The Validation Hierarchy
Complex systems are validated bottom-up: unit-physics, then subsystems, then the full system, because the whole is rarely testable directly.
Building Confidence in Layers
A full engineered system is often impossible to test under operating conditions before it is built, and even then only a few configurations can be measured. The validation hierarchy addresses this by decomposing the system into levels: individual physical phenomena at the bottom, their combinations in subsystems above, and the complete system at the top. Each level is validated against experiments appropriate to it.
The Levels
- Unit problems: a single phenomenon in a controlled setting, where the model can be tested cleanly.
- Benchmark problems: a few coupled phenomena, testing whether the model captures their interaction.
- Subsystem cases: an assembly of physics representative of part of the system.
- Full system: the complete configuration, often only partially testable.
Why Bottom-Up
Confidence flows upward. Validating the unit physics removes it as a source of uncertainty when a discrepancy appears at a higher level, letting the coupling be tested in isolation. Trying to validate the full system directly, without validating the pieces, leaves any discrepancy unattributable: it could be any of the phenomena or any of their couplings. The hierarchy makes discrepancies diagnosable.
Extrapolation to the Untested
The top of the hierarchy is usually where direct full-system validation is thin or absent, so the full-system prediction is an extrapolation supported by the validated lower levels. That extrapolation carries its own uncertainty, which grows with the distance from the validated regime. Honest reporting states the extrapolation explicitly rather than presenting the prediction as if it were directly validated.
For fusion systems not yet built, including the Hyperion breeder and burner concepts, the hierarchy is the natural structure: component and unit physics validated against existing devices and experiments, with full-system behavior presented as a projection whose gates and uncertainties are named rather than hidden.