Hardcoded Data Models

The JS literals embedded in the prototype's /visit/ source define the v1 schema Deepak has in mind. Treat them as the planning surface for our DB models, not as demo dressing.

Patient roster (18 patients · 10 families · 1 doctor)

Provider: Yogini Prajapati MD. The Atlas account that backs this roster reports 81 active patients in production; the prototype demos 18.

Patient ID / Name	Family · Subscription	Age · DOB · Sex	Notes
sofia Sofia Martinez	Martinez · monthly $295 (Lisa Martinez)	7y · 03/14/2019 · F	Peanut allergy (severe), tree nuts (mod), Epi-pen Jr
(sib) Diego Martinez	Martinez	3y · 08/02/2022 · M	Eczema
Aisha Brock	Brock · monthly $295 (Janelle Brock)	5y · 11/22/2020 · F	NKDA
Ethan Brock	Brock	2y · 06/10/2023 · M	NKDA
Liam Chen	Chen · quarterly $885 (Wei Chen)	4y · 09/15/2021 · M	Asthma; dust mites/cat
Maya Chen	Chen	7y · 02/28/2019 · F	NKDA
jaden Jaden Patel	Patel · monthly $295 (Priya Patel)	8y · 01/04/2018 · M	ADHD · Concerta 36mg (CII)
Riya Patel	Patel	6y · 04/19/2020 · F	NKDA
Olivia Kim	Kim	10y · 07/30/2015 · F	NKDA · Well-visit 4 mo overdue
Theo Kim	Kim	12y · 03/02/2014 · M	NKDA
Noah Williams	Williams (new family · enrolling Thu)	3y · 05/12/2022 · M	Transfer record pending
Emma Williams	Williams	5y · 10/08/2020 · F	New patient
mai Mai Nguyen	Nguyen	1d · 05/03/2026 · F	TERM 39+2 · vaginal · 3.42 kg — HOME VISIT
Bao Nguyen	Nguyen	3y · 11/04/2022 · M	NKDA
Chiamaka Okafor	Okafor	14y · 09/12/2011 · F	PHQ-9 / GAD-7 due
mateo Mateo Rodriguez-Hayes	Rodriguez-Hayes	16y · 02/21/2010 · M	T1D pump+CGM · A1c 6.9
camila Camila Rodriguez-Hayes	Rodriguez-Hayes	9y · 08/25/2016 · F	Asthma EIA · PCN allergy
eli Eli Goldberg	Goldberg	11d · 04/23/2026 · M	TERM 38+5 · C-section · 3.21 kg — newborn 2-wk today

Schema implications baked into this roster:

• Family / Household is a first-class entity. Subscription tier (monthly $295, quarterly $885, annual implied) belongs to the Family, not the Patient. One Stripe customer per family, not per patient.
• Guardian linkage — each family row names a responsible parent (Lisa Martinez, Janelle Brock, Wei Chen, Priya Patel, Tasha Williams).
• Routing rule is hard-coded: HOME_VISIT_PATIENTS = new Set(['mai']). The newborn home-visit flow is selected by patient ID in the demo; in production it should derive from patient.age_in_days and an active home-visit encounter.
• Practice constants surfaced elsewhere: 81 active patients on the real panel, +20.9% YoY, $4,752 May MTD charges, 100% Stripe, 1 doctor, 0 in-house dispensing.

Note bank (5 SOAP notes · NOTE_DATA)

Each entry is { title, meta, status, statusLabel, confidence, readonly, signed, sections[] }. Each section is { label, tag, conf, html } where tag ∈ { 'AI', 'auto', 'user' } and conf ∈ { 'high', 'med', 'low' }.

