RMI Vetting Guide — funnel

Live state at every node, and an honest gap wherever there is nothing to read. Ad account 1048601279933853 · pixel 526434314775731
live — a real numbernot instrumented — the event does not existno traffic — it works, nothing has flowednot configured — no credential on this deploymodelled — derived, not observed

What you are launching into

5verifiedobserved working
4assumedbelieved working, never witnessed
3known brokendemonstrated not to work
Of the 12 nodes that exist, 5 have been observed working. Most of the rest are assumed — believed to work because they were built that way, not because anyone has watched a real person traverse them. Tonight’s campaign is what converts assumptions into observations, and it is the reason this row matters more than any metric on the page.
A further 4 nodes are designed but not built, and are excluded from this count deliberately. A node nobody has built is not failing the readiness test — it was never entered into it. Counting it alongside a node that exists and is broken would make this number worse at the only job it has.

Today

Cost per landing page view

not configured
META_ACCESS_TOKEN is not set on this deployment.
Comparison held ready: $1.79–$5.90 per landing page view across three prior Vetting Guide campaigns, $2.17 blended. $11,747.63 spend ÷ 5,404 landing page views across three prior Vetting Guide campaigns.
Source: Meta Ads · act_1048601279933853 · Last updated: just now · Sat, 21 Nov 2026 17:09:01 UTC

Cost per lead

awaiting lead capture
Not a number and not zero. No live cost per lead until the SalesKick → GoHighLevel push path is restored. A denominator missing its recent arrivals overstates CPL, and an overstated CPL is the evidence someone kills a working channel on.
Closed period, to 2026-10-12. The earlier of the last Vetting Guide campaign spend date and 2026-10-12, the date the push path broke. All three Vetting Guide campaigns with spend ran July–September, so they sit safely inside it. Spend across that window is $11,747.63. No figure is shown because the denominator is missing, not because it is zero: Not yet supplied. The dateAdded-bound count for this window has not been run, so no cost per lead is shown — the spend term is known and the denominator is not.
Leads must be dated by contact.dateAdded, NEVER by opportunity createdAt. createdAt is the 2026-05-19 migration timestamp: one verified case has a contact added 2024-09-04 against an opportunity created 2026-05-19, twenty months apart. Using createdAt would pile two years of leads onto a single May date and attribute them to whatever was spending then.
wrong source The Vetting Guide pipeline — Its newest opportunity is 2026-05-31 and nothing has entered since, so none of the Jul–Sep campaign leads are in it at all — an absence, not a small count. More fundamentally it is a migrated snapshot of mixed-vintage leads rather than a time series of opt-ins: contact.dateAdded across it spans 2024-09 to 2026-05. Dividing campaign spend by how many records a May migration happened to sweep up is not a cost per lead under any window you choose.
right source Contacts tagged "stem cell vetting guide" OR "stem-cell-guide", bucketed by contact.dateAdded — A genuine time series. Read from the contacts API rather than pipeline membership, server-side in this app’s own route, with cursor paging — a capped read that silently returns page one is the exact failure this dashboard exists to prevent. Two tags, because the same concept exists under two strings and querying one undercounts by however many carry only the other. The overlap between them is reported rather than assumed: whether they are duplicates or distinct cohorts is a fact to establish. FIRST touch, and on a CONTACT that is attributionSource.campaignId — a single object, no array and no isFirst flag. The attributions[] array carrying isFirst and utmCampaignId is the OPPORTUNITY shape and is absent on contacts; reading it here would have silently returned nothing. Last touch would bill May leads to September campaigns: one contact captured 2026-05-31 carries dateUpdated 2026-10-28, and September webinar campaign ids appear as attribution on May-captured contacts. Built and waiting on GHL_API_KEY. It computes the moment the credential exists.
Closed period. Same treatment as the 32.6% blended win rate: it describes a book that is shut, and it is never a forecast.

Goals

no target set
Outstanding longest of anything on the board. Every goal-relative panel reads "no target set" until these exist.

