06-reference/research

scribble works dmarc quarantine shared resend key

2026-11-13·research-brief·source: deep-research·by Ray Data Co (deep-research synthesis)
scribble-worksdmarcresenddeliverabilitysanity-check

Scribble Works DMARC: is p=quarantine safe with a Resend key shared with Sanity Check?

The question

"Is moving Scribble Works' DMARC record from p=none to p=quarantine safe given the shared Resend sending key with Sanity Check, and what's the actual deliverability/reputation risk?"

Context: the email-domains SOP set a trigger (seven clean days of reports, then propose quarantine) that fired 2026-09-21 and has been carried unactioned through the 09-22, 09-29 and 10-06 lifecycle reviews. The shared key was the stated reason for hesitation.

Short answer: safe, with two cheap checks first. The shared key is not a DMARC risk at all. The real shared exposure sits at the Resend account level (complaint and bounce thresholds, key blast radius), and a DMARC policy change neither helps nor hurts it.

What we already know (from the vault)

Live records and reports (not vault; checked 2026-11-13, read-only)

DNS, by dig:

Name Record
_dmarc.scribbleworks.co v=DMARC1; p=none; rua=mailto:hello@scribbleworks.co; fo=1
_dmarc.news.scribbleworks.co same policy, its own explicit record
resend._domainkey.scribbleworks.co DKIM key present (also present for news.)
send.scribbleworks.co MX feedback-smtp.us-east-1.amazonses.com, SPF include:amazonses.com (Resend return path; same pattern on send.news.)
scribbleworks.co root SPF include:_spf.mx.cloudflare.net ~all, MX Cloudflare Email Routing, cf2024-1._domainkey present
_dmarc.raydata.co v=DMARC1; p=none; pct=100; rua=mailto:…@dmarc.postmarkapp.com; sp=none; aspf=r;
resend._domainkey.raydata.co, send.raydata.co Resend DKIM and return path present

Aggregate reports: about 36 Google reports for scribbleworks.co sit in the founder's mailbox, dated 2026-09-14 through 2026-11-11. I opened 9 of them, spread across the whole period. Every record in all 9 showed the same thing:

Limits of that evidence: 27 reports were not opened; Google is the only reporter found (no Microsoft, Yahoo or Apple reports); there are no reports for news. because it has not sent; and daily volume is tiny, so an occasional sender could be missed.

Code read (worktree at a 2026-09-13 commit, so possibly stale): every outbound path - sign-in and invite mail, the daily playset Worker, the feedback reply script, the guides broadcast script - calls the Resend API with a @scribbleworks.co From address. I found no other service sending as the domain.

What the web says

Convergences and contradictions

Synthesis for RDCO

The quarantine flip is low risk and reversible in minutes. All known Scribble Works mail goes through Resend, signs with the root domain's DKIM key and uses an aligned return path, so it passes DMARC twice over. Quarantine only changes the fate of mail that fails both, which the sampled reports show is none. Sanity Check is untouched because it sends from raydata.co. news.scribbleworks.co is also untouched: it has its own p=none record, which is right until its first broadcasts have produced clean reports. Nonexistent subdomains would inherit quarantine, which is the desired anti-spoofing effect. There is no reputation "reset" from tightening policy, and no deliverability gain to expect at this volume either - the benefit is that nobody can spoof hello@scribbleworks.co to parents.

Two things could break, and both are checkable before the flip. First, the 27 unopened reports: I sampled a quarter of them and found nothing, but the SOP's own bar is "no legitimate-mail failures", and that needs the full set read. Second, any mail sent as @scribbleworks.co outside Resend. The likeliest case is the founder replying to a parent from Gmail using "send mail as hello@scribbleworks.co": the root SPF record authorizes only Cloudflare and there is no Google DKIM key on the domain, so that mail would fail DMARC and land in spam under quarantine. I found no evidence he has set that up (unverified - it is a Gmail setting, not visible from DNS or the vault). Cloudflare Email Routing itself is not a concern: it forwards inbound mail, whose DMARC is judged against the original sender's domain. Stripe or storefront notifications would only matter if a custom sending domain were configured there; no such source appears in the reports.

On staging: skip pct. Resend says receivers largely ignore it, and at 1 to 2 messages a day a percentage sample is meaningless anyway. The staging that matters here is already built in - root first, news. later, reject later still. The proposed record is the current one with only the policy changed: v=DMARC1; p=quarantine; rua=mailto:hello@scribbleworks.co; fo=1. The weak point in the rollout is monitoring: reports land in the founder's inbox where nobody reads them. raydata.co already sends its reports to Postmark's DMARC digest; doing the same for scribbleworks.co would turn raw XML into a weekly summary.

The actual shared-key risks are separate from DMARC and remain open either way. (1) A leaked key can send fully authenticated mail as scribbleworks.co, news. and raydata.co; quarantine does nothing against that, because the attacker's mail passes DMARC. Resend supports per-domain sending_access keys, which would cut that blast radius to one domain per surface. (2) Resend's bounce and spam thresholds are per account, so a Sanity Check broadcast that draws complaints could pause sending for the account, and Scribble Works sign-in links (which double as password recovery) ride that same account. Separate keys do not fix this; only a separate Resend account or team would. Whether the pause is really account-wide is not stated in Resend's docs. Neither item should block the DMARC decision.

Why this is in the vault

It closes the quarantine proposal that the Scribble Works lifecycle email review has carried since 2026-09-22, by showing the shared-key objection does not apply and naming the two checks that remain. It also corrects the blast-radius claim in the email-domains SOP ahead of the first news. broadcast.

Open follow-ups

Related

Sources

Vault:

Primary evidence (2026-11-13):

Web: