Hi Henri,
I'm happy to hear that other people are also thinking about structured/projectional editing. I strongly believe that projectional editing can be done and that they will become more widespread eventually. Unfortunately there are only a very few people who are interested in the development of these tools at the moment, and I myself have very little time allocated for this project.
The document model of ProjecturEd is very flexible, in fact it hardly poses any constraints on the data structures. The implementation simply uses various classes, sequences and primitive types such as strings, numbers, etc. Take a look at
https://github.com/projectured/projectured/tree/master/source/domain page. The most complete parts are the graphics, text, tree, json, xml, and book domains. In the domain directory you will find the domain specific data structure definitions and the data manipulation functions other than construstion and deconstruction (they are mostly added automatically).
I think the interesting parts are not the document data structures but rather the projections. And the most important part is how the generalized read-eval-print-loop, the printer and the reader algorithm works. How primitive, generic, and higher-order projections can be composed to build more complicated ones that can deal with the composition of diverse document data structures in a way the provides useful and predictable behavior.
If you have any further question, suggestion, or idea, please feel free to contact me.