Available estimates for the energy resolution of DUNE vary by as much as a factor of four. To address this controversy, and to connect the resolution to the underlying physical processes, we build an independent simulation pipeline for neutrino events in liquid argon, combining the public tools GENIE and FLUKA. Using this pipeline, we first characterize the channels of non-hermeticity of DUNE, including subthreshold particles, charge recombination, and nuclear breakup. Particular attention is paid to the role of neutrons, which are responsible for a large fraction of missing energy in all channels. Next, we determine energy resolution, by quantifying event-to-event stochastic fluctuations in missing energy. This is done for several sets of assumptions about the reconstruction performance, including those available in the literature. The resulting migration matrices, connecting true and reconstructed neutrino energies, are presented. Finally, we quantify the impact of different improvements on the experimental performance. For example, we show that dropping particle identification information degrades the resolution by a factor of two, while omitting charge deposits from de-excitation gammas worsens it by about 25%. In the future, this framework can be used to assess the impact of cross section uncertainties on the oscillation sensitivity.