Turn "our email is failing authentication" into a specific diagnosis — which mechanism,
which source, which alignment — and exact corrected records.
Climb the Troubleshooting Ladder base skill first: this domain's past authentication
work (a half-finished migration or earlier record change explains most sudden failures),
then the DNS host, who may edit it, and the documented sending-source inventory. An
incomplete inventory blocks any strict-enforcement recommendation.
Then get the evidence — the failing message's full headers (Authentication-Results) or a
DMARC report excerpt: which mechanism failed (SPF, DKIM, or DMARC alignment), for which
source, and what the receiver did (none, quarantine, reject). Then read the live TXT
records — root SPF, _dmarc, the DKIM selectors — and diagnose against what is published,
not the documentation.
Branch:
- SPF — the failing source is missing, or the record is broken. Check: does it include
every legitimate source from the inventory; does it exceed the 10-DNS-lookup limit
(silently fatal, and common); are there multiple SPF records (invalid)? Produce the
exact corrected single record from each vendor's documented include. Prefer ~all;
recommend -all only once the inventory is confirmed complete.
- DKIM — missing or broken signature. Which service should be signing (Microsoft 365,
the gateway, a marketing tool), and is its selector's public key published? The fix is
two-sided: enable signing at the source, publish the selector record the vendor names.
If it can't sign, the client accepts SPF-only alignment there or changes vendor.
- DMARC alignment — SPF and DKIM each pass, yet DMARC fails because neither passes for
the visible From domain. Classic with forwarders and third-party senders on their own
envelope domain; the fix is a custom return-path or DKIM signing as the client's
domain at that vendor. Check alignment explicitly.
- New DMARC rollout — p=none with rua reporting first, monitor a full business cycle,
fix the failing legitimate sources, then quarantine (optionally with pct=), then
reject. Refuse to jump straight to p=reject on a domain with unverified sources: a
wrong policy silently drops the client's mail.
Deliver guidance, don't execute. DNS edits belong to whoever owns DNS access: give the
exact record name, type, value, and where it goes. State the TTL and the honest window:
up to the TTL, practically 24-48 hours at stragglers. One change, then wait and
re-verify; never thrash the record or say "it should work now" right after an edit.
After propagation, re-check the live records and a fresh Authentication-Results header.
If mail still junks once authentication passes, the lever is sender reputation and the
receiver, not more record edits. Note it (apply the PSA Note Discipline base skill):
diagnosis, records before and after, verification, enforcement plan.
Magic Library