There is a well-identified problem of (lack of) reproducibility in experimental Computer Science research, and in particular in High Performance Computing (HPC) research. Running experiments in simulation is one way to lower barriers to reproducibility, and it is used effectively in several areas of Computer Science. Its use in HPC research has gained some traction, but comparatively it is still in its infancy. Part of the reason is that developing simulations models that are sufficiently accurate to produce meaningful results but that are also sufficiently scalable to handle the scale of HPC simulation is challenging. In this presentation we will discuss several advances in the development of such simulation models. These models are implemented as part of the open-source SimGrid simulation framework. We will give an overview of SimGrid, describe its most salient features and capabilities. We will conclude by discussing ways in which SimGrid can be used not as a research tool, but as a tool for debugging, for teaching, and for enabling online decision making.