A block diagram that show the participants and the protocols (like DTLS or RFC4944, etc) would be very helpful to someone new to this field.  Like me.

Sec 1.
"Once a Pledge is enrolled, it can act as constrained Join Proxy between other Pledges and the enrolling Registrar."  Is that a special function of JP-based enrollment, or could anyone in the mesh be a JP? The 1,2 item list has a spurious "that" in the second entry. The "Similar to..." part in the last paragraph is a sentence fragment.

Sec 4.
Oh, you have a diagram here.  Spread out the distance between R and J so that "multi-hop" fits on one line maybe. Consider adding to it and moving it to Sec 1.  Or at least in Sec 1 have a forward pointer. Repeating "(P)" and "(J)" after the first instance is distracting. Type "untill" in last paragraph. Why is "legal" in quotes? "An enrolled device can..." same question as above: ANY enrolled device could?

Sec 5.1
Maybe "such as by" instead of "for example" The parenthetical about "Discovery can also" and the sentence about DNS-SD probably belong in section 6.  In Figure 2, I was briefly confused by the label "Src_IP" and the content having "IP_p" etc.

Sec 5.2
The phrase "but may also reduce" maybe "and may also reduce"? Is are paragraphs 2 and 3 redundant?  Why use JPY and not, say, SJP?  "The registrar should not assume..."  KEY POINT.

Sec 5.3
Why does the text say "ifindex" but the Figure 4 CDDL says "index"? Since there can be more than five elements, what is the meaning of extra elements? Ignore them? Maybe MUST send only five? "Completely opaque to the receiver" really means the receiving Registrar, right?

Sec 6
I was confused about "near" and "remote"  Maybe "near and far" or "local and remote" ? The rest of Sec 6, describing the different discovery methods seems reasonable.  (I am not well-qualified to say more than that)

Sec 7
This could be moved into 5 as a new subsection. If not, sec 5 should have a forward pointer to the comparison.

Sec 8
I like the list of possibilities for evil, and why they're not new. The "enroll itself" item should have the last two sentence fragments merged "With ..., the chance ..."  Next item "Also this is assumed" maybe "This, too, is assumed"  I think you could bundle all of the items which require having the private key, for example, and point out that you depend on the security of DTLS to prevent these things, rather than say "unlikely"