setof/3 discrepancy between tpl and scryer-prolog

[flexoron]

$ tpl
Trealla Prolog (c) Infradig 2020, v2.89.11-1-gb588
?- A=a, setof(A+B,(nonvar(A);var(B)),X), B=b.
   A = a, B = b, X = [a+_A,a+_B].
?- A=a, setof(A+B,B^(nonvar(A);var(B)),X), B=b.
   A = a, B = b, X = [a+_A,a+_B].
?-

$ rustc --version
rustc 1.93.0 (254b59607 2026-01-19)
$ scryer-prolog -v
v0.10.0-84-g453a88f0 % target/release

$ scryer-prolog -f
?- A=a, setof(A+B,(nonvar(A);var(B)),X), B=b.
   A = a, B = b, X = [a+_A,a+_B].
?- A=a, setof(A+B,B^(nonvar(A);var(B)),X), B=b.
   A = a, B = b, X = [a+b]. % correct but ...
?- 

It seems, B^ helps to figure out the set.
Logically correct, other systems do not yield this result.
Is Scryer ahead of others?

[flexoron]

?- A=a, setof(A+B,(nonvar(A);((var(B),B=b))),X), B=c.
   A = a, B = c, X = [a+_A,a+b].
?- A=a, setof(A+B,B^(nonvar(A);((var(B),B=b))),X), B=c.
   A = a, B = c, X = [a+c] % unexpected
;  false.
?- 

[UWN]

Shorter:

?- setof(B,B^(B=a;B=b),Bs).
   B = a, Bs = "a", unexpected
;  B = b, Bs = "b", unexpected.
   Bs = "ab". % expected, but not found