription of its three tables of tests, read from the report's pages: table_III_cold_startup.csv (Table III, SHA-256 067e646ea090e0b5ab072d540436ed0450a582ed9f19d628ed155cef20e64e9b); table_V_hot_startup.csv (Table V, SHA-256 214341b2b05d75d0f2f12dfffb7ac8230eac26242463383e6b1d25ef90f7b4b4); table_VII_hot_standby_operating_power.csv (Table VII, SHA-256 6a73f6c146eadb469a26fae3f6230c2667a3f570dc340b21aa9019a590572b17).
• The conditions the tables do not print: Inputs as the twin's own description (describe_card energy/reactor_spert3_twin) and the comparison it was fitted by give them: the start powers it was fitted with for the three startup series, 11.7719 W cold, 4.04032 W at 260 degF and 12.4711 W at 500 degF (the report gives none: about 50 W, 10 to 50 W); the two pressures it was compared at, 0.0853 MPa (the atmosphere at the Idaho site) cold and 10.44 MPa (1500 psig) hot; the reactivity of Tables III and V as the inhour value of each printed period, with the report's reduced generation times (Table II: 2.15, 2.18 and 2.25 ms at 70, 250 and 500 degF) and six delayed groups (Appendix B, p.132), and Table VII's reactivity as printed; and each run carried 2 s past its first maximum for the startup series and to 1 s for Table VII, as that comparison was made, so the clad maximum after the peak is in the run.

What the rows do not say

• The reactivity of the tests of Tables III and V is the inhour value of each printed period (the period's standard deviation is 2 percent); index.parquet keeps the printed value beside it as cond_reactivity_printed_dollar. Table VII prints no period, and its reactivity is as printed, with a standard deviation of 0.02 to 0.05 dollars.
• Four hot-startup tests (24, 25, 52, 53) were ended by a scram while power was still rising; their measured "peak" is the power at the scram, so they are left out of the comparison above and kept in the rows.
• For tests 79, 80, 84, 85 and 86 the report's timing channel was started by hand (report p.148), so their times of peak are not measured from the start of the transient.
• The report measures no fuel temperature; the clad temperature is from one thermocouple on one corner rod.
• 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 (v2.0): https://doi.org/10.5281/zenodo.23517223. 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.