CONCEPT_RELATIONSHIP table contains not only hierarchical but lateral relationships.
So, if I want to find the descendants - usually needed when creating a concept set - as you actually did, I would use CONCEPT_ANCESTOR, and don’t worry about getting something else through lateral relationships.
Then I thought, that I might look at the relationship_id in this proposed ancestor-relationship table.
In this case to deal with hierarchical relationships, I would need to collapse different relationships into one category: Consists of (RxNorm), Has marketed form (OMOP), Has tradename (RxNorm), Is a (RxNorm) and so on, which is already done in CONCEPT_ANCESTOR.
And lateral relationships such as ‘Has dose form’ and ‘Has brand name’ just don’t have several nesting levels.
I enjoyed the exercise of thinking of possible changes, and satisfied that the current model is the best (at least for me).