Richard Clayton’s experience shows that ARC is not sufficient for
interpreting complex flow schemes.   At the same time, his experience is
useful because it draws attention to the needs of evaluators, instead of
the needs of senders.

The evaluator seeks to accept all wanted messages while obstructing all
unwanted messages, and has these tools:


   - authentication results,
   - identity reputation data, and
   - content filters.

Content is interpreted in the context of the author, so a fraudulent author
always implies unwanted content.   Reputation data is only accurate when
the autho'sr identity is accurate.   Consequently, all email filtering is
dependent on having accurate author information.   As such, email filtering
should ensure that all accepted messages have been authenticated, a result
which can be achieved using a mix of algorithm, delegated trust, and local
policy.  Because every authentication failure introduces a risk, but not a
certainty, of malicious impersonation, the optimal response to
authentication failure is always quarantine, regardless of the domain owner
policy.

Unfortunately, our authentication techniques are attempting to determine
authentication at origin based on information detected upon final
delivery.    This means that authentication results are only reliable when
the originating organization and the delivering organization are the same,
because only the originator and receiver organizations are involved.

Fortunately:

·         Most messages are authentic upon origination, even when they
arrive without authentication credentials.

·         Most in-transit credential upgrades are applied by outbound
gateway services to mail that is free of impersonation, so authentication
results are usually accurate even when mail is not received directly.

Because of this good fortune, authentication failure is often ignored.
DMARC and SPF have been widely interpreted as creating a distinction
between failures that matter and failures that do not matter.   This
distinction is false.

Given the limitations of our authentication techniques, mail flow analysis
becomes a potential fourth technique for email filtering, but it is filled
with difficulty.   Email filtering is already infinitely complex, and the
set of all possible mail flows introduces yet another level of infinity.
ARC attempts to address the complexity of mail flow analysis.

ARC loses reliability easily.   Reliability is lost when intermediary
organizations do not participate, when any participant reports
inaccurately, or any intermediary acts maliciously.  Therefore, ARC may be
useful only in the special case where:

·         the message flow involves only three organizations (originator,
intermediary, receiver),

·         ARC is implemented by the intermediary, and

·         the ARC data provided by the intermediary is trusted by the
receiver, and

·         the intermediary is indifferent to whether the recipient uses the
ARC data or not.

The last criterion suggests why mailing lists felt no incentive to embrace
ARC:   before they can stop munging the From address, they need certainty
that the ARC data is accepted, and they need it accepted by all
participants.   They are not at all indifferent to how ARC data is used by
recipients.

ARC data should be very helpful to the case of a simple auto-forwarding
configuration.    The authentication problems created by auto-forwarding
have been largely ignored because of the fantasy that failures only matter
if they matter to the domain owner.   Once that idea is set aside,
evaluators will seek a way of knowing which messages from a general mail
flow should be accepted and which discarded.    Our two author
authentication techniques are DKIM alignment, which is only possible if the
message is signed, and SPF alignment, which is impossible when the message
is forwarded.   A trusted ARC set allows the recipient’s evaluator to know
with certainty that a forwarding operation has occurred, and allows them to
apply SPF alignment analysis as if the message had been received directly.
That data can be inferred from the Received chain, but ARC ratifies that
data and helps distinguish forwarding flows from outbound gateway flows.


DKIM2's usefulness will be proportional to its deployment.   It may help
with the mailing list problem.   It will not solve the entirety of the
auto-forward problem unless it is deployed universally.   It will help with
other complex flows only to the extent that identity is known with
certainty by the entity which applies the initial signature.  I continue to
have problems with trusting signatures applied by outbound gateways.


Doug Foster




The accuracy of submitted ARC data can be validated against the Received
chain, and ARC helps to ratify what the Received ch

On Fri, Jul 17, 2026 at 6:15 PM Dotzero <[email protected]> wrote:

> I waited until the last minute to respond because I wanted to see as many
> responses as possible before sharing my thoughts. I am personally agnostic
> as to with regard to ARC.and its utility  Having said that I do not see a
> consensus that ARC should be marked historic (at this time). Apologies for
> top posting.
>
> 1) I see a number of people have asserted they find it useful.
>
> 2) I have seen (unrefuted) assertions that the metrics in the original
> document for determining success or failure have not been followed in
> claiming failure.
>
> 3) One of the key assertions for marking ARC historic is premised on DKIM2
> replacing it's functionality. To me this is a specious argument. DKIM2 is
> itself experimental and thus far is only partially baked. It has not been
> deployed and tested in the wild. It is thus at best promiseware.
>
> I see no particular need for this working group to rush to adopt Trents
> document and mark ARC historic. I see 2 reasonable alternative options:
>
> 1) Shut the group down and leave ARC status as-is. Let the supporters of
> ARC continue their experiment and come back at a later point with an
> updated document and data.
> 2)  Shut the group down and leave ARC status as-is until DKIM2 has time in
> the wild and it is shown that it's functionality does in fact replace ARC.
> That would be a more appropriate time to consider marking ARC historic.
>
> Michael Hammer
>
> On Fri, Jul 3, 2026 at 10:29 AM Barry Leiba <[email protected]> wrote:
>
>> OK, this discussion has gone for a while and I don't see new things
>> coming up.  I'd like to take a particular approach now to see where we
>> are and where we're likely to go.  In particular, for right now I do
>> not want to look at the "dmarc-arc-to-historic" document in
>> particular, but instead to just focus on the idea of making ARC
>> "Historic" -- should we do it or not?
>>
>> So, first:
>> If you DO NOT think we should make ARC "Historic", please reply to
>> this message and tell us, *as concisely as you can* why you think we
>> should not do that.  I'm really not looking for long explanations
>> here, so please try to keep it tight and clear.  If, when you reply,
>> you've already seen a reply that says what you want to say... then
>> it's good to just say "I agree with what Ferd said."  Apart from that,
>> please... no comments on what people say here yet.  I want to get the
>> arguments clear first.
>>
>> Please do this before the start of the Vienna Hackathon... so, by the
>> end of everyone's day on Friday, 17 July, two weeks from today.
>>
>> (I'm not going to forbid other discussions at this point, but don't
>> make me do it.  It would be best, to avoid distractions and
>> temptations, if we kept the DMARC mailing list for this purpose for
>> now.)
>>
>> Thanks,
>> Barry, as chair
>>
>> _______________________________________________
>> dmarc mailing list -- [email protected]
>> To unsubscribe send an email to [email protected]
>>
> _______________________________________________
> dmarc mailing list -- [email protected]
> To unsubscribe send an email to [email protected]
>
_______________________________________________
dmarc mailing list -- [email protected]
To unsubscribe send an email to [email protected]

Reply via email to