PROJECTS / DATA NERVE ENGINE
PROBLEM
Building is easy for me. Publishing is where things go to die. The notes piled up. The newsletter didn't.
BUILD
So I wired the brain to the newsletter. Build notes become sourced drafts: brief, draft, my edit, QA, approve. Nothing publishes on its own. If the evidence is missing, the draft pauses instead of filling the gap with generic AI commentary.
WHAT BROKE
The first version drafted in a voice that wasn't mine. Corny, too clean. So I grounded the writing skill in my actual published issues, and after every issue I diff what I shipped against the draft and patch the skill. It learns from what I accept, not from tone instructions.
The honest number: 11 issues sent, and the real baseline was 2 subscribers and 0 clicks. The engine works. Distribution is the part I hadn't built yet.
RECEIPTS
| Issues shipped | 11 |
| Pipeline | brief -> draft -> edit -> QA -> approve |
| Source of drafts | the second brain |
| Auto-published issues | 0 (all human-approved) |
| Total clicks, at baseline | 0 |
All numbers real, pulled from a live Beehiiv API call. Including the zero.