On Fri 15/May/2026 19:41:00 +0200 Richard Clayton wrote:
In message <[email protected]>, Alessandro Vesely
<[email protected]> writes
On Thu 14/May/2026 23:56:08 +0200 Richard Clayton wrote:
In message <CAH48Zfy=jTP3VA_WwYDcEm=BE4hzhOqywSGFwZNDqCzYq=+=o...@mail.gmail.com
, Douglas Foster <[email protected]> writes
[...]
* DKIM2 requires recipient email filters to maintain a list of
trusted mailing lists, so that those DKIM2 transforms can be
accepted. This also has been difficult to achieve to date.
I expect that mailbox providers will apply heuristics as the type (and
scale) of recipes that they will deem appropriate. Note that the
exploded flag allows identification of mailing lists, which current
systems do not do (yes, I know about List-* header field, but I have
done enough analysis of what is seen on incoming email at scale to know
that they are of little practical use at that scale)
This is where I lose track of the reasoning. How can a protocol rely on
unspecified heuristics to decide whether to override a domain's published
policy?
receivers can do whatever they wish ... when this is likely to affect
interworking we draw attention to this by mentioning "local policy" in
the protocol specification
That is precisely the somewhat grey point in DMARC. The straightforward
interpretation of p=reject is to reject unauthenticated messages using that
From: domain. Of course receivers retain discretion, but when receivers
randomly disregard the author's published policy, interoperability suffers.
Protocols work reliably when the expected response is sufficiently well
specified. SMTP itself allows local policy at many stages, but not to the
extent that core semantics become heuristic.
Let me connect this to the upcoming DMARC update. It says that domains whose
users post to mailing lists "SHOULD NOT publish p=reject". That is clearly a
limitation. However, the use of SHOULD rather than MUST suggests that the
mailing-list problem is expected to become solvable eventually.
SHOULD is not a forwarding looking statement ...
This word, or the adjective "RECOMMENDED", mean that there
may exist valid reasons in particular circumstances to ignore a
particular item, but the full implications must be understood and
carefully weighed before choosing a different course.
I see this point has already been explained well by Barry.
Now consider a message from a domain publishing p=reject. The message is
exploded by a mailing list and modified in transit. The mailbox provider must
then decide, precisely, whether to accept or reject it.
quite so
I do not see how f=exploded changes the situation fundamentally.
it allows the mailbox provider to apply a different set of heuristics
than they would do otherwise
This is exactly the point of my disconcert. Once acceptance depends on a
changing set of heuristics, the operational semantics become fuzzy. Different
receivers will inevitably reach different conclusions for the same message stream.
When behavior is specified precisely, interoperability is generally reliable.
When the intended outcome depends on statistical classification and local
inference, interoperability becomes probabilistic.
Now, DMARC had to fit into an already deployed SMTP ecosystem, so it could not
realistically say MUST reject in all cases. We therefore ended up
distinguishing the "means" from the "end", using Murray's terms. But if we
seek a predictable response, we still need sufficiently deterministic means to
synthesize it.
It can be asserted as arbitrarily as List-Id:, and it does not establish
that the modified message still preserves the semantics of the author's
signed text.>
this is true but does not seem to have much to do with ARC
My point is that ARC, like f=exploded, attempts to improve classification of
transformed traffic, but it still leaves the receiver to infer whether the
modified message should be treated as semantically equivalent to the author's
original signed message.
That inference step is exactly where heuristic divergence enters the system.
Best
Ale
--
_______________________________________________
dmarc mailing list -- [email protected]
To unsubscribe send an email to [email protected]