Understanding something is having a good working model stored the way brains store models, which is quite a complex network. Explaining it requires finding a way to serialize it that makes it as easy as possible for a listener to reconstruct the model in his own mind.

I think the effort involved in trying to come up with a serialization causes us to more carefully examine our models, which usually improves them.

But I don't think the lack of a good serialization implies the lack of a good model.