> I suppose atoms could just be cons cells with nil in their cdr slot.
Could they be annotated conses with symbol in the car and value in the cdr (initialized to nil)? nil itself could then be a cons with the nil symbol in the car and nil in the cdr. This should achieve the cons-symbol duality for nil that's usually desired. (Follow-up question: annotate is an axiom, right?)