The funnel

Paid traffic — MetaMetaautoassumed
Ad account 1048601279933853. Feed-only, broad, age 40–65. Six policy-approved existing posts on page 103945574412204. NOT RUNNING as of 2026-11-21: the three active campaigns in this account are Patient Compilation (engagement, created 18 Nov), RMI IG and a July 2025 webinar campaign. None of them is the Vetting Guide, and all recent spend — $20.68 on 19 Nov, $30.23 on 20 Nov, $9.30 on 21 Nov — is Patient Compilation.
not configured
META_ACCESS_TOKEN is not set on this deployment.
Last updated: just now · Sat, 21 Nov 2026 17:09:01 UTC
Verification: Corrected 2026-11-21: the Vetting Guide campaign has NOT launched. Verified against Meta directly — the only three active campaigns in the account are Patient Compilation, RMI IG and a July 2025 webinar, and all recent spend is Patient Compilation. Meta delivery reporting itself is the one solid column in this funnel; there is simply nothing of ours delivering through it.
↓
Step 1 — Landing pageGoHighLevelautoverified
GHL funnel 25Hg6ugZeJwyjCM5simw. Pixel 526434314775731 fires PageView at funnel level. Also served at /vetting-guide — same page.
https://go.rmihealth.com/regen-medicine-guide
not configured
META_ACCESS_TOKEN is not set on this deployment.
Last updated: just now · Sat, 21 Nov 2026 17:09:01 UTC
Verification: Page returns 200 and renders pixel 526434314775731 (PageView). Confirmed by direct fetch.
Open · 2
high"Including ours." must appear above the fold. The best ad in the account (10.12% CTR) ends on that promise and the page never repeats it.
mediumCosta Rica appears nowhere on this page except the legal disclaimer — the loudest unanswered objection.
↓
Decision — Submitted?GoHighLevelautoknown-broken
On yes, four writes must happen at once: contact into the system of record, tag guide-requested, stamp guide_requested_date, and an append-only attribution write of _fbp/_fbc.
not configured
GHL_API_KEY is not set on this deployment.
Last updated: just now · Sat, 21 Nov 2026 17:09:01 UTC
Verification: Form posts with mode:'no-cors' — success is undetectable by design. No Lead event fires.
Open · 3
criticalThe form posts to a Zoho Flow webhook with mode:'no-cors'. The response is opaque, so the page cannot detect success or failure even in principle. A dead webhook looks identical to a working one.
criticalBrowser Lead event does not fire. No event_id, so no CAPI twin to deduplicate against.
high_fbp / _fbc are not captured. Without the carry-forward, every later server-side booking lands unattributed.
↓
Guide deliveryHumanhumanknown-broken
Step 2 promises the guide within 2 minutes. No automated delivery exists. Every copy that went out was source:"app" — a person sending by hand from the GHL UI, days later, only to manually verified addresses. Two automated SMS fire within seconds and promise it regardless.
not configured
GHL_API_KEY is not set on this deployment.
Last updated: just now · Sat, 21 Nov 2026 17:09:01 UTC
Verification: No delivery automation exists. Verified from message history — every send was manual, source:"app".
Open · 2
criticalNo automated guide delivery. The page makes a promise in writing that nothing keeps. Of four May contacts examined in full, three never received it and one got it 10 days late.
highNo canonical guide URL. The PDF sits in the media library three times, byte-identical, under three ids. Needs a redirect RMI controls before any link goes into an email that sends 75 days later.
↓
Rung 1 — Clinic scorecardGoHighLevelautoladdernot-built
The reader scores a clinic against the 31 questions. The act of scoring is what reveals which lane they belong in — and it is the first thing in the funnel that asks them to do something rather than read something.
not configured
GHL_API_KEY is not set on this deployment.
Last updated: just now · Sat, 21 Nov 2026 17:09:01 UTC
Verification: Does not exist yet.
↓
Decision — Whose clinic?GoHighLevelautoladdernot-built
The structurally important branch. Scoring a COMPETITOR routes express — straight to the Candidacy Review, skipping the answers page and the webinar, because someone holding a rival quote is deciding this week and a webinar on Thursday is a delay. Scoring RMI, or nobody, takes the full ladder.
not configured
GHL_API_KEY is not set on this deployment.
Last updated: just now · Sat, 21 Nov 2026 17:09:01 UTC
Verification: Does not exist yet.
↓
Rung 2 — Public answersGoHighLevelautoladdernot-built
RMI answers all 31 of its own questions in public, including the ones where the honest answer is unflattering. The campaign’s whole premise, made checkable.
not configured
GHL_API_KEY is not set on this deployment.
