Our ability to predict the thermodynamic phase behavior of a material system is a direct reflection on our understanding of the relevant interactions. Voorn−Overbeek (VO) theory, which combines Flory−Huggins polymer mixing with Debye−Huckel electrostatics, has been used to describe the associative liquid−liquid phase separation phenomenon known as complex coacervation since the 1950s. The long-standing utility of this theory stems from its simplicity coupled with its apparent agreement with physical systems. VO theory has also served as the starting point for a large class of field theories that predict similar phase behaviors. Recent work using new hybrid simulation methods demonstrates novel coacervate-driven self-assembly is strongly affected by molecular details. Liquid state (LS) theory suggests there are fundamental reasons for this observation and that agreement between VO and experiment is fortuitous. It is hypothesized that VO/experimental matching is due to a cancellation of errors arising from the neglect of monomer-level charge connectivity and excluded volume effects. In this article, we use Monte Carlo (MC) simulations to confirm the earlier predictions from LS theory. We directly observe effects related to connectivity-driven charge correlations. We also observe strong exclusion of salt from the polymer-rich coacervate phase, in direct opposition with VO theory and in near quantitative agreement with experimental results. Strikingly, a comparison of predicted phase diagrams using identical system parameters shows that VO overpredicts coacervate phase behavior and that previous agreement with experiments was likely due to the use of unphysical fitting parameters. This work provides new insights into the mechanisms driving complex coacervation and shows promise for predicting coacervate phase behavior based on resolving molecular level charge structure.