We cannot run all datasets on Ci, nor on a real quantum. We need some heuristics there, like selecting the worst/best split in the dataset.