Front door for the study. The reasoning and the sourcing are in the Consolidated Brief; the per-area evidence is in the chapters.

The one-sentence version

MEDvidi sells a $195 video appointment with a doctor who can prescribe Adderall, takes the money before finding out whether you qualify, and has built a real technology platform underneath it — and the federal rule that makes the whole thing legal expires on 31 December 2026.

What it is

An American online psychiatry clinic. You pick a condition, you pay $195, you have a 30-minute video call with a licensed prescriber, and if they agree, a prescription goes to your pharmacy. Follow-ups are $159 for fifteen minutes. No insurance is accepted. Thirty-five states.

Most of its patients come for ADHD, and most of those want stimulant medication. You can tell without being told: the website is organised by drug name — Adderall, Vyvanse, Concerta, Xanax, Ambien — and it has twenty-eight separate landing pages for "ADHD treatment" in individual states against four for anxiety.

What it actually is

Not one website. The company runs six clinic brands on one shared platform — and the default brand in its own code is not MEDvidi, it is EZCare Clinic. Four separate applications sit behind the consumer site: one for patients, one for doctors, one for the sales team, one for administrators. Thirty-plus engineers make sense in that light; they make no sense if you think the product is a booking form.

The mechanic that explains everything else

You pay before anyone checks whether you qualify. The order is: pay, then fill in your medical history, then wait while the company runs your name through the state controlled-substance database, then get told whether you are accepted.

Nothing in the signup flow shows you the price before you have created an account and handed over six personal details. The price lives halfway down a marketing page and inside a collapsed FAQ.

That sequence is the business's cleverest and most expensive decision at once. It converts far better than showing the price first — people who have already invested effort finish. And it is the direct cause of the company's reputation: about a quarter of its Google reviews are one star, with almost nothing in the middle. Two different journeys share one checkout.

Where the anger actually comes from

Not from being refused. Somebody read two hundred one-star reviews and sorted them, and the single biggest complaint is that the prescription never reaches the pharmacy — the doctor said yes, the visit went fine, and then the electronic prescription sits unsigned, or goes to the wrong pharmacy, or arrives with the wrong state licence on it.

That category has not improved in three years, while two harder problems were solved: doctors not showing up fell from 28% of one-star reviews to 7%, and pharmacies refusing to fill went from 14% to zero. The easy-looking problem is the one left alone.

There is a second thing hiding in the five-star reviews: 43% of them name the support agent who fixed something that had gone wrong. The rating is a service-recovery score wearing a care-quality costume.

How the money works

About 15,000 visits a month, at $195 and $159, with roughly three-quarters of each visit price left over after paying the doctor. Doctors are contractors paid $30–$70 per completed appointment. New patients arrive almost entirely through Google, essentially for free — paid advertising is about half a percent of traffic.

That combination — cheap patients, fat margins — is what lets a company this size carry a 53-person product and engineering team.

And here is the number the whole thing turns on. A stimulant prescription cannot be refilled; American law requires a new prescription every time. So the medicine itself demands roughly twelve visits a year. MEDvidi collects 3.43.

Half of that gap is churn, and the company's own published figures prove it: about 47% of its ADHD patients never come back after the visit they already paid for.

The thing that makes this urgent

Prescribing a controlled substance to someone a doctor has never physically examined is normally illegal in the United States. It has been allowed since 2020 under a temporary emergency rule that has been extended four times. It currently expires on 31 December 2026.

As of August 2026 the DEA has published nothing in 2026 to replace it. The only proposal on the table would end telephone-and-video-only stimulant prescribing outright.

There is one permanent escape: if a patient sees that specific doctor in person once, that pair is exempt forever. MEDvidi built exactly this in 2023, when the rule looked like it would lapse, then quietly deleted it when the extension came.

But the escape is harder than it looks. The law says the prescribing doctor has to do the in-person exam — so a referral from a family doctor, a walk-in clinic or an urgent care is legally worthless. And because MEDvidi's doctors are contractors, every doctor who quits takes their patients' exemption with them.

The trap in the good news

If the rule tightens, MEDvidi's most profitable line shrinks.

If the rule instead makes telehealth stimulant prescribing normal and clean, then Hims & Hers — 2.9 million subscribers, its own pharmacy, and a current policy of not prescribing amphetamines that is a choice rather than a legal constraint — can walk into this market at a price MEDvidi cannot match.

There is no version of the next year in which the regulation is simply good news.

What the AI is for

Four AI features are running: one that writes the doctor's notes by listening to the visit, one that reads every patient chart looking for deviations from clinical protocol, a chatbot and voice agent handling patient contact, and a prescribing assistant announced in April.

The interesting finding is that the note-writing one cannot save the company money, because doctors are paid per appointment, not per hour. Time saved goes to the doctor, not to MEDvidi. It is a recruiting tool, not a margin tool.

The chart-reviewing one is the important one, and not for the reason you would guess. The founder of MEDvidi's closest competitor, Done, was sentenced in July 2026 to six years in federal prison, and the government's case was built on how the company's product was designed: a monthly subscription, automatic refills, half-length first visits, doctors paid by prescription volume. MEDvidi's design is the opposite of every one of those — which is not an accident, because it changed to this design in 2022, in exactly the window its competitors were being investigated.

So a system that reads 100% of charts for protocol compliance is not a productivity tool. It is evidence. The catch is that a flagged problem nobody resolves is better evidence for a prosecutor than for a defendant, and nobody outside the company can see whether those flags get closed.

There is also a gap that costs nothing to fix: the company runs a mental-health chatbot in 35 states and publishes no crisis protocol anywhere — no mention of suicidal ideation, self-harm or 988. California law requires operators to publish one before the chatbot may talk to anyone.

Is there a second business?

MEDvidi runs a separate site selling its AI system to other clinics, health systems and — the form literally offers this as an option — investors.

