Cyclotron Radiation Emission Spectroscopy (CRES) is an electron spectroscopy method developed by Project 8 to make a direct kinematic measurement of the neutrino mass using tritium beta decay. We present a high-resolution CRES analysis of the K, L1, L2, L3, M2, and M3 ^83mKr internal-conversion electron data from Phase II of Project 8. We develop an unbinned event-level likelihood that models the intrinsic Lorentzian line shape, the event-to-event distribution of average magnetic fields sampled by trapped electrons, and the start-frequency reconstruction error. Using this framework, we infer krypton shell binding energies and natural line widths. The analysis improves the precision of the L1 binding energy relative to the literature, while the other inferred binding energies remain consistent with the literature. The binding-energy precision is limited by magnetic-field uncertainty. We therefore perform an analysis of mock data from the Cavity CRES Apparatus (CCA), the next-generation CRES experiment currently being commissioned, based on a simplified model in which improved reconstruction and independent detector-response calibration reduce this magnetic-field uncertainty.