@ak--47/dungeon-master 1.5.3 → 1.6.0

package/dungeons/vertical/healthcare/healthcare.js

@@ -0,0 +1,1488 @@
+// ── IMPORTS ──
+import dayjs from "dayjs";
+import utc from "dayjs/plugin/utc.js";
+dayjs.extend(utc);
+import "dotenv/config";
+import * as u from "@ak--47/dungeon-master/utils";
+import * as v from "ak-tools";
+import { findFirstSequence, scaleFunnelTTC } from "@ak--47/dungeon-master/hook-helpers";
+/** @typedef  {import("../../../types").Dungeon} Config */
+
+// ── OVERVIEW ──
+/*
+ * NAME:       MedConnect
+ * APP:        Telehealth platform connecting doctors, nurses, and patients
+ *             through virtual consultations, prescriptions, and secure
+ *             messaging. Multi-role system; subscription tiers (free/basic/
+ *             premium); feature rollouts for video consultation and AI
+ *             symptom checker; geo-aware (US/EU/LATAM).
+ * SCALE:      10,000 users, ~1.2M events, 121 days (2026-01-01 → 2026-05-01)
+ * CORE LOOP:  sign up → symptom search → book appointment → consultation → prescription → follow-up
+ *
+ * EVENTS (18):
+ *   app session (8) > symptom search (7) > appointment booked (6) > notification received (6)
+ *   > consultation completed (5) > message sent (5) > prescription issued (4)
+ *   > health record accessed (4) > prescription refill (3) > follow up scheduled (3)
+ *   > lab results viewed (3) > payment processed (3) > insurance verified (2)
+ *   > provider rated (2) > profile updated (2) > account created (1)
+ *   > support ticket created (1) > account deactivated (1)
+ *
+ * FUNNELS (5):
+ *   - Onboarding Flow:          account created → insurance verified → symptom search → appointment booked (45%)
+ *   - Booking to Consultation:  symptom search → appointment booked → consultation completed (40%)
+ *   - Full Care Journey:        appointment booked → consultation completed → prescription issued → follow up scheduled (30%)
+ *   - Prescription Lifecycle:   prescription issued → prescription refill → payment processed (55%)
+ *   - Patient Satisfaction:     consultation completed → provider rated → follow up scheduled (25%)
+ *
+ * USER PROPS:  role, specialty, years_experience, preferred_language, has_chronic_condition, age_range, subscription_tier, Platform
+ * SUPER PROPS: subscription_tier, Platform
+ * SCD PROPS:   care_plan (preventive/routine/chronic/acute, monthly fuzzy, max 8)
+ * GROUPS:      none
+ */
+
+// ── HOOK STORIES ──
+/*
+ * NOTE: Cohort effects are HIDDEN — discoverable via raw-prop breakdowns
+ * (HOD, day, tier) or behavioral cohorts. One exception: H6 stamps
+ * no_show=true on flagged bookings (a realistic appointment-status
+ * property, and the only selection-free way to verify per-event thinning
+ * on an activity-selected cohort).
+ *
+ * ───────────────────────────────────────────────────────────────
+ * 1. AFTER-HOURS SURGE PRICING (event hook)
+ * ───────────────────────────────────────────────────────────────
+ *
+ * PATTERN: Consultations between 7PM-7AM (after-hours) have 1.5x
+ * higher consultation_fee. Simulates urgent care premium pricing.
+ *
+ * HOW TO FIND IT IN MIXPANEL:
+ *
+ *   Report 1: After-Hours Fee Premium
+ *   • Report type: Insights
+ *   • Event: "consultation completed"
+ *   • Measure: Average of "consultation_fee"
+ *   • Breakdown: hour of day
+ *   • Expected: hours 19-06 UTC ~ 1.5x avg fee vs hours 07-18
+ *     (after-hours ≈ $112, business ≈ $75)
+ *
+ * REAL-WORLD ANALOGUE: Telehealth platforms charge premiums for
+ * after-hours urgent consultations, a key revenue driver.
+ *
+ * ───────────────────────────────────────────────────────────────
+ * 2. FLU SEASON SPIKE (event hook)
+ * ───────────────────────────────────────────────────────────────
+ *
+ * PATTERN: During days 50-70 (flu season window), appointments
+ * with condition_type "respiratory" get 2x the wait_time and the
+ * condition is forced to "respiratory" 60% of the time.
+ *
+ * HOW TO FIND IT IN MIXPANEL:
+ *
+ *   Report 1: Flu Season Volume
+ *   • Report type: Insights
+ *   • Event: "appointment booked"
+ *   • Measure: Total
+ *   • Filter: condition_type = "respiratory"
+ *   • Line chart by week
+ *   • Expected: Clear spike during flu season window (days 50-70)
+ *
+ *   Report 2: Wait Time During Flu Season
+ *   • Report type: Insights
+ *   • Event: "appointment booked"
+ *   • Measure: Average of "wait_time_hours"
+ *   • Breakdown: "condition_type"
+ *   • Filter: time within flu season
+ *   • Expected: respiratory ~2x wait vs other conditions
+ *
+ * REAL-WORLD ANALOGUE: Seasonal illness creates predictable surges
+ * in appointment demand and wait times.
+ *
+ * ───────────────────────────────────────────────────────────────
+ * 3. EXPERIENCED DOCTOR SATISFACTION (everything hook)
+ * ───────────────────────────────────────────────────────────────
+ *
+ * PATTERN: Users who had >12 consultation events get ALL their
+ * satisfaction_scores redrawn uniform 4.0-5.0 (avg 4.5) vs the declared
+ * baseline weighNumRange(1,5,mode 3) ≈ 3.0. Simulates experienced
+ * doctors earning better reviews.
+ *
+ * HOW TO FIND IT IN MIXPANEL:
+ *
+ *   Report 1: Satisfaction by Consultation Volume
+ *   • Report type: Insights
+ *   • Event: "consultation completed"
+ *   • Measure: Average of "satisfaction_score"
+ *   • Breakdown: behavioral cohort (>12 consultations vs fewer)
+ *   • Expected: heavy consulters ~4.5 avg vs ~3.0 baseline; every one
+ *     of their scores sits in [4.0, 5.0]
+ *
+ * REAL-WORLD ANALOGUE: Experienced providers develop better bedside
+ * manner and patient communication skills over time.
+ *
+ * ───────────────────────────────────────────────────────────────
+ * 4. VIDEO CONSULTATION FOLLOW-UP LIFT (everything hook)
+ * ───────────────────────────────────────────────────────────────
+ *
+ * PATTERN: Each video-mode consultation has a 60% chance to inject one
+ * cloned "follow up scheduled" event 1-7 days later (stamped
+ * consultation_mode="video", fresh days_until_followup 3-14). Users
+ * without an existing follow-up to clone from are skipped.
+ *
+ * HOW TO FIND IT IN MIXPANEL:
+ *
+ *   Report 1: Follow-Up Rate by Consultation Mode
+ *   • Report type: Insights
+ *   • Event: "follow up scheduled"
+ *   • Measure: Total per user
+ *   • Breakdown: "consultation_mode" (from consultation completed)
+ *   • Expected: video-consult users carry ~+0.6 extra follow-ups per
+ *     video consultation vs phone-only users
+ *
+ * REAL-WORLD ANALOGUE: Face-to-face (video) consultations build
+ * stronger patient-doctor rapport, increasing follow-up compliance.
+ *
+ * ───────────────────────────────────────────────────────────────
+ * 5. CHRONIC CONDITION REFILL CHAIN (everything hook)
+ * ───────────────────────────────────────────────────────────────
+ *
+ * PATTERN: Patients with condition_type "chronic" on any prescription
+ * event get additional cloned prescription_refill events injected
+ * every ~30 days after the original. Creates periodic refill cadence.
+ *
+ * HOW TO FIND IT IN MIXPANEL:
+ *
+ *   Report 1: Refill Volume by Condition
+ *   • Report type: Insights
+ *   • Event: "prescription refill"
+ *   • Measure: Total
+ *   • Breakdown: "condition_type"
+ *   • Expected: "chronic" should have ~3-4x more refills than others
+ *
+ * REAL-WORLD ANALOGUE: Chronic conditions (diabetes, hypertension)
+ * require ongoing prescriptions creating predictable refill revenue.
+ *
+ * ───────────────────────────────────────────────────────────────
+ * 6. OCCASIONAL PATIENT NO-SHOWS (everything hook)
+ * ───────────────────────────────────────────────────────────────
+ *
+ * PATTERN: Low-activity users (<15 events — overwhelmingly occasional/
+ * churner patients; providers generate far more) lose 25% of their
+ * "consultation completed" events and get no_show=true stamped on 25%
+ * of their "appointment booked" events. Simulates occasional patients
+ * who book but don't show up.
+ *
+ * HOW TO FIND IT IN MIXPANEL:
+ *
+ *   Report 1: No-Show Rate
+ *   • Report type: Insights
+ *   • Event: "appointment booked"
+ *   • Measure: Total, filtered no_show = true, vs Total overall
+ *   • Expected: no-shows concentrate entirely on low-activity users
+ *     (~25% of their bookings); zero no-shows on active users
+ *
+ *   Report 2: Appointment-to-Consultation Ratio
+ *   • Report type: Funnels
+ *   • Steps: "appointment booked" → "consultation completed"
+ *   • Expected: low-activity users convert visibly worse (engineered
+ *     25% thinning compounded by their organically lower conversion)
+ *
+ * REAL-WORLD ANALOGUE: Infrequent patients have higher no-show rates,
+ * a major operational cost for healthcare providers.
+ *
+ * ───────────────────────────────────────────────────────────────
+ * 7. DOCTOR PROFILE SPECIALIZATION (user hook)
+ * ───────────────────────────────────────────────────────────────
+ *
+ * PATTERN: Users with role "doctor" get specialty set to a specific
+ * value (from the existing array) and years_experience boosted to
+ * senior range (15-30). Nurses get years_experience in mid range.