Last updated: just now · Sat, 21 Nov 2026 17:09:01 UTC
Verification: Does not exist yet — though the content does. All 31 answers are already written.
Open · 1
mediumBuildable today — the content already exists. ../AI SLOP/Vetting Guide/build/content.json (parent of the project root, not inside it) holds all 31 questions, each with its own rmi answer field. Verified.
↓
Decision — Still researching?GoHighLevelautoladdernot-built
Elapsed time on the slow lane, not a failed booking. The distinction is load-bearing: a webinar invitation triggered by not-booking reads as a win-back, which is the wrong posture for an audience buying carefulness.
not configured
GHL_API_KEY is not set on this deployment.
Last updated: just now · Sat, 21 Nov 2026 17:09:01 UTC
Verification: Does not exist yet.
↓
Step 2 — Candidacy ReviewGoHighLevelautoverified
This page IS the Step 1 thank-you URL. Thank-you and candidacy review are one page and cannot be separated by path.
https://go.rmihealth.com/regen-medicine-guide-thank-you-1
not configured
META_ACCESS_TOKEN is not set on this deployment.
Last updated: just now · Sat, 21 Nov 2026 17:09:01 UTC
Verification: Page returns 200 as "Your Candidacy Review". Confirmed by direct fetch.
↓
ApplicationSalesKickautoverifieddormant
SalesKick form-sRQK2TX7H92JEUX3QPdsX3, campaign cmp-bJGcKnCLLcCum6TJygRpxR. The AI agent chases partial submissions to completion, then pushes to book and to show up. CAPI-only — a browser pixel cannot see into a third-party iframe. This surface is not broken: it has conversion history. Meta has simply never seen it.
not instrumented
No live reading here: SalesKick events are not mapped to the Conversions API and nothing polls this surface on a schedule from the dashboard. That is a reporting gap, not an absence of conversions — see the observed figures below, read directly from the SalesKick API.
Observed 2026-09-04 → 2026-11-20 · read 2026-11-20 · not live
18completed applications
11bookings
8marked showed
4bookings attributed to fb_ad
Read-only query against the live SalesKick API: the 50 most recent submissions, newest first.
floor One page of 50, not a census. Paging further returned 403 — rate limiting, since the identical call had just succeeded. Every figure here is a FLOOR.
dormant 48 of the 50 are September, one October, one November. The surface works; nothing has been driving traffic into it since the September ads stopped.
arrival Roughly two-thirds of this already arrived in GoHighLevel. A full GHL census across both tag families — 108 contacts — finds 32 submissions, 12 appcompleted and 9 bookingscheduled inside the same window. The residual that never landed is about six applications and two bookings, all of them after 24 September, which is when the push path was already failing.
Four campaigns share this workspace. Any unfiltered count is four funnels added together.
oursVetting Guide — 28 submissions, 4 bookings
cmp-bJGcKnCLLcCum6TJygRpxR
otherClosers.io Campaign — 12 submissions, 7 bookings
cmp-nU811FuK2VRXqBrkGMTzuK
otherGPM Webinar - V2 — 9 submissions, no bookings
otherGPM Post-Webinar Offer — 1 submissions, no bookings
Verification: Verified with data, 2026-11-20: 18 completed applications over 2026-09-04 → 2026-11-20, read live from the SalesKick API, UTM on every record. A floor, not a census — paging past 50 returned 403. Dormant since September. The org-saleskickDEMOsaleskick concern is withdrawn: the workspace is "RMI Health" with real calendars, and that slug was an asset-path artifact.
Open · 4
criticalSubmitApplication is not mapped. SalesKick has native Meta CAPI — needs the platform-event dropdown options before mapping.
criticalEvery SalesKick read must filter on campaign cmp-bJGcKnCLLcCum6TJygRpxR. Unfiltered, the workspace returns four campaigns at once and reads as roughly four times the truth — the same class of error as the Calendly undercount, pointing the other way.
highPaging past the first 50 submissions returns 403. Until that is resolved every SalesKick figure on this page is a floor of unknown depth, and must be labelled as one.
criticalThe SalesKick → GHL path is DOWN NOW, not broken historically. GHL holds nothing from SalesKick after a 12 October test record, while SalesKick holds a genuine November submission. That is an outage in progress, running unnoticed for roughly six weeks. The path itself was real: a census of 89 found 88 contacts with populated "SK -" fields and 37 carrying SalesKick bkg- booking ids, across 27 custom fields, between 9 August and 22 September. The published API exposes no webhook surface, so whatever drove those writes is not a documented feature — which makes repair an open question rather than a configuration task. The 30-minute poll is what runs today, and it writes to Meta, not to GHL.
↓
Decision — Booked?SalesKickautoverifieddormant
Must write review-booked back to the contact. One of only two signals the entire email journey needs. Schedule fires on booking, NOT attendance.
not instrumented
No live reading here: SalesKick events are not mapped to the Conversions API and nothing polls this surface on a schedule from the dashboard. That is a reporting gap, not an absence of conversions — see the observed figures below, read directly from the SalesKick API.
Observed 2026-09-04 → 2026-11-20 · read 2026-11-20 · not live
11bookings, workspace-wide
4on the Vetting Guide campaign
4attributed to fb_ad
Read-only SalesKick query, 50 most recent submissions.
floor One page of 50. A floor, not a census.
dormant Concentrated in September. Bookings stopped when the ads did.
5 booking surfaces · One live surface nobody designed in, and four designated ones carrying almost nothing.
Bookings are happening — just not where the funnel designed for. Over 26 days, Calendly carried ~455 pixel events and RMI’s own domains carried nine. A booking count built on RMI’s own pixel-bearing pages understates reality by roughly two orders of magnitude. Corrected 2026-11-20: the SalesKick embed is not empty either — it carries 11 observed bookings, 4 on this campaign. It is dormant, not dead.
live-parallelCalendly · ~455 events / 26 days — Where the bookings actually happen. Not in the funnel design, invisible to GHL, Zoho and the SalesKick path.
appt.rmihealth.com/payton · /bailey
live-dormantSalesKick embed — 11 bookings observed across the workspace to 2026-11-20, 4 of them on the Vetting Guide campaign. Works, and quiet since September.
form-sRQK2TX7H92JEUX3QPdsX3
dormantGHL calendar — Correctly configured, round-robins Payton and Bailey. Zero events 1 Sep – 21 Nov.
LsK7Z3HXybAc6RqDf5tw
orphaned/schedule-now-thank-you — Live confirmation page wired to the GHL calendar, reached by almost nobody.
liveAloware by phone — Closer books verbally. Needs a defined write-back or it becomes a sixth invisible surface.
Verification: Reclassified 2026-11-20. Previously known-broken on the strength of an empty GHL calendar; that calendar is empty because bookings land in Calendly and SalesKick, not because none happen. 11 bookings observed in SalesKick, 4 on the Vetting Guide campaign, 4 attributed to fb_ad. A floor. The GHL calendar being dark is still a real gap — it is a write-back problem, not a conversion problem.
Open · 3
criticalWithout review-booked written back, people who have booked keep receiving booking emails — the most damaging failure in the journey.
highStem Cell Vetting Funnel calendar shows zero events between 1 Sep and 21 Nov. Bookings are not reaching GoHighLevel — they reach Calendly and SalesKick instead. GHL seeing nothing is design drift, not an absence of bookings.
highCalendly already fires on pixel 526 (~455 events / 26 days). If a Calendly booking fires Lead and SalesKick later fires Schedule for the same consultation, that booking is counted twice across two systems. Same family as the URL substring trap: a believable inflated number nobody questions. This stopped being a future risk on 2026-11-20: the gateway now polls SalesKick and sends Schedule, while Calendly keeps firing on the same pixel. Nothing deduplicates the two.
↓
Decision — Showed?SalesKickhumanverifieddormant
The SalesKick AI agent drives show-up rate. SalesKick does carry a per-submission showed flag, so a show signal exists in a reachable system — it has simply never been read on a schedule or sent to Meta. The 0.60 show rate inside the modelled booked-call value remains an assumption, not a measurement.
not instrumented
No live reading here: SalesKick events are not mapped to the Conversions API and nothing polls this surface on a schedule from the dashboard. That is a reporting gap, not an absence of conversions — see the observed figures below, read directly from the SalesKick API.
Observed 2026-09-04 → 2026-11-20 · read 2026-11-20 · not live
8marked showed
11of bookings
Read-only SalesKick query, 50 most recent submissions.
floor One page of 50. A floor, not a census.
dormant September-weighted, like everything else on this surface.
vs modelled 8 of 11 is 73%, against a modelled show rate of 0.60. At n = 11 that is evidence the assumption sits in the right region, not a recalibration. The modelled figure stays 0.60 until a ConsultationShowed event produces a measured one.
Verification: Corrected 2026-11-20. A show signal does exist and is reachable: SalesKick marked 8 of the sampled submissions showed. It has never been read on a schedule or sent to Meta, so the 0.60 modelled show rate stays an assumption until a measured one replaces it.
Open · 1
mediumA ConsultationShowed custom event from SalesKick would give both the optimisation signal and the real show rate. The old GHL Appointment Status field sits on a dead pipeline.
↓
Candidacy Review — the human nodeHumanhumanassumed
Samantha Cunningham (Patient Specialist), then Patrick Cunningham. Specialist first, physician second — the email copy calls this person a specialist and never a physician, and that distinction is load-bearing. Aloware carries the telephony and 1:1 sales SMS.
not configured
GHL_API_KEY is not set on this deployment.
Last updated: just now · Sat, 21 Nov 2026 17:09:01 UTC
Verification: Calendar round-robins Payton and Bailey, not Samantha and Patrick. Unresolved.
Open · 1
mediumTwo independent booking systems agree on Payton and Bailey — the GHL calendar round-robins them, and the live Calendly links are /payton and /bailey. The outlier is the campaign brief, which names Samantha and Patrick. Most likely two roles, booker and closer, but unconfirmed.
↓
Decision — Candidate?Humanhumanassumed
On no: tag not-a-candidate, permanent marketing suppression. They keep the guide. This enforces the campaign’s central promise — a declined patient receiving a booking email breaks the premise of the whole campaign.
not instrumented
Zoho is the source of truth for close and payment, but the field names have not been supplied, so nothing can be bound to yet.
Verification: Suppression on decline is specified but not observed.
↓
Handed to Zoho — close and paymentZohohumanassumed
The far side of the Zoho boundary. Proposal Sent → Dates Accepted → Process Payment → Payment Received → Treatment Completed, with status To collect / Confirmed / Refund. The GHL pipeline stops here and models none of it.
not instrumented
Zoho is the source of truth for close and payment, but the field names have not been supplied, so nothing can be bound to yet.
Verification: Zoho holds the close. Field names not yet supplied, so nothing is readable from here.
Open · 2
highBlocking: need the Zoho field names for Payment Received Date, Payment status and Treatment Completed Date, plus the status picklist values. Purchase fires from Zoho and has nothing to bind to.
high25% of completed treatments are comped. Comped treatments must not fire Purchase, or Meta learns comped patients are worth $26k.
↓
CAPI gateway — every event’s route to MetaCAPI gatewayautoknown-broken
capi.rmihealth.co. RMI-owned: its own Vercel project, its own repository (PJCSTRONG/rmi-capi-gateway), its own domain. On 2026-11-20 it delivered its first real server-side event to pixel 526434314775731 and Meta recorded it. Every event this funnel structurally could not send — anything inside the SalesKick iframe, anything days after the click, anything landing when no browser is open — now has a route. Scoped to the Vetting Guide campaign alone.
not instrumented
The gateway exposes no health endpoint this dashboard can read. Its first event is confirmed; its steady state is not observable from here. That is the single most important thing to build next.
Measured at source · 2026-11-21T04:34Z
Vercel runtime logs and account billing for project rmi-capi-gateway, team pjcprojects, read-only.
criticalAccount plan is hobby. Confirmed at source, not inferred from behaviour.
criticalNo invocation at :00 or :30. Three slots observed empty — 03:30, 04:00 and 04:30 — 04:00 and 04:30 by two sessions reading independently. The only polls in a six-hour window were 03:42 and 04:04, and 04:04 is CONFIRMED to have been a manual trigger, not inferred. No deployment was in flight at 04:30, so a build window is not eating it.
criticalThere is no catch-up by design, and that is what turns this from cosmetic into data loss. The poller gates every event on its own timestamp being under 24 hours, deliberately, so that repeat sends stay inside Meta’s 48-hour dedup window. A cron that does not run therefore converts a correct design into a quiet deletion machine: anything ageing past 24 hours between runs is missed permanently, and nothing anywhere records that it happened.
highThe gateway was rejecting events as recently as 03:42: Meta error 190, "Invalid OAuth access token — Cannot parse access token", HTTP 400, retryable false. The 04:04 poll was accepted, so the token was replaced between the two. A rejected event does log at error level, which means the health signal this node needs already has a source to read.
mediumThe campaign filter works. A poll scanning 9 submissions reported inScope 1 and skippedOtherCampaign 8, against cmp-bJGcKnCLLcCum6TJygRpxR.
