# Evidence F: MedicineOS backoffice and operating model

Internal side of the same journey: organic Google click → 4 weeks into tirzepatide. Scope: Catalog, Visitron,
Referrals, Provider Chat/Promptimus, flags, analytics/BI, support tooling, humans, Friendly-PC/CPOM, compliance,
unit economics.

Method: re-read the recovered first-party app and the four internal-portal route inventories; re-fetched public
pages and public APIs on 2026-10-19 (read-only GETs, no accounts, no forms). The three internal portals' own
JavaScript was **not** re-fetched in this file; their operation inventories are taken from the recorded recon
dataset. Where the executive deck asserts operations or SLAs that the underlying dataset does not contain, the
dataset wins and the discrepancy is called out in §7.

Tag legend: `[CODE path:line]`, `[LIVE url, 2026-10-19]`, `[DATA file → key]`, `[DOC file §section]`,
`[PUBLIC url]`, `[INFERENCE]`.

---

## 0. The backoffice estate (what exists, who can reach it)

| App | Host(s) | Auth observed | RTK ops / route decls | Who uses it | Evidence |
|---|---|---|---|---|---|
| **Visitron** (clinician EHR/workstation) | `visitron.internal.generalmedicine.co`, `visitron.staging-nile.com`, `visitron-test.internal…` | Google OAuth `/api/internal_oauth/login`; also `/api/auth/login` / `/api/auth/refresh` / `/api/auth/logout`; edge returns 200 | 91 RTK ops / 98 routes, 33 cache tag types | Clinicians (MD/NP) running video, chat and async encounters; medical directors doing reviews | `[DATA osint/internal_portals_recon.json → visitron.rtk_endpoints, visitron.api_paths]`; `[DOC architecture.md §8.2]` |
| **Catalog** (SKU + CarePathway admin) | `catalog.internal.generalmedicine.co`, `catalog.staging-nile.com` | Edge 307 → `/login` | 106 RTK ops / 47 routes, **one** cache tag (`User`) | Clinical leadership + catalog/product operators authoring SKUs, pathways, questionnaires, criteria, lab panels, ICD-10/CPT, prompts, tenants | `[DATA … → catalog.rtk_endpoints, catalog.api_paths]`; `[DOC architecture.md §8.3]` |
| **Referrals** (concierge / promises queue) | `referrals.internal.generalmedicine.co`, `referrals.staging-nile.com` | Edge 307 → `/login` | 27 RTK ops / 18 routes, 8 tag types | Care coordinators claiming unbooked in-person referral orders, phoning specialist offices, logging outcomes | `[DATA … → referrals.rtk_endpoints]`; `[DOC architecture.md §8.4]` |
| **Provider Chat / Promptimus** | `internal.generalmedicine.co` (200), `promptimus.staging-nile.com` (401) | Edge 200 on prod; prompt workbench auth-gated | 14 RTK ops / 20 routes, 4 tag types | Medical directors / prompt engineers testing prompts, switching models, tuning temperature; operational chat | `[DATA … → provider_chat.rtk_endpoints]`; `[DOC architecture.md §8.5]` |
| **VisitAI + Chrome extension** | `visitai.generalmedicine.co`, `extensions.generalmedicine.co` | `/api/internal_oauth/extension_token` shared by all four apps | — | Clinicians; one extension spanning all portals | `[DOC architecture.md §8.6]` |
| **Provider auth** | `provider-auth.generalmedicine.co` (`awselb/2.0`) | Cognito-consistent ALB | — | Clinician identity, separate from Google-OAuth admin plane | `[DOC architecture.md §8.6]` |

Structural note for the tech plan: Visitron and Provider Chat call **both** `/api/internal/*` (Rails-native) and
`/api/internal_oauth/*` (OAuth-scoped); Catalog and Referrals use `/api/internal_oauth/*` plus `/api/internal/ops/*`.
Two API namespaces are in simultaneous use in one product surface. `[DOC architecture.md §8.9]`

### 0.1 Catalog object model — what a GLP-1 SKU actually is

The Catalog is not a CMS over static pages; it is the clinical protocol editor. Reading the 106 operations as an
object graph:

| Object | Operations | What it controls on the GLP-1 path | Evidence |
|---|---|---|---|
| **SKU** | `createSku`, `updateSku`, `updateSkuGeneral`, `updateSkuIndexable`, `updateSkuVisitConfiguration`, `getSkuStates`, `getSkuCategories`, `getSkus` | Sellable unit (`300543` eval, `300550` follow-up, `201448` Zepbound); indexability in sitemaps; visit configuration; state (`enabled` / `dark_launched`) | `[DATA … → catalog.rtk_endpoints]`; `[DATA extracted_care_plans_and_journeys.json → 300543.state, 400020.state]` |
| **SKU option / modality** | `createSkuPhase`, `updateSkuPhase`, `getSkuPhaseTypes`; `createPhaseModality`, `updatePhaseModality`, `deletePhaseModality` | Video vs chat vs econsult vs in-person; cash vs insurance sub-options and price; `allows_first_fill`, `allows_refill`, `allows_referral`. (`ZeroTouch` is a Catalog display label in the briefing deck, not a confirmed product modality) | `[DATA … → catalog.rtk_endpoints]`; `[DATA … → 300543.modalities, sku_options]` |
| **CarePathway** | `createCarePathway`, `updateCarePathway`, `publishCarePathway`, `returnCarePathwayToDraft`, `archiveCarePathway`, `reorderCarePathwayPhases`, `updateCarePathwayLinkedSkus`, `getCarePathwayPhaseKinds` | Multi-phase protocol; draft → published → archived; phase ordering; which SKUs the pathway links | `[DATA … → catalog.rtk_endpoints]`; `[DOC briefing slide 35 "Authored by Clinical Leadership in Catalog Studio"]` |
| **Criteria set / requirement** | `createCriteriaSet`, `cloneCriteriaSet`, `getCriteriaSetReferences`, `createRequirement`, `updateRequirement`, `deleteRequirement` | Eligibility logic and per-phase requirements (questionnaire, lab, follow-up visit). Cloning implies one criteria set is reused across SKUs | `[DATA … → catalog.rtk_endpoints]` |
| **Questionnaire** | `createQuestionnaire`, `updateQuestionnaire`, `getSkuQuestionnairesJson`, `updateSkuQuestionnairesJson`, `getSkuQuestionnairesPdf`, `downloadBulkSkuQuestionnairesPdf`, `requestQuestionnaireEdits`, `applyQuestionnaireEdit`, `getQuestionnaireUsage` | Screening vs history intake per SKU, versioned with an edit/approval step and PDF export for offline review | `[DATA … → catalog.rtk_endpoints]`; `[DATA … → 300543.screening_questionnaire_id, history_questionnaire_id]` |
| **Lab panel / biomarker** | `getLabPanels`, `saveLabPanel`, `removeLabPanel`, `getAvailableLabPanelBiomarkers`, `getJunctionBiomarkers`, `updateJunctionBiomarker`, `getLabSkuDetail`, `updateLabSkuDetail` | Panel composition from a biomarker catalog; per-lab price/fee fields (`test_price`, `clinical_fee`, `platform_fee`, `order_fee`, `collection_fee`) | `[DATA … → catalog.rtk_endpoints]`; `[LIVE /api/external/lab_skus/500041, 2026-10-19]` |
| **Procedure coding** | `getProcedureCodeMappings`, `createProcedureCodeMapping`, `updateProcedureCodeMapping`, `getProcedureCodeMappingProcedureTypes`, `getProcedureCodeMappingServiceTypes`, `searchIcd10Codes` | ICD-10/CPT mapping per service/procedure type, used for billing and referrals. **No ICD/CPT string appears anywhere in the patient app** — coding is backoffice-only | `[DATA … → catalog.rtk_endpoints, catalog.api_paths]`; `[CODE absence across first-party tree]` |
| **Medication fill defaults** | `getMedicationFillSkuDetail`, `updateMedicationFillSkuDetail`, `updateMedicationDefaults` | Days supply, fill quantity, default pharmacy (Zepbound PDP shows 28-day supply; 6 strengths) | `[DATA … → catalog.rtk_endpoints]`; `[LIVE /s/zepbound-tirzepatide/201448, 2026-10-19]` |
| **Condition taxonomy** | `getCatalogConditions`, `createConditionCategory`, `replaceConditionCategoryCatalogConditions` | Storefront navigation tree; the GLP-1 eval sits under a condition category, Zepbound under medication | `[DATA … → catalog.rtk_endpoints]` |
| **PDP content** | `getProductDetailPage`, `updateProductDetailPage`, `generateSkuImage`, `useSkuImageGeneration`, `uploadSkuReferenceImage` | Page copy, images, AI-generated storefront art | `[DATA … → catalog.rtk_endpoints]` |
| **Prompts and tenancy** | `createPrompt`, `updatePrompt`, `getPromptSummaries`, `getTenants`, `getProviderRoles`, `getEconsultSpecialties` | LLM prompt library; tenant config; provider role taxonomy for routing | `[DATA … → catalog.rtk_endpoints]` |

What this means operationally: launching a new GLP-1-adjacent protocol is a **content operation**, not a deploy —
condition → criteria → questionnaire → SKU → phases → PDP → images → ICD-10/CPT → lab panel, all inside Catalog.
The cost is governance: the same tool has only one cache tag (`User`) for 106 operations, so most mutations rely on
manual refetch, and questionnaire/pathway edits are the only places with an explicit second-person approval step.
`[DATA … → catalog.rtk_endpoints]` `[DOC architecture.md §8.3]`

---

## 1. Steps table — the internal action behind each patient-visible step

Time is from first organic click. "Patient sees/does" is only enough to anchor the internal step; acquisition and
intake detail belong to evidence files A/B/C.

