Enoch chose statistical transparency over deep learning given only 24 radiocarbon-dated training samples