I think the main difference is that the code doesn’t modify the yaml file directly and instead reads in and out json and lets the serialization methods you provide do the work. I don’t 100% follow you warning regarding 2 sub-graphs – do you have an image of the situation you could share and I could test it out?
locutis
