The source and the sink are higher level entities with their > own higher level protocols.
Yes, and they would be examples of the first and second case I gave in my first mail on this thread. The channel between source and sink, which > is the Unicode level and below, should be transparent to PUA characters, > indeed to all characters apart from defined transformations. That surely > is the point of the PUA. If the channel starts messing around with the > characters sent through it, that is what is non-conformant. Such applications would match the third case. The fourth case can be used to build any of the others.