Unexpected bahaviour with respect to "more than one" predicate

[mgttlinger]

Take the following sCASP specification where g is saying roughly that f holds for more than one value of its first argument in any context (value of the second argument of f):

f(a,c).
f(b,d).

equal(a,a).
equal(b,b).

g(A) :- f(A,_), f(B,_), not equal(A,B).

Unexpectedly sCASP happily finds a model for both g(a) and not g(a). Is this due to a misunderstanding on my side or is this considered a bug?

See https://swish.swi-prolog.org/p/strange.pl

Modifying the program to make f unary the behaviour is not observable.

f(a).
f(b).

equal(a,a).
equal(b,b).

g(A) :- f(A), f(B), not equal(A,B).

[mgttlinger]

In the ciao version of sCASP the same behaviour can not be observed. So it sais that not g(a) is not provable in both versions.