+ *
+ * HOW TO FIND IT IN MIXPANEL:
+ *
+ *   Report 1: Experience Distribution by Role
+ *   • Report type: Insights
+ *   • Event: "consultation completed"
+ *   • Measure: Average of user property "years_experience"
+ *   • Breakdown: user property "role"
+ *   • Expected: doctors ≈ 22 years, nurses ≈ 8, patients ≈ 0
+ *
+ * REAL-WORLD ANALOGUE: Provider profiles have specialized expertise
+ * and experience levels that affect patient matching.
+ *
+ * ───────────────────────────────────────────────────────────────
+ * 8. FREE-TIER CONVERSION DROP (everything hook)
+ * ───────────────────────────────────────────────────────────────
+ *
+ * PATTERN: 30% of free-tier users (per-user coin flip) lose ALL their
+ * "consultation completed" events — a per-user cliff, not per-event
+ * thinning. Surviving free users are statistically identical to paid
+ * users, which makes the effect cleanly measurable: the excess
+ * zero-consultation share among free users reads the 30% knob directly.
+ *
+ * HOW TO FIND IT IN MIXPANEL:
+ *
+ *   Report 1: Booking Conversion by Tier
+ *   • Report type: Funnels
+ *   • Steps: "symptom search" → "appointment booked" → "consultation completed"
+ *   • Breakdown: "subscription_tier"
+ *   • Expected: free ≈ 28% vs basic/premium ≈ 40% conversion
+ *
+ * REAL-WORLD ANALOGUE: Free-tier patients face longer wait times
+ * and limited scheduling, reducing completed consultations.
+ *
+ * ───────────────────────────────────────────────────────────────
+ * 9. BOOKING FUNNEL TTC BY TIER (everything hook — property scaling)
+ *
+ * PATTERN: Premium users get shorter wait times and consultation
+ * durations (0.67x); Free users get longer (1.4x); Basic at 1.0x.
+ * Scales `wait_time_hours` on "appointment booked" and
+ * `duration_minutes` on "consultation completed".
+ *
+ * HOW TO FIND IT IN MIXPANEL:
+ *
+ *   Report 1: Wait Time by Subscription Tier
+ *   - Insights > "appointment booked"
+ *   - Measure: Average of "wait_time_hours"
+ *   - Breakdown: subscription_tier
+ *   - Expected: premium ~ 0.67x baseline; free ~ 1.4x baseline
+ *
+ *   Report 2: Consultation Duration by Tier
+ *   - Insights > "consultation completed"
+ *   - Measure: Average of "duration_minutes"
+ *   - Breakdown: subscription_tier
+ *   - Expected: premium ~ 0.67x baseline; free ~ 1.4x baseline
+ *
+ * ───────────────────────────────────────────────────────────────
+ * 10. CONSULTATION-COUNT MAGIC NUMBER (everything)
+ *
+ * PATTERN: Sweet 3-6 consultations → +25% on consultation_fee.
+ * Over 7+ → days_until_followup multiplied by 1.5 (over-consulted
+ * patients wait 50% longer for next visit). No flag.
+ *
+ * HOW TO FIND IT IN MIXPANEL:
+ *
+ *   Report 1: Avg Consultation Fee by Consult-Count Bucket
+ *   - Cohort A: users with 3-6 "consultation completed"
+ *   - Cohort B: users with 0-2
+ *   - Event: "consultation completed"
+ *   - Measure: Average of "consultation_fee"
+ *   - Expected: A ~ 1.25x B
+ *
+ *   Report 2: Follow-Up Wait Time on Heavy Consulters
+ *   - Cohort C: users with >= 7 consultations
+ *   - Cohort A: users with 3-6
+ *   - Event: "follow up scheduled"
+ *   - Measure: Average of "days_until_followup"
+ *   - Expected: C ~ 1.5x A (longer gap before next visit)
+ *
+ * REAL-WORLD ANALOGUE: Engaged patients pay more; over-engaged
+ * patients hit care-fatigue and stretch the gap to next visit.
+ *
+ * ═══════════════════════════════════════════════════════════════
+ * EXPECTED METRICS SUMMARY
+ * ═══════════════════════════════════════════════════════════════
+ *
+ * Hook                        | Metric                        | Expected        | Measured (full fidelity)
+ * ----------------------------|-------------------------------|-----------------|-------------------------
+ * H1 After-Hours Pricing      | fee after-hours / business    | 1.5x            | 1.494 (avg = median)
+ * H2 Flu Season Spike         | respiratory share in-window   | 0.65 (vs 0.125) | 0.654 (out: 0.125)
+ * H2 Flu Season Spike         | resp/other wait in-window     | 2x              | 1.996
+ * H3 Experienced Doctor Sat.  | satisfaction >12-consult users| avg+median 4.5  | 4.499 / 4.500 (0 impure)
+ * H4 Video Follow-Up Lift     | extra follow-ups per video    | +0.6 within 7d  | +0.588
+ *                             |   consult (within-7d diff)    |                 |
+ * H5 Chronic Refill Chain     | surviving clones / model      | ~1.0            | 1.001 (placebo 0.048)
+ *                             |   expectation (survival-adj)  |                 |
+ * H6 Occasional No-Shows      | no_show rate, <15-event users | 0.25 (0 on rest)| 0.248 (0 impure)
+ * H7 Doctor Specialization    | years_experience by role      | 22.5 / 9 / 0    | 22.46 / 9.00 / 0 exact
+ * H8 Free-Tier Cliff          | excess zero-consult share     | 0.30            | 0.313 (survivors 0.986)
+ *                             |   (z_free−z_paid)/(1−z_paid)  |                 |
+ * H9 Wait/Duration by Tier    | free/basic, premium/basic     | 1.4x / 0.67x    | 1.40/0.671, 1.40/0.670
+ * H9 Funnel TTC by Tier       | median TTC free/basic (emu)   | >1 (diluted 1.4)| 1.157 (prem/basic 0.827)
+ * H10 Magic Number            | sweet fee / low fee (median)  | 1.25x           | 1.219
+ * H10 Magic Number            | over/sweet days_until_fu      | 1.5x (phone fu) | 1.500
+ */
+
+// ── SCALE ──
+const SEED = "dm4-healthcare";
+const NUM_USERS = 10_000;
+const DATASET_START = "2026-01-01T00:00:00Z";
+const DATASET_END = "2026-05-01T23:59:59Z";
+const EVENTS_PER_DAY = 1.2;
+const token = process.env.MP_TOKEN || "your-mixpanel-token";
+
+const chance = u.initChance(SEED);
+
+// ── KNOBS (tweak these to reshape stories) ──
+const AFTER_HOURS_START = 19;
+const AFTER_HOURS_END = 7;
+const AFTER_HOURS_FEE_MULT = 1.5;
+
+const FLU_START_DAY = 50;
+const FLU_END_DAY = 70;
+const FLU_RESPIRATORY_LIKELIHOOD = 60;
+const FLU_WAIT_MULT = 2;
+
+const EXPERIENCED_CONSULT_THRESHOLD = 12;
+const EXPERIENCED_SATISFACTION_MIN = 4.0;
+const EXPERIENCED_SATISFACTION_MAX = 5.0;
+
+const VIDEO_FOLLOWUP_LIKELIHOOD = 60;
+
+const CHRONIC_REFILL_MIN = 2;
+const CHRONIC_REFILL_MAX = 4;
+const CHRONIC_REFILL_INTERVAL_DAYS = 30;
+
+const NO_SHOW_EVENT_THRESHOLD = 15;
+const NO_SHOW_DROP_LIKELIHOOD = 25;
+
+const DOCTOR_EXPERIENCE_MIN = 15;
+const DOCTOR_EXPERIENCE_MAX = 30;
+const NURSE_EXPERIENCE_MIN = 3;
+const NURSE_EXPERIENCE_MAX = 15;
+
+const FREE_TIER_DROP_LIKELIHOOD = 30;
+
+const TTC_PREMIUM_FACTOR = 0.67;
+const TTC_FREE_FACTOR = 1.4;
+
+const CONSULT_SWEET_MIN = 3;
+const CONSULT_SWEET_MAX = 6;
+const CONSULT_OVER_THRESHOLD = 7;
+const CONSULT_FEE_BOOST = 1.25;
+const CONSULT_FOLLOWUP_STRETCH = 1.5;
+
+// ── DATA ARRAYS ──
+// Generate consistent doctor/clinic IDs at module level
+const doctorIds = v.range(1, 120).map(() => `DR_${v.uid(6)}`);
+const clinicIds = v.range(1, 25).map(() => `CLINIC_${v.uid(4)}`);
+
+// ── HELPER FUNCTIONS ──
+function handleUserHooks(record) {
+	// H7: DOCTOR PROFILE SPECIALIZATION — doctors get a real specialty and
+	// senior years_experience. Nurses get mid-range experience. Patients
+	// stay at defaults.
+	if (record.role === "doctor") {
+		record.specialty = chance.pickone(["cardiology", "dermatology", "pediatrics", "psychiatry", "general_practice", "pulmonology", "endocrinology"]);
+		record.years_experience = chance.integer({ min: DOCTOR_EXPERIENCE_MIN, max: DOCTOR_EXPERIENCE_MAX });
+	} else if (record.role === "nurse") {
+		record.specialty = chance.pickone(["general_practice", "pediatrics", "emergency"]);
+		record.years_experience = chance.integer({ min: NURSE_EXPERIENCE_MIN, max: NURSE_EXPERIENCE_MAX });
+	} else {
+		record.years_experience = 0;
+	}
+	return record;
+}
+
+function handleEverythingHooks(record, meta) {
+	if (!record.length) return record;
+	const profile = meta.profile;
+	const datasetStart = dayjs.unix(meta.datasetStart).utc();
+	const FLU_START = datasetStart.add(FLU_START_DAY, "days");
+	const FLU_END = datasetStart.add(FLU_END_DAY, "days");
+
+	// ── SUPER-PROP STAMPING ──────────────────────────
+	// Stamp superProps from profile so they are consistent per-user.
+	if (profile) {
+		const tier = profile.subscription_tier;
+		const plat = profile.Platform;
+		record.forEach(e => {
+			if (tier) e.subscription_tier = tier;
+			if (plat) e.Platform = plat;
+		});
+	}
+
+	// HOOK 9: BOOKING FUNNEL TTC BY TIER (property scaling)
+	// Premium users get shorter wait_time_hours (0.67x) and duration_minutes (0.67x).