mediumConfirmed poll parameters: lookbackDays 60, freshnessHours 24.
caveat First read was three missed slots on a project 1.6 hours old. Meta’s side has since closed the question — see the confirmation below.
Confirmed at the destination · 2026-11-21
Checked against the pixel, not recited from the Vercel logs.
Pixel 526434314775731 has recorded nothing since 2026-11-20T20:04:34, and that single event is the only thing the gateway has ever delivered. Both last_fired_time and server_last_fired_time are frozen there. No overnight invocations at all. That timestamp is Meta reporting in Pacific and is the same event as the 04:04:29 UTC poll in the Vercel log — one event, seen from two sides, not two events.
The destination confirms what the source suggested. The cron is dead, not slow, and the gateway has fired once in its life.
3 intake mechanisms · a push stops silently, a poll goes stale silently. Two mechanisms, two watch conditions.
pushLead · Step 1, via Zoho Flow — A push stops silently. Nothing arrives and nothing complains.
pollSubmitApplication · Schedule · ConsultationShowed · SalesKick — A poll of GET /submissions returns stale or truncated pages silently. The 403 seen while paging is exactly that failure in miniature. Declared every 30 minutes; measured below, it is not running at that rate. A push path into GHL did exist 9 Aug – 22 Sep 2026 and broke unnoticed — so "SalesKick can only be polled" is about today, not about always.
pushPurchase · Zoho, on Payment Received — Not wired yet. Blocked on the Zoho field names.
Verification: Demoted from verified on 2026-11-21. The code path is proven: a real server-side event reached pixel 526434314775731 on 2026-11-20 and Meta recorded it. The node is not, because its job is continuous delivery and the scheduled poll does not run — three slots observed empty, two of them confirmed by a second session reading independently. A gateway that delivers only when a human triggers it is not delivering. This should return to verified the moment a scheduled run appears in the logs, and not before.
Open · 6
criticalA third system that can be the reason a number is wrong, and nobody else on the team can see it. Needs a visible health signal, an alert when a poll returns 403 or empty, and a record of what it sent. A silent gateway and a working gateway look identical from Meta’s side until the numbers drift.
criticalSend the submission’s own timestamp, not the poll time. A 30-minute poll makes the gateway the only component that knows when an event actually happened. Get this wrong and every event appears to occur on the half hour.
fixedFIXED in the gateway, commit 0bfb408, and verified here by reading the code rather than taking the report. Schedule had been double-firing on the ordinary path: the poll gates it on the submission’s updatedAt, and marking attendance touches the submission, so book day 0 / show day 7 fired twice — the second outside Meta’s 48-hour dedup window, which Meta records as a new conversion. Book-this-week-show-next-week is the normal case, so it was close to a flat 2x on the most valuable event in the funnel. The guard is now "fire only while the booking has no attendance set". That trades the inflation for a known under-fire: a booking whose attendance is already recorded the first time the poller sees it never fires Schedule at all. Under-firing is the right direction to err, but it is a trade, not a free fix.
criticalvercel.json declares "0,30 * * * *". The Vercel account is on the Hobby plan, and no invocation has landed at :00 or :30 — measured, see below. With a 24-hour freshness window and a poll that does not run every 30 minutes, any submission ageing past 24 hours between runs is missed permanently and silently. This is the mirror of the Schedule defect: that one inflates, this one deletes.
criticalThis is most likely a plan decision, not an engineering problem. The account is confirmed hobby, and a hobby plan is the leading explanation for a sub-daily cron that never runs — which would make the fix roughly $20 a month rather than a code change. Worth confirming in that order, cheapest first: (1) does the plan run sub-daily crons at all; (2) did the crons register against the deployment currently serving production rather than an earlier one; (3) was vercel.json present in the tree that actually built. Only after those three is it worth reading the handler. Project rmi-capi-gateway, prj_tFnGwzWKGHWyx0YCxygBnpmAQkMv, team pjcprojects.
highLead pushed from Step 1 needs a shared event_id with any browser twin, or the two deduplicate against nothing. Step 1 fires no browser Lead today, so this is latent rather than live.

Server-side coverage

No longer 0% — and deliberately not a number yet

no ratefirst event 2026-11-20 · a figure goes here no earlier than 2026-11-27
No longer zero, and not yet a rate. Server-side coverage was 0% — measured, not assumed — for the entire window up to 2026-11-20, when the gateway delivered its first real event and Meta recorded it. One event is not a coverage rate. No percentage will appear here until a week of gateway traffic exists.

Meta custom conversions — last fired

Four of six live conversions on this account had never fired, and nobody noticed for months. This check is one line and would have caught every one of them. It runs on every page load.
not configured

Sync health — GoHighLevel ⇄ Zoho

Contact parity

not configured
Parity cannot be checked: GHL_API_KEY and ZOHO_ACCESS_TOKEN not set. Until this panel reads live, nobody is watching the sync.
Patrick’s hard requirement is that contact data is identical in both systems. An unmonitored sync is one that has already stopped — this funnel has already lost six months to a webhook nobody could tell was failing.
Last checked: just now · Sat, 21 Nov 2026 17:09:01 UTC

GoHighLevel pipeline — lead capture health

StageOpportunities
New Lead619
Contacted0
Scheduled Appt0
Appt Completed0
Proposal Sent0
Closed0
Not Qualified0
This measures that NO AUTOMATION WRITES BACK to this pipeline. It does not measure 619 people being ignored, and an earlier reading of it as "619 unworked leads" was retracted. 87 of the oldest 100 opportunities were created on one day, 2026-05-19, by a migration workflow that swept existing leads in and set the stage once; nothing has maintained it since. Underneath, roughly 4% have since booked or been worked and roughly 11% never asked for the guide — exclusions overlap, so do not add them.
All seven stages queried individually rather than inferred from a sample. Six of seven returning empty against the same query shape is what proves the filter works rather than the number being a passthrough.
A follow-up health panel must be built on contact.dateAdded, not opportunity createdAt. The latter is the migration timestamp, and some of these contacts opted in as early as September 2024.

Metric spine — designed, not yet instrumented

Column spine adopted from Cole Gordon's STA tracking sheet, supplied by Patrick. Columns adopted, numbers rejected: Different funnel, different CPM regime — $264 CPM against the Vetting Guide’s $21.69 in the same window. Transplanting its rates onto a Vetting Guide forecast would import another funnel’s economics wholesale.
Pacing ladder: 2d · 4d · 7d · 14d · 30d · MTD. A narrow audience fatigues inside a week, and a 30-day average hides it.
Delivery
not instrumentedFrequency — Fatigue signal, and the first thing to watch on a narrow audience. Readable from Meta the moment a Vetting Guide campaign delivers.
source when built: Meta
Lead quality
not instrumentedCPNL — cost per NEW lead — Separated from returning leads. A blended cost per lead flatters a list that is recycling.
source when built: Meta × GHL
not instrumentedApproved App % — The quality gate. Volume metrics hide bad-fit traffic entirely — this is where it surfaces.
source when built: SalesKick
not instrumentedCost per Approved App — Cost of a lead worth talking to, rather than cost of a form submission.
source when built: Meta × SalesKick
Mid-funnel
not instrumentedVSL Conversion — Mid-funnel drop-off. This funnel has no equivalent measurement at all, which is a gap rather than a non-applicable column.
source when built: not yet assigned
Booking leakage
not instrumentedTriaged — Where a booking leaks before it becomes a call.
source when built: SalesKick / human
not instrumentedCancelled — Same leak, different cause. Booked and showed are already two facts; this is the third between them.
source when built: SalesKick / calendar
Sales handoff
not instrumentedMQL — The layer between a booking and a sale that nothing currently models.
source when built: Zoho
not instrumentedSets
source when built: Zoho
not instrumentedCP Set
source when built: Meta × Zoho
not instrumentedCPTQO — Cost per truly qualified opportunity.
source when built: Meta × Zoho
Revenue
not instrumentedUF Cash — Upfront cash, held separately from total revenue. On a $30k treatment with payment plans these are different businesses, and it is the same distinction the comped-treatment problem needs.
source when built: Zoho
not instrumentedRevenue — Total contracted, comped treatments excluded at the gate.
source when built: Zoho
not instrumentedROAS — upfront
source when built: Meta × Zoho
not instrumentedROAS — total — Both variants, never one. A single ROAS figure hides whether the cash has arrived.
source when built: Meta × Zoho

