Nanoscale self-assembly is naturally subject to impediments at the nanoscale. The recently developed ability to follow processes at the molecular level forces us to resolve older, coarse-grained concepts in terms of their molecular mechanisms. In this Letter, we highlight one such example. We present evidence based on experimental and simulation data that one of the cornerstones of crystal growth theory, the CabreraVermilyea model of step advancement in the presence of impurities, is based on incomplete physics. We demonstrate that the piercing of an impurity fence by elementary steps is not solely determined by the Gibbs-Thomson effect, as assumed by Cabrera-Vermilyea. Our data show that for conditions leading up to growth cessation, step retardation is dominated by the formation of critically sized fluctuations. The growth recovery of steps is counter to what is typically assumed, not instantaneous. Our observations on mesoscopic impurities for lysozyme expose a nucleation-dominated regime of growth that has not been hitherto considered, where the system alternates between zero and near-pure velocity. The time spent by the system in arrest is the nucleation induction time required for the step to amass a supercritical fluctuation that pierces the impurity fence.