We've found that grounding works reasonably well until the underlying knowledge starts changing frequently, at which point retrieval quality becomes much more of a state-management problem than a search problem.
Are most users treating Tkcore as a knowledge layer for relatively static information, or are you seeing teams use it for fast-changing operational data as well?