How mail for this domain actually works, what cannot be fixed, and what was decided. This file exists because four agents re-derived these facts from scratch in three days (LAC-4545, LAC-4546, LACA-2204, LACA-2252) and two of them independently proposed the same fix that does not work.
Read this before touching DNS, Gmail settings, or DMARC for this domain. If you are about to propose a fix, check it against "Non-fixes" first.
Tickets: LAC-4545 (the bounce), LAC-4546 (the direction decision), LAC-4673
(the rua evidence), LACA-2302 (Resend's delivery log), LAC-4548 (Cal.com),
LACA-2287 (support@), LACA-694 and LACA-1913 (the DMARC ramp programme).
| Piece | Reality |
|---|---|
| DNS | Cloudflare |
| Inbound | Cloudflare Email Routing (route1-3.mx.cloudflare.net) forwarding into Gmail (gojukebox) |
| Google Workspace | None. This domain has no Workspace subscription |
admin@buildandserve.com | A Google account with no Gmail mailbox at all (the Gmail API returns 400 failedPrecondition). It exists to hold a Calendar |
lacy@buildandserve.com | A Gmail send-as alias, routed through smtp.resend.com:465, treatAsAlias: true, verification accepted. Aligns, and that is observed, not inferred (see "Observed mail") |
support@buildandserve.com | A Gmail send-as alias still on smtp.gmail.com:587, treatAsAlias absent. Cannot align. Client-facing. LACA-2287 |
| Outbound that can align | Resend, and only Resend. Not every send-as is on it, see support@ above |
All ten rows on gojukebox@gmail.com, read 2026-10-17:
| Send-as | SMTP relay | Port | Mode |
|---|---|---|---|
gojukebox@gmail.com | (primary, none) | ||
lacy@shipkit.io | smtp.resend.com | 465 | ssl |
lacy@buildandserve.com | smtp.resend.com | 465 | ssl |
me@lacymorrow.com | smtp.resend.com | 465 | ssl |
spin@spinrc.com | server21.abstractdns.com | 465 | ssl |
support@buildandserve.com | smtp.gmail.com | 587 | starttls |
| 4 consumer/edu addresses | (none, correct for them) |
support@ is the only address on a domain we own that is routed through
Google's own SMTP. That is the whole defect, and the table is the argument.
List the rows yourself rather than assuming, and pass --json:
gog gmail settings sendas list --account gojukebox@gmail.com --json
--json is not a preference, it is the difference between right and wrong.
gog gmail settings sendas get in its default format prints eight fields and
silently omits smtpMsa entirely:
send_as_email support@buildandserve.com
display_name Build And Serve Support
reply_to
signature
is_primary false
is_default false
treat_as_alias false
verification_status accepted
Read that output and every alias looks like it has no SMTP relay configured,
which reads as "Gmail sends it directly" for all ten rows. It has already been
reported once as lacy@ having "treatAsAlias: true with no smtpMsa at all."
That is the formatter, not the mailbox. --json shows the field.
Generalise it, because it is the same mistake as the DKIM-selector one below: absence in a tool's output is not absence in the world. Confirm a field is missing from a raw view before you say it is missing. And before calling anything human-only, name the API call and show its error.
Verify any of it yourself in under a minute:
dig +short TXT buildandserve.com # SPF
dig +short TXT _dmarc.buildandserve.com # policy
dig +short TXT resend._domainkey.buildandserve.com # the aligned path
dig +short TXT cf2024-1._domainkey.buildandserve.com # inbound forwarding only
dig +short MX buildandserve.com
Two DKIM selectors exist: resend and cf2024-1. Probing only
google, default, selector1, s1, k1, mail, dkim returns nothing
and looks like "no DKIM is published." That conclusion has been reached and
posted at least once. It is wrong. Probe resend and cf2024-1.
cf2024-1 is Cloudflare's signature on mail it forwards inbound. It does
nothing for outbound alignment, but it is why a bounce DSN can show
dkim=pass header.i=@buildandserve.com.
