Ale,
ARC failed to solve the problem because the available data is insufficient.

SPF Authentication seeks to confirm that the originating server was
authorized to send messages for the original RFC5321.MailFrom.  So I tried
to correct SPF by determining the originating server and the originating
RFC5321.MailFrom address, so that SPF and local policy equivalents could be
evaluated based on original values.   (DKIM seemed of little value for this
purpose because signatures can be added or invalidated in transit, and many
messages lack a relevant signature.)

The first step was to analyze the Received chain.  I tried to partition it
between submission client, originating domain, and intermediary domains 1
to N.   This was a difficult process because some changes in server domain
name are not changes in administrative domain, while some submission
clients are not documented with ASMTP or ESPTMPA.

Determining the originating RFC5321.MailFrom domain is difficult and
sometimes impossible, since it is an attribute of the transmit session, not
the document.   A Received header field may indicate forwarding because of
a "FOR otheruser@otherdomain" term.   The forwarding operation may even be
confirmed if "otherdomain" later becomes the RFC5321.MailFrom domain.   But
knowing that a forwarding step occurred does not indicate the
pre-forwarding address.

I used all available clues, without worrying about a fraudulent header
structure.   Indications of the prior RFC5321.MailFrom address can appear
in ARC-Authentication-Results, Authentication-Results, Recieved-SPF, and
the "(Envelope-From <user@domain>)" term of a few Received headers (those
written by Exim)..  To be useful, I needed to have a header that contained
both an IP address and an RFC5321.MailFrom address, and those values had to
be verifiable against a Received header field.

If the data could be computed reliably, I was prepared to configure
whatever local policy exceptions would be needed before enforcing
authentication failures based on that data.   I just never felt like I had
enough accuracy to use it effectively.

One workaround would be for every Received header to have a FOR term and an
ENVELOPE-FROM term, but I have realized that will not happen.

Maybe DKIM2 will eliminate most of this complexity.  It will clearly
indicate where forwarding has occurred, so that the forwarding point can be
judged for acceptability.   The RFC5321.MailFrom address history becomes
unimportant as long as a DKIM2 signature for the RFC5322.From address is
present, is assumed to be applied without deception of the signer, and is
verificable upon receipt.  It will not solve the problem of authenticating
unsigned messages that are forwarded.   I think it unlikely that it will
change anything for mailing lists, because the problems associated with
transform trust will remain indefinitely.

Dpug


Doug


On Fri, Jul 17, 2026 at 12:40 PM Alessandro Vesely <[email protected]> wrote:

> On Fri 17/Jul/2026 17:31:46 +0200 Murray S. Kucherawy wrote:
> > On Wed, Jul 15, 2026 at 10:40 AM Alessandro Vesely <[email protected]>
> wrote:
> >
> >> I think it's important to understand why the experiment didn't produce
> the
> >> expected results.  However, neither the I-D nor the RFC state exactly
> what
> >> results were expected.  Conveying original authentication assessments
> across
> >> trust boundaries, as a goal, was certainly achieved.  Fixing indirect
> mail
> >> flows certainly not.
> >
> > I thought the Abstract and Introduction sections of RFC 8617 clearly
> > described the expected result: ARC was meant to provide information
> > supplemental to DKIM enough to resolve the indirect mail flow problem.
>
>
> I agree that this was the aspiration.  However, RFC 8617 does not actually
> say
> that the additional information resolves the mailing list problem.  It
> says
> that the additional information can be used to inform message
> disposition.  It
> does not claim that the additional information is, by itself, sufficient
> to
> justify overriding DMARC policy.  In practice, it is not.
>
> In hindsight, this left a significant gap between the protocol mechanism
> and
> its operational use.  The RFC deliberately stops short of specifying how
> receivers should use ARC results in disposition decisions, apparently
> expecting
> operational practice to evolve separately.  That never really happened.
> Ideas
> such as a broadly shared trust database did not materialize, and there was
> no
> common understanding of when ARC evidence should justify overriding a
> DMARC
> failure.
>
>
> > Apparently, broadly, it didn't.
>
>
> Without a shared operational model, deployment remained limited, and
> different
> implementations developed different interpretations of the protocol.  I
> would
> argue that what fell short was not so much the protocol as the operational
> ecosystem around it.
>
>
> Best
> Ale
> --
>
>
>
>
> _______________________________________________
> 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