If you read through the DeadEnds thread there is a comment from mstransky on 2010-12-02T06:28:17-08:00 that kind of hits the nail on the head. And I think I’ve mentioned it here before as well, that I think the actual research data should be kept separate of the conclusions altogether.
So my thinking is you would store both structured and unstructured content outside the conclusional lineage linked model.
A tool like the Forms gramplet wouldn’t update objects in the existing model, it would store the structured data here. This is very much the Clooz approach. Being structured data, the claims are pretty much known because they are inherent in the structure of the document. You know the person name in a census record is the subject, so you know the birth year is a claim about that subject.
You also need to account for unstructured data of course that ideally you would have transcribed from a source. There then needs to be a way to identify the claims in the data or the source, or enter them directly, and the subjects they apply to. If I recall this is discussed a little in the GenTech model somewhere but it may have been elsewhere. Maybe you can highlight the claim in the transcription to identify it for example, stuff like that.
Once you know the claims, and the subjects they apply to, you can do a number of things with them and the data. The subject names are just names, but they basically represent the ‘personas’ without having their own dedicated objects. You might use an object representation for working with the data within Gramps but ideally it would be stored in normal tables in the database so external tools could potentially be used with it and not as pickled or json objects.
One approach for working with the data might be an Evidentia style interface that walks you through extracting the assertions and documenting the analysis and the resulting conclusion and all.
Another approach would be the ability to assemble the claims into linkage bundles, and those into dossiers. This is the methadology documented in Elements of Genealogical Analysis.
Whatever tools are devised to help work with the data, you would be able to correlate or associate any extracted subjects in this model with a conclusional subject in the lineage linked model. If you later find that a claim did not apply to that conclusional person you just uncorrelate the data and optionally document why. You never delete the data, it is a repository that grows over time and contains everything that supports what you enter in the conclusional model and the reasoning behind it.
As mentioned, I picture this sitting in parallel to the existing lineage linked model. Users who do not want to use any of this would be free to do things as they always have, or they could mix and match. It would be just another set of tools in the toolbox.
The research journal idea would tie into this as the search process could include support for extracting and recording the data and not just keeping some high level notes about the search and what was found and what was not.
In the end you would have three submodels in Gramps so to speak. The administrative research submodel, the evidence submodel, and the current conclusional lineage linked submodel but hopefully with a few enhancements like groups and heirarchies for events, groups and sources.
These are my high level thoughts around things and how they might be someday. How well would it all work? Is it really practical? I honestly don’t know.