Generalise that: a passing cf2024-1 anywhere is not evidence that an
outbound path works. Forwarded inbound mail produces a fully aligned
dkim=pass and shows up in the aggregate reports as a clean pass from a
Cloudflare IP (104.30.x). Five of the 34 messages in the Sep 25 to Oct 15
window are exactly that. Only the resend selector means we originated it.
A Google Calendar invitation organised from admin@buildandserve.com is
emitted by Google's own infrastructure, signed d=google.com, with a
google.com envelope sender. Under DMARC both aligned checks fail:
google.com, which does not align to buildandserve.com
even under adkim=r.google.com bounce address, so buildandserve.com's SPF record is never
consulted.No setting changes this. Calendar notifications cannot be routed through an
external SMTP relay; Gmail's "send mail as -> SMTP server" option only affects
mail composed in Gmail. Without a Workspace subscription publishing
google._domainkey, there is nothing to align with.
Both of these have been proposed by agents reading live DNS. Neither works.
include:_spf.google.com to SPF." Buys zero alignment, for the
envelope-sender reason above. It also authorises every Google relay to send
with a buildandserve.com envelope. It was added on 2026-10-16, questioned,
and reverted the same night. Do not re-add it. The same applies to
include:amazonses.com, which was added in the same edit.p=quarantine permanently. A hold is recorded in
dmarc-zone-exceptions.txt and dmarc_ramp_preflight.sh returns HOLD for
this domain. p=reject was never the goal; mail arriving was.admin@buildandserve.com.p=reject, at the cost of a migration that can break mail.p=reject while its mail is unalignable. At
quarantine the failure is a spam folder; at reject it is a hard bounce.
LACA-1913 (lacymorrow.com, identical failure mode) stays held indefinitely.What would flip decision 3: evidence that Lacy's ordinary client mail from
lacy@buildandserve.com is failing DMARC at volume. That test has now been run
twice and came back negative both times: the rua data (LAC-4673) and then
Resend's own delivery log, ten delivered client messages from lacy@ with no
failures (LACA-2302, see "Observed mail"). So decision 3 stands on data, not
inference. Do not reopen the Workspace purchase without new
evidence of the same kind. The authoritative source is the DMARC aggregate
reports arriving at dmarc@buildandserve.com, read as described below, not
another round of DNS inference.
Twenty days of rua, 2026-09-25 to 2026-10-15, 45 reports, 34 messages,
reparsed from the raw XML for auth_results (LACA-2286, LAC-4673):
| Path | Msgs | Aligned | Disposition |
|---|---|---|---|
Resend via SES (54.240.11.x, 54.240.48.x, selector resend) | 25 | pass | none |
Cloudflare inbound forward (104.30.x, cf2024-1) | 5 | pass | none |
| Google relay of a Resend-signed message | 2 | DKIM pass | none |
Google, no DKIM at all, envelope gmail.com | 1 | fail | reject |
sanofi.com forward, not ours | 1 | fail | quarantine |
32 of 33 of our messages aligned-pass. Two recipient-side
Authentication-Results headers show dmarc=pass header.from=buildandserve.com
at policy.dmarc=reject: the Resend path worked even through the reject window.
The one failure is the smtp.gmail.com:587 fingerprint, which is support@.
Three traps when reading this data. All three have already cost a day:
orgs in dmarc_pull.py's summary.json is the reporting org, not the
sender. It cannot tell Resend traffic from Cloudflare traffic.summary.json carries only policy_evaluated. The aligned-vs-raw question
lives in auth_results in the raw XML, so reparse it.dmarc_pull.py SINCE is the mail-search range, not the data range. Ask
for 30 days, get 20, because that is when reports start. Report the window
the data has, not the one you asked for.What rua cannot tell you: Gmail send-as traffic and Resend-API app
traffic are indistinguishable. Both emit SES IPs, selector resend, envelope
send.buildandserve.com. So the 25-message block proves the path aligns; it
does not single out the messages Lacy typed in Gmail. Closing that gap needs
Resend's own delivery log, which is what the next section is.
