Update review · November 16, 2026
DAWN Use Cases -01, Narrower Scope and Two Worked Examples
What changed in -01?
The document went from 14 pages to 26. Almost all of the growth is two new appendices.
The -00 review covered a broad survey of "entities" finding other entities. In -01 the scope is AI resources only, and discovery returns much less.
The title changed too. DAWN is now "Discovery of Agents With Names", and the draft adds "Applicability" to its name.
What is new
- A narrower scope (Sections 1 and 2). Discovery covers a local network, one organisation, or cooperating organisations that already trust each other. Identity management, trust establishment, and Internet-wide indexing of AI resources are now explicitly out of scope.
- Section 4, "Applicability." A three-row table, one row per deployment setting. It names DNS and mDNS as possible mechanisms.
- A thinner discovery result. Every example in Section 7 was rewritten. Discovery returns how to find and contact the resource. Freshness, jurisdiction, availability and policy are checked afterwards by the application (Section 7.2.3).
- A firmer security requirement (Section 9). Outside a local network, all discovery information requires integrity protection.
- Appendix A. Alice's assistant books lunch with Bob at another organisation and reserves a table. It then moves both to Friday using cached discovery results.
- Appendix B. A broadcaster's agent finds edge compute for a personalised live programme.
Are the worked examples useful?
Appendix A is the best part of the document. It draws the line between discovery and application traffic step by step. Finding the booking service is discovery. Searching for a table is not (A.3).
A connection failure does not mean the agent was withdrawn. A replacement agent does not hold the original booking (A.4). An authentic record does not prove the agent may act for Bob (A.6).
Appendix B is not finished. The requirements list ends with a literal "REQ-???" (B.5). There is a stray "[object-based media delivery]" placeholder, and the text has several typos.
The requirements mapping is already stale
Sections A.7 and B.5 map the scenarios to draft-king-dawn-requirements-01. That draft is now at -02, and it was cut from 43 requirements to 17.
The appendices cite eleven requirement IDs. Six no longer exist: REQ-CLASS-2, REQ-PROTO-4, REQ-ARCH-4, REQ-PROP-11, REQ-SEC-4 and REQ-SEC-7. Four more kept their number but now say something different. Only REQ-SEC-1 still matches.
Was the previous feedback addressed?
Not addressed:
- The TBD sections. All five entity class sections are still "TBD" (Section 8). Two of the headings, "Compute Workload" and "Network Function", now sit oddly next to a scope that was narrowed to AI resources.
- The Discovery Pattern section. Still there as Section 6, with two paragraphs added.
- Operational Considerations. Still restates the use cases, and it grew (Section 11).
Partly addressed:
- Overlap with existing federation and trust work. The draft still cites no prior art. It did move the overlap out of scope (Section 2). Every cross-organisation example now starts from a trust relationship the draft never describes.
Mechanisms added or removed
This is a use case document, so there are no protocol mechanisms. Two removals matter. "Trust indicators" are gone from discovery results. So is the paragraph allowing credential, key or attestation references as optional discovery information.
Background of the authors
The author list changed. Rajiv Ramdhany (BBC) was added as a fourth author. Daniel King's affiliation changed from Old Dog Consulting to Lancaster University.
Rajiv Ramdhany has no RFCs on his Datatracker profile, one active draft (this one), and two expired drafts.
Questions for the meeting
The draft has 10 minutes on Thursday, next to draft-jadoon-dawn-layered-agent-discovery. That draft is one use case inside one organisation, and it asks what the minimum discovery information must contain.
- Is this the base for the charter's "Use Cases and Applicability" deliverable, and do single-scenario drafts like Jadoon's fold into it as appendices?
- What do the appendices tell the architecture? A.5 lists name, type, capability identifier and reachability, then defers the field set to the Architecture.
- Fill in Section 8 or delete it. It has been TBD for two revisions.
- Remap to requirements -02. Most of the cited IDs are gone.