Looked at closely, that site is a single page on a bought template, hosted on Netlify, not linked from the main site, with a dead privacy link, a dead terms link and a commented-out brochure button. An earlier version labelled the prescribing AI "Vision — Est. 2027" and stated plainly that it was "not yet FDA-cleared." The rebuilt version deleted that sentence, deleted the disclosure that the press logos had been purchased, and deleted three executives — including the person whose job this is.

The company appears in no FDA database at all. And of 65 open jobs, not one is a sales role.

So: the intention to sell this technology is real and documented. A second business with an actual customer is not. The most likely near-term audience for that page is capital, not customers.

And the thing nobody noticed

MEDvidi is quietly becoming an insurance business. It hired a director in May 2026 to launch commercial insurance billing in Florida and California this year, and told product-manager candidates they would be integrating insurance carriers into the booking flow.

That matters because the arithmetic is unkind: Medicare pays about $171 for a psychiatric evaluation and about $120 for a routine follow-up, nurse practitioners are paid 85% of that by regulation, and commercial rates in mental health tend to sit at or below Medicare. MEDvidi's $169 average cash visit becomes roughly $108. Volume has to rise by half just to stand still.

Insurance here is a volume bet at a lower price, not a margin improvement — and the company is already staffing for it.

The honest caveats

• Most of the operating numbers in this study are MEDvidi's own, and several come from a retrospective marketing page written in 2026 about work done in 2024. They are the best evidence available and they are still marketing.
• Two traffic-estimate vendors disagree about this website by 72% on level, and disagree about whether traffic is going up or down in the same month. No traffic figure here is safe to repeat.
• Nobody outside the company knows the one number that decides how bad December is: what share of visits actually depends on the expiring rule.
• Patient reviews are self-selected. They tell you what goes wrong, not how often.

The rehearsal layer. Everything here is drawn from the Consolidated Brief and the chapters; this document exists because the interview is in days and a study you have to re-derive under pressure is not a study you can use.

1. The ninety-second answer

For when it comes — and it will, in some form: "So what do you make of our business?"

"You're a cash-pay prescribing practice with a real platform underneath it — four applications and six brands on one release train, which is not what the marketing site suggests and is probably why you have thirty engineers.

The thing I keep coming back to is 3.43. A Schedule II script can't be refilled, so the medicine demands roughly twelve visits a year and you collect three and a half. And your own RR1 series says why — it went 41 to 53, which means about half your ADHD patients still don't come back after a visit they already paid for. That's the whole P&L in one ratio, and it's the one thing in the business the product organisation directly owns.

The second thing is that I don't think December is a fire drill. One in-person evaluation permanently cures the practitioner–patient pair, CMS brings in its own in-person requirement for home-based tele-mental-health in January 2028, and you're already staffing a revenue-cycle function for Florida and California. So the in-person capability isn't a hedge against the DEA — it's the prerequisite for the insured version of this company too. That's the one thing I'd fund before I knew how December goes.

And the thing I'd want from you: what share of visits actually needs the flexibility? Everything about the site says it's high, but a business where that's 70 percent and one where it's 30 need different product organisations, and I can't split the difference from outside."

Three moves in that: the platform reframe (shows you looked at the code, not the brochure), a number with its driver (shows arithmetic), and a question that only lands if the arithmetic is real.

2. The three numbers

Not the biggest three — the three whose movement decides whether there is a business in eighteen months.

One · 3.43 visits per patient per year, driven by RR1 at 53%

The only one of the three the product organisation owns directly. Price hasn't moved in 33 months, so every dollar of growth that isn't a new patient comes from this ratio. Moving RR1 from 41% to 53% was worth about +19% on visits per patient — roughly $90 per patient-year, ~$3M against 35,000 patients, from one funnel change with no price change and no new clinician.

If challenged: "120,000 divided by 35,000, both off your About Us page. Both are stated with a plus, so it's a point estimate from two floors."

Two · The share of visits that needs the DEA flexibility, against 134 days

The only number whose movement is binary. Don't guess it — ask for it. Asking is the stronger move, and it signals you know it's the number that sizes the problem.

What you can say about it: the information architecture is built on Adderall, Vyvanse, Concerta and Xanax; 28 programmatic ADHD state pages against 4 for anxiety; a permanent requisition for prescribers with Schedule II authority; the company pays 117 people's DEA registrations.

Three · Paid search at ~0.5% of sessions — acquisition concentration

There is no second channel at scale. Pair it with number one: 47% don't return, so the business depends on new-patient flow just to stand still, and that flow is a single unhedged channel nobody controls. That is what kills companies in this category — not a bad quarter, a Google update.

If challenged: "Third-party modelled, so directional. But 1,400 paid sessions against 153,850 organic isn't a rounding error in the direction that matters."

The line to have ready when he pushes back on a number

"I've built every figure I'm using out of things you've published, and I've written down which ones I don't trust. The two I'd throw out are the traffic numbers — your two vendors disagree by seventy-two percent on level and they disagree on whether it's going up or down — and anything derived from them. The one I'd most like from you is the split of visits by DEA schedule, because that's the number that decides what December is."

A candidate who volunteers his own error bars before being asked for them does not get tested on numbers again.

3. Do not say these

4. The five questions

Ranked, and split by audience — asking a recruiter question in a product interview costs more than it returns.