| Step | Day/time | Patient sees/does | GM system / screen / route | Data captured or emitted | Evidence |
|---|---|---|---|---|---|
| **F1** | T+0 | Lands on `/s/glp-1-medication-evaluation/300543` (or a GLP-1 medication PDP) from a Google result | Catalog-authored SKU record; PDP renderer. `300543` state `enabled`; category `condition`; `allows_video_visit:true`, `allows_chat:true`, `allows_econsult:false`; `availability_promise:"today"` for video | SKU id, category, modality flags, price options, recommended SKU ids (`300513,500041,300550,400012,300312`) | `[DATA osint/extracted_care_plans_and_journeys.json → 300543]`; `[LIVE https://generalmedicine.co/s/glp-1-medication-evaluation/300543, 2026-10-19]` |
| **F2** | T+0–5m | Answers screening questionnaire, then history questionnaire | SKU is bound to two Catalog questionnaire UUIDs: screening `11340c6b-1296-4c46-ad5b-1e9c48474084`, history `3cdcfa37-64a0-4257-964b-9a41209b1f29`. Questionnaire edits go through a second-person review: `requestQuestionnaireEdits` → `applyQuestionnaireEdit` / `discardQuestionnaireEdit`; usage tracked by `getQuestionnaireUsage` | Questionnaire response, response id, per-SKU questionnaire JSON (`getSkuQuestionnairesJson`) | `[DATA … → 300543.screening_questionnaire_id, history_questionnaire_id]`; `[DATA … → catalog.rtk_endpoints: requestQuestionnaireEdits, applyQuestionnaireEdit, getQuestionnaireUsage]` |
| **F3** | T+1–3m | Confirms the state they are physically in before screening | `/s/{slug}/{id}/location-confirmation`; the state is appended to the SKU-options call as `appointment_location_state` | Physical-location state → drives clinician pool filter server-side | `[CODE lib/routes/skuBooking.ts:57]`; `[CODE lib/api/endpoints.ts:132-135]` |
| **F4** | T+2–4m | Sees video $100 cash / ~$25 insurance, chat $30 cash | SKU options endpoint; price comes from Catalog `sku_options` (cash/insurance sub-options), not from the UI | Selected modality, price, `availability_promise` (video "today"; chat next slot e.g. `2026-10-19T10:00:00-06:00`) | `[DATA … → 300543.sku_options]`; `[LIVE …/s/weight-loss-follow-up/300550, 2026-10-19: "Video Visit $25 Est. or $100 w/o insurance · Chat Visit $30 w/o insurance"]` |
| **F5** | T+4–8m | Books the visit, pays | Visit cart (`/api/external/visits/{create,book,add_topic}`); appointment lands on a clinician's Visitron schedule (`getProviderAppointments`) | `visit_id`, `sku_id`, `duration_minutes` (15/30/45 by topic count), `scheduled_start`, `clinician_id`, billing type. Production video slots are **20 min, max 9 per 4-hr block** | `[CODE lib/api/endpoints.ts:82-93]`; `[DOC architecture.md §10.1]`; `[DATA … → visitron.rtk_endpoints: getProviderAppointments, getAppointmentInfo]`; `[LIVE Ashby FT Physician, 2026-10-19]` |
| **F6** | T-15m to T-0 | Receives Zoom link by SMS/email; may be asked for ID | Prewarm: `applyAppointmentLatestPrewarm`, `triggerMetriportSync`, `getNablaStatus`; ID attestation `attestManualIdVerification` | Chart context (`PatientChartContext`), longitudinal records from Metriport/Particle, scribe availability, ID-verification attestation | `[DATA … → visitron.rtk_endpoints: applyAppointmentLatestPrewarm, triggerMetriportSync, getNablaStatus, attestManualIdVerification]`; `[LIVE https://generalmedicine.co/using-zoom, 2026-10-19]` |
| **F7** | T+0 (visit) | Talks to clinician; hears the plan | Visitron encounter: `getAppointmentDraftEncounter` → `updateAppointmentDraftEncounter` → `publishAppointmentDraftEncounter`; diagnosis/order code search (`searchDiagnoses`, `getIcd10Codes`, `getLabCodes`, `getQuestionnaireCodes`); Nabla ambient scribe drafts the note | Draft encounter, ICD-10 diagnosis, orders, note text, `athena_id` on publish | `[DATA … → visitron.rtk_endpoints]`; `[DOC architecture.md §8.2, §8.2 sequence]` |
| **F8** | T+0 (visit) | Is told whether the drug is covered; gets a prescription if appropriate | eRx in DoseSpot (`dose_spot_sso`, `getDoseSpotIdpStatus`, `createDraftPrescription`→`finalizeDraftPrescription`); benefit check through Arrive Health (`launch_benefit_inquiry`, `getInsurancePriceStatus`) and DoseSpot RTPB (`prescription_eligibility_inquiries/*`) | Drug, dose, days supply, pharmacy/NCPDP id, copay/formulary/prior-auth response | `[DATA … → visitron.api_paths: /dose_spot_sso, /arrive_health/launch_benefit_inquiry, /dose_spot/prescription_eligibility_inquiries/]`; `[DATA … → visitron.rtk_endpoints: createDraftPrescription, finalizeDraftPrescription]` |
| **F9** | T+0–30m | Sees a care plan with next steps; receives "why this" text | Care::Plan generation: states `awaiting_curation → awaiting_generation → in_review → sent`; Visitron `generateNextStepReview` → `updateNextStepReview` → `applyReview` / `rejectReview` / `confirmCarePlan` / `confirmNoChange`; then `syncClinicianReviewEncounterToAthena` | `proposed_json` (AI draft) → `final_json` (clinician-approved), `patient_rationale`, next-step cards | `[CODE components/CareActions/careActionSection.ts:64-72]`; `[DATA … → visitron.rtk_endpoints: generateNextStepReview, applyReview, rejectReview, confirmCarePlan, confirmNoChange, syncClinicianReviewEncounterToAthena]`; `[DOC general-medicine-executive-briefing.html slide 35 "AI Draft → Clinician Review → Signed Plan"; slide 320 "AI proposed_json → final_json"]` |
| **F10** | T+0–2d | Books baseline labs (A1c, CMP, TSH, lipids) at a Quest/Labcorp/BioReference draw site | Lab order from a 500xxx SKU; `care_journey/lab_bundle` bundles required panels; result review by a clinician; abnormal → proactive outreach | Lab SKU ids (`500008`, `500041`, `500093`, `500001`), `test_price`, `clinical_fee`, `platform_fee`, `collection_fee`, `order_fee`, `accession_number` | `[CODE lib/api/endpoints.ts:211-215]`; `[LIVE /api/external/lab_skus/{500008,500041,500093,500001,500049}, 2026-10-19]`; `[LIVE https://generalmedicine.co/resources/labs-faqs, 2026-10-19: "If your levels are out of range, a member of our clinical team will reach out"]` |
| **F11** | T+0–3d | Pays (or runs insurance) and picks a pharmacy | Pharmacy order: `selectPharmacy` (NCPDP), `checkoutPharmacyOrderWithInsurance` \| `checkoutPharmacyOrderThroughPharmacy`, `getPharmacyOrderGmCard`; states `created → pharmacy_selected → order_placed → dispensed` | NCPDP provider id, `pharmacy_order_state`, `pharmacy_submission_state`, gm_card BIN/PCN/group/member | `[CODE lib/store/api/pharmacyOrderApi.ts:17-21, 90-113]`; `[DOC data-models.md §1 PHARMACY_ORDER]` |
| **F12** | T+7–28d | Messages the care team about side effects, or opens a chat visit; gets a response within a posted window | Message Center `/account/messages`: `POST /api/external/conversations/token` (Twilio), `GET /conversations/{id}/response_expectation`; categories include `visit`, `lab`, `pharmacy_order` | Conversation, `twilio_channel_sid`, subject category, `response_expectation`, attachments (file upload ids); chat visits carry a **3-hour** clinical response window | `[CODE lib/api/endpoints.ts:59-67]`; `[CODE lib/store/api/messageCenterApi.ts:23, 168]`; `[DOC architecture.md §10.7]`; `[LIVE Ashby FT Physician, 2026-10-19]` |
| **F13** | ~T+28d | Books "Weight Loss Follow Up" for titration (monthly while titrating) | `300550` Weight Loss Follow Up — same price grid; bound questionnaires `e7a4697c-…`, `765b0c3f-…`; Care Journey checklist gates refills on questionnaire + labs + visit | New visit, questionnaire response, phase change (`changePhase`, `changePathway`, `batchChangePatientMedicationPhases`) | `[DATA … → 300550]`; `[LIVE https://generalmedicine.co/resources/using-a-glp-1-for-weight-loss, 2026-10-19: "monthly check-ins at first … then roughly every three months"]`; `[DATA … → visitron.rtk_endpoints: changePhase, changePathway, batchChangePatientMedicationPhases]` |
| **F14** | T+28d onward | Requests a refill (active refills left) or a renewal (none left); "most refill requests reviewed and confirmed the same day" | Refill = direct order; renewal = clinician review. Visitron `createPatientMedicationPrescriptionRenewal`, `approveInboundRequest` / `denyInboundRequest` for inbound pharmacy requests; `updatePatientMedicationDaysSupply`, `clearPatientMedicationDaysSupply`, `updatePatientMedicationCourseType` | Renewal request, inbound eRx request, decision, days supply, course type | `[LIVE https://generalmedicine.co/resources/medication-faqs, 2026-10-19]`; `[DATA … → visitron.rtk_endpoints]` |

### 1.1 GLP-1 baseline monitoring → GM lab SKU (the actual mapping)

The public guide prescribes A1c, a general metabolic panel, thyroid hormone and a lipid panel at treatment
start, then repeat labs every 3–6 months. Those are sold as independent 500xxx SKUs, each priced
`test + $15.00 clinical fee`:

| Guide requirement | GM SKU | Biomarkers in the panel | Quest cash price |
|---|---|---|---|
| Blood sugar (HbA1c) | **500041** A1c: Diabetes and Prediabetes Check Up | Hemoglobin A1c with eAG | **$47.64** |
| General metabolic panel | **500008** Comprehensive Metabolic Panel | CMP with Adjusted Calcium, Plasma | **$48.29** |
| Thyroid hormone | **500093** Thyroid Panel with TSH | TSH, Free T4, Free T3 | **$54.10** |
| Cholesterol (lipid panel) | **500114** Lipid panel | Lipid Panel, Standard | **$49.90** |
| Lipid + inflammation (upsell) | **500012** Lipid panel plus | hs-CRP, Cardio IQ Lipid, Cardio IQ ApoB | **$62.90** |
| All four in one draw | **500001** Annual wellness (12 biomarkers) | CBC, CMP, Cardio IQ Lipid, ApoB, HbA1c, TSH+FT4, hs-CRP, ESR, Iron/TIBC, Ferritin, Vit D, Urine albumin | **$129.92** |
| Same 12-biomarker set, cheaper | **500049** Total Health Panel | identical 12-biomarker workup | **$114.00** |

`[LIVE /api/external/lab_skus/{500041,500008,500093,500114,500012,500001,500049}, 2026-10-19]`
`[LIVE GLP-1 guide, 2026-10-19]` `[DATA live_sitemap_skus.json → sku_urls (500114 lipid-panel, 500012 lipid-panel-plus)]`

Operational consequence: the baseline is **à la carte**, so a cash patient can spend $47.64–$129.92 on labs the
same day as a $100 visit, and the clinician has to decide which panels to order. The Care Journey is what enforces
the repeat interval; there is no "GLP-1 lab bundle" SKU in the public catalog.


---

## 2. Handoffs table — GM ↔ outside party on this path

Only handoffs that touch the GLP-1 path or the backoffice that runs it. "Who pays" is stated only where observed.

| Step | Counterparty (company + product) | Direction | What is exchanged | Mechanism | Trigger | Who pays / fee | Evidence |
|---|---|---|---|---|---|---|---|
| F5/F6 | **Zoom** (video) | GM → patient | Unique meeting URL delivered by SMS/email | Redirect/deep link | Visit booked | Unknown; not itemized to patient | `[LIVE /using-zoom, 2026-10-19]`; `[DOC architecture.md §10.2]` |
| F6 | **Metriport** (HIE) + **Particle Health** | GM ⇄ vendor | Longitudinal records, prior prescriptions, labs | REST (`triggerMetriportSync`, `latest_metriport`), then injected into AI context | Appointment prewarm, `/ai` prewarm flag | Unknown | `[DATA … → visitron.rtk_endpoints: triggerMetriportSync]`; `[DOC architecture.md §10.8]` |
| F7 | **Nabla** (ambient scribe) | vendor → GM | Drafted encounter note text | Embedded/iframe + status check | Clinician starts scribe | Unknown | `[DATA … → visitron.rtk_endpoints: getNablaStatus]`; `[DOC architecture.md §1.1 vendor row 3]` |
| F7/F9 | **Athenahealth** (EHR) | GM → Athena, Athena → GM | Encounter note, diagnoses, care plan, chart summary; `athena_id` | REST sync/resync (`sync_encounter_to_athena`, `encounter_summary_athena_resync`, `care_plan_athena_resync`) | Encounter publish / plan confirm / manual resync | Unknown | `[DATA … → visitron.api_paths: /sync_encounter_to_athena, /care_plan_athena_resync, /encounter_summary_athena_resync]`; `[DOC architecture.md §8.2]` |
| F8 | **DoseSpot** (eRx + RTPB) | GM ⇄ DoseSpot | Patient context, eligibility inquiry, formulary, new-Rx transmission; SSO identity | SSO redirect + REST + NCPDP SCRIPT (attributed) | Prescribe | Unknown | `[DATA … → visitron.api_paths: /dose_spot_sso, /dose_spot/prescription_eligibility_inquiries/{launch,formulary,sync_insurance}]` |
| F8 | **Arrive Health** (RTPB) | GM ⇄ Arrive | Patient + drug + pharmacy → copay, formulary, prior-auth flags; `arrive_rtpbi_inquiry_id` | REST launch + status poll | Benefit check at prescribing | Unknown | `[DATA … → visitron.api_paths: /arrive_health/launch_benefit_inquiry, /patient_context]`; `[DOC architecture.md §8.2 sequence]` |
| F8/F11 | **Waystar** (clearinghouse) | GM ⇄ payer (via Waystar) | Claims/eligibility (EDI 270/271 attributed) | DNS-only hosts `waystar-connector/-bridge.staging-nile.com`; no payload observed | Insurance billing | Unknown | `[DOC architecture.md §7.1 vendor row 7; §7.3 host topology]`; `[DOC adversarial-and-market-report.md §5 vendor table]` |
| F10 | **Quest Diagnostics / Labcorp / BioReference** | GM → lab; lab → GM | Order + requisition; results (`resulted`), `accession_number`, PDF | Lab order API; HL7/FHIR attributed for Quest | Lab booked | Patient pays `test_price + $15 clinical fee`; lab economics not disclosed | `[LIVE /api/external/lab_skus/500041, 2026-10-19]`; `[DOC data-models.md §1 LAB_ORDER]` |
| F10 | **Junction / Vital** (`tryvital.io`) | GM ⇄ vendor | Biomarker catalog, lab-panel composition, ordering through Vital affiliate PCs | REST (`junction_biomarkers`, `getAvailableLabPanelBiomarkers`) | Catalog admin composes a panel | Unknown | `[DATA … → catalog.rtk_endpoints: getJunctionBiomarkers, getAvailableLabPanelBiomarkers]`; `[LIVE /vital-informed-consent, 2026-10-19]` |
| F10 | **OpenLoop / Reliant.MD / MECNB** (Vital affiliate PCs) | patient → PC | Telehealth consult consent, state-specific professional entity | Consent form + state routing | Lab/telehealth consult under Vital consent | Patient responsible for amounts due | `[LIVE https://generalmedicine.co/vital-informed-consent, 2026-10-19]` |
| F11 | **Retail/mail pharmacies** (CVS, Walgreens, Walmart, LillyDirect, NovoCare) | GM ⇄ pharmacy | Rx + patient identifiers; price/cash vs insurance; order state | NCPDP provider id; pharmacy order API; `gm_card` for pre-pay | Patient selects pharmacy | Patient pays cash price, insurance copay, or GM pre-pay; GM margin undisclosed | `[CODE lib/store/api/pharmacyOrderApi.ts:90-113]`; `[CODE components/Refresh/SearchResults/searchPharmaciesUtils.ts:22 (lillydirect, novocare featured)]`; `[DOC architecture.md §9.1]` |
| F11 | **Eli Lilly / Novo Nordisk** (manufacturer direct) | GM → manufacturer hub | Referral of cash Rx to LillyDirect/NovoCare; co-branded Lilly storefront | Pinned pharmacy row + `/p/lilly/weight-loss` | GLP-1 cash fill | Manufacturer sets cash price (Foundayo from $149/mo, Zepbound from $299/mo) | `[LIVE https://generalmedicine.co/p/lilly/weight-loss, 2026-10-19]`; `[CODE …searchPharmaciesUtils.ts:22]` |
| F12 | **Twilio** (SMS/OTP + Conversations) | GM ⇄ Twilio | OTP send/verify (Cognito CUSTOM_AUTH → Twilio in Lambda); ephemeral chat token; message delivery | Lambda + REST; `POST /api/external/conversations/token` | Login, every thread | Per-message/token cost; not disclosed | `[CODE lib/analytics/otpAnalytics.ts:8-9 comment]`; `[CODE lib/store/api/messageCenterApi.ts:168]` |
| F12/F14 | **Zendesk** (clinician chat/triage) | GM internal | Async chat/triage threads, clinician replies, attachments | Zendesk agent workspace (named as the chat platform in the clinician posting); not present in the recovered patient bundle | Chat visit or triage message | Zendesk subscription; not disclosed | `[LIVE Ashby "Full Time Physician - Primary Care", 2026-10-19]` |
| All | **Persona** (KYC) | GM ⇄ Persona | Government ID + selfie; `inquiry.completed` webhook | Webhook + embedded SDK | ID verification at signup/visit | Unknown | `[DOC architecture.md §5.1]` |
| All | **Stripe / Trust Commerce** | GM ⇄ PSP | Card tokens, charges, refunds; Mercy guarantor billing | REST + webhooks | Checkout, cancellation fee, refund | Card fees; PCI DSS L1 | `[DOC architecture.md §1.1, §4]` |
| F1–F14 | **Segment** | GM → Segment | Product events (`sf_`, `ns_`, `ssi_` prefixes), identify/alias | JS SDK, proxied CDN `evs.segment.generalmedicine.co` | Every interaction | Segment contract; not disclosed | `[CODE packages/mantine-theme/analytics/segment.ts:14-19]`; `[CODE packages/mantine-theme/analytics/events.ts:1-90]` |
| F1–F14 | **Datadog** (RUM) | GM → Datadog | Sessions, session replay, API latency, flag evaluations, errors | Browser SDK; Slack replay notifications via Rails job | Every session | Datadog contract | `[CODE lib/api/metrics.ts:1-90]`; `[CODE lib/launchDarklyFlagEvaluator.ts:91]`; `[DOC architecture.md §8.7]` |
| F1–F14 | **LaunchDarkly** | GM ⇄ LD | Flag evaluations, context key (patient id or anonymous UUID) | Client SDK, streaming/report mode | Every page | LD contract | `[CODE contexts/LaunchDarklyContext.tsx:51-66]` |
| Admin | **Google OAuth** + **AWS Cognito** | GM ⇄ IdP | Staff identity, extension tokens | OAuth code flow; `/api/internal_oauth/callback` | Admin login, extension | Unknown | `[DOC architecture.md §7.4 identity and trust chains]` |
| Admin | **Sanity.io** | GM → CMS | Editorial content, images (`cdn.sanity.io`) | Headless CMS API | Content authoring | Sanity plan | `[DOC architecture.md §1.1]`; `[LIVE GLP-1 guide images on cdn.sanity.io, 2026-10-19]` |
| Hiring | **Ashby** | GM ⇄ Ashby | 16 open requisitions, applications | Hosted job board + API | Recruiting | Ashby plan | `[LIVE https://api.ashbyhq.com/posting-api/job-board/general-medicine, 2026-10-19]`; `[CODE lib/hardcoded-data/ClinicianHiring/clinician-hiring-content.tsx:28-33]` |
| BI | **Databricks** + **Metabase** | GM internal | Warehouse tables → dashboards | Metabase Databricks driver | Analyst query | Unknown | `[LIVE https://analytics.staging-nile.com/, 2026-10-19]`; `[DOC architecture.md §8.7]` |
| Ops | **Slack** | GM internal | Session-replay alerts to `#session-replay-new-storefront`; named as the team channel in the clinician posting | `PostSessionReplayNotificationJob`, 1 post/user/hour dedupe | RUM replay | Slack plan | `[DOC architecture.md §8.7]`; `[LIVE Ashby FT Physician, 2026-10-19]` |
| Trust | **LegitScript** | GM → public | Certification badge and listing | Link + badge asset | Marketing/compliance | Certification fee; not disclosed | `[CODE components/Footer.tsx:164, 269]`; `[LIVE https://legitscript.com/websites/generalmedicine.co]` |

**Single points of failure:** DoseSpot is the most fragile dependency in this table — if e-prescribing fails, Visitron
cannot prescribe and the visit-to-pharmacy funnel stops. Identity (Persona) and payments (Stripe) are also
single-sourced, with Trust Commerce as a secondary gateway for Mercy. Labs and pharmacy are multi-sourced.
`[DOC adversarial-and-market-report.md §5]`

**Correction on support tooling:** Zendesk **is** used, but as the clinician chat/triage workspace, not as a
patient help desk — the FT physician posting names "Zendesk (chat)" alongside Athena, Zoom, Slack and Nabla.
No Zendesk SDK or widget appears in the recovered patient bundle, and patient support still runs first-party over
Twilio SMS/Conversations plus a tenant phone number. Whether the same Zendesk instance also backs patient support
agents is not established. `[LIVE Ashby FT Physician, 2026-10-19]` `[INFERENCE on the patient-support boundary]`

---

## 3. Ops touchpoints — where a human acts

### 3.1 Roles observed

| Role | Function on this path | Tool / queue | Evidence |
|---|---|---|---|
| **Full-time physician (MD/DO)** | 36 clinical hours/week of video + async chat; owns the encounter, prescribing and care-plan sign-off | **Athena (EMR) · Zoom (video) · Zendesk (chat) · Slack (team) · Nabla (scribe)**; Visitron on the backoffice side | `[LIVE Ashby "Full Time Physician - Primary Care", 2026-10-19]` |
| **1099 physician (part-time)** | 10–12 hrs/week; video + chat visits alongside an existing practice | Same platform set, Visitron | `[LIVE Ashby "Virtual Physician (Part-Time - 1099)", 2026-10-19]` |
| **1099 nurse practitioner** | 12–15 hrs/week; video + chat, multistate licensure covering ≥30% of the US population; practices under collaborative agreements where required | Same, plus CPA structure | `[LIVE Ashby "Virtual Advanced Nurse Practitioners - 1099", 2026-10-19]` |
| **Medical Director** | Clinical oversight and sign-off; prompt/model tuning | Visitron + Provider Chat/Promptimus | `[LIVE https://generalmedicine.co/bio/dr-pallabi-sanyal-dey, 2026-10-19]`; `[DOC briefing slide 51]` |
| **Clinical operations lead** | Builds clinical teams, policies, partnerships | Cross-portal | `[LIVE https://generalmedicine.co/bio/dr-poushali-bhattacharjee, 2026-10-19]` |
| **Clinician training lead** | Trains every new clinician before first patient | Training program | `[LIVE https://generalmedicine.co/bio/samantha-schisler, 2026-10-19]` |
| **Care coordinator** | Appointment scheduling/reminders, inbound and outbound calls to **patients, provider offices, insurance companies and labs**, care documentation in the EMR, referral/authorization paperwork. Reports to the Operations Manager; works with the Director of Operations, Medical Director and clinicians | Athena + internal tools, phone, Referrals portal | `[LIVE Ashby "Care Coordinator", 2026-10-19]`; `[DATA … → referrals.rtk_endpoints]` |
| **Operations Manager / Director of Operations** | Line management for coordinators; referral and scheduling operations | Referrals + internal tools | `[LIVE Ashby "Care Coordinator" reporting line, 2026-10-19]` |
| **Manager, NP Licensure & Collaborative Practice** | Owns the collaborating-physician bench, CPAs, primary-source verification, state paperwork, and licensure escalations — including **a pharmacy refusing to fill an NP-written prescription in a given state** | Ashby (pre-hire review) + state board portals; no internal portal recovered | `[LIVE Ashby "Manager, NP Licensure & Collaborative Practice", 2026-10-19]` |
| **Catalog operator / clinical product** | Authors SKU, phases, criteria, questionnaires, lab panels, ICD-10/CPT mappings, prompts | Catalog (106 ops) | `[DATA … → catalog.rtk_endpoints]` |
| **Prompt engineer / medical director** | Tests prompts against a model version, sets temperature, saves | Provider Chat `getModelVersions`, `getTemperatures`; Visitron `/api/internal/copilot_prompt_test/*` | `[DATA … → provider_chat.rtk_endpoints; visitron.api_paths]` |
| **RCM Director / VP** | Owns patient access, **authorization**, coding, billing, denials, collections, payer contracts, clearinghouse vendors | Waystar bridge; RCM stack not recovered | `[LIVE Ashby "Revenue Cycle Management, Director" / "…, VP", 2026-10-19]`; `[DOC architecture.md §7.1 vendor row 7]` |
| **Clinical recruiter, physicians** | Full-cycle physician sourcing and screening in Ashby | Ashby | `[LIVE Ashby "Clinical Recruiter, Physicians", 2026-10-19]` |
| **Support agent ("care team")** | Answers SMS/Message Center threads, refunds, cancellation questions | Message Center (Twilio) + tenant phone; SMS 98528 | `[CODE hooks/useContactSupport.ts:10]`; `[LIVE /refund-policy-labs, 2026-10-19]` |

### 3.2 Queues and case states

| Queue | Object | States observed | Assignment model | SLA / timing | Evidence |
|---|---|---|---|---|---|
| **Visitron appointment list** | Appointment | `booked`, `scheduled`, `confirmed`, `awaiting_confirmation`, `ready_for_visit`, `held`, `in_progress`, terminal `canceled`/`hold_expired`/`patient_no_show` | Clinician sees own schedule (`getProviderAppointments`); reschedule same or any provider | Slot promise from SKU config: video `"today"`; chat next slot timestamp | `[CODE components/CareActions/careActionSection.ts:40-92]` `[DATA … → visitron.rtk_endpoints: getProviderAppointments, rescheduleSameProvider, rescheduleAnyProvider]` `[DATA … → 300543.sku_options.top_fields.availability_promise]` |
| **Draft encounter** | Encounter | draft → published → Athena-synced | Clinician owns the encounter | None observed | `[DATA … → visitron.rtk_endpoints: getAppointmentDraftEncounter, updateAppointmentDraftEncounter, publishAppointmentDraftEncounter, syncEncounterToAthena]` |
| **Async clinical review (ACR)** | Clinician review | `getPendingReview` → `submitClinicianReview` → `applyReview` / `rejectReview`; rejection reason captured | Pending list; clinician pulls (no claim operation); can request more photos/files | None observed in client; ACR is the async path for 300xxx protocols | `[DATA … → visitron.rtk_endpoints: getPendingReview, submitClinicianReview, applyReview, rejectReview, updateClinicianReviewRejectionExplanation, getAsyncClinicalReviewFileUploadPreview]` |
| **Care-plan generation** | Care::Plan | `awaiting_curation` → `awaiting_external_work` → `awaiting_generation` → `in_review` → `awaiting_first_send`/`awaiting_curation_send` → `sent` | Clinician reviews AI draft | Guide says monthly → Q3M cadence for GLP-1; no system SLA | `[CODE components/CareActions/careActionSection.ts:64-72]` `[LIVE GLP-1 guide, 2026-10-19]` |
| **Care Journey checklist** | Patient action | `locked`, `available_at`, `completed`; groups Questionnaire / Lab / Visit | Patient-owned, clinician advances phase | Deadlines carried as `soonest_deadline` / card `deadline` | `[CODE lib/store/api/careJourneyApi.ts:46-77]` `[CODE lib/careJourney/checklist.ts:14-20]` `[DATA … → visitron.rtk_endpoints: changePhase, changePathway]` |
| **Referrals promises queue** | In-person referral order | claimable / snoozed / dropped; `QueueEnabled` can switch the queue off globally | Atomic claim (`claimPromise`); `dropOrSnoozeOrder` releases or snoozes | **No SLA value recovered.** Deck's "2-day referral promise" is not supported by the dataset (§7) | `[DATA … → referrals.rtk_endpoints: claimPromise, dropOrSnoozeOrder, getQueueEnabled, recordOutreach, logOutcome, endShift]` |
| **Pharmacy order** | PharmacyOrder | `created` → `pharmacy_selected` → `order_placed` → `dispensed`; cancel/switch flags | Patient-driven; clinician/pharmacy inbound requests via `listInboundRequests` | Refill review "same day" (public claim) | `[CODE lib/store/api/pharmacyOrderApi.ts:17-21, 90-113]` `[DOC data-models.md §1]` `[LIVE medication FAQ, 2026-10-19]` |
| **Async chat / triage visit** | Chat visit | Patient message → clinician reply; Zendesk thread (clinician side) | Zendesk queue; **3-hour response window** | **3 hours**, per the FT physician practice model | `[LIVE Ashby FT Physician, 2026-10-19]` |
| **Video visit capacity** | Video visit | 20-minute appointments, **max 9 per 4-hour block** | Scheduled block; operations handles non-clinical tasks | 20 min per visit; 36 clinical hrs/week for FT physicians | `[LIVE Ashby FT Physician, 2026-10-19]` |
| **Message Center** | Conversation | open thread, unread/read, `response_expectation` | Care-team queue by subject category | Server-computed per thread and per new thread; support hours Mon–Fri 9–9 ET | `[CODE lib/api/endpoints.ts:59-67]` `[CODE lib/store/api/messageCenterApi.ts:59-72]` `[CODE components/Refresh/SupportWidget/SupportWidget.tsx:146]` |
| **Shift** | Clinician shift | `createShiftCheckout` → `getPendingShiftCheckouts` → `shift_checkout_status`; chat shifts extend via `extendChatShift` | Clinician clocks out; pending checkout blocks closure | **4–6 hour blocks, Mon–Fri 7 a.m.–7 p.m. MT**; weekend/evening expansion underway; extension limits not recovered | `[LIVE Ashby FT Physician, 2026-10-19]`; `[DATA … → visitron.rtk_endpoints: createShiftCheckout, getPendingShiftCheckouts, getShiftCheckoutStatus, extendChatShift]` |

**Assignment precision:** Referrals is the only portal with an explicit work-claim operation (`claimPromise`). Visitron
has **no** claim/assignee operation in its inventory — clinicians see their own schedule (`getProviderAppointments`)
and a notification inbox (`getProviderNotifications`, `markProviderNotificationRead`, `resolveProviderNotificationVisit`),
so review work is pulled from a pending list rather than atomically claimed. That is an operational difference worth
copying deliberately: a clinical review queue without a claim primitive can double-work a case.
`[DATA … → visitron.rtk_endpoints; referrals.rtk_endpoints]` `[INFERENCE on pull semantics]`

**Licensure filtering (who may prescribe where):** the state is collected per booking, forwarded as
`appointment_location_state`, and the clinician-pool filter is server-side; the frontend never sees the licensed
set. `[CODE lib/routes/skuBooking.ts:57]` `[CODE lib/api/endpoints.ts:132-135]` `[DOC architecture.md §10.6]`
The public claim is 50 states + DC; care is 18+. `[DOC architecture.md §10.6 quoting the live /faq]`

**Support hours:** Mon–Fri 9 a.m.–9 p.m. ET on both the support widget and the store widget. That is the **same
window** as the clinical chat shift (Mon–Fri 7 a.m.–7 p.m. MT), so support and clinician coverage are aligned in one
9-to-9 Eastern block; the FT posting says weekend/evening expansion is "underway", which the patient-facing widgets
do not yet reflect.
`[CODE components/Refresh/SupportWidget/SupportWidget.tsx:146]` `[CODE components/Store/StoreWidget.tsx:110]`
The GLP-1 guide adds the clinical version of that promise: "Between scheduled visits, you can message your care
team with questions or side effects, with same-day support available during business hours."
`[LIVE https://generalmedicine.co/resources/using-a-glp-1-for-weight-loss, 2026-10-19]`

**Response promises — two different clocks.** The patient **Message Center** fetches a server-computed
expectation per thread and for a new thread (`GET /conversations/{id}/response_expectation`,
`GET /response_expectation`); the rendered string was not recovered, and the dossier's "respond within 2 hours"
is illustrative. Separately, the clinical **chat visit** model posts a fixed **3-hour response window**, with
inbound chat handled in Zendesk. `[CODE lib/api/endpoints.ts:63-65]` `[DOC architecture.md §10.7]`
`[LIVE Ashby FT Physician, 2026-10-19]`

