Look at *differences* between words rather than at the words themselves?

[radapox]

So to to read this you would need to understand the operation that took place and convert this operation (rather than the resulting letters) into your source text.

Correct!

Another question I have is what effect such a code would have on typ-token ratio. Wouldn't you get an abnormal amount of unique words? If "bed" results in "azexiz", what are your odds of ever getting "azexiz" again? (I really don't know, I guess you'd have to test this).

Yes, I suppose that's a possible risk; I'm not sure about that either. Then again, the VM does have an unusually high frequency of unique words, if I'm not mistaken. Not sure if the numbers add up to a method like this; that would need to be tested as you say. 

Also, as You are not allowed to view links. Register or Login to view. remarked to my earlier post, this method would be problematic for Labelese. That's why I'm keeping as open mind a mind as possible regarding the multitude of different organizing principles that may be involved. Nick Pelling illustrated the possible complexities quite nicely in You are not allowed to view links. Register or Login to view.. [EDIT: Hey, wow, that post is exactly 10 years old today.]