For the CEO conversation:

1. "What share of your follow-up book is Schedule II–V? That's the number that sizes December and sizes the Prescribing Assistant, and I couldn't find it anywhere."
2. "Your requisitions name nine teams; the posting says two tracks. Which of those nine still exist, who's the PM on each, and which two would you merge if you had to merge two?" — answerable, flattering in the right way, and whichever two he names tells you what he already thinks is broken.
3. "Your Core PM requisition says that role integrates insurance billing carriers into the core funnel, and you hired a Director of Revenue Cycle Management in May for Florida and California this year. Which product team owns eligibility and prior auth in the booking flow — and does the $195 price survive it?"
4. "When the AI agent last met a patient expressing suicidal ideation, what did it say, who saw it, and how long did it take them?" — if there's an answer, the company is further along than its public surface suggests. If there isn't, your first 30 days are already written.
5. "There are 65 open roles and not one of them sells. If the AI Clinic Track is a product for other clinics, who sells it — and is medvidi.ai for partners or for investors?" Hold this one until late, and ask it plainly. The answer defines the job.

For the recruiter, not the CEO:

• The offer discrepancy. MEDvidi's own careers page lists this role as a contract, B2B, Poland / Remote-EU engagement with a probation period and Polish-market benefits, and does not mention equity at all; the brief you were sent promises "competitive compensation and equity aligned with impact." Both documents are public, so it can't be deflected — but it's an economics question, not a product question. Put it to the recruiters, before or after the CEO round.
• Whether the VP of Engineering reports to you or to the co-founder who is also Smart IT's CEO and MEDvidi's titular CTO. This determines whether the job is real. It is a late-stage question.

5. The traps

The membership trap. If asked how you'd raise lifetime value, do not reach for a subscription without drawing the line first. Done's prosecuted mechanism was medication without an encounter — a monthly fee plus auto-refill explicitly designed to minimise follow-up appointments. A membership that prepays encounters is a different object. Say the distinction before you say the idea:

"The packaging question is real — you're collecting a third of the cadence your own medicine requires. But the shape matters more than the idea. Done sold a subscription that delivered medication without a visit, and that's what the indictment was about. Prepaid encounter bundles are the opposite structure, and you've already priced one — the weight-loss page sells a two-visit bundle at $249."

The "what would you do first" trap. Seven different areas of this business each generate a credible top priority. If you list them you sound like a consultant. Lead with the sequencing principle instead: fund first what pays in every branch of December. Then name one thing, and name what you're not doing.

The AI-autonomy trap. The CEO has publicly said the AI does not make independent decisions and that this is "the architecture we intentionally built." Do not propose more autonomy. Propose the evidence system that would let him earn it when state sandboxes open — which he himself flags as the direction.

The number-correction trap. If he corrects a figure, don't defend it. Ask what it should be and what it's measuring. The study's whole quantitative dispute dissolved once figures were placed next to their dates; the same move works live.

6. First 90 days, if asked

Three things, sequenced, with what you're not doing said out loud.

1. The in-person evaluation capability, scoped to the top five states covering 44% of exposure — about 112 clinician-days at a 45% attendance rate, roughly half a million dollars against a $14.6M revenue line, three-to-six-month payback. Not the 2023 flow: that implemented a proposed rule that never became law, and there is no referral cure — §829(e) requires the prescribing practitioner to do the exam. And because the cure attaches to a practitioner–patient pair held by a contractor, clinician retention becomes a compliance variable.
2. Pre-payment qualification — state, age, licensure match, excluded drug classes, the PDMP-visible patterns that reliably decline. It reduces paid initial visits and you should say so. It is also the December contingency, because that screen is where a lot more people are about to be told no.
3. A clock on the transmission queue — 34% of one-star reviews, the only failure category that never improved in three years, and the platform already has the timestamps. An ageing view, an escalation rule, a status the patient can see.

Not doing: a native app; a new state before December; brand consolidation; touching the editorial/E-E-A-T machine; re-litigating the $195 headline price; an outcome-priced guarantee.

7. If they ask about evals

The posting names AI evals more times than anything else, and the company's own requisitions describe eval as production-behaviour analysis with dashboards, mentioning release gates only as a QA nice-to-have.

"The gap isn't measurement, it's gating. You already watch the AI; nothing currently blocks on it. And three of the things I'd need already exist in production: the AI-mode toggle-off rate in the Doctor Cabinet is an implicit trust metric, the chart reviewer has per-rule scores, and there's a report-ai-mistake endpoint that is a human-in-the-loop label channel nobody appears to be harvesting. Structured reason codes on that endpoint, one queue across the clinician-facing features, a ten-day disposition SLA — two quarters of that produces a better labelled corpus than any golden set you could buy, at zero marginal cost, because the clinicians are already paid per visit."

And the framing that makes it fundable:

"In this category evals aren't a quality programme. Chart Review AI writes a permanent, timestamped, discoverable log of every protocol deviation it detects — and DOJ named clinical protocols as one of the instruments of the Done offence. A flagged deviation nobody closed is a better exhibit for a prosecutor than for a defendant. Closing that loop is the cheapest insurance this company can buy."

8. Five things you know that they may not expect you to

Use sparingly — one or two, where they're load-bearing. Deployed as a list they read as a gotcha reel; deployed singly they read as diligence.

1. The default tenant in your own patient-portal config is EZCare Clinic, not MEDvidi — six brands, four apps, one release train.
2. Your ADHD self-test has its subtype labels transposed in the scoring code. A high-inattentive profile is told it's predominantly hyperactive. Ten characters of JavaScript, on the highest-intent page in the estate.
3. The funnel loads no consent-management platform while firing Google, Microsoft and an OpenAI ad SDK on the screen where the patient declares their condition — while the marketing site loads Cookiebot 82 times. The FTC's Cerebral theory didn't need HIPAA.
4. Your Terms say "we do not offer refunds" and your Refund Policy lists six full-refund conditions, and both sit behind the same single checkbox — along with five other instruments, positioned behind the sticky CTA on a 390px screen.
5. You published a "not yet FDA-cleared" note on the AI Prescriber in September 2025 and deleted it in the April rebuild. Worth knowing before anyone says "FDA pathway" out loud.

Item 3 or 4 is the one to use if you only use one: both are cheap to fix, both are the kind of thing a CEO notices you noticed, and neither requires you to be right about strategy.