### 3.3 What Promptimus / Provider Chat actually drafts

| Draft artifact | Mechanism | Human control | Evidence |
|---|---|---|---|
| **Care-plan / next-step review** | `generateNextStepReview` produces `proposed_json`; clinician edits (`updateNextStepReview`, `updateClinicianReviewEncounterDraft`), applies or rejects (`applyReview`/`rejectReview`), signs (`confirmCarePlan`), then pushes to Athena (`syncClinicianReviewEncounterToAthena`) | Clinician sign-off is a distinct step; rejection carries an explanation field | `[DATA … → visitron.rtk_endpoints]`; `[DOC briefing slide 320 "AI proposed_json → final_json"]` |
| **Patient-facing rationale** | `patient_rationale` field on next-step cards; viewed/feedback events tracked (`PATIENT_RATIONALE_VIEWED`, `PATIENT_RATIONALE_FEEDBACK`) | Clinician-approved text reaches the patient | `[CODE lib/store/api/nextStepCardsApi.ts:34]`; `[CODE packages/mantine-theme/analytics/events.ts:49-50]` |
| **Encounter note** | Nabla ambient scribe (not Promptimus) drafts note from the visit | Clinician edits/signs in Visitron | `[DATA … → visitron.rtk_endpoints: getNablaStatus]`; `[DOC architecture.md §8.2]` |
| **Prompt templates themselves** | Prompt library in Catalog and Provider Chat: `createPrompt`/`updatePrompt`/`getPromptSummaries`; per-model temperature; test harness in Visitron (`/api/internal/copilot_prompt_test/*`) | Medical directors test, switch models, tune settings without a release | `[DATA … → catalog.rtk_endpoints; provider_chat.rtk_endpoints; visitron.api_paths]`; `[DOC briefing slide 51]` |
| **Patient copilot** (the other side of the same prompt substrate) | `/api/external/copilot/*`, ActionCable `ExternalCable::CopilotChannel`; `gen-ai-prewarm-copilot` injects a Metriport review into context | Patient-facing triage/navigation, not diagnosis | `[CODE lib/api/endpoints.ts:310-318]`; `[DATA … → visitron.cable_channels, provider_chat.cable_channels]`; `[DOC architecture.md §10.8]` |

Hard limit on this finding: the **prompt text and model** are not in any recovered artifact. `getModelVersions`
proves the app enumerates model versions; which vendors/models are configured is unknown.

### 3.4 Clinical operating model and labor unit costs (from the job postings)

The full-time physician posting is the most detailed operating document in the public record. It states the
production model directly:

> **Platforms:** Athena (EMR) · Zoom (video) · **Zendesk (chat)** · Slack (team) · Nabla (scribe AI)
> **Visit Types:** Synchronous video (**20-min appointments, max 9 per 4-hr block**) and asynchronous triage/chat
> (**3-hr response window**).
> **Shifts:** **4–6 hour blocks (Mon–Fri, 7 AM–7 PM MT**; weekend/evening expansion underway). Non-clinical tasks
> handled by operations.
> **Scope:** Adult medicine only. No pediatrics, advanced psychiatry, or controlled substances.

`[LIVE Ashby "Full Time Physician - Primary Care", 2026-10-19]`

| Role | Pay | Hours / terms | Evidence |
|---|---|---|---|
| **FT physician, primary care** | **$180,000–$200,000/yr** + equity + malpractice with indefinite tail + licensure/credentialing covered | **36 clinical hours/week**; DEA license required for credentialing | `[LIVE Ashby FT Physician, 2026-10-19]` |
| **1099 physician** | **$50 per completed video visit / $12 per chat visit** | **10–12 hrs/week**; multistate licensure ≥20–30% of US population; CA/TX/FL/NY preferred | `[LIVE Ashby "Virtual Physician (Part-Time - 1099)", 2026-10-19]` |
| **1099 nurse practitioner** | **$35 per completed video visit / $8 per chat visit** | **12–15 hrs/week**; multistate licensure ≥30% of US population | `[LIVE Ashby "Virtual Advanced Nurse Practitioners - 1099", 2026-10-19]` |
| **Care coordinator** | **$22/hr** | Full-time, remote; 1 year healthcare patient-service experience; Athena preferred | `[LIVE Ashby "Care Coordinator", 2026-10-19]` |

Operating consequences that belong in the plan:

1. **Chat is the highest-contribution visit type per clinician dollar.** A $30 cash chat visit pays a $12
   physician or $8 NP fee, leaving **$18 or $22** before overhead; even the $25 insured chat leaves $13/$17.
   `[INFERENCE from observed prices and posted pay]`
2. **Video is the margin product on cash and a loss-leader on a $25 copay.** A $100 cash video visit pays a
   $50 physician fee or $35 NP fee; a $25 insured video visit does not cover either fee (**−$25 MD / −$10 NP**),
   so the insured video path only works with payer reimbursement or with NPs at high throughput. `[INFERENCE]`
3. **Throughput is deliberately capped.** 20-minute slots and max 9 per 4-hour block is a 75% slot-utilization
   ceiling (12 slots would be 100%), leaving slack for charting and the 3-hour chat window. `[INFERENCE]`
4. **Licensure is coverage-based, not literally one clinician per state.** Each NP must cover ≥30% of the US
   population and each 1099 physician 20–30%, so the network as a whole reaches 50 states through overlapping
   multi-state licenses. `[LIVE Ashby, 2026-10-19]`
5. **The NP prescribing layer is contractual, not just technical.** The licensure manager stands up and maintains
   **collaborative practice agreements (CPAs)**, manages the collaborating-physician bench, runs primary-source
   verification, and resolves a **pharmacy refusing to fill an NP-written prescription in a given state** —
   the exact failure the state licensure gate in Visitron cannot prevent. `[LIVE Ashby "Manager, NP Licensure
   & Collaborative Practice", 2026-10-19]`

---

## 4. Feature flags and experiments on the GLP-1 path

| Flag key | Surface | Effect on this journey | Evidence |
|---|---|---|---|
| **`gen-ai-prewarm-copilot`** | `/ai` entry | When on, forces `/ai/health-records` first, pulls Metriport/Particle records, injects a review into copilot context — changes how a GLP-1-curious patient is routed and what the clinician inherits | `[CODE hooks/useGeneralAiStatus.tsx:58]`; `[DOC architecture.md §10.8]` |
| **`sku-sticky-cta-refresh`** | SKU detail page (all GLP-1 SKUs) | Sticky vs inline CTA; tracked to Datadog and Segment as `sku_page_cta_clicked`/`SKU_STICKY_CTA_VARIANT` | `[CODE components/Refresh/SkuPages/SkuBodyRenderer.tsx:82-88]` |
| **`message-center`** | Authenticated dashboard | Server-evaluated; when on, every "contact support" action routes to `/account/messages` instead of SMS and rewrites copy ("Text our care team" → "Message our care team") | `[CODE contexts/AccountDashboardFeatureContext.tsx:26]`; `[CODE hooks/useContactSupport.ts:10, 55-95]` |
| **SKU-page refresh flag** (key name not recovered) | SKU PDP renderer | Chooses refresh vs legacy PDP; emits `SKU_PAGE_VARIANT` with `flag_key`, anonymous id, Datadog session id | `[CODE components/Refresh/SkuPages/SkuBodyRenderer.tsx:58-73]` |
| **Dashboard 1.0 vs 1.5** | Account dashboard | Two dashboards measured separately (`DASHBOARD_1_0_*`, `DASHBOARD_1_5_*`) — the surface a 4-week patient sees next steps on | `[CODE packages/mantine-theme/analytics/events.ts:21-25]` |
| **Email-validation variant** | Signup | `EMAIL_VALIDATION_VARIANT` experiment | `[CODE packages/mantine-theme/analytics/events.ts:18]` |
| **Tenant capabilities** (not LD) | `tenants/walmart`, `tenants/mercy` | White-label overrides, custom SKUs, S3 uploads, Trust Commerce billing | `[DOC general-medicine-executive-briefing.md §Section 03/04]` |

Delivery details worth copying into a tech plan: LD context key is `patientId` from localStorage or an anonymous
UUID cookie; `sendEventsOnlyForVariation: true` and `useReport: true`; every flag evaluation is mirrored into
Datadog RUM via `addFeatureFlagEvaluation`, so a session replay can be filtered by flag state.
`[CODE contexts/LaunchDarklyContext.tsx:51-66]` `[CODE lib/launchDarklyFlagEvaluator.ts:88-95]`

**Rollout modes are a Catalog field, not a flag:** SKUs carry `enabled` / `dark_launched` (live example: `400020`
Prescription Renewal Visit is `dark_launched`). CarePathways have draft → published → archived. `[DATA extracted_care_plans_and_journeys.json → 400020.state]` `[DATA catalog.rtk_endpoints: publishCarePathway, returnCarePathwayToDraft, archiveCarePathway, updateSkuIndexable]`

### 4.1 Analytics and data stack (how this path is instrumented)

```
Browser ──┬─ Segment JS ──► evs.segment.generalmedicine.co (CNAME proxy) ──► Segment ──► destinations
          ├─ Datadog RUM  ──► sessions, replay, API latency, flag evals
          └─ POST /api/external/analytics/{visits,events} ──► Nile-Core
Nile-Core ──► Databricks (warehouse) ──► Metabase dashboards (SSO/LDAP)
Nile-Core ──► PostSessionReplayNotificationJob ──► Slack #session-replay-new-storefront
```

| Layer | Observed | Evidence |
|---|---|---|
| **Product analytics** | Segment, production write key `vge0rHGUKWaaZtBNN0j9bpQBolWSCE3W`, CDN proxied through `evs.segment.generalmedicine.co`; staging key also embedded in the bundle | `[CODE packages/mantine-theme/analytics/segment.ts:14-19]` |
| **Event taxonomy** | Prefixed namespaces: `sf_` storefront (36 events), `ns_` next-steps/care-plan (17), `ssi_` insurance; a separate 18-event `SegmentEvents` set covers copilot, ID verification, SKU/email experiments, dashboard versions and message center | `[CODE packages/mantine-theme/analytics/events.ts:1-90]`; `[CODE packages/mantine-theme/analytics/segment.ts:52-75]` |
| **First-party analytics API** | `/api/external/analytics/visits` and `/api/external/analytics/events` — server-side ingestion alongside Segment | `[CODE lib/api/endpoints.ts:291-294]` |
| **Observability** | Datadog RUM: session replay, `addAction('nile-core-api-request', …)` with normalized route + status + latency, `addFeatureFlagEvaluation`, `setViewContextProperty` | `[CODE lib/api/metrics.ts:1-90]`; `[CODE lib/launchDarklyFlagEvaluator.ts:88-95]`; `[CODE components/Refresh/SkuPages/SkuBodyRenderer.tsx:45-49]` |
| **Warehouse** | **Databricks** advertised as a Metabase engine (host field, `adb-xxxxx.azuredatabricks.net` helper text) | `[LIVE https://analytics.staging-nile.com/, 2026-10-19]` |
| **BI** | **Metabase v1.63.10** (build 2026-08-11), unauthenticated at the edge, `enable-sso-ldap: true`, Google auth enabled; 133,097-byte bootstrap | `[LIVE https://analytics.staging-nile.com/, 2026-10-19]` |
| **Ops alerting** | Rails job posts session replays to Slack `#session-replay-new-storefront`, deduped to one post per user per hour; guest RUM sessions tagged with the same anonymous id Segment sends | `[DOC architecture.md §8.7]` |
| **Attribution** | Campaign-touch attestation token at the edge; UTM handling for LillyDirect landings; `attribution/campaignTouch.ts` | `[CODE app/store/AdLandingTracker.tsx:32]`; `[CODE lib/attribution/campaignTouch.ts]`; `[DOC architecture.md §4]` |

