Ugh, what about this:

template<template<typename = int> typename X, X<> *Y> void f() {}
template<template<typename = float> typename X, X<> *Y> void f() {}

The default template argument here really is part of the signature.

I think what we really need to emphasize here is that signatures only matter up to equivalence. The above two declarations are not equivalent because their signatures are not equivalent (because of [basic.def.odr]/12.5, which presumably is intended to apply to default template arguments too).

We get this wrong in [basic.link]/10.4 ("the same" should be "equivalent"), and somewhat also in [namespace.udir]/7 (but this only applies to non-templated functions, so it's less of an issue). I don't think we can include this in the definition of "signature", though, due to ISO's insistence on definitions being substitutable into their point of use.

Editorial meeting: Not editorial.

Could you please rebase?