Answers are grounded in this paper and cite their sources.
Reduced-order models of the Kilowatt Reactor Using Stirling TechnologY (KRUSTY ) heat-pipe microreactor test are usually scored against curves digitized from published figures. This paper asks which of their apparent gains survive when the heat-load forcing is not measured independently. We reconstruct source-tagged power and outer-fuel-temperature series and grade the evidence at three levels: calibration (8–12 h), transfer to a later window (20–28 h), and a forward prediction (16–19.8 h) driven only by the documented reflector-platen movements. All five 20–28 h heat-load amplitudes are back-inferred from the power record of the same test, so the transfer is a transfer of thermal parameters, not an independent-input validation. Under a common load input, two thermal stores reduce the 20–28 h power root-mean-square error (RMSE) by 63.9–65.7% relative to one store. Normalized by the mean power of that window, the error falls from 15.4% to 5.6%; in absolute terms it falls from 0.278 to 0.101 kW, and the temperature RMSE from 3.60 to 1.89 °C. A third store lowers the power RMSE only to 0.096 kW, while one of its conductances reaches its upper bound, and with the fuel capacity varied the fitted total storage stays near 19.3 kJ/K, so the data constrain the total storage rather than its distribution. Among the model orders tested, two stores therefore give the best-supported balance between residual reduction and parameter stability. In an out-of-window forward evaluation that uses no post-event output and an observed initial state, the two-store model reaches 0.376 kW and 4.99 °C. The contribution is therefore a reproducible benchmark and an evidence-grading procedure, not a claim of accuracy superior to high-fidelity models.
No takes yet. Share an insight, caveat, or question.
Liu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: