The play is now QUAD.

Everything, in order

Read from the commit history and the working notes. Dates are when the change shipped. Names of partners are real; contract numbers are not on this page.

1. May 24. Day 1. It was not a GLP-1 site.First commit: “Lumora, Medvi-style quiz funnel for prescription sermorelin.” A longevity peptide, sold with his funnel shape: landing, quiz, verdict, price. Four portraits flanking the hero because his did. The pattern came before the product.
2. May 26 to June 8. Plumbing.Lead capture and transactional email, GA4 with per-question quiz events, Meta pixel and server-side events wired and deduplicated, a fake “reviewing your answers” interstitial removed within two days of being added. Hero copy rewritten for ad message match before any ad existed.
3. June 20. The blog starts publishing itself.Three SEO articles, then a daily auto-published post from a generator. It would later cost us: twenty of those posts were deleted in September because their cost-comparison and same-ingredient language drew certification questions.
4. June 23. Pivot one: sermorelin to compounded tirzepatide.The demand was in GLP-1 and the partner stocked it. The name became StaveMD. Every image was regenerated to match the new buyer.
5. June 24 and 25. 146 commits in two days.An agent loop tore down the MedVi ad library into 243 ads, ranked ten image archetypes, and produced twenty StaveMD ad variants across the three winning formats. The homepage was rebuilt on his spine: vial band, stat callouts, benefits card. This is the read that became page three of the playbook.
6. June 27. First real price.$149 first month, $299 ongoing. It would change five more times.
7. July. Trust, and what you are allowed to say.Pharmacy and clinician panel, footer commitments on AI imagery and pricing, a paid-creator video wall with a plain-language FTC line, FAQ rewritten against competitor FAQs, “US-licensed” softened to “licensed”, a coaching claim dropped because it was not yet in a contract. Community language mined from forums and fed back into copy and the blog generator.
8. July 21. The first Meta ad account is suspended.Before a single ad ran. Appeal filed. This is the day-one-account-day-one-ads failure page two warns about, and we walked into it.
9. August 9 to 15. We build our own checkout.Plan picker on the result page, objection-handling reviews at the payment step, “charge at payment, refund in full if a provider declines.” Three weeks later all of it is deleted.
10. August 17 to 21. New ad account, new pixel, and the partner goes live.Replacement Meta account, pixel moved into code. Every call to action now deep-links into the clinical partner’s branded intake. OpenLoop supplies the clinicians in fifty states, the prescriptions, the pharmacies and the compliant intake; we set retail above their floor.
11. August 23 to 28. Quiz rework and the launch kit.Weight slider, email step, then a handoff-first rebuild because the funnel was dying inside the quiz. A second ad-library teardown, a swipe file of GLP-1 video patterns, a produced UGC ad, and a twenty-minute runbook for getting an account reinstated.
12. August 29. LegitScript filed, expedited.Healthcare Merchant Certification. Answer sheet built from spreadsheets the partner had pre-filled. Screening passed August 31. Analyst review September 1.
13. September 1. Warm-up begins, and a side door opens.Meta page-likes campaign at $10 a day, no health ads. The same day, a vending lander for a card room in San Jose with Vietnamese, Chinese and Spanish variants. Three days later, a data decision that mattered: clinical outcomes never go to Meta.
14. September 1 to 9. Three rounds of certification questions.Round one: compounding and the FDA’s February letter, social accounts, privacy, payment timing. Round two: same-active-ingredient and cost-savings language; twelve blog posts removed. Round three: the same language found across the blog; eight more posts deleted, twenty-five paragraphs rewritten, and a rule written down: state our own price plainly, never compare to a brand, never say the molecule is the same.
15. September 5 to 7. Design up, testimonials off.The site moved to the same type and design system as this one. A concern-to-commitment ledger replaced the review wall, every line of it a contracted partner behaviour. Then a kill switch: all testimonials and creator video hidden site-wide while certification is under review.
16. September 8. The bank declines Meta’s $6.Flagged as fraud. The ad account goes unsettled and every write fails until the balance is paid. Nothing was running, so nothing was lost except a day.
17. September 9. Pivot two: retire our checkout.Quiz cut to three questions, verdict, straight into the partner’s intake with the plan in the URL. The on-site payment gate, the result page and the checkout were deleted. Prices reset to the September market: semaglutide from $139 a month on a prepaid year, tirzepatide from $199, a $30 first-month code. The partner’s intake was restyled in our palette and set live the same day.
18. September 9. The named reviewer is removed, and the site speaks three languages.No medical bylines, organisation as author. Homepage rebuilt lean around a price table. Vietnamese, Chinese and Spanish homepages. The card-room lander re-aimed from weight loss to longevity, where the margin is.
19. September 10. Stripe asks the pharmacy question.Both Stripe accounts flagged under the restricted-business policy; one had its industry set to pharmaceuticals and both described themselves as a GLP-1 telehealth platform from the sermorelin days. Questionnaires submitted, descriptions corrected. The Ad Library re-read the same day showed MedVi’s own GLP-1 machine nearly dark, which became the new page three.

Seven things we would tell ourselves on May 24

• File the two slow forms in week one. We built for fourteen weeks before filing LegitScript. Everything since has been waiting on it.
• Warm the ad account before you touch it. The July suspension cost a month and a replacement account.
• Do not build a checkout. Three weeks of ours was deleted the day the partner’s intake became the funnel.
• Every generated blog post is a claim you will have to defend. Twenty came down during certification.
• Copy the shape, not the words. His hero shape converts. His “compounded” and “same molecule” drew letters, and drew our certification questions.
• Set the industry field honestly on day one. Two payment accounts were paused over a description written for a product we no longer sell.
• Price moves five times before you have a customer. Write the price table as data, not prose, so the sixth move is one edit.

One hundred and ten days, two employees and an agent, no revenue yet, and the whole machine assembled and waiting on one certificate. That is the honest state of the playbook when you run it yourself.