It's both in every case so far. Machines are made and operated to discover empirical data, but we invent the engineering along the way. Computation is woefull

It's both in every case so far. Machines are made and operated to discover empirical data, but we invent the engineering along the way. Computation is woefully insufficient for particle level simulation. If we had a proper model of turbulence that issue would be lessened. We don't have a full grasp on plasma physics. If we did we could point to a single stellarator geometry and say "that one". The physics and engineering also interplay at many levels. Optimal decisions at every level take both the plasma physics and engineering into account.