The document discusses how well particle simulation codes predict beam performance in linear accelerators. It finds that codes generally agree at the few percent level for core beam properties but differ more in details. While codes provide initial validation of machine designs, real machines have unknown errors, so the most important use of codes is showing how beams change with machine parameters. Code limitations include inaccurate initial beam distributions and transient field errors. Simulations can predict beam properties diagnostics cannot measure like halo distributions with large particle counts.