The Dynamics of the MJO (DYNAMO) field campaign took place in the Indian Ocean during boreal fall and winter of 2011–2012 to collect observations of Madden-Julian Oscillation (MJO) initiation. Hindcast experiments are conducted with an atmospheric general circulation model with varying values of a dilute CAPE entrainment rate parameter for the first two MJO events of DYNAMO from 1 October 2011 to 15 December 2011. Higher entrainment rates better reproduce MJO precipitation and zonal wind, with RMM skill up to 20 days. Simulations with lower entrainment rapidly diverge from observations with no coherent MJO convective signal after 5 days, and no MJO predictive skill beyond 12 days. Analysis of the tropical Indian Ocean column moist static energy (MSE) budget reveals that the simulations with superior MJO performance exhibit a mean positive MSE tendency by vertical advection; inconsistent with reanalysis that indicates a weak negative tendency. All simulations have weaker mean MSE source tendency and significantly weaker cloud-radiative feedbacks. The vertical gross moist stability (VGMS) is used to interpret these MSE budget results in a normalized framework relevant to moisture mode theory. VGMS in the high entrainment runs is far too low compared to ERAi, indicating that it cannot be used in isolation as a measure of model success in producing a realistic MJO hindcast, contrary to previous studies. However, effective VGMS that includes radiative feedbacks is similar among the high entrainment runs and ERAi. We conclude that the MJO is erroneously improved by increasing the entrainment parameter because moistening by vertical MSE advection compensates for the overly weak cloud-radiative feedbacks.