Two questions:

1. Do you want to explicitly forbid the case where both the subject_id and the object_id are sssom:NoTermFound? Currently this would be perfectly allowed.

Such a mapping could be used to explicitly indicate that two domains (vocabularies, ontologies, databases, whatever) are completely disjoint and have nothing in common. I suppose this would be a very infrequent need, but a priori I see no reason to forbid that.

2. Do we need special rules for inverting mappings that refer to sssom:NoTermFound, when the mapping predicate has a distinct inverse predicate (e.g., the mapping predicate is skos:narrowMatch, which is inverted into skos:broadMatch)?

Consider the following mapping (justification omitted for brevity):

subject_id   subject_source   predicate_id       object_id           object_source
HP:1234      obo:hp           skos:narrowMatch   sssom:NoTermFound   obo:mp

Applying the classic inversion rules, this would be inverted into:

subject_id          subject_source   predicate_id       object_id   object_source
sssom:NoTermFound   obo:mp           skos:broadMatch    HP:1234     obo:hp

I think this is fine: the first mapping says “HP:1234 has no narrow match in MP”, the second says “nothing in MP has a broad match to HP:1234” – this seems correct to me, what do you think?

Such a mapping could be used to explicitly indicate that two domains (vocabularies, ontologies, databases, whatever) are completely disjoint and have nothing in common. I suppose this would be a very infrequent need, but a priori I see no reason to forbid that.

I like the thinking. Do you think we should add this as a comment to the spec?

Do we need special rules for inverting mappings that refer to sssom:NoTermFound, when the mapping predicate has a distinct inverse predicate (e.g., the mapping predicate is skos:narrowMatch, which is inverted into skos:broadMatch)?

Thank you for thinking of this! The example makes me was hard to think through mentally, but I think you are right, there is no problem. If HP:123 does not have a narrow match in MP, No term in MP can be a broad match to it.

I like the thinking. Do you think we should add this as a comment to the spec?

To the user-facing documentation, maybe. To the spec, no need, there’s nothing that implementations should be aware of.