The AMIE/DYNAMO 2011-12 field campaign collected unprecedented observations of evolution in cloud population during initiation of the Madden-Julian Oscillation (MJO). The field observations provide ample information to advance our understanding of interaction between convection and its large-scale environment. But it remains a challenge as how to use such information to benefit the development of global cloud-permitting models, which have emerged to demonstrate the future of climate models. Cloud-permitting models, by explicitly resolving meso-scale convective systems, partially shift sources of errors and uncertainties from cumulus parameterization to parameterization of cloud microphysics and shallow clouds. When a field campaign does not provide all needed observations of microphysics, it still helps evaluation and development of cloud-permitting models through assessing their simulations of macro-scale cloud behavior using a hierarchy of model configurations. In this study, a three-tier modeling-observation comparison strategy is designed to channel field observations, essentially at a point from a global viewpoint, to evaluation and development of global cloud-permitting models. The three tiers consist of conventional cloud-resolving models (CRMs) of small domains covering an area of field observations with strong observational constraints and relaxed constraints based on various theoretical assumptions, cloud-permitting limited area models (CLAM) with weak constraints through lateral boundary conditions, and global cloud-permitting models (GCPM) with observational constraint only at the lower boundary. Comparisons are made between observations and models and between models to identify their major sources of discrepancies: parameterization of microphysics, grid spacing, observational constraints, etc. MJO events observed during the field campaign are used as targets for model evaluation against observations. Results at the mid-point of this project show general agreement between observed and simulated (by CRMs and CLAM) cloud distributions and evolution. But the models produced excessive stratiform rain and insufficient shallow clouds. Sensitivity of the simulations to cloud microphysics is large, as expected. The GCPM (NCAR MPAS) simulations produced some most unexpected results: Simulations with finer grid spacing (3 km) without cumulus parameterization (NP03) do not outperform those with coarser grid spacing (15 km) and parameterization of shallow convection (SP15) in terms of the diurnal cycle and the MJO. This and other results suggest that shallow convection is very important to overall simulations of precipitation but only over the ocean. Even without cloud-permitting grid spacing, shallow convection cannot be explicit resolved and must be parameterized because of their importance.