Review · November 22, 2026
Locality in Agent Discovery Summary
The Problem
DAWN's charter covers three discovery settings: a local network, an organization, and cooperating parties over the public Internet. The MDI draft concludes that the minimum discovery information is the same in all three.
This draft says the minimum may be the same, but three other things are not:
- how much a client can learn at or right after discovery
- what the client assumes from where it is looking
- how well the client can check who published the result
The third one is the interesting observation. Source checking does not get steadily better or worse from local to remote. DNSSEC and TLS tie a record to a cooperating party's domain. A Discovery Proxy or catalog can only be authenticated as itself. mDNS advertisements can't be authenticated at all.
The result is a mismatch. Clients assume the most on the local link, where the source is hardest to check. The author calls this the "locality gradient".
How It Fits In
This is a discussion contribution. It defines no protocol, wire format, or registry, and has no IANA actions.
It fits the charter. It is input to the architecture deliverable and to the minimum information work. The draft frames itself around locality instead of trust, which keeps it clear of the charter's exclusion of trust management and trust evaluation.
It creates no new roles. It borrows its terms from the terminology draft and adds three plain-language ones: known context, binding, and serving system.
The dependencies are DNS-SD RFCs (6762, 6763, 8766, 9665) and six individual DAWN drafts. Nothing it cites is an adopted DAWN document, because DAWN has none yet. There is no OAuth or WIMSE dependency.
High-Level Summary
The draft raises four discussion points.
Extra detail beyond the minimum. A minimal result doesn't stop local deployments from offering more. The risk is a single result that mixes a signed minimum with unsigned local detail, and a client that trusts all of it equally.
Where a result came from. The client already knows the scope it searched, the carrier it used, and the network it is attached to. The draft suggests the architecture describe a record the client keeps per result: scope, carrier, serving system, and whether each value was authenticated. Publishers would not publish this. A locality claim inside the result would just be the serving system vouching for itself.
Matching and choosing between local and remote results. Local and remote discovery can return different endpoints for "the same" agent. Recognizing they are the same is hard, because names only mean something within their scope. Choosing is harder. Preferring local exposes the client to forged mDNS answers. Preferring remote exposes it to poisoned or stale records.
Physical proximity as trust. Being on the same link is a stand-in for shared administrative control. On public Wi-Fi it means very little. The draft lists three ways to meet the charter's integrity requirement for unauthenticated advertisements: authenticate the agent at the follow-up connection, protect the delivery hop, or protect the advertisement back to its origin.
Section 6 turns these into suggestions for the architecture, MDI, and binding documents.
Is This Document Necessary?
There is nothing to implement, so interoperability isn't the right test. The question is whether this needs to be its own document.
It probably doesn't. Section 5 is candid that the MDI draft already covers most of the ground:
- MDI Section 9 already requires each binding to say how integrity and origin are protected.
- MDI Section 8.5 already says an unverified instance does not displace a verified one.
- MDI Section 13 already lists the same three answers for unauthenticated advertisements.
- MDI Section 11 already notes the correlation cost of a Stable Identifier.
So what is new? Two things.
The first is the cross-scope matching gap. MDI says equal Stable Identifiers denote the same resource within one naming scope. It says nothing across scopes. A client that finds an agent over mDNS and again in the organization's DNS zone has no defined way to know they are the same agent. This will affect real deployments, and it is worth an issue against the MDI draft and the framework draft.
The second is the client-kept provenance record. It is a reasonable architecture suggestion. It is also client-internal state, which doesn't affect interoperability and doesn't need standardizing. A paragraph of guidance in the architecture document is the right size for it.
The draft would be more useful if it took positions. Every point ends in "could" or "feedback is welcome". On the tie-break question it offers one cautious policy and then notes that policy doesn't help with two unverified results. A recommendation the working group can argue with would move the discussion further than a list of open questions.
One thing to watch: Section 5 is pinned to MDI -02 section numbers. MDI is already at -03, and the terminology draft has moved past the cited -06. The section-by-section mapping will go stale quickly.
Writing Quality
The writing is good. It is 10 pages, with short sentences and plain words. It has no BCP 14 keywords, which is correct for a document with no normative content.
Isolating the MDI-specific material in Section 5 is a smart structural choice. The rest of the document survives changes to that draft.
There is some repetition. Section 5 restates Section 4 through the MDI lens, Section 6 restates it as suggestions, and Section 8 restates it as threats. The document could lose a page or two without losing content.
The "gradient" name is slightly off. The draft's main point is that source checking is not monotonic from local to remote, which is the opposite of a gradient.
Author Background
Osama Abboud (Huawei Technologies Duesseldorf) has no published RFCs. This is his only active Internet-Draft. He has two expired drafts, both in the sustainability and energy management space (draft-csha-sustain-reporting-arch and draft-ahc-green-smartpdu-yang). He is a relatively new IETF author, and this is his first draft in DAWN.