The numbers, adjudicated

Proven rule. Every claim carries a tag. Every estimate shows its arithmetic and its assumptions. Where two sources cannot be reconciled, this document says so and gives the range plus the rule for which figure to quote in which context.

Kicker. The seven Findings printed three revenues ($20M, $27M, $30M), three benches (9, 50, 117), two traffic levels 72% apart and two opposite verdicts on clinical capacity — and almost all of it dissolves once each figure is put back next to the date and the measurement basis it came from. The revenue conflict is not a conflict: $20M / $27M / $30M is one dated ARR series (≈Nov 2025 / Apr 2026 / May 2026), and ARR is a forward-annualised run rate, not a trailing-year total. At the company's own stated 100% year-over-year growth, ARR always runs about 1.39x trailing-twelve-month billings — so $27M ARR and "120,000 visits a year at $195/$159" are the same business measured two ways, not two claims about one quantity. The bench is four different populations, not four answers: 117 is the 2025 roster, 100+ the current marketing floor, ~50 a partner-site deployment case, 9 a marketing showcase — and a capacity bridge shows 120,000 visits needs about 59–63 clinicians at the contractual 15-hour minimum, which is why 50 and 117 are both true at once. That bridge also settles the supply argument that split A3 and A5: supply was slack in 2025 (roster ~50% used) and is tight now (~75%), and the 117-person roster at its contractual minimum tops out around 220,000–240,000 visits, roughly $37–40M of ARR — four to seven months away at the claimed growth rate. Two things do not reconcile and should not be quoted: third-party traffic (Semrush and Similarweb disagree by 72% on level and disagree in sign on direction for the same month) and any conversion rate derived from it. And the one figure that survives every version of the revenue dispute, because it is a ratio rather than a level, is revenue per patient per year: $570–$581. Its driver, joined here for the first time from two Findings that each held half, is the company's own ADHD return rate RR1 moving 41% → 53% — which, run through the retention arithmetic, is exactly what moved visits per patient from ~2.9 to 3.43. The 3.43 is a churn number.

Source legend

1. What we found

1.1 The master ledger — every published claim about scale, with the date it refers to

The first pass's failure was not sloppiness. It was that seven Scouts each found a subset of these rows and none of them saw the date-referred-to column, which is the only column that makes the set coherent. Note that this is not the date a Scout found the figure — every row was found in the same week.

Two rows in that table were misread in the first pass and matter a lot:

V5 is not a volume. A5 read "130,000+ psychiatric visits" on [AI-SITE] as an August 2026 annual visit count and used it to argue current scale. It is a training corpus size, phrased in all three sources as what the model was built on: "Built on data from 130,000+ psychiatric visits, the tool is already cutting 30+ hours…" [PR-APR26]; "MEDvidi AI is trained on 130,000+ real psychiatric visits" [INT-MAY26]; "Trained on 130,000+ psychiatric visits and 15,000+ monthly patient encounters" [AI-SITE]. And it has not moved in four and a half months — identical on 8 April, 4 May and 19 August 2026. A live volume counter moves. A dataset-size marketing claim does not. Do not use 130,000 as a visit count.

V4 is not a volume either. "Trained on 10,000+ real patient visits per month" [PR-APR26] is the rate at which training data accrues, which is a floor on clinical volume but is stated as an AI claim. It is compatible with anything above 120,000 a year.

1.2 The revenue conflict is a units error, and it dissolves completely

Three of the four revenue figures are ARR. One is not. That is the entire dispute.

ARR is an instantaneous, forward-annualised run rate. Trailing-twelve-month billings are what a company actually collected over the last year. For a company growing, the two are never equal, and the gap is a pure function of the growth rate.

The ARR-to-TTM factor at a stated growth rate                          [REASON]

  Model growth as smooth exponential at g per year: r(t) = r0 * (1+g)^t
  ARR at time T          = r(T) x 1 year          (the instantaneous rate, annualised)
  TTM revenue at time T  = integral of r over [T-1, T]

  Ratio TTM / ARR = g / ((1+g) x ln(1+g))

  At MEDvidi's own stated g = 100% YoY [PR-APR26]:
     TTM / ARR = 1 / (2 x ln 2) = 1 / 1.3863 = 0.7213
     i.e.  ARR = 1.386 x TTM

  Assumption: growth is smooth rather than lumpy. Lumpy growth widens the band but does not
  change the sign or the order of magnitude. This is arithmetic on the company's own stated
  growth rate, not a forecast.