Every claim below is tagged to the Area that proves it. [N] is the number set, which adjudicates every quantitative dispute in the study. Where two Areas disagreed, §7 says how it was resolved. Where nothing resolved it, §9 says so.

Kicker. MEDvidi looks like a $195 online psychiatry clinic and is actually a six-brand multi-tenant prescribing platform whose per-visit price is not a pricing decision but a compliance artefact — the thing it changed in 2022 to avoid becoming the company whose founder was sentenced to six years in federal prison in July 2026. That architecture is its best asset and its binding constraint: it caps revenue per patient at 3.43 visits a year against a clinical clock that demands twelve. Everything the product organisation could usefully do lives in that gap, and it has 134 days to do it before the federal rule the whole controlled-substance line runs on expires.

1. Bottom line

Six things are true at once, and holding all six is what the job requires.

1. The price architecture is a compliance artefact, not a positioning choice. MEDvidi sold a $49/$199/$219 subscription with therapy and unlimited messaging in May 2022, went to variable per-visit pricing from $25 by December 2022, and settled on flat $195/$159 by November 2023 — which is the exact window in which Cerebral retreated and Done Global was investigated [A1] [A6]. Done's founder was sentenced on 7 July 2026 to 72 months for a business model DOJ described in product terms: monthly subscription, auto-refill, half-length initial visits, volume-based clinician pay [A4]. MEDvidi's current design inverts every one of those. That is the most valuable thing the company owns and it is invisible to patients, to clinicians and to the market [A6].

2. The same architecture caps the business. A Schedule II stimulant cannot be refilled; a new prescription is legally required every time [A3] [A4]. So the clinical clock says twelve visits a year and the commercial system collects 3.43 [N]. MEDvidi realises about 30% of the revenue its own medicine dictates [A3].

3. Half of that gap is churn at the first visit, and the company knows it. MEDvidi's own roadmap publishes an ADHD return rate moving 41% → 53% [A10]. Run through a retention model that is exactly what moved visits per patient from ~2.9 to 3.43 [N]. Which means roughly 47% of ADHD patients never return after a visit they already paid $195 for [N].

4. The largest source of patient anger is not the clinical gate — it is the prescription failing to arrive. The one-star corpus, hand-coded, puts fulfilment failures at 34.1% against 18.4% for "the provider declined to prescribe" [A9]. It is the only complaint category that has never improved in three years, while provider no-shows (28% → 7%) and pharmacy network problems (14% → 0%) were both solved [A9].

5. The federal rule expires on 31 December 2026 and there is no referral cure. DEA has published nothing in 2026 — no extension, no final rule, no new proposal [A4]. The only proposal on the table would end telemedicine-only Schedule II prescribing outright [A4]. And 21 U.S.C. §829(e) requires that the prescribing practitioner personally conduct the in-person evaluation, so a PCP referral, a MinuteClinic visit and an urgent-care note are all legally worthless [A11].

6. The company is already changing its business model, and the study nearly missed it. MEDvidi hired a US Director of Revenue Cycle Management in May 2026 to build a commercial insurance revenue cycle "beginning with Florida and California in 2026", and told Core-Track PM candidates two months earlier they would "integrate insurance billing carriers into the core funnel" [A13]. Every other agenda in this study was argued on a cash-pay architecture.

The synthesis. MEDvidi is not facing a demand problem or a supply problem. It is a business that grew for three years by adding clinicians and states to a funnel it never optimised, and both of those runway items expire inside a year — the 117-person roster tops out around 220–240k visits (~$37–40M ARR), four to seven months away at the claimed growth rate [N], and the content engine is at terminal inventory with net −8 posts since August 2025 [A2]. Neither thing that produced the last three years can produce the next one. The leverage has to come from the middle.

2. What MEDvidi actually is

2.1 Four apps, six brands, one release train

The public production config in the patient-portal bundle names six consumer clinic brands sharing one codebase — default tenant EZCare Clinic, plus MMJ Doctor, ESAcare, MEDvidi, Mango Clinic and MDBerry — on a platform domain dctr.app [A1] [A2]. Four separate Angular applications (patient, clinician "Doctor Cabinet", CRM, admin) plus a video app deploy from one release train against one API [A1]. BBB independently lists EZCare Clinic and Mango Clinic as alternate names of MEDvidi, Inc. [A1].

Consolidation is in flight: mangoclinic.com now 301-redirects to medvidi.com, and ezcareclinic.io already serves the page title "MEDvidi | Mental Health Clinic USA" [A2]. But EZCare is still a live revenue surface with its own analytics, its own funnel at start.ezcareclinic.io/signup, a different 20-state list including New Jersey and Oklahoma that MEDvidi does not serve, and — contradicting three Findings' claim that the company exited therapy — a live offer of 60-minute therapy sessions with "a personal licensed therapist" [A7] [A10].

The estate is configured as Website × Service × State × Plan [A1]. That is why thirty engineers make sense; it is not a WordPress site plus a booking form.

The correction this forces. If the candidate walks in describing medvidi.com, he is describing the marketing veneer [A1]. The reframe — "this is a multi-tenant clinical operations platform with a consumer front end" — is not in the job posting and is checkable in one click.

2.2 The mechanic: pay, then qualify

The portal's own analytics event names run appointment flow → payment step → successful payment page → intake step, and the lead only enters WAITING_FOR_INTAKE once isPaid is true [A1]. After payment comes a PDMP and intake verification gate with named rejection reasons — verification_issues_pdmp_drugs, verification_issues_pdmp_medication, provider_state_not_matching_patient_residence — and the admin app carries a first-class "Bamboo Health Credentials" entity, Bamboo being the PDMP gateway operator [A1].

