case-p-lex-np-no identifies label with wrong constituent?

[goodmami]

I found that rel_p shares a label with the thing in the relative clause and not the thing outside of it that's being modified, which seems to me to be backwards. LTDB tells me that this is introduced by case-p-lex-np-no.

Consider: 尻尾 の 長い 犬 が 吠え た

I would expect that rel_p to share its label with _inu_n instead of _nagai_a, or, even better, just remove rel_p and equate the labels of _inu_n and _nagai_a, like the ERG does. But possibly I'm not seeing the nuances that make rel_p required.