+	// Free users get longer wait_time_hours (1.4x) and duration_minutes (1.4x).
+	// Basic users stay at baseline. SQL-measurable via AVG(wait_time_hours) broken by tier.
+	if (profile) {
+		const userTier = profile.subscription_tier;
+		const ttcFactor = userTier === "premium" ? TTC_PREMIUM_FACTOR : userTier === "free" ? TTC_FREE_FACTOR : 1.0;
+		if (ttcFactor !== 1.0) {
+			// Timestamp shift: affects Mixpanel funnel TTC
+			const bookingSeq = findFirstSequence(
+				record,
+				["appointment booked", "consultation completed", "follow up scheduled"],
+				60 * 24 * 30
+			);
+			if (bookingSeq) scaleFunnelTTC(bookingSeq, ttcFactor);
+			// Property scale: affects Insights AVG reports
+			record.forEach(e => {
+				if (e.event === "appointment booked" && typeof e.wait_time_hours === "number") {
+					e.wait_time_hours = Math.round(e.wait_time_hours * ttcFactor * 10) / 10;
+				}
+				if (e.event === "consultation completed" && typeof e.duration_minutes === "number") {
+					e.duration_minutes = Math.round(e.duration_minutes * ttcFactor);
+				}
+			});
+		}
+	}
+
+	// HOOK 1: AFTER-HOURS SURGE PRICING — consultations 7PM-7AM
+	// UTC get consultation_fee 1.5x. No flag — discover via HOD chart.
+	// Only "consultation completed" declares consultation_fee ("appointment
+	// booked" was a dead branch — its guard on e.consultation_fee never held).
+	// Runs after H9's timestamp shift, so the hour check sees final times.
+	record.forEach(e => {
+		if (e.event === "consultation completed") {
+			const hour = new Date(e.time).getUTCHours();
+			if ((hour >= AFTER_HOURS_START || hour < AFTER_HOURS_END) && e.consultation_fee) {
+				e.consultation_fee = Math.floor(e.consultation_fee * AFTER_HOURS_FEE_MULT);
+			}
+		}
+	});
+
+	// HOOK 2: FLU SEASON SPIKE — d50-70 respiratory dominates, wait_time doubles.
+	// UTC parses throughout — a machine-local dayjs() here would move the
+	// window boundaries by the generating machine's TZ offset, breaking the
+	// same-seed-same-output determinism promise across machines.
+	record.forEach(e => {
+		if (e.event !== "appointment booked") return;
+		const t = dayjs.utc(e.time);
+		if (t.isAfter(FLU_START) && t.isBefore(FLU_END)) {
+			if (chance.bool({ likelihood: FLU_RESPIRATORY_LIKELIHOOD })) e.condition_type = "respiratory";
+			if (e.condition_type === "respiratory") {
+				e.wait_time_hours = Math.floor((e.wait_time_hours || 12) * FLU_WAIT_MULT);
+			}
+		}
+	});
+
+	// ── HOOK 8: FREE-TIER CONVERSION DROP ────────────
+	// Free-tier users lose ~30% of "consultation completed" events
+	// (last step of Booking to Consultation funnel), simulating
+	// lower conversion for non-paying patients.
+	if (profile && profile.subscription_tier === "free" && chance.bool({ likelihood: FREE_TIER_DROP_LIKELIHOOD })) {
+		record = record.filter(e => e.event !== "consultation completed");
+	}
+
+	// ── HOOK 3: EXPERIENCED DOCTOR SATISFACTION ──────
+	// Users with >12 consultation events get boosted satisfaction scores.
+	let consultCount = 0;
+	record.forEach(e => {
+		if (e.event === "consultation completed") consultCount++;
+	});
+
+	if (consultCount > EXPERIENCED_CONSULT_THRESHOLD) {
+		record.forEach(e => {
+			if (e.event === "consultation completed") {
+				e.satisfaction_score = chance.floating({ min: EXPERIENCED_SATISFACTION_MIN, max: EXPERIENCED_SATISFACTION_MAX, fixed: 1 });
+			}
+		});
+	}
+
+	// ── HOOK 4: VIDEO CONSULTATION FOLLOW-UP LIFT ────
+	// Patients with video consultations get 2x follow-up events.
+	const hasVideoConsult = record.some(e =>
+		e.event === "consultation completed" && e.consultation_mode === "video"
+	);
+	if (hasVideoConsult) {
+		const templateFollowUp = record.find(e => e.event === "follow up scheduled");
+		if (templateFollowUp) {
+			const videoConsults = record.filter(e =>
+				e.event === "consultation completed" && e.consultation_mode === "video"
+			);
+			videoConsults.forEach(vc => {
+				if (chance.bool({ likelihood: VIDEO_FOLLOWUP_LIKELIHOOD })) {
+					record.push({
+						...templateFollowUp,
+						time: dayjs(vc.time).add(chance.integer({ min: 1, max: 7 }), "days").toISOString(),
+						user_id: vc.user_id,
+						consultation_mode: "video",
+						days_until_followup: chance.integer({ min: 3, max: 14 }),
+						// fresh insert_id: the engine stamps insert_id at generation
+						// (lib/generators/events.js), so a bare spread copies the
+						// template's id and Mixpanel's $insert_id dedupe would
+						// silently drop every clone after the first
+						insert_id: chance.guid(),
+					});
+				}
+			});
+		}
+	}
+
+	// ── HOOK 5: CHRONIC CONDITION REFILL CHAIN ───────
+	// Patients with chronic prescriptions get refills every ~30 days.
+	const chronicRxs = record.filter(e =>
+		e.event === "prescription issued" && e.condition_type === "chronic"
+	);
+	if (chronicRxs.length > 0) {
+		const templateRefill = record.find(e => e.event === "prescription refill");
+		if (templateRefill) {
+			chronicRxs.forEach(rx => {
+				const rxTime = dayjs(rx.time);
+				const refillsToAdd = chance.integer({ min: CHRONIC_REFILL_MIN, max: CHRONIC_REFILL_MAX });
+				for (let i = 1; i <= refillsToAdd; i++) {
+					record.push({
+						...templateRefill,
+						// clones past datasetEnd are dropped by the engine's
+						// unconditional future-time guard — late-window chronic
+						// prescriptions keep fewer of their refills by design
+						time: rxTime.add(CHRONIC_REFILL_INTERVAL_DAYS * i + chance.integer({ min: -3, max: 3 }), "days").toISOString(),
+						user_id: rx.user_id,
+						condition_type: "chronic",
+						medication_type: "chronic_maintenance",
+						refill_count: i,
+						// fresh insert_id — same $insert_id dedupe rationale as H4
+						insert_id: chance.guid(),
+					});
+				}
+			});
+		}
+	}
+
+	// HOOK 6: OCCASIONAL PATIENT NO-SHOWS — low-activity users (< 15 events
+	// at this point in the pipeline, clones included) lose 25% of their
+	// consultations and get no_show=true stamped on 25% of their bookings
+	// (they booked but didn't show). no_show is DECLARED [false] on
+	// "appointment booked" (schema-first rule), so flipped rows are the only
+	// true values in the dataset. Because the flag is decided before any
+	// later deletion and every subsequent step only shrinks a user's stream,
+	// users with >= 15 output events provably carry zero no_show=true rows.
+	if (record.length < NO_SHOW_EVENT_THRESHOLD) {
+		for (let i = record.length - 1; i >= 0; i--) {
+			if (record[i].event === "consultation completed" && chance.bool({ likelihood: NO_SHOW_DROP_LIKELIHOOD })) {
+				record.splice(i, 1);
+			}
+		}
+		record.forEach(e => {
+			if (e.event === "appointment booked" && chance.bool({ likelihood: NO_SHOW_DROP_LIKELIHOOD })) {
+				e.no_show = true;
+			}
+		});
+	}
+
+	// HOOK 10: CONSULTATION-COUNT MAGIC NUMBER (no flags)
+	// Sweet 3-6 consultations → +25% on consultation_fee. Over 7+ →
+	// days_until_followup stretched 1.5x (over-consulted patients wait
+	// longer for the next visit). Counts run AFTER all filters (H8/H6)
+	// and nothing drops consultations later, so output-side consult
+	// counts rebuild these cohorts exactly.
