Given the subject, it is probably a mistake to reply, but it's a new year...

You seem to be acting under some misapprehensions, some of which are
reasonable.

Yes, Google is well aware of JoeJobs, that's why we were involved in and
early proponents of the various authentication schemes (SPF/DKIM/DMARC/ARC).

Yes, Google only generates domain reputations based on authenticated
domains.

The other thing is that our error messages and suggestions pages are
non-specific.  I've mentioned this before on this list, but the gist
of the matter is that any spam determination is the result of one or more
spam rules which combine up to 20 or so spam signals, and
some of those signals are generated from machine learning models of past
spam based on even more individual mail transaction/message features.

These rules are constantly being monitored, updated and added to in
response to updated spammer tactics.

None of this is unique to Google, any major spam target either does this
themselves or outsources it to a third party antispam provider.

There is no one signal to mention in the error message.  There is no one or
limited list of suggestions on how to fix it, and the more specific
the suggestion the more out of date it might be.  Even if it were possible
to be more specific, we also wouldn't want to be specific
enough to help the spammers.

The error message is chosen from a limited list of options by the spam rule
writer and generally picked based on what they feel is
the primary signal or best way to represent it.  They are not updated very
often, and sometimes entire new classes of rules may roll out
without new/better error messages... fighting the spam and limiting the
false positives tends to take precedence over the limited utility
of the rejection message.  Our user studies have shown that the more
complicated the message, the harder it is to understand for the
typical user anyways, hence our own attempts to have very general bounce
messages we generate to users based on the rejections other
providers give us.

I haven't been on the Gmail team in quite a while now, and so I have no
particular insight into the specific rules involved here.  You can
look at postmaster.google.com for information about the reputation of your
domains... but that only helps if there is "enough" information
available (ie, high enough authenticated volume from a particular domain).
I know one historical way that "low reputation" has been used
incorrectly is for low volume or low nonspam volume or "unknown" domains.
Reputation heavily favors recency.  All of these things also
update in real time, so truly no spam evaluation is ever the same.

There have been extensive discussions on here about how some of these
things affect low volume senders and why based on our experience
with compromised hosts sending spam.  Low volume also makes it hard to find
good senders in bad neighborhoods.  This is the real issue with
the modern large scale email ecosystem, while it doesn't require you to be
one of the major providers, there is a relatively high bar for volume to
create the signal necessary for modern antispam systems to have good
discrimination.

The end result is that spam rejections can seem very arbitrary and
capricious.   In the large, false positives are low... but the rate will
always appear
high for any rejections on low volume domains.

hmm, it occurs to me given work we've done on my current team
that Mantel–Haenszel analysis would be useful to see how the false positive
rate
appears to different buckets, it has become more common across Google, but
I have no idea if the spam team is using that or not.

Anyways, this is a long winded way of saying sorry for your experience.

Brandon

On Thu, Jan 8, 2026 at 3:43 AM Peter N. M. Hansteen via mailop <
[email protected]> wrote:

> On Thu, Jan 08, 2026 at 07:31:22AM +0100, Peter N. M. Hansteen via mailop
> wrote:
> > If anyone here knows a sane person at Google, please point them at
> >
> > https://nxdomain.no/~peter/google-abuse/20260108_google_bounce.txt
> >
> > vs
> >
> > https://nxdomain.no/~peter/eighteen_years_of_greytrapping.html
> >
> > and ask them for conclusions or even better, to contact me.
>
> It looks like either one or more people here alerted someone in
> charge of these things, or perhaps there are even Googlers lurking here,
> since a resend to the same addresses succeeded just now (well, delivered,
> but directed to the spam folder, so there is room for improvement still).
>
> It also dawned on me that I had ranted in public about a similar episode
> some years back,
> https://nxdomain.no/~peter/does_your_email_provider_know_what_a_joejob_is.html
> ,
> a piece which is also linked in the retrospective article.
>
> To whoever set the fixing action in motion: Thank you!
>
> And I hope that the parties involved will preserve the knowledge acquired.
>
> All the best,
> Peter
>
> --
> Peter N. M. Hansteen, member of the first RFC 1149 implementation team
> https://nxdomain.no/~peter/blogposts
> https://nostarch.com/book-of-pf-4th-edition
> "Remember to set the evil bit on all malicious network traffic"
> delilah spamd[29949]: 85.152.224.147: disconnected after 42673 seconds.
> _______________________________________________
> mailop mailing list
> [email protected]
> https://list.mailop.org/listinfo/mailop
>
_______________________________________________
mailop mailing list
[email protected]
https://list.mailop.org/listinfo/mailop

Reply via email to