I can reproduce this now.  The problem involves calling INTERNAL:make on one of these new type of expanded's that inherit from ANY.  Apparently, that doesn't work.
   Randy

class
	ROOT_CLASS

inherit
	INTERNAL

create
	make

feature
	make is
	local
		l_int: INTERNAL
		l_spec: SPECIAL[ANY]
		l_field: ANY
	do
		l_spec ?= (<<False, "hi">>).area

		-- This code represents what happens in SED_OBJECT_GRAPH_TRAVERSABLE:internal_traverse.
		l_field := l_spec.item(0)
		-- 'l_field' now refers to a BOOLEAN.  But this isn't an ordinary expanded BOOLEAN, nor is it
		-- a BOOLEAN_REF.  It is this new-fangled "reference BOOLEAN" (a lot like a BOOLEAN_REF).
		-- Apparently, you can't 'mark' this type of object.

		mark(l_field)
		print(is_marked(l_field).out + "%N")			-- This should be True but isn't.
		print(is_marked(l_spec.item(1)).out + "%N")		-- This should be True but isn't.
--		unmark(l_spec.item(1))							-- This will fail a require.

		-- This code better represents how I discovered the problem.
		store(<<False, "hi">>)
	end

	my_field: ANY

	store(a_object: ANY) is
	local
		l_file: RAW_FILE
		l_writer: SED_MEDIUM_READER_WRITER
	do
		create l_file.make_open_write("k.dat")
		create l_writer.make(l_file)
		l_writer.set_for_writing
		sed.independent_store(a_object, l_writer, False)
	end

	sed: SED_STORABLE_FACILITIES
	once
		create Result
	end

end