The Scope Creep Automated reviews of new AI-related working group submissions

Update review · November 21, 2026

MDI Model -03, a Third Author and a Fixed DNS-AID Citation

draft-cui-dawn-mdi-model -03View on Datatracker →

What changed since -02?

Very little. The model is the same.

The -02 review covered a rewrite. The -03 is a cleanup, dated 19 November, the same Thursday the draft had its ten-minute slot under Discovery Architecture.

Every element, Status value, constraint, and open question is unchanged. Tables 1 and 2 are identical. The draft still has no change log.

The real changes are these:

  • A third author. Chenguang Du moves from the acknowledgements to the author list.
  • A shorter abstract. It drops the MVD explanation and says the same thing in half the words.
  • A fixed DNS-AID citation. Every reference now points at draft-mozleywilliams-dawn-dnsaid in place of the older dnsop version.
  • Two wording changes, in Section 6 and in the Freshness definition.

Was the previous feedback addressed?

One item out of six.

  • Addressed: the stale DNS-AID mapping. Appendix B said DNS-AID puts the agent protocol in alpn or bap. It now says DNS-AID uses proto and alpn the same way the common parameters draft does. That matches the DAWN version of DNS-AID.
  • Not addressed: ARD is still missing. draft-bu-dawn-ard-discovery is not cited or mapped. The abstract now calls this document "a mapping target for existing discovery mechanisms." ARD requires a globally unique identifier and a display name, and has no field for the agent protocol. That is the opposite of this minimum, and Table 3 still has no column for it.
  • Not addressed: the proto gap. The model says the agent protocol is needed in every context. The DNS parameters draft makes proto a SHOULD. Appendix B maps the parameter and does not mention the difference.
  • Not addressed: integrity. Section 9 still says a binding MUST protect the integrity of all MDI. DNS-AID still says records SHOULD be DNSSEC-signed. What counts for mDNS is still an open question in Section 13.
  • Not addressed: value spaces. Entity Type is still "left to DAWN, bindings, or profiles." Endpoint purpose still has three meanings in prose and no defined values.
  • Not addressed: stability classes. Static, mainly-static, and dynamic still carry no obligation.

Section 8 still says "All Status values are provisional."

Mechanisms added or removed

None.

Two sentences changed meaning slightly.

Section 6 used to define "using a result" as choosing one result, then fetching, talking, or asking a catalog. It now starts with "determining which Endpoint and agent protocol to use." That lines the test up with the two mandatory elements. It also takes selection among results out of the definition, which fits the charter putting selection semantics out of scope.

The Freshness definition now says the digest on a description Endpoint "is not a protocol version or a trust signal." That closes off a misreading. Rule 4 of Section 8.5 already said integrity is not trust.

Background of the authors

One author was added.

  • Chenguang Du (Zhongguancun Laboratory) is a new IETF author with no RFCs. Du has six active drafts and no expired ones. Five are AI agent or LLM drafts, and four of those are co-authored with Yong Cui. Du was credited in -00 and -02 for contributions to the design and text.
  • Yong Cui (Tsinghua University) has 11 RFCs in softwires and DHCPv6, and 21 active drafts.
  • Kehan Yao (China Mobile) has one RFC, 13 active drafts, and is listed as Tech Advisor for DAWN.

The acknowledgements now thank Lori Jakab and Sumit Ahuja for comments on using discovery scenarios to clarify requirements, and Adrian Farrel and Linda Dunbar for discussion of discovery inputs and outputs. Jakab co-authors the mDNS and common parameters drafts. Farrel co-authors the terminology and requirements drafts. The authors of the drafts this model depends on are reading it.

Questions worth raising

The same four as last time, with one closed.

  1. Is Section 6 the method the working group wants for deciding the minimum? The draft still asks this as its first open issue.
  2. Is an optional Stable Identifier right, when ARD requires one and the DNS drafts have none?
  3. Does "integrity MUST be protected" survive mDNS and unsigned zones?
  4. Can this be a mapping target for existing mechanisms without mapping ARD? The DNS-AID half of this question is fixed. The ARD half is now more visible, because the abstract makes the claim directly.