Then nothing. Macros work.

Silly example:

    COALTON-USER> (defmacro rpn (x y op)
                    `(,op ,x ,y))
    RPN

    COALTON-USER> (coalton-toplevel
                    (define (double x)
                      (rpn 2 x *)))
    ;; DOUBLE :: ∀ A. NUM A ⇒ (A → A)

    COALTON-USER> (coalton (double 3.0))
    6.0

One concern I’d have is that any type errors would be reported on the macro expanded code and thus be pretty much inscrutable outside of toy examples.

I think you're right that debugging errors involving complicated macros can get difficult, but to at least make the situation more tolerable, when Coalton expands a macro, it remembers where the expansion came from, so an error will be reported in the right place with the right source code. For example, using the RPN macro from the sister comment, here's an intentional type error:

    COALTON-USER> (coalton (rpn "x" "y" +)) 
      --> <macroexpansion>:1:9
       |
     1 |  (COALTON (RPN "x" "y" +))
       |           ^^^^^^^^^^^^^^^ expression has type ∀. (NUM STRING) => STRING with unresolved constraint (NUM STRING)
       |           ^^^^^^^^^^^^^^^ Add a type assertion with THE to resolve ambiguity
       [Condition of type COALTON-IMPL/TYPECHECKER/BASE:TC-ERROR]

Statically checked macros are useful to introduce new syntax at no runtime cost.

No, macros run at compile time (cf https://www.lispworks.com/documentation/HyperSpec/Body/03_bb...), you may be confusing macros and reader macros.

Fair that I was definitely mixing them in my thinking. My general point was supposed to be simply that "compile time" is probably not what people are thinking of when coming from other languages. I was clearly a bit too eager to try and say that, though. :(