Apply it:

Reconciling R2 with R4                                                 [REASON]
  $27M ARR (8 Apr 2026)  x 0.7213  =  $19.5M of trailing-twelve-month billings
  A2 and A3, independently, from 120,000 visits x published list prices:
      35,000 initials x $195  = $ 6,825,000
      85,000 follow-ups x $159 = $13,515,000
      total                    = $20,340,000
  Gap between the two:  20.34 / 19.48  =  1.04, i.e. 4%.

Four percent. Two figures the study treated as irreconcilable are the same number to within the rounding on "120,000+" and "$27 million". A3's $20.3M and the April press release's $27M ARR are not competing estimates of one quantity — they are a trailing total and a forward run rate for the same business in the same period, and the ratio between them is exactly what a 100% growth rate predicts.

The rest of the series falls into place on the same arithmetic:

The dated ARR series, and what each implies                            [REASON]
  Blended price per visit, on the company's own 20/80 visit mix [PR-APR26] and prices [WEB]:
      0.20 x $195 + 0.80 x $159 = $39.00 + $127.20 = $166.20

  R1  $20M ARR  ->  $20,000,000 / 166.20 = 120,337 annualised visits/yr
  R2  $27M ARR  ->  $27,000,000 / 166.20 = 162,455 annualised visits/yr  (13,538/month)
  R3  $30M ARR  ->  $30,000,000 / 166.20 = 180,505 annualised visits/yr  (15,042/month)

  And R3 lands on the company's own August statement:
      "15,000+ monthly patient encounters" [AI-SITE] x 12 = 180,000+/yr
      180,000 x $166.20 = $29.9M

Two things fall out of that block and both are worth saying out loud in the room.

First, R1's date can be recovered. "$20M ARR" corresponds to an annualised rate of 120,337 visits — i.e. the moment the run rate equalled the 2025 calendar-year visit count. Working backwards from April 2026 at 100% growth: 27 / 2^(5/12) = 20.2. So the CEO-bio figure is a roughly November 2025 number carried forward in a career-summary bio. It is not a contradiction of anything; it is a bio, and bios lag.

Second, R3 is corroborated twice and R2 is corroborated once. $30M appears in the unite.ai introductory bio [INT-MAY26] and, independently, in the co-founder's own public LinkedIn headline [LI]. $27M appears as a direct company statement in a wire release [PR-APR26] and is independently reproduced by dividing it by a published price into a published visit count. R5 — A6's $15M–$45M from headcount — is a 20x-sensitive estimate [V-QUANT] and is superseded by four company statements. Retire it.

The rule for the room. Say the series, not a number. "You've said $20M, then $27M in April, then $30M in May — that's an ARR series doubling year on year, which puts trailing billings around $20M last year and somewhere north of $21M now. Which of those two is the number you manage to?" That sentence cannot be corrected, because it contains no claim the company has not published, and it demonstrates that the arithmetic was done.

1.3 One caution on the series that the arithmetic does not remove

The April 2026 release states, in the same document, "$27 million in annual recurring revenue and 100% year-over-year growth" and "operates across 36 US states, supporting more than 120,000 patient visits annually" [PR-APR26]. The About Us year tab attaches that same 120,000 to 2025 [WEB]. So one of two readings is true:

• (a) 120,000 is calendar 2025 and the PR reused it as boilerplate. Then the April 2026 run rate is 162,455 annualised visits and 120,000 is stale within its own press release. This is the more likely reading, because the number is identical to the About Us 2025 tab.
• (b) 120,000 is genuinely the trailing twelve months to April 2026. Then the arithmetic in §1.2 is not a coincidence — it is the mechanism — but calendar 2025 must have been below 120,000, contradicting the About Us tab.

Either reading gives the same conclusion (ARR ≠ TTM; the series is coherent), so nothing downstream depends on picking one. But do not present the 4% match in §1.2 as proof of reading (b). Present it as what it is: a cross-check that the two figures are mutually consistent at the company's own growth rate. [REASON]

1.4 The bench: four numbers, four populations, one question each

A5's defence of the 50 — that 1,500 ÷ 50 = 30 hours/month matches the April release's "30+ hours of administrative work per provider each month" [PR-APR26] — is a valid internal consistency check and it is not evidence about scope. It proves the 1,500 and the 50 refer to the same group. It says nothing about whether that group is the whole network. [V-QUANT] reached the same conclusion independently.

1.5 The capacity bridge — how many working clinicians 120,000 visits actually requires

This is the arithmetic neither A3 nor A5 did, and it is what makes 50 and 117 both true.