ID	Title	Date	Status	Source
today-sick-draft	Sick visit · cough × 4 days	Today 5/4/2026	draft (94% confidence)	AI scribe
wcc-6y	Well-child · 6 year annual	6/12/2025	signed	Hand-written
eczema-followup	Follow-up · eczema flare	2/14/2026	signed	Hand-written
otitis-2024	Sick visit · ear infection	11/4/2024	signed	Hand-written
wcc-5y-K	Well-child · 5y kindergarten physical	7/22/2024	signed	Hand-written

The three-way provenance tag (AI / auto / user) is the v1 contract for note authorship. It enables:

• Per-section "View AI original" + "↻ Regenerate" buttons.
• Inline diff highlighting (<span class="ne-edited"> mint) when Yogini edits an AI section.
• Confidence chips ("High confidence" / "Medium · review") shown next to AI sections.

A signed note carries an signed audit string (e.g. "Electronically signed by Yogini Prajapati MD · 06/12/2025 at 11:42 AM · IP 192.168.1.84") and is readonly: true. Amendments to signed notes require an audit-logged amendment workflow.

Detail-drawer modals (8 entries · DETAIL_CONTENT)

Opened by openDetail(id) from anywhere on the chart. Slides in from the right. Each entry: { icon, cls, title, body }.

ID	Type	Subject
sms	Text thread	Lisa Martinez — "Sofia coughing 4 days"
appt	Appointment	Today 10:30 AM sick visit · 30 min · booked from SMS
note	Signed SOAP	Well-child 6/12/2025 (Sofia at 6y)
rx-epi	Rx detail	Epi-pen Jr 0.15 mg/0.15 mL · 1 refill · H-E-B Pharmacy · DoseSpot Rx ID 20250612-EPI-001
imm	Immunizations	4 vaccines from 6y well-child (DTaP #5 / IPV #4 / MMR #2 / Varicella #2) with lot numbers
icd-J069	ICD-10	J06.9 · Acute upper respiratory infection (with cross-references)
icd-R059	ICD-10	R05.9 · Cough
icd-Z00121	ICD-10	Z00.121 · Routine well-child exam

AI canned responses (6 entries · AI_RESPONSES)

Keyed by lowercased question (substring fallback). Define the quality bar for production Claude calls.

Trigger keyword	Response gist
what should i prioritize this morning?	Top-3 ranked: Sofia 10:30 sick visit · Camila refill · Eli newborn 2-wk + home-visit prep ready note
prep for sofia's 10:30 sick visit	Chart context recap → ask points → differential dx → flag well-visit overdue + Epi-pen 47-day expiry
draft replies to all unread messages	7 of 9 auto-drafted, HEEADSSS-confidential threads excluded, batch send option
anything overdue i should book today?	2 well-visits overdue · window-opening Rx alerts · Friday CII handling
drug interactions on sofia?	Cross-check active meds (Epi-pen Jr + Cetirizine 5mg) against likely cough Rx; warn diphenhydramine + pseudoephedrine
summarize this week	15 visits · 1 home visit · 1 prospect call · ~4.2 hrs AI scribe savings · MTD charges $4,752

All responses include <span class="ai-action" onclick="..."> deep-links back into the app (openPatient('sofia'), showView('rx'), etc). The contract: AI replies should be action-linked, not just informational.

AI panel context

updateAiContext() runs on every view change. The "I see X" pill below the assistant follows the user's screen automatically:

today    → Today screen · 4 visits · 2 AI flags
list     → Patients list · 18 patients · 10 families
calendar → Calendar · this week · 15 visits booked
chart    → Sofia Martinez chart · sick visit today 10:30
home     → Mai Nguyen home visit · 1 day old · birth history loaded
docs     → Documents · 8 unfiled · 79% AI auto-routed
inbox    → Inbox · 9 unread · 2 urgent · 7 AI replies drafted
rx       → Prescriptions · 2 refills awaiting sign-off
billing  → Billing · May $4,752 charged · MRR $2,863
reports  → Reports · clinical-quality · 78% Bright Futures
marketing→ Marketing · 47 leads · 2.1% conversion

Engineering implication: every screen must emit a view_context blob (currentView, filters, selectedPatientId, etc.) so the AI panel and ambient features can RAG against the same context the user sees.

Command palette (22 entries · CMD_ITEMS)

Each entry: { label, icon, action }. Triggered by ⌘K. Typing > switches to AI mode.

