Domain Email Warming
A domain that has never sent mail has no reputation. The first hundred cold messages from it are judged on nothing, and receivers resolve nothing to spam. Warming builds that record deliberately: a small, growing volume of real correspondence that gets opened and replied to, from mailboxes you control, measured at the destination.
The rule that governs this skill: a send receipt is not a result. Every provider will tell you a message was accepted. None of them will tell you it reached the inbox. If you are not reading the delivered message on the receiving side, you are not warming a domain, you are generating traffic and hoping.
Be honest about what warm-up does and does not buy
State this to the user before they spend weeks on it.
It helps with: establishing a sending history so volume is not itself suspicious; proving your authentication works before real prospects see it; finding out which message formats get filtered; building engagement signals (opens, replies) that receivers weigh heavily.
It does not fix: bad list quality, irrelevant messages, or a domain with an existing bad reputation. Warm-up is not a laundering step. A domain warmed for a month and then used to blast a purchased list is back in the spam folder within days.
Its main limit: warming to mailboxes you own at one provider measures one
provider. Gmail's verdict tells you little about Outlook or a corporate filter. Plus
addresses (you+tag@gmail.com) all land in one mailbox, so they give you parallel
sends, not recipient diversity. Say so rather than implying broader coverage.
The procedure
1. Preflight before a single message goes out
Warming a domain that fails authentication is worse than not warming it: you pay to teach every receiver that your domain sends unverifiable mail.
node scripts/warmctl.mjs preflight --config <config>.json
Fix every FAIL first. See references/authentication.md for what each check means, and for the DMARC inheritance trap that silently kills a new sending subdomain.
2. Decide the sending identities
Warm the mailboxes you will actually send from — hello@, a founder address, the
address outreach replies to. Spreading volume across several mailboxes is closer to how
a real company sends than funnelling everything through one.
On sending subdomains: a separate subdomain (mail.example.com) isolates outreach
reputation from transactional mail, so a spam complaint on outreach does not put
password resets at risk. The cost is that it starts from zero reputation and must be
warmed separately. A subdomain is usually send-only — it has no inbound route — so give
every subdomain identity a replyTo on the apex domain, or replies bounce.
3. Ramp the volume
node scripts/warmctl.mjs plan --config <config>.json # inspect
node scripts/warmctl.mjs send --config <config>.json --apply # send
Start small, grow gradually, never step. The shape of the curve matters less than the absence of jumps. references/ramp-and-volume.md covers sizing and why most of each day must go to a mailbox you can read.
4. Close the loop on the receiving side
This is the step people skip, and it is the one that does the work.
node scripts/warmctl.mjs engage --config <config>.json --apply
For each delivered message it records where the message landed, pulls it out of spam if it is there, marks it read, and replies to a share of them. Rescuing from spam tells the provider its classification was wrong. A reply is the strongest positive signal a mailbox can produce. See references/placement-measurement.md.
5. Read the result, per format
node scripts/warmctl.mjs report --config <config>.json
Formats are filtered differently, so the tool sends several shapes and reports each separately. From a real day-one run on a freshly warmed domain:
by format variant
plain 100% primary primary:4
html_simple 100% primary primary:3
html_logo 67% primary updates:1 primary:2
html_rich 67% primary primary:2 updates:1
newsletter 0% primary updates:3
The gradient is the point. Nothing was marked spam, so a naive "delivered" check would have called this a total success — while the broadcast-shaped message, the one that most resembles real outreach, never reached the Primary tab once. Test the shape you intend to send, not a plain-text stand-in for it.
Consent is a separate gate, and it is not optional
Warming is self-mail and raises no consent question. The outreach it leads to does. In Canada, CASL governs commercial electronic messages and requires a consent basis, sender identification and a working unsubscribe — with penalties that apply per message. Do not let a warmed domain become the reason someone skips that gate. references/consent-and-law.md has the specifics.