+	const consultCt = record.filter(e => e.event === "consultation completed").length;
+	if (consultCt >= CONSULT_SWEET_MIN && consultCt <= CONSULT_SWEET_MAX) {
+		record.forEach(e => {
+			if (e.event === "consultation completed" && typeof e.consultation_fee === "number") {
+				e.consultation_fee = Math.round(e.consultation_fee * CONSULT_FEE_BOOST);
+			}
+		});
+	} else if (consultCt >= CONSULT_OVER_THRESHOLD) {
+		record.forEach(e => {
+			if (e.event === "follow up scheduled" && typeof e.days_until_followup === "number") {
+				e.days_until_followup = Math.round(e.days_until_followup * CONSULT_FOLLOWUP_STRETCH);
+			}
+		});
+	}
+
+	return record;
+}
+
+// ── CONFIG ──
+/** @type {Config} */
+const config = {
+	version: 2,
+	seed: SEED,
+	datasetStart: DATASET_START,
+	datasetEnd: DATASET_END,
+	avgEventsPerUserPerDay: EVENTS_PER_DAY,
+	numUsers: NUM_USERS,
+	format: "json",
+	gzip: true,
+	credentials: {
+		token,
+	},
+	switches: {
+		hasSessionIds: true,
+		alsoInferFunnels: false,
+		hasLocation: true,
+		hasAndroidDevices: true,
+		hasIOSDevices: true,
+		hasDesktopDevices: true,
+		hasBrowser: false,
+		hasCampaigns: false,
+		isAnonymous: false,
+		hasAdSpend: false,
+		hasAvatar: true,
+	},
+	identity: {
+		avgDevicePerUser: 2,
+	},
+	concurrency: 1,
+	writeToDisk: false,
+	scdProps: {
+		care_plan: {
+			values: ["preventive", "routine", "chronic", "acute"],
+			frequency: "month",
+			timing: "fuzzy",
+			max: 8
+		}
+	},
+	mirrorProps: {},
+	lookupTables: [],
+
+	// ── Events (18) ──────────────────────────────────────────
+	events: [
+		{
+			event: "account created",
+			weight: 1,
+			isFirstEvent: true,
+			isAuthEvent: true,
+			properties: {
+				referral_source: ["organic", "doctor_referral", "insurance_partner", "social_media", "search"],
+			},
+		},
+		{
+			event: "symptom search",
+			weight: 7,
+			properties: {
+				search_term: ["headache", "fever", "cough", "back pain", "fatigue", "anxiety", "rash", "nausea", "chest pain", "joint pain"],
+				results_count: u.weighNumRange(0, 25, 0.5),
+			},
+		},
+		{
+			event: "appointment booked",
+			weight: 6,
+			isStrictEvent: false,
+			properties: {
+				doctor_id: chance.pickone.bind(chance, doctorIds),
+				clinic_id: chance.pickone.bind(chance, clinicIds),
+				condition_type: ["general", "general", "general", "respiratory", "dermatology", "mental_health", "chronic", "pediatric"],
+				wait_time_hours: u.weighNumRange(1, 72, 0.4),
+				appointment_type: ["new_patient", "follow_up", "follow_up", "urgent", "routine", "routine"],
+				// declared false; H6 flips to true on 25% of low-activity users' bookings
+				no_show: [false],
+			},
+		},
+		{
+			event: "consultation completed",
+			weight: 5,
+			isStrictEvent: false,
+			properties: {
+				doctor_id: chance.pickone.bind(chance, doctorIds),
+				consultation_mode: ["phone", "phone", "video"],
+				duration_minutes: u.weighNumRange(5, 60, 0.6, 15),
+				consultation_fee: u.weighNumRange(25, 200, 0.4, 75),
+				satisfaction_score: u.weighNumRange(1, 5, 0.8, 3),
+				condition_type: ["general", "general", "respiratory", "dermatology", "mental_health", "chronic", "pediatric"],
+			},
+		},
+		{
+			event: "prescription issued",
+			weight: 4,
+			isStrictEvent: false,
+			properties: {
+				medication_type: ["antibiotic", "antiviral", "painkiller", "anti_inflammatory", "antidepressant", "inhaler", "topical", "chronic_maintenance"],
+				quantity: u.weighNumRange(1, 90, 0.3, 30),
+				condition_type: ["general", "respiratory", "dermatology", "mental_health", "chronic", "chronic", "pediatric"],
+				refill_count: u.weighNumRange(0, 3),
+			},
+		},
+		{
+			event: "prescription refill",
+			weight: 3,
+			isStrictEvent: false,
+			properties: {
+				medication_type: ["antibiotic", "antiviral", "painkiller", "anti_inflammatory", "antidepressant", "inhaler", "topical", "chronic_maintenance"],
+				quantity: u.weighNumRange(1, 90, 0.3, 30),
+				condition_type: ["general", "respiratory", "dermatology", "mental_health", "chronic", "chronic", "pediatric"],
+				refill_count: u.weighNumRange(1, 6),
+			},
+		},
+		{
+			event: "follow up scheduled",
+			weight: 3,
+			isStrictEvent: false,
+			properties: {
+				doctor_id: chance.pickone.bind(chance, doctorIds),
+				days_until_followup: u.weighNumRange(3, 30, 0.5, 7),
+				condition_type: ["general", "respiratory", "dermatology", "mental_health", "chronic", "pediatric"],
+				consultation_mode: ["phone", "phone", "video"],
+			},
+		},
+		{
+			event: "message sent",
+			weight: 5,
+			properties: {
+				message_type: ["question", "question", "update", "result_inquiry", "prescription_question", "scheduling"],
+				recipient_role: ["doctor", "doctor", "nurse", "support"],
+				response_time_hours: u.weighNumRange(0.1, 48, 0.3, 4),
+			},
+		},
+		{
+			event: "lab results viewed",
+			weight: 3,
+			properties: {
+				test_type: ["blood_panel", "urinalysis", "imaging", "allergy_test", "metabolic_panel", "thyroid"],
+				result_status: ["normal", "normal", "normal", "abnormal", "pending"],
+			},
+		},
+		{
+			event: "health record accessed",
+			weight: 4,
+			properties: {
+				record_type: ["visit_summary", "lab_results", "prescriptions", "immunizations", "billing"],
+				access_method: ["app", "app", "web_portal"],
+			},
+		},
+		{
+			event: "insurance verified",
+			weight: 2,
+			properties: {
+				insurance_type: ["private", "private", "employer", "medicare", "medicaid", "self_pay"],
+				verification_status: ["approved", "approved", "approved", "pending", "denied"],
+				copay_amount: u.weighNumRange(0, 75, 0.5, 20),
+			},
+		},
+		{
+			event: "payment processed",
+			weight: 3,
+			properties: {
+				amount: u.weighNumRange(10, 500, 0.3, 75),
+				payment_method: ["credit_card", "credit_card", "insurance_claim", "hsa_fsa", "debit"],
+				payment_status: ["success", "success", "success", "success", "failed"],
+			},
+		},
+		{
+			event: "notification received",
+			weight: 6,
+			properties: {
+				notification_type: ["appointment_reminder", "appointment_reminder", "lab_ready", "prescription_ready", "message_received", "billing"],
+				channel: ["push", "push", "email", "sms"],
+				opened: [true, true, true, false],
+			},
+		},
+		{
+			event: "provider rated",
+			weight: 2,
+			properties: {
+				doctor_id: chance.pickone.bind(chance, doctorIds),
+				rating: u.weighNumRange(1, 5, 0.7, 4),
+				would_recommend: [true, true, true, true, false],
+			},
+		},
+		{
+			event: "support ticket created",
+			weight: 1,
+			properties: {
+				category: ["billing", "technical", "scheduling", "prescription", "insurance", "other"],
+				priority: ["low", "low", "medium", "medium", "high"],
+				resolution_hours: u.weighNumRange(1, 96, 0.4, 24),
+			},
+		},
+		{
+			event: "profile updated",
+			weight: 2,
+			properties: {
+				field_updated: ["insurance", "address", "phone", "emergency_contact", "allergies", "medications"],
+			},
+		},
+		{
+			event: "app session",
+			weight: 8,
+			properties: {
+				session_duration_sec: u.weighNumRange(10, 1800, 0.4, 120),
+				pages_viewed: u.weighNumRange(1, 15, 0.5, 3),
+			},
+		},
+		{
+			event: "account deactivated",
+			weight: 1,
+			isChurnEvent: true,
+			returnLikelihood: 0.15,
+			isStrictEvent: true,
+			properties: {
+				reason: ["switched_provider", "cost", "no_longer_needed", "poor_experience", "insurance_change"],
+			},
+		},
+	],
+
+	// ── Funnels (5) ──────────────────────────────────────────
+	funnels: [
+		{
+			name: "Onboarding Flow",
+			sequence: ["account created", "insurance verified", "symptom search", "appointment booked"],
+			conversionRate: 45,
+			order: "sequential",
+			isFirstFunnel: true,
+			timeToConvert: 72,
+			weight: 3,
+		},
+		{
+			name: "Booking to Consultation",
+			sequence: ["symptom search", "appointment booked", "consultation completed"],
+			conversionRate: 40,
+			order: "sequential",
+			timeToConvert: 48,
+			weight: 5,
+		},
+		{
+			name: "Full Care Journey",
+			sequence: ["appointment booked", "consultation completed", "prescription issued", "follow up scheduled"],
+			conversionRate: 30,
+			order: "sequential",
+			timeToConvert: 168,
+			weight: 3,
+		},
+		{
+			name: "Prescription Lifecycle",
+			sequence: ["prescription issued", "prescription refill", "payment processed"],
+			conversionRate: 55,
+			order: "sequential",
+			timeToConvert: 720,
+			weight: 2,
+		},
+		{
+			name: "Patient Satisfaction",
+			sequence: ["consultation completed", "provider rated", "follow up scheduled"],
+			conversionRate: 25,
+			order: "sequential",
+			timeToConvert: 72,
+			weight: 2,
+		},
+	],
+
+	// ── SuperProps ──────────────────────────────────────────
+	superProps: {
+		subscription_tier: ["free", "free", "free", "basic", "basic", "premium"],
+		Platform: ["ios", "android", "web"],
+	},
+
+	// ── UserProps ──────────────────────────────────────────
+	userProps: {
+		role: ["patient", "patient", "patient", "patient", "patient", "patient", "patient", "patient", "nurse", "doctor"],
+		specialty: ["none"],
+		years_experience: u.weighNumRange(0, 5, 0.5),
+		preferred_language: ["en", "en", "en", "en", "es", "pt", "de", "fr"],
+		has_chronic_condition: [false, false, false, true],
+		age_range: ["18-25", "26-35", "26-35", "36-45", "36-45", "46-55", "56-65", "65+"],
+		subscription_tier: ["free", "free", "free", "basic", "basic", "premium"],
+		Platform: ["ios", "android", "web"],
+	},
+
+	// ── Personas ──────────────────────────────────
+	personas: [
+		{
+			name: "doctor",
+			weight: 5,
+			eventMultiplier: 5.0,
+			conversionModifier: 1.8,
+			churnRate: 0.01,
+			properties: {
+				role: "doctor",
+				segment: "provider",
+			},
+		},
+		{
+			name: "nurse",
+			weight: 10,
+			eventMultiplier: 3.0,
+			conversionModifier: 1.5,
+			churnRate: 0.03,
+			properties: {
+				role: "nurse",
+				segment: "provider",
+			},
+		},
+		{
+			name: "patient_active",
+			weight: 40,
+			eventMultiplier: 1.0,
+			conversionModifier: 1.0,
+			churnRate: 0.05,
+			properties: {
+				role: "patient",
+				segment: "active_patient",
+			},
+		},
+		{
+			name: "patient_occasional",
+			weight: 30,
+			eventMultiplier: 0.5,
+			conversionModifier: 0.7,
+			churnRate: 0.12,
+			properties: {
+				role: "patient",
+				segment: "occasional_patient",
+			},
+		},
+		{
+			name: "patient_churner",
+			weight: 15,
+			eventMultiplier: 0.3,
+			conversionModifier: 0.3,
+			churnRate: 0.4,
+			properties: {
+				role: "patient",
+				segment: "churner",
+			},
+			activeWindow: { maxDays: 21 },
+		},
+	],
+
+	hook(record, type, meta) {
+		if (type === "user") return handleUserHooks(record);
+		if (type === "everything") return handleEverythingHooks(record, meta);
+		return record;
+	},
+};
+
+// ── STORIES ──────────────────────────────────────────────────────────────
+// Machine-checkable contract for the 10 numbered hooks. Evaluate with:
+//   node scripts/verify-stories.mjs dungeons/vertical/healthcare/healthcare.js --data-prefix verify-healthcare
+
+const EV = `read_json_auto('{{PREFIX}}-EVENTS*.json', sample_size=-1, union_by_name=true)`;
+const US = `read_json_auto('{{PREFIX}}-USERS*.json', sample_size=-1, union_by_name=true)`;
+
+// Identity prelude. account created is both isAuthEvent and isFirstEvent, so
+// born users auth on their very first event and user_id should be present on
+// every record; the prelude still resolves through the device pool
+// (avgDevicePerUser: 2, "anonymousIds" is the legacy USERS-shard key) as
+// belt-and-braces for any device-only edge.