Clinician-equivalents required, at the contractual minimum              [REASON]

  INPUTS
    visit lengths            30 min initial, 15 min follow-up          [WEB]
    visit mix                20% initial / 80% follow-up               [PR-APR26]
      -> weighted contact time per visit = 0.20 x 0.50h + 0.80 x 0.25h = 0.30 h
    calendar utilisation     85% (gaps, no-shows, over-runs, chart close-out)   ASSUMPTION
    contractual minimum      ~15 h/week                                 [JOBS]
    working year             48 weeks                                   ASSUMPTION
      -> 15 x 48 = 720 booked hours per clinician per year

  AT 120,000 VISITS (calendar 2025)
    contact hours needed      120,000 x 0.30      = 36,000 h
    booked-calendar hours     36,000 / 0.85       = 42,353 h
    clinician-equivalents     42,353 / 720        = 58.8

  Sensitivity on the visit mix. Using A3's mix instead (35,000 initials of 120,000, i.e.
  29.2% initial, weighted contact 0.3229 h):
    contact 38,751 h -> booked 45,589 h -> 63.3 clinician-equivalents

  RESULT: 120,000 visits/year needs about 59-63 clinicians working the 15-hour minimum.

Against a 2025 roster of 117, that is 50–54% roster utilisation. And 59–63 is the same number as the partner site's "50-provider team" to within the difference between working the contractual minimum and working a little above it.

So the reconciliation is:

117 people held a MEDvidi-sponsored licence during 2025. Roughly 50–60 of them carried the clinical load, at 15–19 hours a week each. Both numbers are the company's own and both are true, because they count different things: a roster and a working bench. [REASON]

That is corroborated from a third direction by the employment model itself — 1099 contractors, "work on your own schedule", a 5-to-50-hour slider on the careers page, MEDvidi paying for the state and DEA licences, and a permanent open requisition in 33 of 35 served states [JOBS] [F:A7]. A pool that is licensed at company expense and paid only per completed appointment will always have more names on it than shifts on it.

1.6 Is clinical supply the binding constraint? Settled — it depends on the year, and the year has changed

A3 divided by 117, got 1,026 visits per provider, and concluded "supply is not what is limiting revenue" with 2.7x headroom. A5 divided by 50, got 3,249, and concluded the binding constraint is licensed-clinician-hours. [V-QUANT] corrected A3's arithmetic to 48% / 2.1x on A3's own visit mix. All three are computing different things.

The like-for-like series, on one denominator and one method:

Visits per roster member, at each dated volume                          [REASON]
  120,000 (CY2025)                / 117 = 1,026
  162,455 ($27M ARR, Apr 2026)    / 117 = 1,389
  180,000 (15,000/mo, Aug 2026)   / 117 = 1,538

Roster capacity ceiling and headroom                                    [REASON]
  Ceiling = 117 clinicians x 720 booked h/yr x 0.85 utilisation / 0.30 h per visit
          = 84,240 booked h -> 71,604 contact h -> 238,680 visits/yr
  On A3's visit mix (0.3229 h/visit): 221,741 visits/yr

  So the ceiling is 220,000-240,000 visits a year, which at $166.20 blended is
  $36.9M - $39.7M of ARR.

  Headroom against each dated volume:
                        20/80 mix      A3 mix
    at 120,000 visits     1.99x          1.85x     (roster ~50-54% used)
    at 162,455 visits     1.47x          1.36x     (roster ~68-73% used)
    at 180,000 visits     1.33x          1.23x     (roster ~75-81% used)

  Months of runway at the company's own 100% YoY growth rate:
    from 162,455 to ceiling:  5.4 - 6.7 months
    from 180,000 to ceiling:  3.6 - 4.9 months

The verdict. Supply was not the constraint in 2025 and is becoming the constraint now. Both Findings were right about a different year and neither said which year. The defensible statement, and the one that is far more useful than either original:

"On your own numbers the roster is 117 and the load is about 15,000 visits a month. At the contractual fifteen hours a week, that roster tops out somewhere around 220–240 thousand visits a year — call it $37–40M of ARR. At 100% growth you hit it inside four to seven months. That's why 35 of your 59 live requisitions are clinical and why you're recruiting prescribers in 33 of the 35 states you serve." [REASON] on [WEB] [AI-SITE] [JOBS] [F:A7]

Three independent corroborations that the ceiling is real and close:

• The CEO's own quote: "within three months of joining us, most providers are 80% booked with follow-up patients" [INT-MAY26]. A bench that fills in a quarter is a bench with no slack.
• 35 of 59 live requisitions are clinical, covering 33 of the 35 served states — 94% of markets recruiting at once [JOBS] [F:A7].
• The AI programme itself. Chart Review AI, AI Scribe and the AI Prescribing Assistant are all pointed at clinician minutes, and the April release's headline benefit is "enabling clinicians to see up to 10X more patients" [PR-APR26]. Companies do not spend an engineering programme on capacity they already have.

