Large-scale simulations often use multiple agent representations to permit the study of specific multi-agent phenomena, and to find a balance between run-time performance and level of detail of the simulation. Although these approaches are effective, they do not always offer the desired level of analysis, especially when this level is between the resolutions of the models available. In this paper, we aim at offering a finer method in exploring this tradeoff by introducing an intermediate level between two given resolutions, which can apply to all agent models and allows a more progressive transition to offer the desired level of analysis. We introduce a framework for such a methodology and evaluate it through the extension of an existing approach, along two criteria: its impact on computational resources, and an estimate of the dissimilarity between a simulation using our methodology and one without. Initial experiments show that consistency is almost maintained while CPU gain varies from low to significant depending on the context.