+const ID_CTE = `dmap AS (SELECT unnest("anonymousIds") AS device_id, distinct_id FROM ${US}),
+ev AS (SELECT coalesce(m.distinct_id::VARCHAR, e.user_id::VARCHAR, e.device_id::VARCHAR) AS uid,
+  e.time::TIMESTAMP AS t, e.* FROM ${EV} e LEFT JOIN dmap m ON e.device_id = m.device_id)`;
+
+// Temporal boundaries computed from the same knobs the hooks use (the hook
+// parses in UTC, so these UTC timestamps are exact window edges)
+const FLU_IN_START_TS = dayjs.utc(DATASET_START).add(FLU_START_DAY, "day").format("YYYY-MM-DD HH:mm:ss");
+const FLU_IN_END_TS = dayjs.utc(DATASET_START).add(FLU_END_DAY, "day").format("YYYY-MM-DD HH:mm:ss");
+const END_TS = dayjs.utc(DATASET_END).format("YYYY-MM-DD HH:mm:ss");
+// H4 window guard: consultations in the last 7 days can't be credited with a
+// clone that would land past datasetEnd (future-time guard drops it)
+const END_MINUS_7_TS = dayjs.utc(DATASET_END).subtract(7, "day").format("YYYY-MM-DD HH:mm:ss");
+
+// Per-user consultation counts. H10 (and H3) classify on counts taken AFTER
+// all filters (H8 free-tier cliff, H6 no-show thinning) and nothing drops
+// consultations later, so output-side counts rebuild the hook cohorts exactly.
+const CONSULT_CTE = `cc AS (SELECT uid, count(*) AS ct FROM ev WHERE event = 'consultation completed' GROUP BY 1)`;
+
+/** @type {import("../../../types").DungeonStory[]} */
+export const stories = [
+	{
+		id: "H1-after-hours-pricing",
+		hook: "H1",
+		archetype: "temporal-inflection",
+		narrative: `consultations between ${AFTER_HOURS_START}:00 and ${AFTER_HOURS_END}:00 UTC carry consultation_fee × ${AFTER_HOURS_FEE_MULT}. H10's sweet-spot fee boost rides both HOD bins equally (consult-count cohorts are hour-independent), so both the avg and median ratios read the ${AFTER_HOURS_FEE_MULT} knob directly (Math.floor bias < 1%)`,
+		assertions: [
+			{
+				breakdown: {
+					type: "duckdb",
+					sql: `WITH ${ID_CTE}
+SELECT CASE WHEN extract(hour FROM t) >= ${AFTER_HOURS_START} OR extract(hour FROM t) < ${AFTER_HOURS_END} THEN 'after' ELSE 'business' END AS grp,
+ count(*) AS event_count, count(DISTINCT uid) AS user_count,
+ avg(consultation_fee) AS avg_fee, median(consultation_fee) AS med_fee
+FROM ev WHERE event = 'consultation completed' GROUP BY 1`,
+				},
+				select: { a: { where: { grp: "after" } }, b: { where: { grp: "business" } } },
+				expect: { metric: "a.avg_fee / b.avg_fee", op: "between", target: [1.35, 1.65] },
+				minCohort: 400,
+			},
+			{
+				breakdown: {
+					type: "duckdb",
+					sql: `WITH ${ID_CTE}
+SELECT CASE WHEN extract(hour FROM t) >= ${AFTER_HOURS_START} OR extract(hour FROM t) < ${AFTER_HOURS_END} THEN 'after' ELSE 'business' END AS grp,
+ count(*) AS event_count, count(DISTINCT uid) AS user_count,
+ avg(consultation_fee) AS avg_fee, median(consultation_fee) AS med_fee
+FROM ev WHERE event = 'consultation completed' GROUP BY 1`,
+				},
+				select: { a: { where: { grp: "after" } }, b: { where: { grp: "business" } } },
+				// scaling a whole bin scales every quantile: median ratio = knob too
+				expect: { metric: "a.med_fee / b.med_fee", op: "between", target: [1.35, 1.65] },
+				minCohort: 400,
+			},
+		],
+	},
+	{
+		id: "H2-flu-season",
+		hook: "H2",
+		archetype: "temporal-inflection",
+		narrative: `days ${FLU_START_DAY}-${FLU_END_DAY}: bookings are forced respiratory at ${FLU_RESPIRATORY_LIKELIHOOD}%, and every in-window respiratory booking gets wait_time_hours × ${FLU_WAIT_MULT}. Expected in-window respiratory share = 0.60 + 0.40 × 1/8 = 0.65 (declared mix is 1-in-8 respiratory); out-window share stays at the declared 0.125. H9's tier scaling rides all conditions equally, so the in-window resp/other wait ratio reads the ×${FLU_WAIT_MULT} knob`,
+		assertions: [
+			{
+				breakdown: {
+					type: "duckdb",
+					sql: `WITH ${ID_CTE}
+SELECT CASE WHEN t > TIMESTAMP '${FLU_IN_START_TS}' AND t < TIMESTAMP '${FLU_IN_END_TS}' THEN 'in' ELSE 'out' END AS grp,
+ count(*) AS event_count, count(DISTINCT uid) AS user_count,
+ count(*) FILTER (WHERE condition_type = 'respiratory')::DOUBLE / count(*) AS resp_share,
+ avg(wait_time_hours) FILTER (WHERE condition_type = 'respiratory') AS resp_wait,
+ avg(wait_time_hours) FILTER (WHERE condition_type <> 'respiratory') AS other_wait
+FROM ev WHERE event = 'appointment booked' GROUP BY 1`,
+				},
+				select: { i: { where: { grp: "in" } } },
+				expect: { metric: "i.resp_share", op: "between", target: [0.58, 0.72] },
+				minCohort: 200,
+			},
+			{
+				breakdown: {
+					type: "duckdb",
+					sql: `WITH ${ID_CTE}
+SELECT CASE WHEN t > TIMESTAMP '${FLU_IN_START_TS}' AND t < TIMESTAMP '${FLU_IN_END_TS}' THEN 'in' ELSE 'out' END AS grp,
+ count(*) AS event_count, count(DISTINCT uid) AS user_count,
+ count(*) FILTER (WHERE condition_type = 'respiratory')::DOUBLE / count(*) AS resp_share
+FROM ev WHERE event = 'appointment booked' GROUP BY 1`,
+				},
+				select: { o: { where: { grp: "out" } } },
+				// purity: forcing happens only inside the window
+				expect: { metric: "o.resp_share", op: "between", target: [0.09, 0.16] },
+				minCohort: 200,
+			},
+			{
+				breakdown: {
+					type: "duckdb",
+					sql: `WITH ${ID_CTE}
+SELECT CASE WHEN t > TIMESTAMP '${FLU_IN_START_TS}' AND t < TIMESTAMP '${FLU_IN_END_TS}' THEN 'in' ELSE 'out' END AS grp,
+ count(*) AS event_count, count(DISTINCT uid) AS user_count,
+ avg(wait_time_hours) FILTER (WHERE condition_type = 'respiratory') AS resp_wait,
+ avg(wait_time_hours) FILTER (WHERE condition_type <> 'respiratory') AS other_wait
+FROM ev WHERE event = 'appointment booked' GROUP BY 1`,
+				},
+				select: { i: { where: { grp: "in" } } },
+				expect: { metric: "i.resp_wait / i.other_wait", op: "between", target: [1.7, 2.35] },
+				minCohort: 200,
+			},
+		],
+	},
+	{
+		id: "H3-experienced-doctor-satisfaction",
+		hook: "H3",
+		archetype: "cohort-prop-scale",
+		narrative: `users with >${EXPERIENCED_CONSULT_THRESHOLD} consultations get every satisfaction_score redrawn uniform [${EXPERIENCED_SATISFACTION_MIN}, ${EXPERIENCED_SATISFACTION_MAX}] (avg AND median 4.5 — both quantile reads of the uniform). Purity is exact: later hooks only DELETE consultations, so any user still >${EXPERIENCED_CONSULT_THRESHOLD} in the output was boosted — all surviving scores sit in the redrawn range. No ratio-vs-baseline assertion: the declared weighNumRange(1, 5, 0.8, 3) baseline is a 3-value seeded pool (the 4th arg is POOL SIZE, not mode), so the organic mean is not derivable from the schema`,
+		assertions: [
+			{
+				// deterministic purity — a single sub-4.0 score on an
+				// output->12-consult user is a hook bug, not sampling noise
+				breakdown: {
+					type: "duckdb",
+					sql: `WITH ${ID_CTE}, ${CONSULT_CTE}
+SELECT count(*) FILTER (WHERE e.satisfaction_score < ${EXPERIENCED_SATISFACTION_MIN}) AS below_min,
+ count(*) AS scores, count(DISTINCT c.uid) AS exp_users
+FROM cc c JOIN ev e ON e.uid = c.uid AND e.event = 'consultation completed'
+WHERE c.ct > ${EXPERIENCED_CONSULT_THRESHOLD}`,
+				},
+				assert: (rows) => {
+					const r = (rows || [])[0];
+					if (!r || Number(r.exp_users) === 0) return { pass: false, verdict: "NONE", detail: "no >12-consult users" };
+					const clean = Number(r.below_min) === 0;
+					return {
+						pass: clean,
+						verdict: clean ? "NAILED" : "INVERSE",
+						detail: `below-4.0 scores=${r.below_min} of ${r.scores} across ${r.exp_users} experienced users (must be 0)`,
+					};
+				},
+			},
+			{
+				breakdown: {
+					type: "duckdb",
+					sql: `WITH ${ID_CTE}, ${CONSULT_CTE},
+lab AS (SELECT uid, CASE WHEN ct > ${EXPERIENCED_CONSULT_THRESHOLD} THEN 'exp' WHEN ct <= 9 THEN 'base' ELSE 'mid' END AS grp FROM cc)
+SELECT l.grp, count(DISTINCT l.uid) AS user_count, count(*) AS event_count, avg(e.satisfaction_score) AS avg_sat
+FROM lab l JOIN ev e ON e.uid = l.uid AND e.event = 'consultation completed' GROUP BY 1`,
