Ref

• T66818: Mitigate strict DMARC policy on the mailing lists
• T168467: Improve lists.wikimedia.org DMARC compatibility

When lists.wikimedia.org processes incoming emails (let's say revi.email for the purpose of this ticket), it strips DKIM-Signature from incoming mail and injects lists.wm.o DKIM-Signature before firing outgoing messages. This breaks DMARC (even for p=none) because end recipients only get DKIM Signature stamped by lists.wikimedia.org, not revi.email. (SPF will of course fail, so that's out of question.)
Maybe consider using ARC (it records the DKIM signature status when received by lists.wikimedia.org, thus allowing DMARC to pass even after mailing list processing)? (For example, kernel.org does ARC correctly, albeit not using mailman3...) (I originally intended to say "consider stop stripping senders' DKIM signature" but realized we tend to add footers, thus we are doomed to fail DKIM by editing body in transit.)

For some real life data, here's my 30 days statistics for revi.email. (Of this, kernel.org, wikimedia.org, sr.ht, fsfe.org, topicbox.com are mailing lists.)

• Out of 1963 emails sent from kernel.org (It is actually this single email), only 4 emails failed DMARC, which all of them seems to be Outlook unhappy with something, based on my reading of XML files. (Well, kernel.org mailing lists does not strip senders' DKIM signatures, at least).
• Out of 348 emails sent from wikimedia.org, all 348 emails failed DMARC.
• Out of 239 emails sent from sr.ht, only 1 email failed DMARC. My understanding of that 1 failure is double redirect.
• FSFE and topicbox lists addes footer, thus it is destined to fail DKIM.

w3.org mailing lists keeps senders' DKIM and adds w3.org DKIM. IETF mailing lists also does the same for ietf.org. Both uses mailman 3. (I had no chance to send email to those lists, so I have no real life data)
Also when you send email to -owners from revi.wiki (which has dmarc p=quarantine), it just appends lists.wikimedia.org DKIM on top of existing revi.wiki DKIM, so this is not a generic mailman 3 issues.

As far as I can see, toolforge.org does ARC signing too, so ARC signing doesn't sound like an 'impossibility'.