Tony’s STEMMA data model is very well thought out and it would be nice to see Gramps extended with similar capabilities.
Adding Group has already been mentioned here before, although the term Sets was used. I know people have brought up being able to have sub-events of events before but really that is just looking for a way to group events.
I think adding Document or Content as a child of a Source provides the means for capturing and properly storing all or a subset of the information from a given Source. I really wish the Forms Gramplet stored data in that manner. I know Gramps grew out of the conclusion based Gedcom model but in my mind we really should have some way to reliably record and separate the source data and content from the conclusions we draw from it in a structured manner. By doing so we can then better classify the data and associate the relevant peices of data as Evidence with the final conclusion or conclusions they support. A Citation would naturally be associated with the Evidence.
Two other research subjects it would be nice for Gramps to support are Artifact for family heirlooms and Animal for family pets.
It would be nice if Researcher was not the singular owner of the database and we could add researchers and associate them with data or conclusions they may have contributed. I can see this being useful for Gramps.js It might be useful to have Agent as well for documenting DNA matches of interest. Actually the former probably is an instance of the later and the Agent would have a type.
Finally instead of keeping trees in separate databases I think adding the concept of a Tree to the data model itself would be useful. When creating an object you could choose if it was going to be global, and hence visible to any tree, or local to that tree only. In that sort of model your Place, Source and Repository objects, and maybe others, would typically be global in nature and shared across all the trees in your database. You could “copy” a tree to work out a new line, and if it is later disproven discard it. Here I am thinking along the lines of the copy on write technique used in virtual memory managers and of course storage subsystems for snapshots. You could also have the ability to fold different trees together, something similar to a git merge.
One closing thought. Gedcom has played an important role in the community for many years, there is no questioning that. But I think an argument could be made that it and the mindset it has instilled have held back the evolution of software in the field for two decades now. And one could argue the online providers are having much the same effect. True there are packages like Evidentia, Centurial and Clooz but in the big picture they have small user bases.