Establishing atomistic models of amorphous materials remains a central challenge in materials science. Here, we examine three representative multicomponent monolayer systems using first-principles-trained machine-learning potentials (MLPs) and an energy-guided Monte Carlo structural-search workflow. For monolayer amorphous boron nitride (maBN), independent density-functional-theory (DFT) single-point calculations confirm that the ensemble generated by the MLP-driven search is lower in energy on average than an extended-Tersoff-generated ensemble. Across the sampled maBN structures, local-ring classification combined with strict connected-domain analysis identifies h-BN-like, o-B₂N₂-like, and o-B₄N₄-like crystallite domains together with mixed medium-range order. Only individual samples containing at least two distinct phase-pure crystallite-domain types are classified as polymorphic crystallites. The same analysis finds coexisting hexagonal and tetragonal domains in monolayer amorphous LiCl and graphene-like, h-BN-like, borophene-like, and mixed-order regions in monolayer amorphous BCN. Five-nanosecond MLP molecular-dynamics trajectories at 300 K show that the candidate structures are dynamically persistent on the simulated timescale. These results support a polymorphic-crystallite description for the selected model systems, while experimental discrimination from alternative continuous-random-network descriptions will require medium-range-order-sensitive measurements.