ARKONE
Twelve upright marks on a dark ground, one for each stage of the engine, with mark 7 lit in amber and the book title beneath.

How do you actually send without getting banned, so that reading the replies happens by itself?

September 8, 2026 · 6 min read

S
Sobin George Thomas

Bursts are what the platforms watch for. Email goes out under a per-mailbox budget that doubles weekly and halves after a bounce day; LinkedIn stays by hand.

A job in its own right. Domains, inboxes, SPF, DKIM, DMARC, warm-up, and a dashboard somebody watches all day. At a thirty-person consultancy that somebody is the founder, on a Sunday, between two other jobs.


By hand today

“Set up sending domains and inboxes”; “Configure SPF, DKIM, and DMARC”; “Run inbox warm-up systems”.

A job advert for a Cold Email Deliverability Expert, on OnlineJobs.ph, read 7 September 2026

That is the advert for the person an agency hires to do this stage. The whole job is domains, mailboxes, authentication, warm-up and watching numbers.

A company without that hire does the same work in bursts: forty emails on a Tuesday evening, none for a fortnight, then eighty on a Monday. On LinkedIn it is the same shape. Twenty connection requests in one sitting.

Bursts are what the platforms watch for. One LinkedIn user on Hacker News: “they told me that I am using an automation tool. I said that I am not… This time, no back and forth. Permanent ban with no explanation.” (tarek_computer)

The account was the asset. Nobody had written down a cap, so nobody knew it had been crossed.


What it costs you each week

0.10%. The spam complaint rate Google tells senders to stay under. Past it, the domain is what gets blocked, not the message. One complaint in a thousand sends, and you are at the line.

Source: Google, Email sender guidelines, read 7 September 2026

The rate is the number a mailbox provider reads, however good the message. A domain that crosses the line goes to spam for everyone who uses it.

Most companies under fifty people have never opened Google Postmaster Tools. The one number the platform judges you on is the one number nobody on the team looks at.


The build

Claude Code: a budget, not a blaster. Per mailbox, per day: a cap that starts low, doubles weekly, and halves after a bounce day. Email goes out from your own mailboxes, inside the published limits. LinkedIn is offered to you one name at a time. You click.

Start with the rule that shapes everything else. LinkedIn forbids automation. Its page on invitation restrictions says “If you send an excessive number of invitations and we suspect the use of an automation tool, we may suspend or restrict your account”. LinkedIn names no weekly number; “100 a week” is a vendor claim.

So in this build only email is sent, from your own mailboxes, within the published limits. LinkedIn is done by hand, and the tool only keeps count.

The email limits, quoted from the primary pages. Google: senders of “more than 5,000 messages per day to Gmail accounts” must meet its bulk rules, and every sender must “Keep spam rates reported in Postmaster Tools below 0.10% and avoid ever reaching a spam rate of 0.30% or higher”. Google Workspace caps each account at “Maximum messages per day: 2,000”. Microsoft: “domains sending more than 5,000 emails per day” must have SPF, DKIM and DMARC “at minimum p=none”, and from 5 May 2025 Outlook will “reject messages that don’t pass the required authentication requirements” with “550; 5.7.515 Access denied”. Nobody in this book sends five thousand a day. Authenticate the domain anyway; the reject code above is what happens without it.

Open Claude Code in the folder that holds the approved sequences from chapter six and a file listing your mailboxes. Type:

Read accounts.json and leads.csv. Build a daily send budget per mailbox.
Week one: ten emails a day per mailbox. Double each week up to the ceiling
I give you. If a mailbox has a bounce today, halve its cap tomorrow. Weekends
send nothing. Stop a mailbox entirely on a 429, a bounce spike or a provider
warning, and stay stopped until I reset it. For LinkedIn, print five names a
day with their drafted note; never send them. Show me today's list and wait.

What comes back is a table: one row per mailbox, today’s cap, today’s names, and the step each is on. The budget is saved as budget.json, and every send is appended to sends.log with the mailbox, the day, the lead and the step. When you type yes, the emails go out from your own mailboxes. The LinkedIn names sit in the terminal with their notes; you open the profile, paste, and the tool counts it.

The pattern is ArkOne’s own operator ramp, at plans/linkedin-operator-extraction/notes_plan.md line 36: a floor of five invitations and ten messages a day, doubling weekly to forty and eighty, a weekly invitation ceiling of one hundred, and a breaker that trips on a 429 or a checkpoint page. ArkOne’s CRM carries the other half at docs/crm-module.md line 104: nothing sends anything; a send is logged after a human sends it. The reader builds their own version of both. There is no demo to download.


What it feeds

Reading every reply. Every send carries its mailbox, day, lead and step, so a reply can be matched to the message it answers. Proof: sends per mailbox against its cap, complaints under the line, days since a warning.

The next stage reads replies, and it can only sort a reply if it knows which send it belongs to. That is what sends.log is for. Three numbers prove this stage worked: sends per mailbox per day against its cap, the complaint rate under the line, and days since the last restriction.


What stays with a person

The yes before the first send, and the stop. The tool prints the list. A person approves it. When the breaker trips, a person decides whether to start again.

Approving the sequence before the first send is the last moment the words can change before they carry your domain’s name.

The second decision is the stop. When a mailbox bounces or LinkedIn shows a warning, the tool halts that account and waits. Whether to resume is a judgement about your reputation with a platform that does not explain itself. A tool that resumed on its own is how the ban with no explanation happens.


The tools you already pay for

Nothing new to buy.

Mailbox: your own Google Workspace or Microsoft 365 seats, under their published caps.

Postmaster Tools: free from Google; the complaint rate, read once a week.

CRM: the send logged after it happened, by mailbox, day and step.

LinkedIn: your own account, by hand, at your own pace, with the tool counting.

No warm-up subscription and no sending platform.


Send from your own accounts, under the published limits. Then the replies are worth reading.

Ask: How many emails did each of your mailboxes send yesterday, and what was its cap?

This is chapter seven of twelve, after writing one message per person and before reading and sorting every reply. The book, Building a real outbound engine with Claude Code and Codex, end to end, runs the whole engine in order, from writing down the market to reading the weekly number, one chapter per stage. A free starter edition for firms under fifty people is coming through the 100k programme.

هل أنتم مستعدون لبدء الحوار؟

اكتشفوا كيف تبني ArkOne هياكل الحوكمة وإدارة البرامج لتحقيق عوائد قابلة للقياس من الذكاء الاصطناعي.

احجز مكالمة استكشافية