 We developed a new method to measure the level of atomic disorder in materials called sodas. This method uses graph neural networks to calculate a continuous disorder field based on thermal perturbation data. By comparing sodas to other common methods, we found that it accurately tracks the evolution of interfaces between different phases of matter. Additionally, we were able to use the gradient of the disorder field to better understand the relationship between local atomic structure and material properties. This article was authored by James Chapman, Tim Hsu, Xiao Chen, and others.