+				},
+				select: { x: { where: { grp: "exp" } } },
+				// uniform [4.0, 5.0] → 4.5
+				expect: { metric: "x.avg_sat", op: "between", target: [4.35, 4.65] },
+				minCohort: 30,
+			},
+			{
+				breakdown: {
+					type: "duckdb",
+					sql: `WITH ${ID_CTE}, ${CONSULT_CTE}
+SELECT 'exp' AS grp, count(DISTINCT c.uid) AS user_count, count(*) AS event_count, median(e.satisfaction_score) AS med_sat
+FROM cc c JOIN ev e ON e.uid = c.uid AND e.event = 'consultation completed'
+WHERE c.ct > ${EXPERIENCED_CONSULT_THRESHOLD}`,
+				},
+				select: { x: { where: { grp: "exp" } } },
+				// median of uniform [4.0, 5.0] = 4.5 — independent quantile read
+				expect: { metric: "x.med_sat", op: "between", target: [4.35, 4.65] },
+				minCohort: 30,
+			},
+		],
+	},
+	{
+		id: "H4-video-followup-lift",
+		hook: "H4",
+		archetype: "cohort-count-scale",
+		narrative: `each video consultation has a ${VIDEO_FOLLOWUP_LIKELIHOOD}% chance to inject one cloned follow-up 1-7 days later. Per-consultation attribution: counting follow-ups within 7d after each consultation, video minus phone reads the 0.6 knob with per-EVENT attribution that cancels user-level activity selection (organic near-rates are mode-blind: a consultation's mode is an iid per-event draw, so both bins sample the same users' timelines). Attenuation: a clone can also land within 7d of a neighboring phone consultation of the same user, inflating the phone bin — hence the band floor below 0.6. Cohort restricted to users with ≥1 follow-up (clone requires an organic template) and consultations ≥7d before datasetEnd (clones past the end are future-guard dropped). Deliberately single-assertion: user-level composites (video-users vs phone-only fu-per-consult) were tested and rejected — conditioning on fus>0 inflates the low-activity phone-only group, and tier/persona sampling coupling plus the video_consultation feature's launch-gated mode mix make any user-level band underivable from knobs`,
+		assertions: [
+			{
+				breakdown: {
+					type: "duckdb",
+					sql: `WITH ${ID_CTE},
+fu_users AS (SELECT DISTINCT uid FROM ev WHERE event = 'follow up scheduled'),
+cons AS (SELECT e.uid, e.t, e.consultation_mode AS mode
+  FROM ev e JOIN fu_users f ON f.uid = e.uid
+  WHERE e.event = 'consultation completed' AND e.t <= TIMESTAMP '${END_MINUS_7_TS}'),
+cnt AS (SELECT c.uid, c.mode, c.t, count(fu.uid) AS fu7
+  FROM cons c LEFT JOIN ev fu ON fu.uid = c.uid AND fu.event = 'follow up scheduled'
+    AND fu.t > c.t AND fu.t <= c.t + INTERVAL 7 DAY
+  GROUP BY 1, 2, 3)
+SELECT mode AS grp, count(*) AS consults, count(DISTINCT uid) AS user_count, avg(fu7) AS avg_fu7
+FROM cnt GROUP BY 1`,
+				},
+				select: { v: { where: { grp: "video" } }, p: { where: { grp: "phone" } } },
+				expect: { metric: "v.avg_fu7 - p.avg_fu7", op: "between", target: [0.33, 0.78] },
+				minCohort: 150,
+			},
+		],
+	},
+	{
+		id: "H5-chronic-refill-chain",
+		hook: "H5",
+		archetype: "cohort-count-scale",
+		narrative: `each chronic prescription spawns ${CHRONIC_REFILL_MIN}-${CHRONIC_REFILL_MAX} cloned refills at ~${CHRONIC_REFILL_INTERVAL_DAYS}d intervals (condition_type=chronic, medication_type=chronic_maintenance, refill_count=i); clones past datasetEnd are future-guard dropped. The assertion rebuilds the survival model per prescription from its actual date (attempt i at +${CHRONIC_REFILL_INTERVAL_DAYS}·i days; P(n≥3)=2/3, P(n≥4)=1/3 from the uniform 2-4 draw), subtracts the organic chronic∧chronic_maintenance baseline measured on non-chronic-rx users (declared mix: 2/7 × 1/8 ≈ 0.036), and checks measured clones ÷ model expectation ≈ 1. Cohort restricted to chronic-rx users with ≥1 refill (the hook needs an organic template)`,
+		assertions: [
+			{
+				breakdown: {
+					type: "duckdb",
+					sql: `WITH ${ID_CTE},
+refill_users AS (SELECT DISTINCT uid FROM ev WHERE event = 'prescription refill'),
+crx AS (SELECT e.uid, e.t FROM ev e JOIN refill_users ru ON ru.uid = e.uid
+  WHERE e.event = 'prescription issued' AND e.condition_type = 'chronic'),
+cohort AS (SELECT uid FROM crx GROUP BY 1),
+exp_calc AS (SELECT sum(
+   CASE WHEN t + INTERVAL 30 DAY <= TIMESTAMP '${END_TS}' THEN 1.0 ELSE 0 END
+ + CASE WHEN t + INTERVAL 60 DAY <= TIMESTAMP '${END_TS}' THEN 1.0 ELSE 0 END
+ + (2.0/3) * (CASE WHEN t + INTERVAL 90 DAY <= TIMESTAMP '${END_TS}' THEN 1.0 ELSE 0 END)
+ + (1.0/3) * (CASE WHEN t + INTERVAL 120 DAY <= TIMESTAMP '${END_TS}' THEN 1.0 ELSE 0 END)) AS expected_clones
+  FROM crx),
+r AS (SELECT e.uid, (e.condition_type = 'chronic' AND e.medication_type = 'chronic_maintenance') AS is_cm,
+  (c.uid IS NOT NULL) AS in_cohort
+  FROM ev e LEFT JOIN cohort c ON c.uid = e.uid WHERE e.event = 'prescription refill'),
+agg AS (SELECT count(*) FILTER (WHERE in_cohort) AS t_coh,
+  count(*) FILTER (WHERE in_cohort AND is_cm) AS cm_coh,
+  count(*) FILTER (WHERE NOT in_cohort) AS t_non,
+  count(*) FILTER (WHERE NOT in_cohort AND is_cm) AS cm_non FROM r)
+SELECT 'all' AS grp, (SELECT count(*) FROM cohort) AS user_count,
+ a.cm_non::DOUBLE / nullif(a.t_non, 0) AS organic_cm_rate,
+ ((a.cm_coh - (a.cm_non::DOUBLE / nullif(a.t_non, 0)) * a.t_coh)
+   / (1 - (a.cm_non::DOUBLE / nullif(a.t_non, 0)))) / nullif(x.expected_clones, 0) AS clone_yield
+FROM agg a, exp_calc x`,
+				},
+				select: { all: { where: { grp: "all" } } },
+				// ±3d jitter and boundary effects keep this near but not at 1.0
+				expect: { metric: "all.clone_yield", op: "between", target: [0.7, 1.35] },
+				minCohort: 80,
+			},
+			{
+				breakdown: {
+					type: "duckdb",
+					sql: `WITH ${ID_CTE},
+refill_users AS (SELECT DISTINCT uid FROM ev WHERE event = 'prescription refill'),
+crx AS (SELECT e.uid, e.t FROM ev e JOIN refill_users ru ON ru.uid = e.uid
+  WHERE e.event = 'prescription issued' AND e.condition_type = 'chronic'),
+cohort AS (SELECT uid FROM crx GROUP BY 1)
+SELECT 'all' AS grp, count(*) AS event_count, count(DISTINCT e.uid) AS user_count,
+ count(*) FILTER (WHERE e.condition_type = 'chronic' AND e.medication_type = 'chronic_maintenance')::DOUBLE / count(*) AS cm_rate
+FROM ev e LEFT JOIN cohort c ON c.uid = e.uid
+WHERE e.event = 'prescription refill' AND c.uid IS NULL`,
+				},
+				select: { all: { where: { grp: "all" } } },
+				// placebo: non-chronic-rx users' refills carry only the declared
+				// organic chronic∧chronic_maintenance mix (2/7 × 1/8 ≈ 0.036)
+				expect: { metric: "all.cm_rate", op: "between", target: [0.015, 0.06] },
+				minCohort: 200,
+			},
+		],
+	},
+	{
+		id: "H6-occasional-no-shows",
+		hook: "H6",
+		archetype: "cohort-count-scale",
+		narrative: `users with <${NO_SHOW_EVENT_THRESHOLD} events (at hook time, clones included) lose ${NO_SHOW_DROP_LIKELIHOOD}% of consultations and get no_show=true on ${NO_SHOW_DROP_LIKELIHOOD}% of bookings. The flag gives selection-free verification of a per-event effect on an activity-selected cohort: flagged ⇒ hook-count ≤ 14 ⇒ output count ≤ 14 (everything after only deletes), so users with ≥15 output events provably carry ZERO no_show=true rows (exact purity), and the no_show rate among ≤14-event users reads the knob (diluted slightly by unflagged users who slipped under 15 when future-dated clones were guard-dropped). The consultation-drop side is asserted as a direction-only composite: the flagged cohort is dominated by occasional/churner personas whose conversionModifier (0.7/0.3) organically lowers consult-per-booking, and H8's free-tier cliff skews zero-consult users into the small bin — the engineered 25% thinning is inseparable from that selection, which is exactly why the no_show flag exists`,
+		assertions: [
+			{
+				// deterministic purity
+				breakdown: {
+					type: "duckdb",
+					sql: `WITH ${ID_CTE},
+tot AS (SELECT uid, count(*) AS ct FROM ev GROUP BY 1)
+SELECT count(*) FILTER (WHERE e.no_show = true AND t2.ct >= ${NO_SHOW_EVENT_THRESHOLD}) AS big_noshows,