Gaps that matter for the technology plan: no event schema, no dbt/lineage, no warehouse table names, and no
retention/consent configuration were recovered. The Metabase instance is **publicly reachable without auth**, which
means the BI layer's access control is the first thing to fix if this stack is copied. `[INFERENCE from live 200 response]`

---

## 5. Money — every price, fee and refund rule on this path

### 5.1 Patient-facing prices (observed today)

| Item | Price | Evidence |
|---|---|---|
| GLP-1 medication evaluation (300543) — video cash / insurance / chat cash | **$100.00 / $25.00 / $30.00** | `[DATA extracted_care_plans_and_journeys.json → 300543.sku_options]` |
| Weight Loss Follow Up (300550) — video cash / insurance / chat cash | **$100.00 / $25.00 / $30.00** | `[DATA … → 300550.sku_options]`; `[LIVE /s/weight-loss-follow-up/300550, 2026-10-19]` |
| Zepbound (201448) — video / insurance / chat | $100.00 / $25.00 / $30.00; `allows_refill:true` | `[DATA … → 201448]` |
| Lilly storefront GLP-1 cash drug price | **Foundayo from $149/mo; Zepbound from $299/mo** (pen and KwikPen); "$25 typical visit copay"; "no subscription" | `[LIVE https://generalmedicine.co/p/lilly/weight-loss, 2026-10-19]` |
| Zepbound strengths / supply | 2.5, 5, 7.5, 10, 12.5, 15 mg/0.5 mL; **28-day supply**; "Available after visit" | `[LIVE https://generalmedicine.co/s/zepbound-tirzepatide/201448, 2026-10-19]` |
| Baseline lab — CMP (500008) | **$48.29** Quest / $48.92 Labcorp / $53.25 BioReference (cash); `insurance_orderable: true`, `insurance_price: null` | `[LIVE /api/external/lab_skus/500008, 2026-10-19]` |
| Baseline lab — A1c (500041) | **$47.64** Quest / $48.28 Labcorp / $48.71 BioReference | `[LIVE /api/external/lab_skus/500041, 2026-10-19]` |
| Baseline lab — TSH panel (500093) | **$54.10** Quest / $73.08 Labcorp / $57.18 BioReference | `[LIVE /api/external/lab_skus/500093, 2026-10-19]` |
| Baseline lab — Lipid panel (500114) | **$49.90** Quest / $51.04 Labcorp / $45.10 BioReference | `[LIVE /api/external/lab_skus/500114, 2026-10-19]` |
| Baseline lab — Lipid panel plus (500012) | **$62.90** Quest / $76.57 Labcorp / $71.54 BioReference | `[LIVE /api/external/lab_skus/500012, 2026-10-19]` |
| 12-biomarker Annual Wellness (500001) | **$129.92** Quest; $196.05 Labcorp; $202.82 BioReference; not insurance-orderable | `[LIVE /api/external/lab_skus/500001, 2026-10-19]` |
| Total Health Panel (500049) | **$114.00** Quest; **$180.13 Labcorp; $186.90 BioReference** | `[LIVE /api/external/lab_skus/500049, 2026-10-19]` |
| Lilly store lab insight panel | "$30 typical copay" for A1c/lipids/thyroid/kidney/liver | `[LIVE /p/lilly/weight-loss, 2026-10-19]` |
| Cancellation / no-show fee | Applies to video visits only (`video_visit`, `video_visit_with_specialist_consultation`), inside 24 h; **$25.00 fallback** when the amount is null; Medicaid patients exempt | `[CODE utils/cancellationFeeUtils.ts:6, 22, 26-28]`; `[LIVE /cancellation-policy, 2026-10-19]` |
| Lab refund | Full refund if sample not collected, up to **6 months** from purchase; **"No Fees Deducted … no administrative or physician order fees"**; same across lab partners | `[LIVE /refund-policy-labs, 2026-10-19]` |

Fee triggers that exist as first-class reasons (useful for an ops taxonomy): `late_customer_cancel`, `no_show`,
`no_show_predicted`, `no_confirmation`, `no_id_verification`, `missing_insurance`.
`[CODE utils/cancellationFeeUtils.ts:44-49]`

### 5.2 The lab fee stack (the one place GM's cost structure is partly visible)

Every sampled lab SKU satisfies `total_cash_price = test_price + clinical_fee`, where `clinical_fee` is **$15.00**
and `clinical_fee_enabled` is true. The payload also carries three fields that are **not** in the total:
`platform_fee`, `order_fee`, `collection_fee`.

| SKU | test_price (Quest) | clinical_fee | platform_fee | order_fee | collection_fee (Q/L/B) | total_cash_price |
|---|---|---|---|---|---|---|
| 500041 A1c | $32.64 | $15.00 | $15.00 | $6.00 | $9 / $5 / $3 | $47.64 |
| 500008 CMP | $33.29 | $15.00 | $15.00 | $6.00 | $9 / $5 / $3 | $48.29 |
| 500093 TSH | $39.10 | $15.00 | $15.00 | $6.00 | $9 / $5 / $3 | $54.10 |
| 500001 Annual Wellness | $114.92 | $15.00 | $40.00 | $6.00 | $9 / $5 / $3 | $129.92 |
| 500049 Total Health | $99.00 | $15.00 | $24.08 | $6.00 | $9 / $5 / $3 | $114.00 |

`[LIVE /api/external/lab_skus/{500041,500008,500093,500001,500049}, 2026-10-19]`

Two $30 constants in client code are **fallbacks only**, not live prices: `STANDARD_CLINICAL_FEE = 30` (used when the
API omits `clinical_fee`) and `ESTIMATED_INSURANCE_COPAY = 30` (shown when the insurance price is unavailable — which
is exactly the state of every sampled lab SKU, whose `insurance_price` is `null`). The Lilly storefront's "$30 typical
copay" for labs is consistent with that estimate, not with a quoted benefit.
`[CODE types/enums/lab-pricing-constants.ts:8, 13]` `[CODE components/LabPanelPricing.tsx:69]`

Interpretation limit: the arithmetic is observed; **who keeps `platform_fee`/`order_fee`/`collection_fee` is not
observed**. They may be GM take, partner pass-through, or a mix. Do not book them as gross margin.
`[INFERENCE]`

### 5.3 What GM actually earns on this path

| Layer | Figure | Status |
|---|---|---|
| Visit platform fee | $25–$100 per visit (insured vs cash), $30 chat | Observed price; **take rate unknown** |
| Lab clinical review fee | **$15.00** per lab order | Observed |
| Pharmacy routing | `gm_card` pre-pay (BIN/PCN/group/member), price comparison across 50K+ pharmacies | Mechanism observed; **margin unknown** |
| Drug cash price | Foundayo $149/mo, Zepbound $299/mo (manufacturer-set direct hubs) | Observed |
| Enterprise licensing | Walmart Better Care Services, Mercy white-label | Contract terms undisclosed |
| Company revenue | "early revenue … $1M–$5M" mid-2025 | Third-party estimate, low-medium confidence |
| Total equity | $152M ($32M May 2025 + $120M Oct 2026); post-money $669.68M | Third-party private-market estimate; **no SEC Form D** |

`[LIVE /p/lilly/weight-loss, 2026-10-19]` `[DOC alphasense-company-primer-2026-10-19.md §Revenue Model, §Estimated Revenue Trajectory, §Funding]`
`[DOC adversarial-and-market-report.md §3.2, §3.4]`

### 5.3.1 Clinician labor per visit (the missing cost line)

The postings make the variable labor cost of a visit explicit for 1099 clinicians:

| Visit type | Patient price (cash) | Physician fee | NP fee | Contribution before platform/overhead |
|---|---|---|---|---|
| Video | **$100** | **$50** | **$35** | 50% (MD) / 65% (NP) |
| Chat | **$30** | **$12** | **$8** | 60% (MD) / 73% (NP) |
| Video, insured | **$25 copay** | **$50** | **$35** | **−$25 (MD) / −$10 (NP)** before payer reimbursement |
| Chat, insured | **$25 copay** | **$12** | **$8** | +$13 (MD) / +$17 (NP) |

`[LIVE Ashby 1099 physician and NP postings, 2026-10-19]` `[DATA 300543/300550 sku_options]`

For a full-time physician, the posting's own throughput rules give a loaded labor cost per visit: 36 clinical
hours/week at 9 video visits per 4-hour block = **81 visits/week**; at 48 working weeks that is **3,888
visits/year**, so a $180k–$200k salary implies **≈$46–$51 of physician labor per visit** before malpractice,
credentialing, benefits or no-show slack. `[INFERENCE: arithmetic on the posted salary, hours and 20-min/9-per-block
cap; assumes full video density and 48 working weeks]`

That is the number to beat: a $100 cash video visit clears it; a $25 insured video visit does not, unless the payer
pays the claim. The posted **$50 contractor video fee sits almost exactly on the salaried per-visit cost**
($46–$51), which is what you would expect if the 1099 rate was priced off the employed model. `[INFERENCE]`

### 5.4 Unit-economics benchmarks (not GM actuals)

AlphaSense explicitly reports that mature membership/revenue/CAC metrics **are not available** for General
Medicine because it is early post-funding. Everything below is sector benchmark, not company disclosure.

