subtypep doesn't understand EQL/MEMBER/KEYWORD type

[Jim Newton]

I'd like to request an enhancement.

In SBCL (subtypep '(eql :x) 'keyword) returns NIL,NIL
but
(subtypep '(eql "x") 'string) returns T, T
and
(typep :x 'keyword) returns T
and
(subtypep '(member 1 2 3) 'fixnum) returns T, T

According to the spec subtypep IS ALLOWED to return NIL, NIL if the computation involves EQL or MEMBER,
so the behavior is not a bug.

In any case I'd like (subtypep '(eql :x) 'keyword) and (subtypep '(member :x :y :z) 'keyword) to return T, T

According to a conversation with Douglas Katzmann, this is a known issue related to a limitation of the KEYWORD type.

[Stas Boukarev (stassats)]

KEYWORD is a deftype for (and symbol (satisfies keywordp)), and subtypep can't reason well about satisfies.

[Jim Newton (jimka-issy)]

Am I wrong? why would SUBTYPEP need to reason about satisfies when its first argument is (EQL ...) or (MEMBER ...) ?
It seems to me, (without understanding the details of course, hence the question), that in this case
SUBTYPEP, should call TYPEP on the argument of EQL or the arguments of MEMBER.
Perhaps there's a bootstrapping or metacircularity issue with my suggestion?