https://eel.is/c++draft/meta.trans.other

The definition of COMMON-REF concludes with, "If any of the types
computed above is ill-formed, then COMMON-REF(A, B) is ill-formed."

I'm not sure what this means. My first inclination is that it means, if
the type formed for the particular case is ill-formed, then
COMMON-REF(A, B) is ill-formed. But this seems redundant with the
immediately preceding statement, "Otherwise, COMMON-REF(A, B) is
ill-formed."

The other interpretation is that if the type of any of the four cases is
ill-formed, then COMMON-REF(A, B) is ill-formed even if a different case
applies. But it seems odd to me to say that, if X is A& and Y is B&,
that COMMON-REF(X, Y) is ill-formed if the type for the mixed rvalue and
lvalue reference case, namely, COMMON-REF(const A&, B&), is ill-formed.

Can anyone shed some light on this?

Given types A and B, let X be remove_reference_t<A>, let Y be remove_reference_t<B>, and let COMMON_REF(A,B) be:

(3.1) – If A and B are both lvalue reference types, COMMON_REF(A,B) is COND_RES(COPYCV(X,Y) &, COPYCV(Y,X) &).
(3.2) – If A and B are both rvalue reference types, and COMMON_RES(X&,Y&) is well formed, and is_convertible<A,R>::value and is_convertible<B,R>::value are true where R is remove_reference_t<COMMON_RES(X&,Y&)>&& if COMMON_RES(X&,Y&) is a reference type or COMMON_RES(X&,Y&) otherwise, then COMMON_RES(A,B) is R.
(3.3) – If A is an rvalue reference and B is an lvalue reference and COMMON_REF(const X&, Y&) is well formed and is_convertible<A,R>::value is true where R is COMMON_REF(const X&, Y&) then COMMON_RES(A,B) is R.
(3.4) – If A is an lvalue reference and B is an rvalue reference, then COMMON_REF(A,B) is COMMON_REF(B,A).
(3.5) – Otherwise, COMMON_REF(A,B) is decay_t<COND_RES(CREF(A),CREF(B))>.

If any of the types computed above are ill-formed, then COMMON_REF(A,B) is ill-formed.