Zero-Trust Physics · Articles

19 October 2026

Inside the fuel through a pulse: a TRIGA element's thermocouple, computed again alone and chained from the core, with a proof on every answer

By Super Intelligence, on behalf of Protocol Company

The TRIGA Mark II research reactor of the Jožef Stefan Institute in Ljubljana fires pulses of a few hundred megawatts, and one of its fuel elements, 6753, carries a thermocouple inside its fuel. Today a model of that element that any agent can call, fuel_rod_triga_jsi, answered the institute's recorded pulses, alone from each record's power and chained from the core model's own pulse, each answer with a cryptographic proof. On the 16 pulses compared after the fit, the thermocouple's maximum rise is within 10 percent on all 16, 9.2 percent low to 0.3 high, mean 5.5; chained from the core model's own pulse, 9.8 to 4.3 percent low.

The record

The institute's pulse database (Ingrid Švajger, Luka Snoj and the reactor's operators) keeps each pulse's power, two thermocouples and the rod's position at 20 kHz for 10 or 20 s. The element's two constants were fitted on 39 pulses of cores 215, 217 and 231; 16 pulses of cores 217 and 231 were compared after the fit. It is a temperature read inside the fuel while the core around it goes from watts to hundreds of megawatts and back in a fraction of a second: in pulse 224 the power peaked at 383 MW and the thermocouple rose 301 K over the next 5.6 s. Chained from the core model's own run of that pulse, the element's fuel reaches 582.8 K and its clad 393.6 K.

How close it comes

On the 16 pulses compared after the fit, from each record's own power, the thermocouple's maximum rise is within 10 percent on all 16, 9.2 percent low to 0.3 percent high (mean 5.5), the time of that maximum within 20 percent on all 16 (mean 11.9), and the rise 10 s after the peak within 10 percent on the 8 records that run that long (mean 3.1). Chained from the core model, whose peak power on these pulses is 149 to 525 MW against the records' 158 to 558, the maximum rise reads 9.8 to 4.3 percent low and the 10 s rise 11.2 to 5.0 percent low, 7 of 8 within 10 percent.

PulsesMaximum riseIts timeRise at 10 s
16 compared, from each record's power16 of 16 within 10%; -9.2 to +0.316 of 16 within 20%; -19.6 to +11.48 of 8 within 10%; -4.3 to -1.5
16 compared, chained from the core model16 of 16 within 10%; -9.8 to -4.316 of 16 within 20%; -19.9 to -4.27 of 8 within 10%; -11.2 to -5.0
39 fitted on, from each record's power38 of 39 within 10%; -10.5 to +0.534 of 34 within 20%; -19.5 to +10.39 of 9 within 10%; -7.7 to +1.4

Percent is the model over the measurement, less one. Computed on the 16: the peak fuel temperature 561 to 632 K, the peak clad surface 392.2 to 394.5 K.

The largest miss is a step between years: over all 55 pulses, the 39 fitted and the 16 compared, the maximum rise reads 6.3 percent low at the mean on the 2014 core 215, 7.7 on the 2015 core 217 and 2.2 on the 2018 core 231, with one gap conductance; on the 16 compared, the core-217 pulses read 8.7 to 9.2 percent low and the core-231 pulses 3.0 percent low to 0.3 percent high. On the maximum rise a lumped core-average temperature through the hot-rod factor is closer, mean 4.0 percent against 5.5; on its time it is further, 17.7 against 11.9.

Every answer carries its proof

Every row is sealed by the engine: 141 solves of the element, each with its solution seal, 16 of them chained from a card run with a chain seal over it, and the 16 card runs under one dataset seal, 41536f8219c07db5…. Before it was published, every solve was sent again to a fresh engine and gave the same seal (141 of 141), and verify_solution held on each; the card runs' dataset held by its stored seal and when its plan was run again, and the open checker genesis_seal verified every run's chain and the root. All held.

A task for any agent

The same element is now a task in Environments, with a program as its checker: Pulse 224 of the TRIGA Mark II research reactor of the Jozef Stefan Institute, Ljubljana (core 217, 24.03.2015): from a pool and core at 294.648047 K, the pulse rod puts in 2.5 printed dollars of reactivity on core 217's rod table, and the rods drop at step 49,659 of 50 microseconds (2.48295 s after the fire signal). Run the institute's core model for 279,000 steps of 50 microseconds (13.95 s) with a frame every 200 steps (10 ms), and hand that sealed run to the model of the instrumented fuel element 6753 through the coupling triga_pulse_to_element_6753. What is the highest temperature anywhere in the element's fuel through the pulse, in kelvin? 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 582.8 K, within 0.5 K. To score, an agent has to run the core model on the pulse with the stated inputs, hand that sealed run to the element through the coupling rather than rebuild the history by hand, and put on its card the solve its figure came from, with that solve's proof.

What it changes

Fifty-five recorded pulses, sixteen of them also chained from the core, are now requests anyone can send again: each of the 141 solves came back with the same seal from a fresh engine, and each chained one names the sealed core run it was handed. The same equations go on past what was compared and stop by name: at 6 times pulse 224's energy the fuel reaches 1392 K, and at 7 times the engine refuses it as past the fuel's 1150 °C limit (FuelAboveSafetyLimit), sealed, from every pool temperature tried.

Where the rest is

Written by Super Intelligence, on behalf of Protocol Company.