Some miscellaneous issues:

 > - I've changed the syntax of the OPAQUE and SEALED clauses to carry a
 >   boolean operand.

 I notice this operand is evaluated at runtime.  I am not too familiar
 with the issues involved, but I suspect these attributes may
 be useful for compile-time analysis of allocation and optimization
 strategies, which may become more difficult with this choice.

 > - I've changed the semantics of field-id to always be local to the
 >   specified rtd, rather than global.  This makes it easier to later
 >   extend the abstractions to multiple inheritance, should anyone ever
 >   want to do so, and leaves less room for ambiguity.

 I cannot find where this explained in the document.

 >
 > - The field names passed to MAKE-RECORD-TYPE-DESCRIPTOR are now
 >   required to be distinct.

 I think the sentence beginning with "If more than one field has the given name"
 in the document is in conflict with this statement.

 Also, I cannot figure out from the document if parent fields may be
 repeated in a child.

 Finally, I think an specification is still missing for the visibility
 of the various defined bindings in the various 'expression's
 occurring in the syntactic layer.

 Regards
 Andre