I guess this result is derived from the agreement that logical operation inside the phrase operator is treated as
"both operands a and b are in the _same_ position just before c".
select '(a & b) <-> c'::tsquery @@ 'a:1 b:1 c:2'; ?column? ---------- t (1 row)
Though it's not clear what it means if there is another phrase operator inside logical. Result positions of (a <2> c) and (b <-> c) are different, I guess. It's not clear to me how should this behave in the case of a chain of nested phrase-logical-phrase operations.
I would expect that the "position" of a match should depend on whether it's the LHS or RHS of the phrase operator. If it's the LHS, I'd expect the "position" to be the "end" of the match. If it's the RHS, I'd expect the "position" to be the "beginning" of the match.
Maybe that's the wrong intuition and it's always the beginning of the match?
