What are the characteristics of Labelese?

[Anton]

If all of the VMs text is somehow the result of a random process, in the style of Rugg or Timm, then we would see something different.

That's a good point!

And yes, numbers also remain a valid option in my opinion, especially for the labels.

I think that's an interesting avenue of research, one could take "labels as numbers" as a starting hypothesis and test its validity (using some method which I can't invent right now)