Walked in a browser on 2026-08-19 with every write request blocked at the network layer, the funnel is six screens on the ADHD assessment path, four on ADHD transfer, and exactly two for anxiety, depression, insomnia, weight loss, ESA and OCD — one tap from a checkbox to the account wall [A8]. ADHD is the only service with any qualification screens at all, and its two questions are preference questions ("what bothers you most"), not clinical or eligibility ones [A8].

No price appears anywhere in the funnel — not on the entry screen, not on any qualification screen, not at the account wall. The published price lives roughly halfway down a service page and inside a collapsed FAQ accordion, hedged as "the majority of our services" [A8].

Reconciliation note. [A1] read this sequence out of a JavaScript bundle and called it "behaviourally sophisticated"; [A8] walked it and found the dominant mechanism is neither scarcity nor urgency — MEDvidi is cleaner on the standard dark-pattern taxonomy than its reputation suggests, with no countdown, no pre-checked opt-in, no confirmshaming, no hidden-cost reveal [A8]. The mechanism is sequence: everything upstream of the card form exists to maximise disclosed sunk effort, and a price shown at 25% progress with six fields already filled is a decision made on sunk cost rather than on price [A8]. Resolved in favour of [A8], which saw it.

2.3 Where the exposure actually sits

Not in persuasion. In three specific artefacts, all measured [A8]:

• One consent checkbox bundles seven legal instruments — Terms of Use, Privacy Policy, Terms and Conditions, Telehealth Consent, AI Usage Consent, Refund Policy, SMS delivery — and at 390×844 it occupies y 751–877 with the sticky CTA at y 721–769, i.e. entirely behind the button and partly off-screen.
• The two documents behind that one checkbox contradict each other. The Refunds & Fees Policy lists six full-refund conditions; Terms and Conditions §(f) says flatly "We do not offer refunds." The same T&C authorises auto-renewal in capitals and never states any cancellation method — while the public FAQ says "we have no subscription model."
• The funnel loads no consent-management platform at all while firing Google enhanced conversions, Microsoft UET and an OpenAI advertising SDK on the screen where the patient declares their condition. The marketing site loads Cookiebot 82 times; the funnel loads it zero times. This is the structural shape of the FTC's Cerebral theory, which did not require HIPAA [A8] [A4].

That last item also contains the study's only genuinely new commercial signal: the OpenAI ads SDK is present on join.medvidi.com and absent from medvidi.com [A8]. Somebody has already instrumented an assistant-surface acquisition channel.

3. How the money works

3.1 The number set, adjudicated

Seven Scouts printed three revenues, three clinical bench sizes and two traffic levels 72% apart. Almost all of it dissolves once each figure is placed next to its date and its measurement basis [N].

The ARR-versus-trailing reconciliation is the load-bearing move. ARR is a forward-annualised run rate; trailing-twelve-month billings are what was collected. At the company's own stated 100% YoY growth, TTM/ARR = g / ((1+g)·ln(1+g)) = 0.7213, so ARR runs 1.386× trailing [N]. Apply it: $27M ARR × 0.7213 = $19.5M trailing, against $20.34M computed independently from 120,000 visits at published prices — a 4% gap [N]. Two figures the study treated as irreconcilable are the same number measured two ways.

The rule for the room: say the series, not a number.

3.2 The 3.43, and why it is a churn number

Two Findings each held half of this and neither knew [N]. Joining them: with RR1 at 53% and observed visits per patient at 3.43, the post-visit-2 retention rate solves to 78.2%; roll RR1 back to 41% and expected visits fall to 2.88 — a +19% lift, worth about $90 per patient-year, roughly $3M a year against 35,000 patients, produced by one funnel change with no price change and no new clinician [N].

And the headline stands even if the model is wrong in detail: 47% of ADHD patients do not come back, on the company's own metric. A days-supply artefact cannot make half the patients disappear after the visit they already paid for. Days supply shapes the tail; churn cuts the head [N].

3.3 The unit economics, and what they buy

Clinicians are 1099 contractors on a published per-completed-visit rate card — psychiatrist $70/$50, physician $60/$40, PMHNP $45/$30 [A5]. Against $166.20 blended revenue and Stripe fees, contribution is 64–77% of the visit price, most likely 72–74% [N]. The PMHNP follow-up at ~81% clinical margin is the most profitable unit MEDvidi sells [N].

That margin plus near-zero paid acquisition is what lets a ~$30M clinic carry a 53-person product and engineering organisation [A3] [N]. It also names the fragility precisely: the business is a levered bet on a single organic search position [A3].

Do not use revenue-per-head as an argument about org size. [A3] computed it on a whole-company denominator and attributed it to the 54-person org; corrected, it is $376k–$556k against LifeMD's audited $499k, and the comparison is not like-for-like on either side because LifeMD's affiliated medical group is excluded from its employee count [N]. The interesting question survives without the ratio: a clinic doing 15,000 visits a month does not need thirty engineers for a booking flow, so what are they building?

3.4 The membership the FAQ denies

The shipped patient portal contains a fully built, Stripe-backed membership — MEDvidi Prime — with cards, billing history, dunning, a cancellation flow with a save-offer, reactivation, and a 28-event funnel including a Stayed With Membership event, behind a feature flag Membership_Launch_Test_1, plus a targeted offer that fires at the exact moment a patient owes a $50 late-reschedule fee [A1] [A3]. The public FAQ says, in the present tense, "we have no subscription model" [A2] [A5].

Reconciliation note — the study's sharpest internal conflict. [A6] warns that "a VP who arrives and proposes a membership to improve LTV will be proposing the business model the DOJ successfully prosecuted." [A3] makes Prime its load-bearing recommendation. Both are right about different objects, and the distinction is the whole difference between Prime and the indictment: Done's prosecuted mechanism was medication without an encounter — a monthly fee plus auto-refill explicitly designed to minimise follow-up appointments [A4]. Prime, on its own copy, covers monthly visits. A membership that prepays encounters is a different object from one that delivers pills without them. Nobody in the study stated this; it is the sentence that makes the recommendation safe.

