Example of a perfectly nested bipartite network (upper box), outline of phenology model, and model assessment scheme (lower box). A link is included in an empirical network if an interaction is recorded during the data aggregation window between two actors of different types, node sets {A,B,C} and {x,y,z}, representing, e.g. different plant and pollinator species, respectively. In the corresponding incidence matrix, filled entries indicate links between nodes (rows and columns) and nestedness is characterized by an upper triangular or staircase pattern, as shown. The phenology model starts with a description of when actors are present and active in the system during the data aggregation window, which in this case is 6 days (a). This temporal information dictates the number of copresences between each pair of nodes (b), which is used to determine the probability of each link in model-generated networks (c). Specifically, the probability of an interaction given a copresence is assumed to be p and the probability of no interaction $q=1−p$⁠, and so the probability of at least one interaction during the data aggregation window given n copresences is $1−qn$⁠. Link probabilities are scaled such that the expected number of links for an ensemble of model-generated networks equals the number in the empirical network (see Materials and methods). Each model realization returns an incidence matrix (d) and each link can be assigned to one of four categories: true positive (TP)—link present in both model and empirical networks; true negative (TN)—link absent in both model and empirical networks; false positive (FP)—link present in model network but absent in empirical network; false negative (FN)—link absent in model network but present in empirical network.