The Epistemics of Simulation
Simulation produces knowledge only under conditions; understanding those conditions is what separates a result from a rendered picture.
What a Simulation Actually Knows
A simulation is an argument, not an oracle. It says: if these equations hold, if these parameters are right, and if the numerical method converges, then the world should behave like this. Every conclusion inherits the strength of the weakest link in that chain. Treating output as measurement, rather than as conditional inference, is the most common epistemic error in computational science.
Three Sources of Error
A simulated prediction carries three distinct error sources that must not be conflated. Model-form error comes from choosing equations that only approximate reality. Numerical error comes from discretizing continuous equations onto a finite mesh and finite time step. Parametric uncertainty comes from imperfectly known inputs such as transport coefficients or material properties.
- Model-form error is reduced by better physics, not finer meshes.
- Numerical error is reduced by refinement and bounded by verification.
- Parametric uncertainty is reduced by measurement and propagated by uncertainty quantification.
Why Agreement Is Not Proof
A simulation can match data for the wrong reasons, when model-form error and numerical error cancel by accident. Such agreement is fragile: it dissolves the moment the mesh, the regime, or the parameter set changes. Genuine epistemic weight comes from agreement that survives refinement and holds across regimes the model was not tuned to.
At Kronos
The Hyperion breeder and the burner are design and simulation work, not built hardware. Their published physics claims are therefore conditional statements about verified solvers and stated assumptions, kept distinct from claims about operating hardware. No hardware net-gain claim is made before FOAK first tritium. The discipline is to report what is inferred and under which conditions, never to present a rendering as a reading.