+ count(*) FILTER (WHERE e.no_show = true) AS all_noshows,
+ count(DISTINCT t2.uid) FILTER (WHERE t2.ct < ${NO_SHOW_EVENT_THRESHOLD}) AS small_users
+FROM ev e JOIN tot t2 ON t2.uid = e.uid WHERE e.event = 'appointment booked'`,
+				},
+				assert: (rows) => {
+					const r = (rows || [])[0];
+					if (!r || Number(r.all_noshows) === 0) return { pass: false, verdict: "NONE", detail: "no no_show=true bookings at all" };
+					const clean = Number(r.big_noshows) === 0;
+					return {
+						pass: clean,
+						verdict: clean ? "NAILED" : "INVERSE",
+						detail: `no_show=true on ≥15-event users: ${r.big_noshows} of ${r.all_noshows} total (must be 0; small-bin users=${r.small_users})`,
+					};
+				},
+			},
+			{
+				breakdown: {
+					type: "duckdb",
+					sql: `WITH ${ID_CTE},
+tot AS (SELECT uid, count(*) AS ct FROM ev GROUP BY 1),
+bk AS (SELECT e.uid, count(*) AS bookings, count(*) FILTER (WHERE e.no_show = true) AS noshows
+  FROM ev e JOIN tot t2 ON t2.uid = e.uid
+  WHERE e.event = 'appointment booked' AND t2.ct < ${NO_SHOW_EVENT_THRESHOLD} GROUP BY 1)
+SELECT 'small' AS grp, count(*) AS user_count,
+ sum(noshows)::DOUBLE / nullif(sum(bookings), 0) AS ns_rate
+FROM bk`,
+				},
+				select: { s: { where: { grp: "small" } } },
+				expect: { metric: "s.ns_rate", op: "between", target: [0.15, 0.3] },
+				minCohort: 150,
+			},
+			{
+				// composite direction check (selection + engineered thinning)
+				breakdown: {
+					type: "duckdb",
+					sql: `WITH ${ID_CTE},
+tot AS (SELECT uid, count(*) AS ct FROM ev GROUP BY 1),
+per AS (SELECT t2.uid, (t2.ct >= ${NO_SHOW_EVENT_THRESHOLD}) AS big,
+  count(*) FILTER (WHERE e.event = 'appointment booked') AS bk,
+  count(*) FILTER (WHERE e.event = 'consultation completed') AS cons
+  FROM tot t2 JOIN ev e ON e.uid = t2.uid GROUP BY 1, 2)
+SELECT CASE WHEN big THEN 'big' ELSE 'small' END AS grp, count(*) AS user_count,
+ sum(cons)::DOUBLE / nullif(sum(bk), 0) AS cons_per_bk
+FROM per WHERE bk > 0 GROUP BY 1`,
+				},
+				select: { s: { where: { grp: "small" } }, b: { where: { grp: "big" } } },
+				expect: { metric: "s.cons_per_bk / b.cons_per_bk", op: "between", target: [0.2, 0.85] },
+				minCohort: 150,
+			},
+		],
+	},
+	{
+		id: "H7-doctor-specialization",
+		hook: "H7",
+		archetype: "cohort-prop-scale",
+		narrative: `user hook: doctors get specialty from a real list and years_experience uniform [${DOCTOR_EXPERIENCE_MIN}, ${DOCTOR_EXPERIENCE_MAX}] (avg 22.5); nurses uniform [${NURSE_EXPERIENCE_MIN}, ${NURSE_EXPERIENCE_MAX}] (avg 9); patients pinned to 0. Deterministic per-role ranges — range violations are hook bugs, not noise`,
+		assertions: [
+			{
+				breakdown: {
+					type: "duckdb",
+					sql: `SELECT role AS grp, count(*) AS user_count,
+ avg(years_experience) AS avg_yx, min(years_experience) AS min_yx, max(years_experience) AS max_yx,
+ count(*) FILTER (WHERE specialty = 'none') AS none_ct
+FROM ${US} GROUP BY 1`,
+				},
+				assert: (rows) => {
+					const by = Object.fromEntries((rows || []).map(r => [r.grp, r]));
+					const d = by.doctor, n = by.nurse, p = by.patient;
+					if (!d || !n || !p) return { pass: false, verdict: "NONE", detail: `missing role rows (${(rows || []).map(r => r.grp).join(",")})` };
+					const bad = [];
+					if (Number(d.min_yx) < DOCTOR_EXPERIENCE_MIN || Number(d.max_yx) > DOCTOR_EXPERIENCE_MAX) bad.push(`doctor yx [${d.min_yx}, ${d.max_yx}] outside [${DOCTOR_EXPERIENCE_MIN}, ${DOCTOR_EXPERIENCE_MAX}]`);
+					if (Number(d.none_ct) !== 0) bad.push(`${d.none_ct} doctors with specialty='none'`);
+					if (Number(n.min_yx) < NURSE_EXPERIENCE_MIN || Number(n.max_yx) > NURSE_EXPERIENCE_MAX) bad.push(`nurse yx [${n.min_yx}, ${n.max_yx}] outside [${NURSE_EXPERIENCE_MIN}, ${NURSE_EXPERIENCE_MAX}]`);
+					if (Number(p.min_yx) !== 0 || Number(p.max_yx) !== 0) bad.push(`patient yx [${p.min_yx}, ${p.max_yx}] not pinned to 0`);
+					return {
+						pass: bad.length === 0,
+						verdict: bad.length === 0 ? "NAILED" : "INVERSE",
+						detail: bad.length ? bad.join("; ") : `ranges exact: doctor [${d.min_yx}, ${d.max_yx}], nurse [${n.min_yx}, ${n.max_yx}], patient pinned 0 (${d.user_count}/${n.user_count}/${p.user_count} users)`,
+					};
+				},
+			},
+			{
+				breakdown: {
+					type: "duckdb",
+					sql: `SELECT role AS grp, count(*) AS user_count, avg(years_experience) AS avg_yx FROM ${US} GROUP BY 1`,
+				},
+				select: { d: { where: { grp: "doctor" } } },
+				expect: { metric: "d.avg_yx", op: "between", target: [21, 24] },
+				minCohort: 40,
+			},
+			{
+				breakdown: {
+					type: "duckdb",
+					sql: `SELECT role AS grp, count(*) AS user_count, avg(years_experience) AS avg_yx FROM ${US} GROUP BY 1`,
+				},
+				select: { n: { where: { grp: "nurse" } } },
+				expect: { metric: "n.avg_yx", op: "between", target: [8, 10] },
+				minCohort: 80,
+			},
+		],
+	},
+	{
+		id: "H8-free-tier-cliff",
+		hook: "H8",
+		archetype: "funnel-conversion-by-segment",
+		narrative: `${FREE_TIER_DROP_LIKELIHOOD}% of free-tier users lose ALL consultations (per-user cliff). Estimator: (z_free − z_paid) / (1 − z_paid) where z = zero-consultation user share — the natural-zero baseline z cancels, and tier-blind processes (H6 thinning) cancel too, so the statistic reads the 0.30 knob directly. Sharp discriminator vs per-event thinning: SURVIVING free users are untouched, so their consult counts must match basic users (ratio ≈ 1.0); thinning would read ~0.7. The survivor comparison is SEGMENT-STANDARDIZED: tier and persona are sampled from the same seeded stream and come out measurably correlated (free skews occasional, premium skews provider), and persona eventModifier drives volume — raw cross-tier count comparisons are confounded by composition, standardizing on the persona-stamped segment removes it (thinning would still read ~0.7 within every segment)`,
+		assertions: [
+			{
+				breakdown: {
+					type: "duckdb",
+					sql: `WITH ${ID_CTE},
+per AS (SELECT u.distinct_id::VARCHAR AS uid, u.subscription_tier AS tier FROM ${US} u),
+cons AS (SELECT uid, count(*) AS ct FROM ev WHERE event = 'consultation completed' GROUP BY 1),
+j AS (SELECT p.tier, coalesce(c.ct, 0) AS ct FROM per p LEFT JOIN cons c ON c.uid = p.uid),
+z AS (SELECT count(*) AS user_count,
+ count(*) FILTER (WHERE tier = 'free' AND ct = 0)::DOUBLE / nullif(count(*) FILTER (WHERE tier = 'free'), 0) AS z_free,
+ count(*) FILTER (WHERE tier <> 'free' AND ct = 0)::DOUBLE / nullif(count(*) FILTER (WHERE tier <> 'free'), 0) AS z_paid
+ FROM j)
+SELECT 'all' AS grp, user_count, z_free, z_paid,
+ (z_free - z_paid) / nullif(1 - z_paid, 0) AS cliff_share FROM z`,
+				},
+				select: { all: { where: { grp: "all" } } },
+				expect: { metric: "all.cliff_share", op: "between", target: [0.24, 0.36] },
+				minCohort: 500,
+			},
+			{
+				breakdown: {
+					type: "duckdb",
+					sql: `WITH ${ID_CTE},
+cons AS (SELECT uid, count(*) AS ct FROM ev WHERE event = 'consultation completed' GROUP BY 1),
+surv AS (SELECT u.subscription_tier AS tier, u.segment AS seg, c.ct
+  FROM ${US} u JOIN cons c ON c.uid = u.distinct_id::VARCHAR),
+seg AS (SELECT seg,
+  avg(ct) FILTER (WHERE tier = 'free') AS f_avg, count(*) FILTER (WHERE tier = 'free') AS f_n,
+  avg(ct) FILTER (WHERE tier = 'basic') AS b_avg, count(*) FILTER (WHERE tier = 'basic') AS b_n
+  FROM surv GROUP BY 1)
+SELECT 'all' AS grp, sum(f_n + b_n)::BIGINT AS user_count,
+ sum(f_n * f_avg / b_avg) / sum(f_n) AS std_ratio
+FROM seg WHERE f_avg IS NOT NULL AND b_avg IS NOT NULL AND b_n >= 10`,
+				},
+				select: { all: { where: { grp: "all" } } },
+				// per-user cliff, not thinning: survivors untouched → ratio ≈ 1.0
+				// (b_n >= 10 is a stability guard against tiny-segment blowup)
+				expect: { metric: "all.std_ratio", op: "between", target: [0.9, 1.1] },
+				minCohort: 500,
+			},
+			{
+				// the documented Mixpanel funnel report, through the emulator.
