The abundance of primordial deuterium formed after Big Bang Nucleosynthesis has been measured to percent-level precision.  Predictions for the primordial deuterium abundance have achieved similar precision, but different teams have reported different central values, sometimes resulting in a mild (~2sigma) disagreement with measurement.  I will introduce a new method to predict the primordial deuterium abundance using Gaussian Processes, which is robust against analysis choices like model selection, dataset selection, and theory prior.  I will illustrate why different groups have found different results, and briefly discuss the new physics implications of these findings.