Note also that [N] could not reproduce the $159 price or the monthly interval from the public bundle. Say "the membership is shipped and flag-gated; its price and interval are not observable from outside" — do not quote the $581 → $1,908 figure [N].

4. December, correctly understood

4.1 What actually expires

MEDvidi's controlled-substance business runs on a temporary rule expiring at the end of 31 December 2026, and as of August 2026 DEA has published nothing in 2026 to replace it — no fifth extension, no final rule, no new proposal [A4]. The only replacement on the table is the January 2025 Special Registration proposal, which as written would end telemedicine-only Schedule II prescribing outright: it caps telemedicine Schedule II scripts at under 50% of a prescriber's monthly Schedule II total and requires the prescriber to be physically in the patient's state [A4].

4.2 The cure exists, is smaller than feared, and is structurally worse

One in-person evaluation permanently exempts that practitioner–patient pair from the remote-prescribing rules [A4]. MEDvidi already built this once: a May 2023 Wayback capture shows a fully-formed continuity product — get an in-person appointment, obtain a qualifying telemedicine referral, send it before the deadline — quietly deleted when the extensions came [A4].

Three corrections from the feasibility pass, and they change the plan [A11]:

• Smaller. Roughly 13,600 active Schedule II patients need an in-person evaluation — five to eleven clinician-days each across the panel. A top-five-state programme covering 44% of the exposure costs about half a million dollars against a $14.6M revenue line: a three-to-six-month payback.
• There is no referral cure. §829(e) requires the prescribing practitioner to have conducted the evaluation. The 2023 flow implemented a proposed rule that never became law, so rebuilding it as designed rebuilds a compliance artefact for a regulation that does not exist.
• The cure is held by a contractor. Because the exemption attaches to a practitioner–patient pair held by a 1099 clinician, every clinician who quits re-orphans their entire cured panel — at 30% churn, roughly 4,000 patients a year, forever.

That last point converts a compliance project into an org-design argument: if the cure is the asset, the company has to own the people who hold it [A11]. Nobody in the room will have said that.

4.3 The asymmetry nobody states

If the framework tightens, MEDvidi's Schedule II franchise shrinks. If it normalises telehealth controlled-substance prescribing with clean rules, then Hims & Hers — 2.9 million subscribers, $753M a quarter, its own pharmacy, and a stated policy of not prescribing amphetamines that is a choice rather than a constraint — can enter at a price MEDvidi cannot match [A6].

There is no version of the next four quarters in which regulation is simply good news [A6]. Which is the argument for converting the current advantage into something a rule change cannot take away.

4.4 And the deadline is not a one-off

CMS has its own in-person requirement for home-based tele-mental-health starting January 2028 [A11]. A payer strategy inherits a second in-person obligation twelve months after the DEA one. So the in-person capability is not a 2026 fire drill — it is the prerequisite for both the regulated present and the insured future. That is the strongest available reason to fund it now rather than hope for a fifth extension.

5. The AI programme, and what it is for

5.1 What is actually shipped

Four features, three named in the posting [JD] and one not:

5.2 The reframe that lands

AI Scribe cannot save MEDvidi money, and that is fine. Because clinicians are paid per completed appointment, a minute of documentation removed is a minute returned to the clinician, not a dollar returned to MEDvidi — it is a ~31% effective raise at zero cost [A5]. That makes it a supply-acquisition and retention product, which is exactly what a company with open clinician requisitions in 33 of its 35 states needs [A5]. The honest KPI is not minutes saved per note; it is visits completed per offered clinician-hour, and clinician 6-month retention [A5].

Also: do not repeat the "10× clinician capacity" claim. The published ambient-documentation range is 0.76–2.1 minutes saved per note, and documentation is at most a third of a 15-minute follow-up [A5]. Repeating it in a room with anyone who has read the literature is a credibility loss.

5.3 The argument worth making

Every one of the four features is a governance artefact before it is a product feature [A12]. The Scribe writes the record a board or a DEA investigator reads. Chart Review AI writes a permanent, timestamped, discoverable log of every protocol deviation it detected. The agent writes the transcript of what the company said to a patient in distress. In a category where the peer's founder was convicted and where DOJ's theory was that clinical protocols were the instrument of the offence [A4], MEDvidi has built four systems that generate exactly the evidence a prosecutor would want and has not built the loop that makes that evidence exculpatory [A12].

Evals are not a quality programme here. They are the mechanism that turns a liability into a defence [A12]. That sentence reframes a cost centre as the cheapest insurance the company can buy — and it maps one-to-one onto the posting's most-repeated requirement.

Chart Review AI is the specific case: coverage went from 2–3% to 100% [A5], which is a real risk purchase. But coverage is now a solved problem and continuing to report it is complacency. The metric that matters is flag closure — of the deviations flagged, what share were reviewed, remediated and closed, by whom, in what time. A flagged deviation left unresolved is a better exhibit for a prosecutor than for a defendant [A5].

5.4 The compliance gap that is cheap and live

MEDvidi runs a patient-facing voice and chat agent across 35 states and its AI disclosure page contains no crisis protocol — no mention of suicidal ideation, self-harm, 988 or any crisis line [A12]. California SB 243 §22602(b) requires an operator to maintain such a protocol and publish it on its website before its chatbot may engage users at all, with a private right of action at $1,000 per violation [A12]. Six of the 35 served states now restrict AI in mental or behavioural health, four enacted in 2026; Maine expressly bars AI from "independently interacting with patients" [A12]. Illinois and Nevada prohibit AI from delivering mental or behavioural health care and both are served states [A4].

Whether the customer-service exclusion applies is a lawyer's argument. The product argument does not depend on it: a mental-health agent will meet suicidal ideation, and the only question is whether it was designed for that day or discovered it [A12]. Two-week build; removes the one failure mode that could end the AI programme in a single news cycle.