+				// Window = funnel's 48h × H9's max stretch 1.4 (the free-tier
+				// timestamp scaling rides this funnel's booked→consult gap).
+				// Composite: the cliff (×0.7) compounds with H9 window censoring
+				// on free — band sits below the pure-cliff 0.70
+				breakdown: {
+					type: "timeToConvert",
+					steps: ["symptom search", "appointment booked", "consultation completed"],
+					breakdownByUserProperty: "subscription_tier",
+					conversionWindowMs: Math.round(48 * TTC_FREE_FACTOR * 3600 * 1000),
+				},
+				assert: (rows) => {
+					const by = Object.fromEntries((rows || []).map(r => [r.segment_value, r]));
+					const f = by.free, b = by.basic;
+					if (!f || !b) return { pass: false, verdict: "NONE", detail: `missing tier rows (${(rows || []).map(r => r.segment_value).join(",")})` };
+					const cf = f.step_counts[2] / f.step_counts[0];
+					const cb = b.step_counts[2] / b.step_counts[0];
+					const ratio = cf / cb;
+					const pass = ratio >= 0.55 && ratio <= 0.8;
+					return {
+						pass,
+						verdict: pass ? (Math.abs(ratio - 0.7) <= 0.07 ? "NAILED" : "STRONG") : (ratio < 1 ? "WEAK" : "INVERSE"),
+						detail: `funnel conversion free=${cf.toFixed(4)} basic=${cb.toFixed(4)} ratio=${ratio.toFixed(3)} (expect ~0.70, band [0.55, 0.80]; entered free=${f.step_counts[0]} basic=${b.step_counts[0]})`,
+					};
+				},
+			},
+		],
+	},
+	{
+		id: "H9-ttc-by-tier",
+		hook: "H9",
+		archetype: "funnel-ttc-by-segment",
+		narrative: `premium × ${TTC_PREMIUM_FACTOR} / free × ${TTC_FREE_FACTOR} on (a) wait_time_hours and duration_minutes (iid property scale — avg ratios read the knobs exactly; H2's flu doubling is tier-blind and cancels) and (b) the first booked→consult→follow-up sequence's timestamps (scaleFunnelTTC). The TTC assertions run through the Mixpanel-aligned emulator at a 2016h conversion window = max stretch ${TTC_FREE_FACTOR} × (2 gaps × 30d per-gap cap in findFirstSequence) — the window must cover the stretched support or censoring dilutes the free tier (the ai-platform H9 lesson). Only each user's FIRST sequence is scaled and the emulator's greedy first-conversion aligns with findFirstSequence's greedy scan, but organic re-conversions still dilute the measured ratio toward 1 — bands assume ≥25% of the full effect survives`,
+		assertions: [
+			{
+				breakdown: {
+					type: "duckdb",
+					sql: `WITH ${ID_CTE}
+SELECT subscription_tier AS grp, count(*) AS event_count, count(DISTINCT uid) AS user_count, avg(wait_time_hours) AS avg_wait
+FROM ev WHERE event = 'appointment booked' GROUP BY 1`,
+				},
+				select: { f: { where: { grp: "free" } }, b: { where: { grp: "basic" } } },
+				expect: { metric: "f.avg_wait / b.avg_wait", op: "between", target: [1.26, 1.54] },
+				minCohort: 300,
+			},
+			{
+				breakdown: {
+					type: "duckdb",
+					sql: `WITH ${ID_CTE}
+SELECT subscription_tier AS grp, count(*) AS event_count, count(DISTINCT uid) AS user_count, avg(wait_time_hours) AS avg_wait
+FROM ev WHERE event = 'appointment booked' GROUP BY 1`,
+				},
+				select: { p: { where: { grp: "premium" } }, b: { where: { grp: "basic" } } },
+				expect: { metric: "p.avg_wait / b.avg_wait", op: "between", target: [0.6, 0.74] },
+				minCohort: 300,
+			},
+			{
+				breakdown: {
+					type: "duckdb",
+					sql: `WITH ${ID_CTE}
+SELECT subscription_tier AS grp, count(*) AS event_count, count(DISTINCT uid) AS user_count, avg(duration_minutes) AS avg_dur
+FROM ev WHERE event = 'consultation completed' GROUP BY 1`,
+				},
+				select: { f: { where: { grp: "free" } }, b: { where: { grp: "basic" } } },
+				expect: { metric: "f.avg_dur / b.avg_dur", op: "between", target: [1.26, 1.54] },
+				minCohort: 300,
+			},
+			{
+				breakdown: {
+					type: "duckdb",
+					sql: `WITH ${ID_CTE}
+SELECT subscription_tier AS grp, count(*) AS event_count, count(DISTINCT uid) AS user_count, avg(duration_minutes) AS avg_dur
+FROM ev WHERE event = 'consultation completed' GROUP BY 1`,
+				},
+				select: { p: { where: { grp: "premium" } }, b: { where: { grp: "basic" } } },
+				expect: { metric: "p.avg_dur / b.avg_dur", op: "between", target: [0.6, 0.74] },
+				minCohort: 300,
+			},
+			{
+				breakdown: {
+					type: "timeToConvert",
+					steps: ["appointment booked", "consultation completed", "follow up scheduled"],
+					breakdownByUserProperty: "subscription_tier",
+					// 2016h = 1.4 × 2 gaps × 30d per-gap cap (covers stretched support)
+					conversionWindowMs: 2016 * 60 * 60 * 1000,
+				},
+				select: { f: { where: { segment_value: "free" } }, b: { where: { segment_value: "basic" } } },
+				expect: { metric: "f.median_ttc_ms / b.median_ttc_ms", op: "between", target: [1.04, 1.44] },
+				minCohort: 150,
+			},
+			{
+				breakdown: {
+					type: "timeToConvert",
+					steps: ["appointment booked", "consultation completed", "follow up scheduled"],
+					breakdownByUserProperty: "subscription_tier",
+					conversionWindowMs: 2016 * 60 * 60 * 1000,
+				},
+				select: { p: { where: { segment_value: "premium" } }, b: { where: { segment_value: "basic" } } },
+				expect: { metric: "p.median_ttc_ms / b.median_ttc_ms", op: "between", target: [0.6, 0.97] },
+				minCohort: 150,
+			},
+		],
+	},
+	{
+		id: "H10-consult-count-magic-number",
+		hook: "H10",
+		archetype: "frequency-sweet-spot",
+		narrative: `sweet ${CONSULT_SWEET_MIN}-${CONSULT_SWEET_MAX} consultations → consultation_fee × ${CONSULT_FEE_BOOST}; over ${CONSULT_OVER_THRESHOLD}+ → days_until_followup × ${CONSULT_FOLLOWUP_STRETCH}. Both are property-only mutations on cohorts the output rebuilds exactly (counts run after all filters). Median ratios are selection-free: scaling a whole cohort's iid draws scales every quantile by the knob. H1's after-hours boost rides all count-cohorts equally (hours are count-independent). The days assertion filters to consultation_mode='phone' follow-ups — H4's injected clones are always video with a different days distribution, and the over-cohort receives more clones`,
+		assertions: [
+			{
+				breakdown: {
+					type: "duckdb",
+					sql: `WITH ${ID_CTE}, ${CONSULT_CTE},
+coh AS (SELECT uid, CASE WHEN ct BETWEEN ${CONSULT_SWEET_MIN} AND ${CONSULT_SWEET_MAX} THEN 'sweet'
+  WHEN ct >= ${CONSULT_OVER_THRESHOLD} THEN 'over' ELSE 'low' END AS grp FROM cc)
+SELECT c.grp, count(DISTINCT c.uid) AS user_count, count(*) AS event_count, median(e.consultation_fee) AS med_fee
+FROM coh c JOIN ev e ON e.uid = c.uid AND e.event = 'consultation completed' GROUP BY 1`,
+				},
+				select: { s: { where: { grp: "sweet" } }, l: { where: { grp: "low" } } },
+				expect: { metric: "s.med_fee / l.med_fee", op: "between", target: [1.12, 1.4] },
+				minCohort: 60,
+			},
+			{
+				breakdown: {
+					type: "duckdb",
+					sql: `WITH ${ID_CTE}, ${CONSULT_CTE},
+coh AS (SELECT uid, CASE WHEN ct BETWEEN ${CONSULT_SWEET_MIN} AND ${CONSULT_SWEET_MAX} THEN 'sweet'
+  WHEN ct >= ${CONSULT_OVER_THRESHOLD} THEN 'over' ELSE 'low' END AS grp FROM cc)
+SELECT c.grp, count(DISTINCT c.uid) AS user_count, count(*) AS event_count, median(e.days_until_followup) AS med_days
+FROM coh c JOIN ev e ON e.uid = c.uid AND e.event = 'follow up scheduled' AND e.consultation_mode = 'phone'
+GROUP BY 1`,
+				},
+				select: { o: { where: { grp: "over" } }, s: { where: { grp: "sweet" } } },
+				// Math.round on small integer days adds up to ~5% bias
+				expect: { metric: "o.med_days / s.med_days", op: "between", target: [1.3, 1.75] },
+				minCohort: 60,
+			},
+		],
+	},
+];
+
+export default config;