[Ballot discuss]
I appreciate the changes that were put into the -25 version. However, the change to 3.2 made things very confusing and didn't address the original problem, and the change for 4.2 that was agreed in email didn't get made. The items regarding section 5.2, 5.5, and 13 were never addressed addressed at all. Here are the remaining bits:

Section 3.2 paragraph 1: The first sentence still includes the unnecessary MUST, and it's now so complex that I can't make heads or tails of it. It needs rewriting. There's no also need for the second sentence; that gets taken care of in the subsequent paragraphs.

4.2, last paragraph: "MUST include" should just be "includes"

5.2, paragraph 1: "MUST follow" -> "will follow". I'd also like to see "and hence MUST NOT be used" just deleted; it doesn't add anything as far as I can tell.

If an answerer wishes to accept the loopback request it MUST
include both the loopback role and loopback type attributes in the
answer.

Instead:

An answerer includes both the loopback role and loopback type
attributes in the answer to indicate that it will accept loopback
requests.

Also: "MUST be loopback-mirror" -> "will be loopback-mirror"

5.5:

This entire section should be removed; it is not important to this particular document as it is a general problem for any protocol.

Furthermore, if the mirroring entity is behind a NAT, it MUST send
some packets to the identified address/port(s) of the peer, in
order to open the NAT pinhole.

That MUST is completely bogus. Even if being behind a NAT always required sending packets to pinhole, the MUST is still wrong: This is not a requirement of *this* protocol; it's just stating a fact. But the fact is that being behind a NAT does *not* require pinholing: If the entity is behind a PCP server, it MUST do no such thing.

Note that for any form of NAT traversal to function, symmetric
RTP/RTCP [RFC4961] MUST be used. In other words both agents MUST
send packets from the source address and port they receive packets
on.

Again, completely false in the case of PCP. Please remove section 5.5.

13:

All implementations MUST at least support the rtp-pkt-loopback mode
for loopback-type, with direct media loopback payload encoding.

I don't understand: Are you saying that there is no environment under which I might *only* implement rtp-media-loopback? Why is this a MUST?