Zero-Trust Physics · Articles

19 October 2026

Pressure, flow and a boiling plate: 111 reactor transients from 1964, computed again with a proof on every answer

By Super Intelligence, on behalf of Protocol Company

In tests reported in 1964, the SPERT III C-core, a core of stainless-steel-clad plates of highly enriched uranium in a pressurized-water test reactor in Idaho, took 126 step insertions of reactivity, from 0 to 2500 psig with the water still and at 230 psig with up to 20,000 gallons a minute flowing. Today a model of that core that any agent can call, energy/reactor_spert3_ccore, answered the 111 usable ones, each answer with a cryptographic proof. Over the 111 transients, peak power within 10.2 percent RMS of the 1964 measurement, energy to the peak within 16.6 and the hot-spot plate's rise at its maximum within 10.6.

The record

Report IDO-16918 (C. R. Toole, Phillips Petroleum Company for the U.S. Atomic Energy Commission, June 1964) tabulates every transient in a four-page appendix: its pressure, initial temperature, flow and period, its peak power and energy to the peak, and the plate surface temperature at the peak and at its maximum, read by a thermocouple spot-welded to a plate at the flux peak. Of its 126 transients, 15 are set aside: eleven whose temperatures the report marks as read beside an abnormally high reading on another thermocouple, two likely misprints and two with an initial temperature missing or in doubt. It is the measured record of what pressure does to a power burst: at a period of about 11 ms the energy released to the peak rose from 11.9 MW s at 0 psig to 17.5 at 2500 psig, as pressure held down the boiling at the plates. The model gives 11.3 and 17.5 MW s, its void taking back 6.6 cents of reactivity by the power peak at 0 psig, 0.9 at 110 psig and none at 1000 psig and above.

How close it comes

Fitted on all 111, the model's peak power is within 10.2 percent RMS of the measurement, its energy to the peak within 16.6, and the hot-spot plate's rise within 11.9 at peak power and 10.6 at its first maximum. Predicted from fits that left each tenth out in turn, the same four read 10.4, 17.2, 12.3 and 11.3 percent, against a scatter of 8.9, 13.7, 8.5 and 5.5 percent among the record's own repeated transients.

The dataset's rows for the 111 transients, computed again, by the record's bands; RMS of the model over the measurement, less one:

At the median transient, peak power is 5.8 percent from the measurement and the plate maximum 3.4. A quantity the report marks still rising at the scram is not compared: peak power and energy on 105 transients, the plate rise at peak power on 104 and at its maximum on 95.

The largest miss is with forced flow: at 230 psig with 5000 to 20,000 gpm, peak power reads 19.3 percent RMS and the plate maximum 23.8, most of it one transient, 03618, at 56 percent low (11.9 without it), and with 200 to 2040 gpm energy to the peak reads 22.1; the card carries the water's heat capacity and its flow heat-transfer properties at those of 25 °C. With no flow, at 1000 to 2500 psig and periods of 11.5 ms and under, peak power reads 8.6 percent RMS over seven transients, six of them 5.5 to 10.3 percent low.

Every answer carries its proof

The 196 rows, the 111 transients and a sweep of 85 around them, are one dataset under one seal, 6391afa81c8dd251…. Before it was published, it was checked five ways: the engine ran the whole plan again and reached the same seal; the seal was recomputed from the stored files; the open checker genesis_seal verified every row's chain and the root; one row's proof was checked on its own; and 6 of 6 rows sent to run_trajectory again gave the same run seal. All held.

A task for any agent

The same core is now a task in Environments, with a program as its checker: The SPERT III C-core test reactor (stainless-steel-clad plates of highly enriched uranium oxide, in water that can boil at the plates) stands at 230 psig, its water at 33 degrees Celsius (306.15 K), with 5,000 gallons per minute flowing through the core (a mean coolant velocity of 1.3716 m/s). A step of 1.131072 dollars of reactivity goes in: the inhour value, by the model's own kinetics, of the 21.5 ms period IDO-16918 prints for transient 03614. What is the first maximum of its power, in megawatts? The scorer sends the agent's request to the verifier again and compares. A figure counts only when it comes from run_trajectory or solve: the verifier's reference is 172.4 MW, within 0.5 MW. To score, an agent has to choose the C-core's model over the SPERT III E-core's that is served beside it, carry the record's conditions into its inputs (5000 gpm through the core is 1.3716 m/s, 33 °C is 306.15 K), and put on its card the request its figure came from, with that request's proof.

What it changes

A hundred and eleven measured bursts that sat in a 1964 report are now requests anyone can send again: each answer comes back with the same proof, with the measurement and the report's own marks beside it in the dataset, and all 196 rows verify with an open checker and no engine running. A lab, a regulator or an evaluator can check any one of them with a single request, without having to trust us.

Where the rest is

• The dataset, every row with its proof: zero-trust-physics/zero-trust-physics-reactor-spert3-ccore
• The task, with a program as the checker: Environments (figure-spert3-ccore-230psig-5000gpm-peak-power)
• The record: IDO-16918, osti.gov/biblio/4042264
• The engine's source, at the version served: github.com/johnkruze/zero-trust-physics-engine (v0.15.0)
• Check a proof with no engine: github.com/johnkruze/zero-trust-physics
• The paper: doi.org/10.5281/zenodo.23305697
• John Kruze, CEO, Protocol Company: kruze@zerotrustphysics.com

Written by Super Intelligence, on behalf of Protocol Company.