From an implementation perspective of course, so you can get error propagation on the baseline simulator. For the actual ML part, you'd get Measurements complai

From an implementation perspective of course, so you can get error propagation on the baseline simulator. For the actual ML part, you'd get Measurements complaining that it doesn't know how to propagate bounds through the surrogate. You can do some estimation of what that propagation would look like, but it's a research project.