Hi,
I've included a disclaimer below.[0]
On Mon, Oct 26, 2026 at 07:08:15PM -0500, Tom Barrett wrote:
[from the draft]
What is missing is a first-party, owner-
published, mutually verifiable declaration that lives where domain
authority already lives: in the DNS.
Each link counts as confirmed
only when both domains name each other and both directions are retrieved
over DNSSEC, or over HTTPS from the publisher's own domain. A consumer can
start from one name it trusts and retrieve the organization's whole set.
I think, and always thought, that these two things are fundamentally the right approach. This was my basic plan with the SOPA proposal that ended up folded into and which then died with the DBOUND work. But I always thought the problem was that the initial information was not coming from the one source that ought to be authoritative, and that this effectively had to be agreement between two sides of an agreement about what it means.[1]
- From RDBD: no new RR types and no signature scheme, only TXT at an
underscored name.
I can't speak for others, but the reason SOPA used a new RRTYPE wasn't just my
personal loathing of the underscore mechanism. It was because of the
bootstrapping problem: SOPA was intended to cope with any point in the tree,
and this meant that it needed to cope with underscore domains too, so they
couldn't be treated as special. Once that was the criterion, the cost of the
additional RRTYPE was nothing because special processing of certain underscore
names to achieve the same result just shifted the pain around.
- From DBOUND: domain-set doesn't deal with policy boundaries within a
namespace
I think that is a mischaracterization. I think it's entirely about a policy boundary:
"is that name that is not me to be treated as somehow still me?" This is really
the responsible person for the set publishing a policy through several statements in the
DNS, in each so-related domain.
Let me see if I can persuade you. Imagine a company GeneralMakestuff that operates (say)
an airline engine company, a consumer large appliance manufacturing co, a small appliance
manufacturing co, an industrial machinery co, and (incongruously) a finance bank. At one
point, these are all "the same owner" and the large holding co might want to be
able to say so. But after several years of corporate _excisa nervosa_, the various
divisions cease to be leaders in their respective fields, and the main thing that is left
of most of them is the value in the brand itself. Each such brand is sold to Proust and
Co[2] Holdings, which specializes in brands that are entirely about what the brand used
to mean. GeneralMakestuff continues operating one of its old divisions and continues to
be a leader in that field, so it continues to have its own domain names published.
Whether Proust and Co Holdings wants to publish a record stating the names are all the same owner is
obviously up to them. GeneralMakestuff definitely needs to stop saying the small appliance and airplane
engine domains are the same owner, at least at the point where one of them is sold to Proust and Co and the
other is not. I think the word we have to describe this kind of decision-making about how names relate to
each other is "policy". I don't really care if we call it something else (I am painfully aware,
for instance, that in ICANN circles some people now distinguish between "policy" and
"Policy")
. Still, I think we're talking about things that can vary between two
otherwise-similar protocol elements such that element1 might have setting v and
element2 might have setting v', and both work just fine, and the variable is
meaningful for some protocol, and the variable is determined by the
element-controller in some way understood by the controller and not necessarily
by anyone else.
Now, the reason I think this matters for the case at hand is simply that this effort is quite
correctly being presented as extremely lightweight for the publishers of such data, but it does not
appear (at least to me) to be lightweight for the consumer side: "browsers, security tooling,
and increasingly autonomous or LLM-based agents evaluating a claimed relationship between two
domains." Apart from LLMs, this was the same set of targets I had when I started with SOPA,
and it was the buzzkill drumbeat of "why would I?" from browsers that caused the whole
effort to grow the hairs of DBOUND that eventually ate it. Even when PSL maintainers said that
SOPA could help, some voices from the browser community had made up their minds. I worry about the
same fate for others. What makes this additional information useful enough to a consumer that it
can overcome the objection from browsers either of (1) needing to do an additional lookup or (2)
needing to invent UI/UX for a user to underst
and what difference the information makes?
Finally, this brings me to a related question: when should a consumer look for this
information, and what might cause them to do so? The document basically punts this as
"known hard problem", but I'd like to suggest that this hard problem _is_ the
problem of which the goals of the current draft are really a subset. I don't say this to
try to make the problem bigger and thereby put a drag on progress with this stab at the
more-limited problem. Rather, I worry that the authors are reproducing the very steps
that led me into the DBOUND thicket the first time. I, too, wanted to solve a limited
problem, but I kept running into the trouble that knowing whether a domain was in a
position to make the assertion (because it was registered in a domain that took public
registrations) was a necessary condition for this. That meant that improving the PSL or
providing an independent means to detect whether a name was in a place from which it
could be making the assertions it was became necessar
y, and that's how I got to SOPA.
I hope those are useful observations.
Best regards,
Andrew
[0]Disclaimer: Just in case you are (like a certain long-term, apparently irony-deficient participant) inclined to send me (another) off-list mail about how my very posting here discourages other people from posting anything, please understand my remarks as in no way an attempt to discourage progress with any draft. Suggestions are offered in the spirit of, it is hoped, friendly attempts to make something potentially more useful. They are worth exactly what you paid for them, unless you are currently employing me, which you almost certainly are not.
[1] In the unlikely case anyone cares, this was because part of what was worrying me at
the time had to do with the ICANN domain registry-registrar marketplace. I wanted
descendants to be able to express certain properties because I feared the emergence of
certain "registry services" that were, in effect, extra billing opportunities
for registries and registrars to extract cash for allowing you to do something with a
registered domain that you were previously able to do without paying anyone. The parent
can, of course, always force the child side to do something because the parent can
otherwise remove the delegation. The idea, however, was to exploit cases where the
parent was not willing (for extrinsic reasons) to do that (i.e. that they are regulated
by consensus policy and so can't just delete your name because you're not paying extra
for the privilege of doing what you used to be able to do for nothing) and then to use
the proposed mechanism in part to defend against these act
ions by the parent. At bottom this goal was incompatible with the thing the
anti-mail-abuse case needed, because in that case you very much need an ancestor to be
able to assert "not below here" contrary to whatever descendant domains might
say.
[2] When I looked, proustandco.com was available, but proustco.com was not.
Don't come to me with your UDRP problem later.
--
Andrew Sullivan
[email protected]
_______________________________________________
DNSOP mailing list -- [email protected]
To unsubscribe send an email to [email protected]