Much of the atomistic modeling being undertaken still uses either simple central-force models of interatomic interactions, e.g., the embedded-atom method or Finnis-Sinclair potentials, or models which include angular interactions in the most simple or ad hoc way. Such models typically constitute considerable simplifications of the underlying quantum mechanics and as a result often fail to reproduce the structures and energetics of even the most basic lattice defects. This somewhat disparaging situation makes it difficult to gain confidence in the predictive capabilities of empirical atomistic calculations, particularly when the results are used as microscopic input to higher length- and time-scale modeling methods; a regime which can give fundamentally different results when input energies might vary by only a few tenths of eV.