Assignment of objects to data models based on attributes. I am not convinced this is that robust; consider the option of assigning different data models to different contexts/endpoints.
"Two objects with compound identifiers (unique) are linked if their identifiers are identical (e.g. subject + year level for syllabus and for lesson)". Enable compound link keys instead (unless that's already implemented.)
Namespace types in the global GQL schema by the data model they come from; e.g. SIF_Name vs IMS_Name.
In filters, we need true JSON Path, including * to match any list child: "TeachingGroupPeriodList.TeachingGroupPeriod.*.DayId", not "TeachingGroupPeriodList.TeachingGroupPeriod.1.DayId". Without that, the restriction on queries is too onerous.
Links in traversal need to have the option of naming what field the link uses, or else the traversal has limited value. That includes properly specifying multiple fields for links, rather than the presumption in Unique that all objects will have the same compound primary key.
Add not-equals, boolean operators, less than, greater than to filters.