Twin Validation and Verification
Verification asks whether the twin solves its equations correctly; validation asks whether those equations match reality.
Two different questions
A twin makes decisions, so its trustworthiness must be earned, not assumed. Two distinct questions govern that trust. Verification: are the models implemented and solved correctly, free of numerical and coding error? Validation: do the models actually reproduce the behavior of the real system? A twin can be perfectly verified and still wrong, if its physics does not match the world.
Verification methods
- Compare against known analytical solutions where they exist
- Check convergence as the numerical resolution is refined
- Cross-check independent codes against each other on the same problem
- Test conservation of mass, energy, and charge as internal consistency checks
Validation methods
Validation compares model predictions to measured data the model did not see during its construction. It is graded: a twin validated on routine operation is not thereby validated for rare instabilities. Honest validation states its domain, the range of conditions over which the twin has actually been tested against reality.
Validation before hardware
Because the Kronos machines are not built, validation against their own operating data is not yet possible. The available substitutes are validation of the component physics codes against existing experiments and published data, cross-code verification, and observing-system simulation experiments, in which the twin assimilates synthetic data from a high-fidelity reference model to prove the assimilation machinery works. These build confidence in method while being explicit that plant-level validation waits for first tritium near 2030.
The honest posture
No net-gain claim is made for either machine before the first-of-a-kind breeder Hyperion operates. The twin's predictive skill is likewise unproven at the plant level until then. Stating the validation domain plainly, and refusing to extrapolate the twin's authority beyond it, is the discipline that keeps the program credible. See uncertainty quantification and fidelity levels.