Rainfall manipulation experiments are among our best physical approximations of future scenarios and serve as important benchmarks for measuring the performance of terrestrial biosphere models. By synthesizing results from rainfall manipulation experiments across a range of biomes and environments, a new study by Paschalis et al.(2020) diagnoses current limitations in model structure, identifies new data needs, and provides lessons for future model development.Predicting vegetation response to rainfall changes is an inherently challenging problem that manifests at multiple temporal and spatial scales. Water availability, as controlled by rainfall, can be extremely unpredictable. At the individual level, the interaction of rainfall with plants at multiple timescales-from the sub-daily to the interannual-combined with the nonlinear ways plants tend to respond to water availability, means that we cannot easily extrapolate plant responses at one timescale to another: Our usual rules of mathematical averaging and large numbers simply do not apply (Katul, Porporato, & Oren, 2007). At the site level, the effects of rainfall-as a physically conserved resource, unlike temperaturecan actually be moderated or magnified by intermediate controls on subsurface water availability (e.g., McLaughlin et al., 2020), plant water use and water potentials (e.g., Feng et al., 2018), and carbon allocation strategies (e.g., Martínez-Vilalta et al., 2016), all of which can more directly influence vegetation functions. Finally, at the ecosystem level, the effects of rainfall on overall ecosystem fluxes can be further modulated by other ecological and demographic processes, such that the net outcomes of competition, mortality, and recruitment must also be considered (Fisher et al., 2018). Thus, translating rainfall into ecosystem-level water, energy, and carbon fluxes is a nontrivial and cross-scale process, with the effects of rainfall at various spatial scales typically captured within terrestrial biosphere models by coupling a land surface scheme with a hydrological modeling component and a dynamic vegetation module. The main question that drives Paschalis et al. (2020) is whether ecosystem responses to rainfall variability-subject to all the scaling challenges above-have been adequately represented within terrestrial biosphere models. This is done by synthesizing results from 10 terrestrial ecosystem models applied to rainfall manipulation experiments at a number of sites spanning a wide range of biomes and environments.Any evaluation of model performance must first acknowledge that the accuracy of our predictions is still very much limited by our current and foreseeable computational capabilities, as well as the uncertainties in the observations used for benchmarking. Due to computational constraints, we must carefully consider the tradeoff between computational cost and model abstraction and aim to find the "just right" amount of model detail that can reproduce the expected responses at the appropriate temporal and spatial scales.Howe...