The Intelligence Layer
The intelligence layer is the software and models that turn raw plant data into predictions, decisions, and design insight.
What the layer is
Above the sensors and actuators of a fusion machine sits a layer of software whose job is to make sense of data and inform choices. It spans physics models, statistical estimators, machine-learning surrogates, optimizers, and the control logic that ties them together. We call this the intelligence layer because its purpose is inference, not just measurement.
What it does
- Estimates quantities that cannot be measured directly, such as internal plasma profiles.
- Predicts near-future behavior fast enough to act on it.
- Searches large design spaces for configurations that meet many constraints at once.
- Flags anomalies against a learned or modeled baseline.
Layers within the layer
At the bottom are first-principles models grounded in physics equations. Above them sit reduced-order and surrogate models that trade some fidelity for speed. On top are decision components — optimizers and controllers — that consume model outputs. Each level has different accuracy, latency, and trust characteristics.
Trust and traceability
Because outputs may influence expensive or safety-relevant decisions, every component must be verifiable. That means known inputs, recorded versions, and reproducible results. An intelligence layer that cannot explain how it reached a number is a liability, not an asset.
Design and operations, one stack
The same models used to design the breeder Hyperion and the D–³He burner appear later in operation, adapted for speed. Keeping design and operations on a shared modeling stack means insight flows both ways: field data refines design models, and design physics anchors operational estimates.