Yes, both occurrences of "instance" should be replaced by "specialization".

Hm, that would result in allocation functions being templates, while deallocation functions would be functions. Perhaps it would be better for them to be congruent.

I don't understand. The actual allocation/deallocation function is always a function (perhaps instantiated from a template).

Sure, but the current spec refers to allocation functions as being templates:

An allocation function can be a function template

Which isn't correct either

There are two levels here: One is restrictions on the declaration of the allocation / deallocation function, which also applies to any function templates called "operator new" / "operator delete".
The other level is restrictions on the behavior, which applies to any specialization actually instantiated. (We can't talk about runtime behavior of templates; templates don't exist at runtime.)

[basic.stc.dynamic] has more problems, though: It specifies the name of global allocation/deallocation functions, but not of class member functions.

In my view, we should apply a larger change here:

• Move most of the hanging paragraph of [basic.stc.dynamic] Into a new subclause "Implicitly-declared global allocation and deallocation functions" at the end of [basic.stc.dynamic].
• Change [basic.stc.dynamic.allocation] to say "A function or function template whose name is operator new or operator new[] shall satisfy the following rules (syntax rules as bulleted list). Such a function or a specialization of such a function template is called an allocation function."
• Similar for [basic.stc.dynamic.deallocation].

Editorial meeting: Agreed with larger change. Just call it "operator new" function. Give it a try.