Since a few decades, research is going through the so-called replication crisis. It is known that different software versions or CPU architectures will impact the performance and results of an algorithm. This awareness has a positive impact on the research as more and more published papers pay more attention at the reproducibility aspect by giving more details about the proposed algorithms and experiment parameters. Yet, implementation choices are often left out of the publishing process and using the exact same machine and parameters as the original authors is usually impossible. With this work, we give some information on how the modeling language JuMP passes the constraints to the solvers and we present a new julia tool called SortModel which allows for reordering the constraints of a JuMP model with respect to multiple criteria. Using SortModel, we show the performance variability [Lodi and Tramontani] induced by constraint ordering and provide a simple way to experiment with.