Applications focused on extremes require robust sampling to yield meaningful results. In the absence of adequate historical sampling, practitioners often employ empirical distributions to extrapolate to the tails of a probability distribution. With climate simulations, one can conduct multiple realizations of a simulation with the same forcing, to sample more of the internal variability in the model climate, which is one of the ways in which single-model large ensembles are used. These allow one to separate internal variability from the structural uncertainties that arise from model disagreement in a multi-model ensemble. With dynamically downscaled climate models, the ensembles of either of these types are generally small. A growing set of simulations, unprecedented in scale for the western U.S., is being generated using the Weather Research and Forecasting (WRF) regional climate model to downscale global climate models (GCMs) from the Coupled Model Intercomparison Project - phase 6 (CMIP6). Despite the care going into its formulation, downscaling a range of better-performing GCMs and multiple realizations of some of these, the ensemble is insufficient to sample structural uncertainties in some extremes. Thus strategies for appropriate sampling with the downscaled data are employed on an application specific basis. This can be a challenge for data users who are unfamiliar with the concept of internal variability, and its relative importance as a source of uncertainty, which is central to the proper interpretation of the downscaled data in practice. In one example, we pool data across the entire WRF ensemble to achieve adequate sampling of extreme events to characterize 99.99th percentile rainfall in Los Angeles County. In a second example, we discuss creating tools on the Cal-Adapt: Analytics Engine to help users determine when such a pooling approach may be appropriate, and when a single simulation sufficiently samples the population of interest so that a direct multi-model comparison is also possible. In both cases, it is important to communicate the reasons for these choices to data users who have less familiarity with climate modeling. This communication is achieved through visualizations that provide context, providing appropriate defaults, and additional climate education, as needed to avoid mis-use.