5.5 The B2B question, deflated

The completeness critic's strongest thesis was that MEDvidi is building an operating system for regulated prescribing clinics with its own clinic as customer zero. The dedicated pass does not overturn it — it re-times it [A10].

medvidi.ai is a single Astro page on a bought SaaS template, hosted on Netlify, unlinked from medvidi.com, with a dead Privacy link, a dead Terms link, a commented-out one-pager button and a contact form posting to Netlify Forms [A10]. Its Wayback history shows two states: in September 2025 the AI Prescriber was labelled "Vision", dated "Est. 2027", and carried an explicit note reading "The AI Prescriber is not yet FDA-cleared." The post-March-2026 rebuild deleted that note, deleted the "Verified by BrandPush.co" disclosure marking the press wall as purchased, deleted three of seven executives including the SVP of Product and Engineering whose seat this role replaces, deleted the 2027 date, and added a retrospective dated roadmap [A10].

Calibrated verdict [A10]:

• Intent to sell the stack: ~85%. Documented, not inferred.
• A live B2B business with an external paying customer: ~10%. Eleven checkable absences, the loudest being 65 live requisitions and not one sales, partnerships, solutions-engineering or customer-success role.
• Near-term audience is capital rather than customers: ~55%. Four paid wire releases in eight months, an "Investor" option on the only form, an ARR-and-profitability claim in the CEO bio, a deleted regulatory disclaimer, and an unfalsifiable "FDA pathway" phrase.

MEDvidi appears in no FDA database — not 510(k), De Novo, PMA, registration and listing, nor the AI-Enabled Device List [A10]. Meanwhile the regulator actually authorising AI to touch prescribing in the US today is a state office (Utah's Office of AI Policy, four authorised pilots, MEDvidi not among them) and every one of its agreements excludes controlled substances — which is MEDvidi's entire revenue core [A10].

Every operating metric this study leaned on — PDMP 23 min → one click, in-visit prescribing 40% → 55%, end-of-call booking 22% → 67%, RR1 41% → 53% — comes from that retrospective block and did not exist in public before ~April 2026 [A10]. They remain the best evidence available and they are company marketing about its own past. Quote them as the company's own claims, with the date, never as measurements.

6. The thing the study nearly missed

MEDvidi is already building an insurance line, and every agenda in this Inquiry was argued on a cash-pay architecture [A13].

A cash-pay company hired a US Director of Revenue Cycle Management in May 2026 to build commercial insurance revenue cycle "beginning with Florida and California in 2026", and two months earlier told Core-Track PM candidates they would "integrate insurance billing carriers into the core funnel" [A13]. The member terms already authorise MEDvidi to bill insurers on a patient's behalf [A7]. The FAQ has promised insurance for thirty-three months [A3].

Eligibility checks, benefit verification, prior authorisation, denial handling and a second price for the same visit are the largest single change to the booking funnel since payment-before-intake [A13], and nobody in the study costed it.

The arithmetic, done [A11]: Medicare's published 2024 national non-facility allowed amounts are $171.35 for a psychiatric diagnostic evaluation (90792) and $119.71 for a moderate-complexity established visit (99214); nurse practitioners are paid 85% of that by regulation (42 CFR 414.56); behavioural-health commercial rates historically sit at or below Medicare. So MEDvidi's blended $169.51 cash visit becomes about $107.68 net with an NP-heavy panel — volume must rise 1.53× just to stand still.

That makes insurance a volume bet at a ~35% lower unit price, not a margin improvement [A11], and Teladoc's Q2 filing is the manual for how it fails: demand for covered care arrived faster than credentialed capacity could serve it, so a company with $774M of cash cut marketing 17% and watched segment EBITDA fall 88% [A6]. The payer transition is a supply-and-workflow programme, not a billing integration [A6].

7. Contradictions and how they resolved

8. The agenda: two quarters, sequenced, with the cut list

Seven Scouts produced seven top priorities against 53 people, one data engineer and a 134-day regulatory clock. This is the adjudication.

The sequencing principle. Fund first what pays in every branch of the December decision. Everything else waits behind the branch it depends on.

Q4 2026 — build the thing that pays in every branch

1. The in-person evaluation capability. (Existential, dated, and not a fire drill.) Rebuild it as a permanent capability, not the 2023 flow — that implemented a proposed rule that never became law [A11]. Scope it to the top five states covering 44% of exposure: ~112 clinician-days at a 45% attendance rate, roughly half a million dollars against a $14.6M revenue line [A11]. Every converted patient becomes an annuity that survives every future cliff [A4], and the same capability is the prerequisite for the CMS January 2028 in-person requirement the insurance line inherits [A11]. This is the one item that is correct whether December tightens, extends or relaxes. Cost to name out loud: it costs conversion today, and the cured pair is held by a contractor, so clinician retention becomes a compliance variable [A11].

2. Pre-payment eligibility qualification. (The December contingency in disguise.) State of residence, age, provider-licensure match, the drug classes the platform will never prescribe, and the PDMP-visible patterns that reliably lead to declination — moved in front of the card [A1] [A9]. Three independent Areas converge on it, and the patient corpus supplies the demand-side proof: the clinical-gate complaint is the fastest-growing one-star category (13% → 36%) and the complaint is never "the doctor said no" — it is always "you could have told me before you took $195" [A9]. And with the flexibility expiring, the same screen is where a lot more people are about to be told no [A9]. Cost: it reduces paid initial visits. Say the number; a VP who proposes this without saying so will be caught [A1].

3. The transmission queue. (Largest patient-facing failure, cheapest instrumentation.) 34.1% of one-star reviews, the only category that never improved in three years [A9]. It is three problems with three owners: an asynchronous provider-signature step with no SLA and no visible queue, a manual patient-driven pharmacy re-route, and a transmission layer partly still on fax [A9]. None is a model problem; all three are a workflow with no clock on it. The platform already has the timestamps — instrumentation is a day [A9].