| Metric | Benchmark | Source |
|---|---|---|
| CAC, digital primary care | **$75–$400** | `[DOC alphasense-company-primer-2026-10-19.md §Digital Health Unit Economics Benchmarks]` (neolytix) |
| CAC, direct primary care | **$82–$400**; solo primary care $145–$185; premium concierge $265–$487 | `[DOC alphasense-operational-financial-analysis-2026-10-18.md §Unit Economics, CAC, and LTV Metrics]` (macbach) |
| Cost per lead, clinicians | **$56.83**, 19% below cross-industry average | `[DOC alphasense-company-primer-2026-10-19.md §Digital Health Unit Economics Benchmarks]` (imatrix) |
| Healthcare search ad conversion | **~11.6%** | `[DOC alphasense-company-primer-2026-10-19.md §Digital Health Unit Economics Benchmarks]` |
| Revenue per virtual visit | **$150–$300** → payback on first visit | `[DOC alphasense-company-primer-2026-10-19.md §Digital Health Unit Economics Benchmarks]` (medicalppcagency) |
| LTV, insurance-based primary care | **$2,000–$6,000**; LTV:CAC 7.5:1–20:1 | `[DOC alphasense-company-primer-2026-10-19.md §Digital Health Unit Economics Benchmarks]` (neolytix) |
| DPC tenure / LTV:CAC | **5.2 years**, **15.7x**; concierge 6.8 years, 19.4x; traditional PCP 3.0x | `[DOC alphasense-operational-financial-analysis-2026-10-18.md §Unit Economics, CAC, and LTV Metrics]` |
| Lab operating cost structure | reagents/consumables **~80%** of direct cost, labor **~20%**; Quest/Labcorp cost base **~20% lower** than hospital labs | `[DOC alphasense-operational-financial-analysis-2026-10-18.md §Lab Costs and Operational Infrastructure]` |
| At-home kit failure rate | **5–10%** sample failure | `[DOC alphasense-operational-financial-analysis-2026-10-18.md §Lab Costs and Operational Infrastructure]` |
| SG&A drag for owned-clinic "high flyers" | **30–40% of revenue** | `[DOC alphasense-operational-financial-analysis-2026-10-18.md §Lab Costs and Operational Infrastructure]` |
| GM capital per employee | **$1.46M** ($152M / 104) | `[DOC adversarial-and-market-report.md §3.4]` |
| Retention, established luxury primary care | **>95% year-on-year** in established luxury networks; **82% overall**, with the first year the weak point | `[DOC alphasense-operational-financial-analysis-2026-10-18.md §Membership, Revenue, and Unit Economics]` |
| Retention lift from structured workflows | **+64% median** patient LTV | `[DOC alphasense-company-primer-2026-10-19.md §Digital Health Unit Economics Benchmarks]` |

**Illustrative per-patient economics for this journey (`[INFERENCE]`, arithmetic on observed prices and posted
clinician fees only):** T+0 eval + one monthly follow-up + baseline labs (one to three panels).

| Path | Revenue | Clinician fees | Visit contribution | Labs |
|---|---|---|---|---|
| Cash, MD | $100 + $100 + $47.64–$129.92 | $50 + $50 | **≈$100** | $47.64–$129.92 revenue, cost unknown |
| Cash, NP | same | $35 + $35 | **≈$130** | same |
| Insured, MD | $25 + $25 + lab rail | $50 + $50 | **≈−$50** before payer reimbursement | `insurance_price: null` on every sampled lab SKU |
| Insured, NP | $25 + $25 + lab rail | $35 + $35 | **≈−$20** before payer reimbursement | same |

The sampled lab SKUs return `insurance_price: null`, so the insured lab price is not observable from the public
payload; `500008` is `insurance_orderable: true`, while `500001` and `500049` are not. Payback-on-first-visit is
plausible on the cash path and negative on the insured path at a $25 copay unless the payer pays the claim.
This is arithmetic on GM's own price list and its posted contractor fees, not a GM disclosure.

---

## 6. Failure modes and branches

| Failure / branch | What the system does | Who acts | Evidence |
|---|---|---|---|
| **BMI below the FDA threshold** | FDA criteria: BMI ≥30, or ≥27 with a weight-related condition. Below 27 the guide calls treatment off-label, not insurance-covered, "considered case by case"; 27–29.9 without a qualifying condition does not meet the stated criteria | Clinician in Visitron | `[LIVE GLP-1 guide, 2026-10-19]` |
| **Contraindication** (MTC/MEN2 history, pancreatitis, gastroparesis, eating disorder, pregnancy) | Not a hard system block in the storefront; screening questionnaire + clinician review decides | Clinician | `[LIVE GLP-1 guide, 2026-10-19]`; `[DATA … → 300543.screening_questionnaire_id]` |
| **Patient state not covered by a licensed clinician** | State confirmed per booking; server filters the pool. If empty → no slots (frontend has a `booking_no_availability` event) | None; patient sees no availability | `[CODE lib/routes/skuBooking.ts:57]`; `[CODE packages/mantine-theme/analytics/events.ts:84]` |
| **No insurance / Medicaid** | Separate price rails; Medicaid excluded from cancellation fees; `booking_no_medicaid_option` event exists | Billing | `[LIVE /cancellation-policy, 2026-10-19]`; `[CODE packages/mantine-theme/analytics/events.ts:83]` |
| **Prior auth denied** | DoseSpot/Arrive RTPB returns PA requirement; GM markets "we even take care of prior authorizations"; guide says PA "generally takes several business days — your care team will follow up" | RCM / care team. RCM Director/VP scope explicitly includes **authorization** and denials; no dedicated PA queue appears in the four portals | `[CODE lib/hardcoded-data/CategoryPages/medications.ts:43]`; `[LIVE GLP-1 guide, 2026-10-19]`; `[LIVE Ashby RCM Director/VP, 2026-10-19]`; `[INFERENCE on which system executes PA work]` |
| **Drug out of stock / pharmacy cannot fill** | Order state stays at `order_placed`; `can_switch_pharmacy`, `can_choose_new_pharmacy`, `can_cancel_pharmacy_order` flags drive the UI; patient can switch | Patient + care team | `[CODE lib/store/api/pharmacyOrderApi.ts:17-21]` |
| **Lab result abnormal** | Clinician reviews results; "a member of our clinical team will reach out to you directly" | Clinician / care team | `[LIVE labs FAQ, 2026-10-19]` |
| **Lab refund requested pre-collection** | Full refund, no fee deduction, 6-month window; request by SMS 98528 or support@navhealth.co | Support | `[LIVE /refund-policy-labs, 2026-10-19]` |
| **No-show / late cancel** | Fee trigger set; fee applies only to video visits; $25 fallback; Medicaid exempt | Billing | `[CODE utils/cancellationFeeUtils.ts:6, 22, 26-28, 44-49]` |
| **ID verification fails / not confirmed** | `no_id_verification` is a cancellation-fee trigger; Visitron has manual attestation | Clinician | `[CODE utils/cancellationFeeUtils.ts:48]`; `[DATA … → visitron.rtk_endpoints: attestManualIdVerification]` |
| **Care-plan AI generation fails or is rejected** | Plan sits in `awaiting_generation` or goes back through `rejectReview` with an explanation; unknown states fall to "Updates" in the patient UI rather than History | Clinician | `[CODE components/CareActions/careActionSection.ts:64-72, 104-113]` |
| **Clinician rejects an ACR** | `rejectReview` + `updateClinicianReviewRejectionExplanation`; can escalate to a synchronous follow-up video (SKU 300591) | Clinician | `[DATA … → visitron.rtk_endpoints]`; `[DOC architecture.md §10.3]` |
| **Chat queue overloaded / after hours** | `availability_promise` is a timestamp, not "now"; support is Mon–Fri 9–9 ET; clinician chat shifts can be extended | Clinician / support | `[DATA … → 300543.sku_options.top_fields.availability_promise]`; `[CODE components/Refresh/SupportWidget/SupportWidget.tsx:146]`; `[DATA … → visitron.rtk_endpoints: extendChatShift]` |
| **Refill vs renewal** | Refill = direct order; renewal = clinician review (new prescribing decision), "most … same day" | Clinician | `[LIVE medication FAQ, 2026-10-19]` |
| **Referral queue disabled** | `QueueEnabled` / `/promises/queue/enabled` can turn the concierge off globally without a deploy | Ops | `[DATA … → referrals.rtk_endpoints: getQueueEnabled]` `[DOC architecture.md §8.4]` |
| **No referral on this path** | GLP-1 SKUs set `allows_referral:false`, so the promises queue is **not** on the GLP-1 medication path | — | `[DATA … → 300543.modalities, 300550.modalities, 201448.modalities]` |

---

## 7. Corrections to the existing dossiers (use these, not the deck)

1. **The deck's Referrals appendix invents operations.** Appendix F.03 lists `getPromisesQueueEnabled`,
   `setPromisesQueueEnabled`, `getAssignees`, `updateAssignee`, `getOrders`, `getAllOrders`, `updateNotes`,
   `getDocuments`, `getDocumentDownloadUrl`. The underlying dataset contains none of these; it has
   `claimPromise`, `dropOrSnoozeOrder`, `getQueueEnabled`, `getOrder`, `getReferrals`, `listAllOrders`,
   `recordOutreach`, `logOutcome`, `endShift`. `[DOC briefing slide 320 vs DATA internal_portals_recon.json → referrals.rtk_endpoints]`
2. **"2-day referral promise" is unsupported.** No SLA value exists in the dataset or in any live page.
   `[DOC briefing slide 51]`
3. **Same problem in the ACR table:** `presignAcrFileUpload`, `confirmAcrFileUpload`, `completeClinicianReview`,
   `preassigned` appear in the deck but not in the portal inventory, which has
   `getAsyncClinicalReviewFileUploadPreview`, `submitAsyncClinicalReview`, `getClinicianReviews`, `applyReview`,
   `rejectReview`. `[DOC briefing slide 320 vs DATA visitron.rtk_endpoints]`
4. **"38-state" is 37 in the client.** `careActionSection.ts` explicitly maps **37** states. The 38th state, if it
   exists, is not in the recovered client map. `[CODE components/CareActions/careActionSection.ts:40-92]`
5. **"Three Friendly PCs" understates the public entity set.** The current privacy notice (Sep 2, 2026) names
   **five** PCs by state — Emerald P.C. (AZ), Emerald P.A. (KS), Emerald P.C. (CA), Centennial P.C. (ID),
   Ruby P.C. (NJ) — which only overlaps NPPES on New Jersey. NPPES recovered four NPIs in **UT, FL, NJ**.
   `[LIVE /privacy-policy, 2026-10-19]` `[DATA infra_npi_catalog.json → npi.friendly_pc_assessments, npi.organizations]`
6. **The lab-fee appendix is stale on 500049.** Deck says $114.00 at Quest, Labcorp and BioReference; live shows
   $114.00 Quest, $180.13 Labcorp, $186.90 BioReference. `[DOC briefing slide 324]` `[LIVE lab_skus/500049, 2026-10-19]`
7. **Four legal names are in concurrent use.** Code footer: "Path Healthcare Systems Corp."
   (`components/Footer.tsx:182`); live pages: "© 2025 General Medicine Inc."; privacy notice: Path Healthcare
   Systems Corp.; refund page support address: `support@navhealth.co`. Pick one for contracts, BAAs and consent copy.
   `[CODE components/Footer.tsx:182]` `[LIVE /refund-policy-labs, /support, /privacy-policy, 2026-10-19]`
8. **The `POST /support` page is partly unrendered.** It serves a literal `{heroText?.descriptionLeft}` string in
   production HTML. `[LIVE https://generalmedicine.co/support, 2026-10-19]`
9. **Self-correction on this file's own inference.** I first inferred "no help-desk SaaS" from the patient bundle.
   That was too broad: the FT physician posting names **Zendesk** as the chat platform. The bundle absence is real;
   the conclusion about the operating stack was not. Vendor claims must be checked against postings, not just code.
   `[LIVE Ashby FT Physician, 2026-10-19]`

