Update review · November 16, 2026
MDI Model -02, a Smaller Minimum With a Method Behind It
What changed since -00?
Almost everything. This is a rewrite, not a revision.
The -00 review covered a list of thirteen elements with three of them mandatory. The -02 has a new title, a second author, and more than twice the pages. The draft has no change log, and this compares -00 against -02.
The big shift is that the draft now argues for its list. Section 6 asks one question of every candidate field: could the client still use the result if this were missing? Section 7 runs that question over the charter's three deployment contexts. Table 1 records the answer for each field and where else the client could get it.
The draft now presents itself as input to the charter's Discovery Architecture and Protocol deliverables. It no longer claims to define MDI by itself.
What is the minimum now?
Smaller than before.
- Name (MUST). The name a result is found under, such as a DNS owner name.
- Endpoint (MUST). A locator and one or more agent protocols, plus new optional
purposeanddigestmembers. - Freshness (SHOULD).
- Entity Type, Display Name, Authentication Hint, Stable Identifier (MAY).
"Mandatory" also means less than it did. Section 8 says a value can come from the response, from what the client already knows, or from a binding rule. A record can meet the MUSTs while carrying almost nothing.
Was the previous feedback addressed?
- Addressed: Scope Hint, Operational Hint, Provenance. They are out of the minimum. Section 8.3 moves them, with Capability Summary and Trust Reference, to information a binding may carry as an Extension.
- Answered in the opposite direction: too much optional surface. I suggested promoting some SHOULDs to MUST. The draft demoted instead. Entity Type went from MUST to MAY. The old Entity Identifier split into a mandatory Name and an optional Stable Identifier. The reasoning is now on the page. Section 8 still says "All Status values are provisional."
- Not addressed: value spaces. Entity Type is still "left to DAWN, bindings, or profiles" (Section 8.2). The new Endpoint
purposehas three meanings described in prose and no defined values. - Not addressed: stability classes. Table 2 still labels elements static, mainly-static, or dynamic. No obligation follows from a label.
Mechanisms added or removed
Added:
- A four-stage flow of publication, serving, lookup input, and response (Section 5).
- Endpoint
purpose: retrieve a description, interact, or continue discovery at a catalog. This replaces the old Capability Reference. - A rule that a binding MUST protect the integrity of all MDI it conveys (Section 9). The charter says work items "will provide integrity of all the discovery information exchanged."
- Five open questions for the working group (Section 13).
Removed: Capability Summary, Capability Reference, Trust Reference, Provenance, Scope Hint, and Operational Hint as elements.
Do the protocol proposals carry this minimum?
Mostly, but not as the model states it.
The DNS parameters draft (draft-blwm-dawn-common-discovery-parameters) carries the agent protocol in proto, and its Table 1 makes proto a SHOULD. The model says the protocol is needed in every context.
DNS-AID (draft-mozleywilliams-dawn-dnsaid) uses those same parameters on SVCB records. The gap is integrity: its design section says records SHOULD be DNSSEC-signed, and the model's Section 9 says MUST protect. The model also cites the older dnsop version of DNS-AID and says it puts the agent protocol in alpn or bap. The DAWN version uses proto, so Appendix B is out of date.
ARD (draft-bu-dawn-ard-discovery) is not mapped or cited at all. Its required fields (Section 3.2.1) are a globally unique URN identifier, a displayName, a media type, and a url or inline data. That inverts the model: ARD requires the two things the model makes optional, and has no field for the agent protocol.
Background of the authors
The author list changed. -00 listed Yong Cui alone. -02 adds Kehan Yao.
- Kehan Yao (China Mobile) has one RFC, RFC 10054 on Computing-Aware Traffic Steering, 13 active drafts, and is listed as Tech Advisor for DAWN.
- Yong Cui (Tsinghua University) has 11 RFCs in softwires and DHCPv6, and 21 active drafts.
Questions worth raising
The draft has ten minutes under Discovery Architecture on Thursday.
- Is Section 6 the method the working group wants for deciding the minimum?
- Is an optional Stable Identifier right, when ARD requires one and the DNS drafts have none?
- Does "integrity MUST be protected" survive mDNS and unsigned zones, given DNS-AID only says SHOULD for DNSSEC?
- Should Appendix B map ARD and the current DNS-AID before anyone calls this a common model?