In this work we use mathematical modeling and complementary experiments to study the dynamics of modulation in the accessory radula closer (ARC) neuromuscular system of Aplysia. Here we join a dynamic model of the modulation from the preceding paper to a model of the basal neuromuscular transform (NMT). The resulting complete model of the NMT allows us to predict, test, and analyze the actual modulated contraction shapes in different types of feeding behavior, through entire quasi-realistic meals. The model reproduces a variety of published and new experimental observations. We find that components of the modulatory network act in interdependency and mutual complementarity, one or another playing a key role depending on the behavior and its past history. The history is remembered by slow dynamical components whose persistence prepares the system for future behavior of the same kind. The persistence becomes counterproductive, however, when the behavior suddenly changes. Superposition of fast dynamical components alleviates the problem under most, but not all, circumstances. In the quasi-realistic meals, the modulation improves functional performance on average, but degrades it after certain behavioral switches, when the model predicts sharp contraction transients. These are indeed seen in the real muscle. We propose that the real system does not switch the underlying motor neuron firing patterns abruptly, but relaxes them gradually, matching the relaxation of the peripheral modulatory state, through such behavioral transitions. We model food-induced arousal, a known phenomenon of this kind. The peripheral dynamics of the modulated NMT thus constrain the motor commands of the CNS.