⌂  Today
👥 Patients (18)
📅 Calendar · this week
✉️  Inbox (9 unread)
📎 Documents (8 unfiled)
💊 Prescriptions (2 awaiting sign-off)
💳 Billing · May 2026
📊 Reports · clinical-quality
📣 Marketing · pipeline

🩺 Sofia Martinez · 7y · sick visit today
👶 Mai Nguyen · 1d · home visit Tue
👶 Eli Goldberg · 11d · newborn 2-week today
🩺 Mateo Rodriguez-Hayes · 16y · T1D follow-up Thu
💊 Camila Rodriguez-Hayes · 9y · refill pending
🩺 Jaden Patel · 8y · ADHD med review

🎙️  Start visit · AI scribe (on current chart)
💊 Send Rx (on current chart)
⚡ AI: draft note for current visit
⚡ AI: summarize this week

⚙️  Settings · Office Hours
⌨️  Keyboard shortcuts
🚪 Sign out

This is the v1 keyboard-shortcut surface. Pressing ? shows: ⌘K = command palette · g+t = today · g+p = patients · g+c = calendar · ESC = close.

State machine captions (STATE_CAPTIONS)

Office scribe setState(n):

n	Caption
1	Step 1 of 5 Sofia just arrived. Yogini reviews her chart — last visit, allergies, what's overdue.
2	Step 2 of 5 Yogini taps record. Whisper transcribes the conversation in real time.
3	Step 3 of 5 Recording stopped. Claude drafted a SOAP note in 18 seconds. Yogini reviews + edits.
4	Step 4 of 5 Note signed. Action plan fires: Rx routes via DoseSpot, follow-up task auto-creates, billing event logs.
5	Step 5 of 5 Lisa's phone buzzes. Plain-language summary in the parent app — what we did, what to do at home, when to call.

These read like spec text and are perfect for production tutorial overlays / onboarding tooltips.

Hardcoded summary stats

Sprinkled throughout — treat as v1 demo data, not configuration:

• Marketing: 47 leads / week · 2.1% conversion (industry avg DPC peds 1.4%).
• Inbox: 38% inbound auto-replied · 12 min avg response · under-1-hr SLA.
• Reports: 78% Bright Futures completion · 94% immunization compliance · 100% asthma action plans · 100% T1D A1c-in-target.
• AI time savings (per week): 32 min/day SOAP scribe + 8 min/day message triage + 3 min/day eRx + 30s × 5 refills = ~4.2 hrs.
• Billing (May 2026): $4,752 charged · $450 outstanding (Williams first charge fires Thu 5/7) · $0 overdue · 100% Stripe · MRR $2,863.
• ARPU: $286 (Starlight family-level) vs $264 Atlas kid-level.

Plan pricing (canonical from this prototype)

For a 2-child family. Reconcile the Onboarding rate ($140/mo for 2 kids) vs the Patient-list family rate ($295/mo) before committing — likely intro-pricing vs steady-state.

Cadence	2-kid rate	Per kid implied
Monthly	$140/mo	$70/mo
Quarterly	$385 / 3 mo ($128.33/mo)	$64.17/mo
Annual	$1,440/yr ($120/mo)	$60/mo

Implications for our codebase

Hardcoded entity	Implementation hook
Family + Guardian + Patient roster	Three-table schema, Family owns Stripe customer, Guardian owns auth, Patient is medical record. Split-billing flag on Family.
NOTE_DATA provenance	Section table per-note with tag, confidence, original_ai_version_id for diff/regenerate.
DETAIL_CONTENT modals	Common drawer renderer keyed by entity type; one query per type.
AI_RESPONSES	Production Claude prompt templates with chart-RAG; same action-link contract.
AI panel ctxByView	Per-view view_context emitter on the client.
CMD_ITEMS	Command palette feed = views + patients + AI commands; pluggable per role.
HOME_VISIT_PATIENTS set	Replaced by encounter.type === 'newborn_home_visit' check at chart open.
State-machine captions	Tutorial / onboarding overlay copy.