Levels of Twin Fidelity
Twins span a ladder from status mirrors to full predictive physics; the right rung depends on the question being asked.
Fidelity is a choice, not a maximum
More physics is not automatically better. A twin exists to answer questions within a time and accuracy budget, and higher fidelity costs computation, data, and validation effort. Engineering a twin means choosing the lowest fidelity that answers the question reliably, and reserving heavy physics for the questions that demand it.
A ladder of fidelity
- Descriptive: mirrors current sensor values and derived indicators, no forward prediction
- Diagnostic: adds state estimation to infer quantities that are not directly measured
- Predictive: rolls the estimated state forward to forecast behavior and remaining life
- Prescriptive: evaluates candidate actions and recommends or executes the best
Most operational twins are a federation running different rungs at once: a descriptive layer for the control room refreshing many times per second, a diagnostic layer reconstructing plasma or thermal state, and a slower predictive layer forecasting component life over campaigns.
Matching fidelity to the question
Reconstructing the magnetic equilibrium of the Hyperion breeder plasma for control needs a fast reduced model that runs inside the control cycle. Predicting fatigue accumulation in a magnet support after thousands of pulses can use an offline high-fidelity structural model run overnight. Forecasting tritium inventory across the breeding blanket needs neutronics coupled to chemistry, updated per campaign rather than per millisecond.
Fidelity and honesty
Because both Kronos machines are pre-construction, every twin fidelity claim rests on validated simulation, not on operating data. The honest posture is to state that clearly: the twin's predictive skill will be established only after the first-of-a-kind breeder produces data near 2030. Until then fidelity is bounded by how well the underlying physics codes have been verified and cross-checked, not by how well the twin has tracked a running plant.
See surrogate models for how high fidelity is made fast enough for the control cycle, and validation and verification for how fidelity claims are tested.