---

## 8. Compliance artifacts (what exists, what is only implied)

| Artifact | Status | Evidence |
|---|---|---|
| **LegitScript certification** | Badge + outbound link to `legitscript.com/websites/generalmedicine.co` in desktop and mobile footer | `[CODE components/Footer.tsx:164, 269]` |
| **HIPAA Notice of Privacy Practices** | Live, last updated **2025-11-03**, issued by the five Emerald/Ruby/Centennial PCs collectively | `[LIVE /hipaa-privacy-practices, 2026-10-19]` |
| **Privacy notice** | Live, last updated **2026-09-02**, Path Healthcare Systems Corp.; states PHI is handled under **Business Associate Agreements "that General has in place with Emerald Providers"**; discloses third-party business associates for technology, transcription, billing, collection | `[LIVE /privacy-policy, 2026-10-19]` |
| **Telehealth consent** | `/vital-informed-consent` names 15 delegated professional entities across 3 networks | `[LIVE /vital-informed-consent, 2026-10-19]` |
| **Vendor BAAs** | **Not observed.** No BAA text, subprocessor list, or trust page recovered. The vendor set implies BAAs with Athena, DoseSpot, Nabla, Metriport, Particle, Junction/Vital, Arrive, Waystar, Twilio, Segment, Datadog, LaunchDarkly, Databricks, Metabase, Sanity, Persona, Stripe, Trust Commerce | `[INFERENCE from vendor list; DOC architecture.md §3]` |
| **Pharmacy licensure** | No license-verification UI or field recovered. Routing is by NCPDP provider id; public claim is 50,000+ pharmacies | `[CODE lib/api/endpoints.ts:339-350]`; `[DOC adversarial-and-market-report.md §3.2]` |
| **NP collaborative practice agreements (CPAs)** | A dedicated manager maintains state- and provider-specific CPA templates, a collaborating-physician bench, primary-source license verification, and state filings; handles escalations such as a pharmacy refusing an NP-written script | `[LIVE Ashby "Manager, NP Licensure & Collaborative Practice", 2026-10-19]` |
| **Prescriber licensure** | State-per-booking gate; 50-state + DC claim; 18+; server-side clinician filter. Coverage is achieved by overlapping multi-state licenses (each NP ≥30% of US population), not one license per state per clinician | `[CODE lib/routes/skuBooking.ts:57]`; `[DOC architecture.md §10.6]`; `[LIVE Ashby postings, 2026-10-19]` |
| **CPOM / Friendly PC** | Utah PC has a **non-physician President** (Ashwin Muralidharan, CPO); FL and NJ PCs have the Medical Director as President; all four NPIs share phone `307-840-9834` and a Park City mailing address. CA SB 351 / OR SB 951 tighten MSO control from 2026-01-01 | `[DATA infra_npi_catalog.json → npi.organizations]`; `[DOC adversarial-and-market-report.md §1.4]` |
| **PCI** | Stripe and Trust Commerce (Mercy guarantor), both PCI-scoped | `[DOC architecture.md §1.1, §4]` |
| **Identity** | Persona KYC with webhook `inquiry.completed`; manual attestation available in Visitron | `[DOC architecture.md §5.1]`; `[DATA … → visitron.rtk_endpoints]` |

**Entity map as it affects prescribing:** MSO (Path Healthcare Systems Corp.) owns the platform and employs the
non-clinical staff; state PCs are the contracting/provider entities; the delegated networks (OpenLoop, Reliant.MD,
MECNB) cover states where the PCs are not. A clinician may prescribe only where licensed, which is enforced by the
server-side pool filter on `appointment_location_state`. `[LIVE /privacy-policy, /vital-informed-consent, 2026-10-19]` `[CODE lib/api/endpoints.ts:132-135]`

---

## 9. Staffing numbers (observed vs estimated)

| Fact | Value | Confidence | Evidence |
|---|---|---|---|
| Employees | **~104** | Third-party estimate (PitchBook) | `[DOC adversarial-and-market-report.md §3.2]` |
| Operating hubs | **3** — San Francisco HQ (128 Spear St), Cambridge/Boston, Salt Lake City/Park City (clinical panel) | Deck, consistent with live HQ address | `[DOC briefing deck slide titled "~104 staff, 3 hubs, 16 open roles"]`; `[CODE components/Footer.tsx:47-48]` |
| Open roles | **16** (live today): **Clinical 4** (2 physician + 1 NP + the NP licensure manager), Ops 2 (Care Coordinator, Clinical Recruiter), Engineering 4, Product 3, Finance 3 (Strategic Finance + RCM Director + RCM VP) | High | `[LIVE api.ashbyhq.com/posting-api/job-board/general-medicine, 2026-10-19]` |
| Public clinician/leadership bios | **20** pages | High | `[DATA live_sitemap_skus.json → bio_pages]` |
| Public LinkedIn profiles | **87** (13 engineer, 8 clinician, 7 executive, 7 designer, 2 advisor, 1 ops, 1 product, 48 unknown) | Medium — search snippets only | `[DATA linkedin_team_and_posts.json → role_breakdown, coverage_note]` |
| Capital per employee | **$1.46M** | Arithmetic | `[DOC adversarial-and-market-report.md §3.4]` |
| Clinician model | 1099 NP/MD panel + FT primary care MD; "community of practice", partner logos Mass General Brigham, UCSF, Stanford, Kaiser | High | `[LIVE Ashby board]`; `[CODE lib/hardcoded-data/ClinicianHiring/clinician-hiring-content.tsx:53-72]` |
| Clinician pay | **1099: $50/video + $12/chat (MD); $35/video + $8/chat (NP)**. **FT: $180k–$200k + equity + malpractice tail + credentialing** | High — posted | `[LIVE Ashby 1099 and FT postings, 2026-10-19]` |
| Clinician hours | FT **36 clinical hrs/wk**; 1099 MD **10–12 hrs/wk**; 1099 NP **12–15 hrs/wk** | High — posted | `[LIVE Ashby postings, 2026-10-19]` |
| Licensure coverage requirement | Each NP must cover **≥30%** of the US population; each 1099 physician **20–30%**; CA/TX/FL/NY preferred | High — posted | `[LIVE Ashby postings, 2026-10-19]` |
| Care coordinator pay | **$22/hr**, full-time, remote | High — posted | `[LIVE Ashby "Care Coordinator", 2026-10-19]` |

No number exists anywhere for the size of the **care-coordinator**, **support**, **RCM**, or **pharmacy/PA**
teams. That is the single biggest staffing unknown for the operating plan. No pharmacy-technician, prior-auth
specialist, or benefits-verification role appears among the 16 open requisitions, and no such queue exists in the
four portals — so if that function exists today it is either unadvertised or absorbed into the clinician and RCM
roles. `[LIVE Ashby board, 2026-10-19]` `[INFERENCE from absence]`

---

## 10. Open questions

1. **Prompt text and model vendor** for Promptimus/Provider Chat/copilot — `getModelVersions` proves enumeration,
   not contents. No recovered artifact contains a prompt.
2. **Which LLM drafts the care plan** (`generateNextStepReview`) and whether the same model serves the patient
   copilot. Two ActionCable channels (`ExternalCable`/`InternalCable::CopilotChannel`) suggest one substrate, not
   one model.
3. **CarePathway phase vocabulary** — `getCarePathwayPhaseKinds` and `getSkuPhaseTypes` exist; no values recovered.
   The "5-phase GLP-1 progression" is a **synthesis of the public guide + SKU prices**, not an observed pathway object.
4. **Whether a single named GLP-1 CarePathway exists** in Catalog and which SKUs it links (`updateCarePathwayLinkedSkus`).
5. **Referrals SLA.** No value anywhere. The queue is real; the promise is not documented.
6. **Prior-auth execution system.** RCM Director/VP scope explicitly includes authorization; the system the team works in is unknown, and no PA queue exists in the four portals.
7. **Support headcount and tooling split.** Patient support is first-party over Twilio/SMS; Zendesk is confirmed only as the clinician chat workspace. Whether Zendesk also backs patient support agents, and whether a BPO sits behind SMS 98528, is unknown.
8. **Coverage beyond the posted window.** 4–6 hour blocks, Mon–Fri 7 a.m.–7 p.m. MT are documented; when weekend/evening expansion goes live, and how the 3-hour chat window is enforced (Zendesk SLA vs Visitron), is not.
9. **BAA inventory and subprocessor list.** None published.
10. **Pharmacy licensure verification** and how a non-licensed pharmacy is excluded.
11. **Databricks workspace and event schemas.** Metabase advertises the driver; no table/lineage evidence.
12. **`platform_fee` / `order_fee` / `collection_fee` accounting treatment** — GM revenue, partner pass-through, or both.
13. **Take rate on pharmacy routing** and whether `gm_card` pre-pay is GM margin or a pass-through.
14. **Enterprise contract values** (Walmart, Mercy) and whether GLP-1 is in scope for either.
15. **How the 104-employee count splits** between clinical, ops, support, RCM and engineering.

---

## Handoffs to other evidence files

- **Cross-reference to evidence A (Acquisition):** the GLP-1 PDP is Catalog-authored; the CTA a visitor clicks is
  flag-controlled (`sku-sticky-cta-refresh`) and the ad landing path carries UTMs (LillyDirect note in
  `app/store/AdLandingTracker.tsx:32`). Any CAC/CPL claim must be labeled benchmark — GM discloses none.
- **Cross-reference to evidence B (First month):** the 37-state Care::Plan map and `careActionSection` are the canonical
  state vocabulary for what the patient sees; use them rather than the deck's "38".
- **Cross-reference to evidence C (Intake):** screening/history questionnaire UUIDs per SKU are in
  `extracted_care_plans_and_journeys.json`; questionnaire edits require a second-person approval
  (`requestQuestionnaireEdits` → `applyQuestionnaireEdit`).
- **Cross-reference to evidence D (Checkout/labs):** lab `total_cash_price = test_price + $15.00 clinical_fee`; the
  platform/order/collection fields are outside the total. Lab refunds are full, fee-free, 6-month, pre-collection.
- **Cross-reference to evidence E (Clinical/Rx):** DoseSpot + Arrive RTPB is the benefit/PA truth source at prescribing time;
  Athena is the system of record after signing. PA follow-through is human and sits outside the recovered portals.
- **Handoff to Main:** the executive deck's appendix tables contain operations that do not exist in the recon
  dataset (§7). If the final journey report is built from the deck, it will overstate the Referrals and ACR
  tooling. Build those two sections from `internal_portals_recon.json`.