lacy@buildandserve.comPulled 2026-10-17 (LACA-2302). lacy@ is not just configured correctly, it is
observed sending real client mail through Resend, the same standard of
proof support@ now has:
curl -s -H "Authorization: Bearer $RESEND_API_KEY" \
'https://api.resend.com/emails?limit=100'
63 messages logged, 2026-09-21 to 2026-10-17. By sender:
| From | Msgs |
|---|---|
invoice@buildandserve.com | 21 |
notify@buildandserve.com | 28 |
lacy@buildandserve.com | 10 |
support@buildandserve.com | 2 (the LACA-2287 tests) |
noreply@buildandserve.com | 2 (contact form) |
All ten lacy@ messages show last_event: delivered. They are ordinary
correspondence to named clients, dated 2026-09-25 through 2026-10-17: a
Jellyfish proposal, a Klaviyo account transfer, three Fwd: of a BigCommerce
migration quote, Re: RCB - Google ad account, a Town of Harmony listing
update. GET /emails/:id returns the body, and it is hand-written prose.
That is enough to attribute them to the Gmail compose path. The app sends
as team@ (RESEND_FROM_EMAIL), invoice@, notify@ and noreply@. Nothing
in the codebase sends as lacy@, and no app generates a Fwd: of a client
quote. So the only origin for these ten is Lacy typing in Gmail and Gmail
relaying through smtp.resend.com:465.
Two limits worth stating. The log records the message, not the submission
protocol, so it does not literally stamp these "arrived over SMTP". And
Resend's log is the only place this exists: the rua data cannot show it, for
the reason in the paragraph above.
Cost: one authenticated GET, no DNS change, none of Lacy's time. The earlier
note that this was not worth pulling was wrong, and so was any claim that
observed proof for lacy@ could not be obtained.
Ten days at p=reject (2026-10-05 to 2026-10-15) produced one bounce:
the copy of the Jellyfish kickoff invite addressed to lacy@buildandserve.com
itself. The client copy of that same invite reached klberrie@gmail.com and
she accepted it.
The one live defect in twenty days of reports is support@buildandserve.com
on smtp.gmail.com:587: header-from buildandserve.com, envelope gmail.com,
zero DKIM signatures, both aligned checks fail, disposition reject
(2026-10-10, reported by Enterprise Outlook). Same fingerprint as the Oct 9
action=oreject bounce to hadas.milstein@emburyjarvis.co. It is client-facing
and it needs no DNS change, only the send-as MSA swap to Resend. Tracked on
LACA-2287.
The key is no longer the blocker and never needed a human. Resend already had a
per-address convention (Gmail - lacy@buildandserve.com, Gmail - jonathan@buildandserve.com), so POST /api-keys with
permission: sending_access and the buildandserve.com domain_id minted
Gmail - support@buildandserve.com on 2026-10-17. No coupling to the site's
production key. Enumerate with GET /api-keys before concluding a key must be
reused. A sending_access key cannot read GET /emails; it returns
401 restricted_api_key.
The actual blocker is that send-as cannot be written by API at all.
gmail.users.settings.sendAs.patch returns:
403 PERMISSION_DENIED
Access restricted to service accounts that have been delegated
domain-wide authority
This is not specific to smtpMsa. A no-op displayName-only patch against
gmail.settings.basic returns the identical 403, which is the control that
makes this a real negative rather than an assumption. gojukebox@gmail.com is a
consumer account, so domain-wide delegation cannot exist for it. Do not hunt
for a scope, a service account, or a re-auth. The Gmail web UI is the only
path, permanently. Budget it as a two-minute human paste, hand over the exact
field values, and include the rollback config.
Cloudflare DNS, the Gmail account, and Invoice Ninja all live on the LACA
side (Zero, io). No LAC agent holds a Cloudflare secret. Changes go over
paperclip-bridge, not by minting new tokens.