Build rule for comparisons — §4.2b

This is enforced in code rather than written down. lib/comparison.js builds every comparison row, and a row missing spend or impressions returns a visible refusal instead of a cost-per-result figure. If any row is withheld the whole comparison is marked unsafe — one side without volumes makes the other side’s number meaningless as a comparison, however well-formed it is alone. Delivery imbalance is computed and flagged at 3× rather than left for the reader to notice.
Required on every row: costPerResult, spend, impressions, window, count. Recommended: frequency — Frequency earns its place: the one apparently-real effect found that day was a saturation artifact — 1.06 against 1.21 frequency, which lifts CPM on its own and accounted for the whole gap.
window Every row carries its window on the face, and two figures from different windows never render in the same comparison. Volumes alone would not have caught this: one pair’s CTR relationship INVERTS between windows — 0.44% against 0.45% at last_90d, effectively tied, and 0.42% against 0.51% at lifetime, 21% apart in the opposite direction. Only pulling a second window exposed it.
count Every rate carries the count it was computed on, not merely spend and impressions. One pair in this dataset reads −3.6%, which looks like a result; it is computed on 82 and 122 landing page views, where the noise floor is around 29%. The percentage hides its own unreliability and the raw count exposes it instantly. Below ~100 events the count renders prominently with a thin-data marker. No confidence intervals on the page — the right move is making the reader see the count, not handing them a statistic they will also over-read.
do not mine This account’s historical creative numbers contain no replicable creative finding and must not be mined for one. Any panel surfacing them carries this warning rather than inviting the inference.

Deliberately not shown

Each of these could be computed today and each would be wrong. A metric that is simply missing looks like an oversight and gets rebuilt by the next person, so they are listed here with the reason.
MetricRenders asWhy not
Pipeline value / revenue forecastnot recordedEvery Vetting Guide opportunity carries monetaryValue 0, 26% of opportunity values are blank generally, and 25% of treatments are comped. Any figure here is wrong in three compounding directions.
Opt-in rate, as a percentagecounts onlyThere is no honest denominator. GHL funnel reporting 404s and Meta landing page views are attributed-only. Opt-in counts sit beside landing-page-view counts; dividing them and calling it a conversion rate would invent a number.
Email open rateclick counts insteadGHL reports 100% open on a newsletter where delivered (1,284) does not equal sent (7,474). The rate is computed against a broken denominator.
Any end-to-end funnel conversion rateneverIt multiplies four systems’ floors, assumptions and broken attribution into a single number that looks authoritative and is not.

SalesKick mirror — a second, independent read

SalesKick funnel state is already mirrored into GoHighLevel as tags. This does not depend on the gateway cron running, which matters because the cron does not run. It is a cross-check against SalesKick, never a replacement — and where the two disagree, the disagreement is itself the sync-health signal.
not configured
GHL_API_KEY is not set on this deployment.
Six tags are ready to read: sk-closers-appcompleted, sk-closers-bookingscheduled, sk-closers-partialapp, sk-websiteform-appcompleted, sk-websiteform-bookingscheduled, sk-websiteform-partialapp.
floor A floor with a known end date — the SalesKick → GoHighLevel push path has been down since 2026-10-12, so nothing after that date is mirrored here.

Modelled — not observed

Modelled value of a booked candidacy review

$1,800.00modelled · 2 of 3 inputs assumed
$19,875 effective revenue per won × 0.60 show × 0.15 cold close
InputValueBasisNote
Effective revenue per won$19,875measured$26,516 mean across 100 most recent won opportunities (1 Apr – 24 Sep 2026) × 0.75, because 25 of 100 completed treatments are comped rather than charged. Raw GHL revenue overstates by about 33%.
Booking → show0.60ASSUMPTIONNot measured, and deliberately unchanged. SalesKick shows 8 of 11 bookings marked showed — 73% — which at n = 11 says the assumption sits in the right region, not that it should become 0.73. A ConsultationShowed event on real volume replaces this.
Cold show → treatment0.15ASSUMPTIONHand-discounted from a blended 32.6% win rate that includes referrals and repeat patients.
GHL "01. Sales Pipeline", which stopped receiving new opportunities on 2026-09-29 when close and payment moved to Zoho. The sample runs to within five days of that cutover, so these inputs are recent but closed.
Three figures existed for this event. They are the same figure at different stages.
FigureStatusWhy
$2,500retiredA rounded placeholder in an earlier spec. Never derived. Retired 2026-11-21.
$2,385pre-comp$26,500 × 0.60 show × 0.15 close. Correct arithmetic, but it bills every completed treatment.
$1,800in use2,385 × 0.75 = 1,789, rounded. The comp-adjusted figure, and the one the CAPI gateway sends.

Data cutover — 2026-09-29

GHL "01. Sales Pipeline" stopped receiving new opportunities on this date; last stage movement 2026-10-27. Every GHL-sourced revenue and win-rate figure ends here. Anything after it lives in Zoho.