The Amazon basin is the largest rainforest in the Earth and of great relevance for the global hydro-climate and biodiversity (Marengo, 2006;Phillips et al., 2008). It is also a region, like many in the tropics, where climate model precipitation biases are both large and systematic. These biases are evident in every aspect of the representation of precipitation, from its spatial and temporal distribution, to its intensity and form. Models systematically have too little precipitation over the northern Amazon (e.g., Fiedler et al., 2020;Yin et al., 2013). The diurnal cycle is characterized by a too early precipitation peak (Betts & Jakob, 2002;Tang et al., 2021) and evidence of convective organization (Mapes & Neale, 2011), which has been estimated to account for up to 50% of the total Amazon rain (Feng et al., 2021), is effectively absent. In this study, we use kilometer-scale "storm-resolving" simulations over large domains to assess the degree to which they reduce these biases and the extent to which this depends on the explicit representation of organized convective systems (OCS). In doing so our premise is that convective features which are not improved, or for which remaining biases show no clear sign of improvement with increases in resolution, are indicative of an important role for non-convective, for example, cloud microphysical, small (sub hectometer) scale mixing or land-surface processes.