O.k., o.k., as Kent and Mark have pointed out, I've already managed to make my first significant error of the new year.
The intent and wording of the PDUTR #28 text on the CGJ is best stated in the Article II.3.9 Application of Combining Marks -- a section I overlooked in responding previously to Eric Muller's query. The problem, of course, is that if you start to apply ordinary combining marks to entire grapheme clusters comprised of sequences with the CGJ, you run afoul of canonical equivalences involving those combining marks. The same thing does not apply for the enclosing combining marks, since there are no canonical equivalences involving those combining marks. So, taking that into consideration, here is my restatement of what I think ought to happen for the three possible cases for the ng-tilde: 1. <U+006E, U+0360, U+0067> 2. <U+006E, U+FE22, U+0067, U+FE23> 3. <U+006E, U+034F, U+0067, U+0303> 1. uses the double-diacritic tilde, which nominally applies merely to the U+006E, but would be designed to lay over the top of a following base character on display. 2. uses the compatibility combining double-tilde halves. These occur in legacy bibliographic data records. In principle, 2 should display in the same way as 1, but would be recommended only for interoperating with the legacy data. 3. uses the grapheme joiner to create a "grapheme cluster", which in this case would be the digraph "ng". Unlike 1 and 2, the tilde would apply only to the "g", so that 3 would not display the same as 1 or 2. To illustrate the canonical equivalence question, consider: 1a. <U+0061, U+0061, U+0301> ==> aá 1b. <U+0061, U+00E1> ==> aá 1a and 1b are canonically equivalent sequences, and should display the same. 2a. <U+0061, U+034F, U+0061, U+0301> 2b. <U+0061, U+034F, U+00E1> Now if we insert a CGJ between the two a's, the sequences are still canonically equivalent, and should display the same. If, however, we say that the creating an "aa" grapheme cluster changes the context over which the following acute accent will display, then we have a situation where canonically equivalent sequences have consistently different display (and possibly interpretation). That wouldn't be a good thing -- hence the wording in PDUTR #28 to preclude the application of combining marks to other than the base character they follow (except for enclosing combining marks or other specified exceptions). --Ken