The send-once rule

Volume is not what gets outreach accounts and sending domains banned. Repetition is. A system that writes to a thousand people once is a system. A system that writes to one person three times because the campaign restarted is spam, and the platforms are right to treat it that way.

Where repetition comes from

Almost never from intent. It comes from state that lives in the wrong place:

• A suppression list scoped to a campaign, so a new campaign starts from zero.
• A “sent” flag in memory, lost when the process restarts.
• Deduplication on a slug or a subject line, so the same person with a slightly different record is contacted again.
• A cooldown measured from the last run instead of from the last contact.

Each of these is a correct-looking design that produces a second message to someone who already had one.

The rule we build to

1. Key suppression on the person, with a stable identifier, not on the campaign or the message. A handle, an email address, a domain for partner outreach.
2. Make the store permanent and shared. It lives in a durable store every runner reads before acting, and it outlives restarts, redeploys and campaign changes.
3. One opener per person, for life. Follow-ups are a separate class with their own counter and their own cooldown, measured from the last contact.
4. Pace in code. A daily ceiling in the runner, well inside platform norms, with human-shaped timing. The prompt cannot raise it.
5. Check deliverability before sending, and stop when placement drops, because a burnt domain is permanent and a missed day is not.

The same rule, elsewhere

The social operator we run keeps one opener per person for life and a permanent close state per conversation. The marketplace publisher holds slug locks so a listing cannot appear twice under two names. The search-page engine deduplicates on the query it is answering before it writes a page. Different surfaces, one rule: remember what you did, in a place that will still exist tomorrow, and check it before you do it again.