[PD] "Structured" dataflow?

[Dudley Brooks]

Can anyone direct me to articles on constructing clear, modular, 
non-spaghetti patches in pd or other visual dataflow languages? 
Especially if the articles derive their recommendations from theoretical 
analysis (as with the investigations that led to structured programming 
in imperative languages), rather than just rules-of-thumb -- although 
the latter are useful also.

Or is some amount of spaghetti unavoidable in dataflow languages, 
perhaps because it is inherent in the situation being modeled, rather 
than being an artifact of the language?

Thanks.