Automatically unmarshalling the C++ vtkJSON object into a Python dict, list, or value is ergonomic, but not as efficient as having Python simply having a copy of the C++ object itself. Also the latter comes essentially for free if we just let the wrappers do their thing.
In general, I’ve aimed for the Python wrappers to be faithful to VTK’s C++ API, with ergonomics as secondary. The reasons for this are 1) it’s efficient and facilitates automation, 2) it results in fewer C++ VTK capabilities being unavailable from Python, and 3) ergonomics can often be added on top in Python.
I don’t mean to derail the conversation about vtkJSON, I just wanted to clarify my thought process with regards to the wrapping.