4. Week-one hygiene that needs no engineer. Split the seven-instrument consent and get it out from behind the CTA; reconcile the two refund statements; publish a cancellation method before Prime ships; fix the transposed ASRS subtype labels (ten characters of JavaScript, telling a high-inattentive patient they are "predominantly hyperactive") [A8]; fix the insurance sentences live on the California and New York ADHD pages and the five wrong-state sentences [A2]; display Trustpilot's number under Trustpilot's logo instead of a 4.85 the wordmark does not report [A9].

Q1 2027 — branch on December, and settle the follow-up

5. The follow-up repricing, as one decision. Prime and the AI Prescribing Assistant are two repricings of the same unit found by two Scouts who never compared them [N]. If the assistant works as advertised it deletes the $159 fee across ~77% of revenue; break-even is either $159 per asynchronous renewal or ~$42 per patient per month as a membership [A5]. Decide the price before shipping the automation. And if Prime launches, insist the success metric is 3-month retained revenue, not join rate — the late-reschedule-fee cohort will produce a flattering conversion rate and an unflattering retention curve [A3].

6. Chart-review flag closure and the eval gate. Change the dashboard from coverage to closure [A5]; turn report-ai-mistake into a labelled corpus with structured reason codes and a 10-day disposition SLA [A12]. The CEO's own requisitions describe eval as production-behaviour analysis with dashboards and mention release gates only as a QA nice-to-have — the gap between "we watch our AI" and "our AI cannot ship without passing" is the cheapest defensible thing a new VP can close [A13].

7. Insurance, as a bounded experiment. One or two states, a named segment MEDvidi cannot reach today, priced as a volume bet at ~35% lower unit revenue [A11]. First hire is credentialing operations; first product is eligibility-and-matching, not a claims connector [A6].

The cut list — argued as loudly as the build list

• No native mobile app. Provably none exists, the patient product is responsive web, and nothing in the observable failure modes is caused by its absence. It would be the most visible and least useful thing a new VP could do [A1].
• Do not build the referral-collection flow. There is no referral cure in the statute [A11].
• Do not buy pharmacy inventory data. It does not exist for any US prescriber; retail stock lives in the chain's ERP. Build the fill-outcome half instead — DoseSpot's HasRxFill, the Surescripts medication-history feed, the FDA shortage feed, and the FullFill/PartialFill/ NoFill model the portal already has [A11].
• Stop opening new states until December. Each new state is another cured-pair matrix to maintain, and depth beats breadth: one provider went from 25 to 80 available hours a week after picking up additional state licences [A5] [A11].
• Do not touch the E-E-A-T machine. Published editorial policy, named author and named medical reviewer on 24 of 24 sampled posts, 9.1 cited sources per post — the only durable moat in YMYL search [A2].
• Do not consolidate the brands. EZCare is live revenue; this is a six-quarter programme that would eat everything else. Note it, price it, do not start it [A7].
• Do not re-litigate the $195/$159 headline price this quarter. They already ran the membership experiment and the $25 loss-leader and killed both [A2]. This resolves the study's own conflict: the headline price stays; the follow-up is not a price test but a packaging decision tied to the automation (item 5).
• Do not outcome-price. A "no prescription, pay less" guarantee puts a financial incentive on the prescribing decision in a Schedule II business four months before the flexibility expires — the exact structure that made Done a criminal case [A2] [A4].
• Weight loss: qualify it or cut it. Today it goes checkbox → account wall in one tap with no BMI, comorbidity, pregnancy or contraindication question [A8], competing on evaluation intent against medication-inclusive players with no drug margin [A6].

The org, if asked to whiteboard it

The requisitions name nine units against a posting that says two tracks [A13]. The shape is a discovery organisation wearing a delivery organisation's clothes: 3.75 engineers per PM, 0.38 designers per PM, 7 analysts per data engineer, roughly one PM per named unit [A13]. The proposal is five teams, one per system of record and per queue, at 6+ engineers per PM — Acquisition & Booking; Continuity & Money (which owns the insurance funnel integration); Clinician Workspace; Patient Operations Agents; Platform & Trust — plus a data platform with one net new data engineer [A13]. Unowned today and defensible to raise: the 476-URL marketing site, the second consumer brand, the video app, pharmacy fulfilment, and AI evaluation as a release gate [A13].

9. Open questions — what only the CEO can answer

1. What share of visits requires the DEA flexibility, by schedule? Not public. It is the number that sizes December, and a business where it is 70% needs a different product organisation from one where it is 30% [N] [A4].
2. Is $30M a run rate off the last month, or the last twelve months billed? The whole reconciliation stands on ARR-versus-trailing [N].
3. Why is revenue reportedly doubling while the one internal vendor series shows organic sessions down 11% from May to July? The sharpest question in the Inquiry and not answerable from outside [N].
4. Are the published figures MEDvidi-brand or whole-house? Six brands share the platform; if "120,000 visits" is the house, every per-patient figure moves [N].
5. What killed the 2022 subscription, and what is different about Membership_Launch_Test_1? [A1]
6. Is "35,000 patients" annual or cumulative? Every per-patient figure depends on it [N].
7. Which entity employs the 30+ engineers — MEDvidi or Smart IT? Smart IT's own case study says it was "responsible for the entire development of MEDvidi" and supplied a CTO as a service; its CEO is MEDvidi's co-founder and titular CTO, and its recruiters are running this search [A7].

10. Confidence map

Every earlier Scout in this Inquiry described MEDvidi's signup flow by reading its JavaScript bundle. These were taken by walking it in a real browser on a real 390×844 mobile viewport, with all write requests blocked. They are the reason several claims in the study changed.