Skip to content
Technology How it works Breeder — Hyperion Burner — Aegis Burner — MetroVolt AI-Native Architecture Magnets Fuel cycle Safety Roadmap
Solutions AI & Data Centers Defense & Government Grid & Baseload Neutron Detection Quantum
Learn Technical Library
Proof Publications Whitepapers Technical Library Open Science & Reproducibility The Honest Gates
Company About / Mission Leadership Environment Health & Safety Investors Careers Press Contact
3D Model
AI Architecture › L0 · Foundation
L0 · Foundation

The Twin Refinement Loop

How discrepancies between the KRONOS-CTRL twin and reality are captured, diagnosed offline, and folded back into better models.

THE STACK · click to jumpL7Ecosystem & StrategyL6Experience & VisualizationL5Applications & CopilotsL4OrchestrationL3Twin Modeling & AIL2Data FabricL1Control PlaneL0Foundation▲tlmctl▼L0 · FOUNDATIONThe offline compute substrate — multi-physics & batch training.1Cloud HPCelastic burst2Bare-Metal ClusterGPU / CPU3Supercomputingmulti-physics runs4Batch Trainingmodel builds5Simulation FarmGrad-Shafranov · MHD6Object StorecheckpointsMACHINE TIETrains the models that ship UP to L3 — no real-time path to the machine.KRONOS FUSION ENERGYAI-NATIVE S.M.A.R.T. GENERATORFOUNDATIONSHEET 02REV. 2026-08L0 · AI-NATIVE STACK
L0 · Foundation — its place in the stack (left, click any layer) and its internal components (right). Telemetry rises; control descends.

Closing the loop between fast and slow

The twin refinement loop is the feedback path that keeps the real-time twin honest. When the KRONOS-CTRL twin's 50 to 100 ms prediction diverges from what actually happens, that residual is a signal. L0 collects these residuals, diagnoses their cause, and refines the models so the next version predicts better. It is how the twin learns from its own mistakes.

Residual as data

Every predicted-versus-actual mismatch, in the Power, Neutronics, Thermomechanics, or MHD module, is logged with full context: the inputs, the prediction, the outcome, and the diagnostic state. Aggregated over many pulses, these residuals reveal where a surrogate is systematically wrong, which is far more useful than any single miss.

Diagnosis before retraining

Not every residual means the model is wrong. It might reflect a bad sensor, a data-fabric issue, or genuinely new machine behavior. The loop diagnoses the cause first, because retraining on mislabeled or corrupted residuals degrades the model. Only cleanly attributed discrepancies are folded into the next retraining cycle.

The loop is deliberately offline. Refinement requires re-solving physics, curating data, and revalidating, none of which can happen inside a real-time control budget. The twin runs fast and fixed during operation; it improves only between deployments, through this slow, careful L0 process.

For both machines, the loop is what turns operating experience into better prediction. Until FOAK around 2030 it runs against simulated pulses and study campaigns, exercising the refinement machinery so that when real breeder and burner pulses arrive, the loop is already trusted and ready to learn from them.

Content reviewed August 2026 · design-and-simulation stage