> Does SICP work in Common Lisp these days?

With a few macros and compatibility functions there is not much in SICP Scheme which does not work in similar ways in Common Lisp. SICP is using a rather primitive version of Scheme - which actually does not even run in newer Scheme dialects without changes. SICP does not even use macros. Later Scheme languages diverge much more and thus form their own language family. There is still a core in Scheme, which is relatively close to Lisp.

Take for example this function from SICP:

    (define (make-stack)
      (let ((s '()))
        (define (push x)
          (set! s (cons x s)))
        (define (pop)
          (if (null? s)
              (error "Empty stack -- POP")
              (let ((top (car s)))
                (set! s (cdr s))
                top)))
        (define (initialize)
          (set! s '())
          'done)
        (define (dispatch message)
          (cond ((eq? message 'push) push)
                ((eq? message 'pop) (pop))
                ((eq? message 'initialize) (initialize))
                (else (error "Unknown request -- STACK"
                             message))))
        dispatch))

The equivalent Common Lisp is roughly this:

    (defun make-stack ()
      (let ((s '()))
        (labels ((push (x)
                   (setq s (cons x s)))
                 (pop ()
                   (if (null s)
                       (error "Empty stack -- POP")
                     (let ((top (car s)))
                       (setq s (cdr s))
                       top)))
                 (initialize ()
                   (setq s '())
                   'done)
                 (dispatch (message)
                   (cond ((eq message 'push) #'push)
                         ((eq message 'pop) (pop))
                         ((eq message 'initialize) (initialize))
                         (t (error "Unknown request ~a -- STACK"
                                   message)))))
          #'dispatch)))

You can see that CL uses two namespaces and some of the operators are named differently (like EQ in Lisp and EQ? in Scheme, SETQ in Lisp and set! in Scheme).

But beyond that and even without a compatibility layer, the SICP code has 1 to 1 direct equivalents in CL: function definition, s-expressions, local functions, local lexical variables, setting variables, creating cons cells, testing for a NIL value, symbols, equivalence of symbols, raising an error, cond operator, ....

With minor compatibility macros we can write:

    CL-USER 6 > (define (make-stack)
                        (let ((s '()))
                          (labels ((push (x)
                                     (set! s (cons x s)))
                                   (pop ()
                                     (if (null? s)
                                         (error "Empty stack -- POP")
                                       (let ((top (car s)))
                                         (set! s (cdr s))
                                         top)))
                                   (initialize ()
                                     (set! s '())
                                     'done)
                                   (dispatch (message)
                                     (cond ((eq? message 'push) #'push)
                                           ((eq? message 'pop) (pop))
                                           ((eq? message 'initialize) (initialize))
                                           (else (error "Unknown request ~a -- STACK"
                                                        message)))))
                            #'dispatch)))

    CL-USER 7 > (make-stack)
    #<interpreted function (SUBFUNCTION (LABELS DISPATCH) :UNKNOWN) 406000135C>

    CL-USER 8 > (let ((s (make-stack)))
                  (funcall (funcall s 'push) 1)
                  (funcall (funcall s 'push) 2)
                  (print (funcall s 'pop))
                  (print (funcall s 'pop)))

    2 
    1 

Above still has to respect that CL is a Lisp 2, but we are getting closer.

If we have a more complete Scheme emulation in Common Lisp, we can use it directly and unchanged from SICP:

    CL-USER 2 > (ps:scheme)
    This is Pseudoscheme 2.12.

    SCHEME 3 > (define (make-stack)
                 (let ((s '()))
                   (define (push x)
                     (set! s (cons x s)))
                   (define (pop)
                     (if (null? s)
                         (error "Empty stack -- POP")
                       (let ((top (car s)))
                         (set! s (cdr s))
                         top)))
                   (define (initialize)
                     (set! s '())
                     'done)
                   (define (dispatch message)
                     (cond ((eq? message 'push) push)
                           ((eq? message 'pop) (pop))
                           ((eq? message 'initialize) (initialize))
                           (else (error "Unknown request -- STACK"
                                        message))))
                   dispatch))
    make-stack defined.

    SCHEME 4 > (let ((s (make-stack)))
                 ((s 'push) 1)
                 ((s 'push) 2)
                 (s 'pop)
                 (s 'pop))
    1

OTOH if you look at Clojure, it lacks the basic cons cell data structure and then one needs to use operators from a different data structure. LET / SET! does something else in Clojure - reading the documentation about that is confusing enough.

> A clean room Common Lisp couldn't meet that standard, for example.

It's not about that an implementation actually uses code from others (though it might), it is about actually being able to run Lisp applications/libraries or have them ported without the need to rewrite them.

Take for example Macsyma. It has its roots in mathematics code written in Lisp in the 60s. It was then further developed in Maclisp. It was then ported to Zetalisp and Franz Lisp (IIRC). Then it was ported to Common Lisp and still runs in various Common Lisp implementations. I have myself ported it to a Common Lisp implementation it never ran before in a day. This is an example of a substantial and non-trivial Lisp program which didn't need to be reimplemented, since it has been moved to new Lisp dialects to be able to be run on current hardware. A version of it now runs on Android devices...

The LOOP macro was actually maintained at MIT as a single file for multiple Lisp dialects for some time.

You can take a Lisp book from the 60s and run its examples in CL without effort. There is code from McCarthy from 1960 which runs mostly unchanged today, with a minimal compatibility layer. You can't run anything from McCarthy in Clojure because it lacks the most basic things like cons cells. All the list processing stuff needs to be rewritten to fit into Clojure's idea for 'persistent sequences' or other concepts.