Provider terms
Check the sending provider's terms before running warm-up through it. Cloudflare Email Service, for instance, is documented as transactional-only, and the same account often carries real transactional mail. If the provider suspends sending, the penalty lands on that stream, not on the warm-up. Keep warm-up volumes modest and the content genuinely 1:1, or use a provider intended for bulk.
The toolkit
scripts/warmctl.mjs — Node 22, no dependencies, one JSON config, one JSON state file.
Everything that sends, replies or changes DNS is a dry run unless you pass --apply.
| Command | Does |
|---|---|
preflight |
SPF, DKIM, DMARC, MX per sending domain, against authoritative nameservers |
plan |
The day's schedule, without sending |
send |
The day's messages, idempotent within a day |
engage |
Classify placement, rescue from spam, mark read, reply |
report |
Placement overall, by day, and by format variant, plus provider quota |
scripts/provision-domain.py — idempotent Cloudflare setup for a new domain:
sending-domain registration, the DNS each asks for, apex MX and routing rules
to the inbound worker. Sending registration alone leaves a domain able to send
but unable to receive (see references/multi-domain.md).
Fleet sweeps
One config directory (*.warmup.json, identities only) feeds every sweep. Each
is a dry run unless passed --apply, dedupes every send in state, and spreads
work across mailboxes and days rather than bursting.
| Script | Does |
|---|---|
signup-sweep.py |
Signs warmup mailboxes up for agent-facing services (OTP loop); trashes consumed codes so the seed inbox stays clean |
subscribe-sweep.mjs |
Subscribes mailboxes to announce-only newsletters (59 lists) and completes confirmations |
vendor-outreach.mjs |
Sends pre-sales questions to vendor support/sales inboxes; each vendor hears from at most 6 domains with unique copy each |
contact-sweep.mjs |
Submits vendor contact/sales web forms; targets ship unpiloted until --pilot proves the field map |
invite-sweep.mjs |
Trades same-domain calendar invites (.ics) between mailboxes, 2 per domain per week |
confirm-subscriptions.mjs |
Completes double-opt-in confirmations for list mail |
reply-inbound.mjs |
Replies to real human mail that reaches warmup boxes |
dmarc-report.mjs |
Reads DMARC aggregate reports from the seed inbox, summarizes pass/fail per domain |
validate-newsletters.mjs |
Checks subscribed newsletters still arrive |
digest-sweep.mjs |
Subscribes the seed to newsletters and alerts via a real browser; publications ship unpiloted until --pilot proves the loop |
notify-sweep.mjs |
Nudges activated form endpoints for on-demand notification drips |
engage-seed.mjs |
Seed-side engagement: rescues warmup mail from spam, stars a sample, reads the stale tail |
reputation-monitor.mjs |
Standing watch: fleet DNS auth plus SURBL/Spamhaus listings, exits 2 when action is needed |
inbox-backstop.py |
Re-applies the warmup filter rule to mail Gmail's own filter skipped |
Note: the cross-mailbox self-mail loop (warm-tick.sh/warm-all.sh plus
warmctl send/engage) is currently parked by operator decision — configs
carry no seeds and the schedule is disabled. The sweeps above are the active
program. Re-enable by restoring seeds and the schedule.
Browser and web sweeps follow the pilot-gate convention: a target ships with
pilot:false plus the observed evidence, and --apply refuses it until an
attended --pilot run proves the full loop (submit, confirmation mail,
verified receipt). Probe before you scale.
Seed-inbox mail from all of the above stays out of the operator's face with
Gmail filters: the warm-up rule (Skip Inbox, apply the warmup label) plus
a Glama-only rule (see running-unattended.md). The warm-up query is
generated, not hand-written: build-filter-query.py builds it from the live
configs, and inbox-backstop.py re-applies it hourly to anything Gmail's own
filter skipped. Regenerate the query when warmed domains or warmup senders
change, then update the filter to match.
Configure with examples/private-repo-f304.warmup.json as the template. The Cloudflare adapter is
one file (scripts/lib/cloudflare.mjs); another provider means replacing that file, not
the tool.
Related
growth-cold-email— the campaign this warm-up is preparing for.lifecycle-email— classifying transactional against marketing mail.