The one condition under which A3's original "supply is not the constraint" is still true: if the 2025 volume and the 2025 roster are both current — i.e. if the $27M and $30M ARR figures are forward-looking or overstated and real volume is still 120,000 a year. Say the condition, do not assume it away.

1.7 Traffic: not quotable, and the reason is worse than a disagreement about level

Both vendors were re-read live on 2026-08-19 for this Finding. Semrush's page still reports the July 2026 period; Similarweb's likewise.

The size of the disagreement                                            [REASON]
  Level:      268,740 / 156,267 = 1.72   -> Semrush is 72% higher
  3-mo avg:   Semrush (268.74+293.27+301.18)/3 = 287,730/month
              Similarweb 468,800/3            = 156,267/month
              ratio 1.84
  Direction:  Semrush -8.37% MoM, Similarweb +13.71% MoM  -> opposite signs
  Engagement: bounce 58.2% vs 40.2%; duration 11:03 vs 2:36 (4.2x)

Two modelled vendors disagreeing 72% on level is normal and survivable. Disagreeing in sign on direction, and by 4.2x on session duration, for the same site in the same month, is not. No absolute traffic figure from either vendor is safe to say to this CEO, who has Google Search Console open on a second monitor.

Consequently, neither conversion rate in the study is safe:

The two conversion rates, and why both die                              [REASON]
  A2:  35,000 new patients / (156,267 x 12) = 1.87%
  A3:  25,000-35,000 initials / (268,740 x 12) = 0.78% - 1.09%
  Spread 0.78% - 1.87%, a factor of 2.4 -- entirely produced by the denominator dispute,
  not by any real range in the business.
  Both also use a medvidi.com-only denominator while ezcareclinic.io runs its own live
  funnel at start.ezcareclinic.io/signup [F:A3], which depresses the true figure further.

What survives. Three statements, all defensible:

1. Organic search is the largest channel — both vendors agree, and they are the only two.
2. Paid search is small. Semrush puts it at 0.52% of sessions; Similarweb's 11.99% is a share of its keyword panel, not of traffic, which [V-QUANT] established. A channel can rank third on a long tail and still be half a percent of sessions. The honest form is "paid search exists and is small", which resolves C5 without picking a vendor.
3. The ratio claims inside a single vendor are more robust than any level. Semrush's own three-month series is down 11% from May to July; that is one vendor's internal comparison and it is the strongest traffic statement available — and it points the opposite way to the revenue story, which is the most interesting open question in the whole Inquiry (§4.3).

1.8 Revenue per patient — the one figure that survives every version of the revenue dispute

Because it is a ratio, it is invariant to the ARR-versus-TTM argument entirely.

Revenue per patient per year                                            [REASON]
  Route 1, from the company's own summary block [WEB]:
      120,000 visits / 35,000 patients = 3.43 visits per patient per year
      3.43 x $166.20 blended (20/80 mix)     = $570.07
      3.43 x $169.50 blended (1 initial + 2.43 follow-ups) = $581.37

  Route 2, from total billings:
      $20,340,000 / 35,000 = $581.14   (upper visit mix)
      $19,980,000 / 35,000 = $570.86   (lower visit mix)

  Route 3, sanity check at the 2026 run rate:
      162,455 visits at 3.43 visits/patient = 47,363 patients
      $27,000,000 / 47,363 = $570.06        -- identical, by construction

  ANSWER: $570 - $581 per patient per year. Call it ~$575.

Three caveats, all of which should be said before the number if challenged:

• Both inputs are floors — "35,000+" and "120,000" — so 3.43 is a point estimate from two lower bounds. If the real figures were 40,000 and 125,000, the ratio is 3.13. [WEB]
• The patient count sits in an undated summary block while the visit count is in a 2025 year tab. They are presented together ("35,000+ patients & 120,000 online visits every year") so treating them as one period is reasonable, but it is an assumption.
• Fee revenue is excluded. A3 modelled $0.19M–$0.55M collected in no-show and late-reschedule fees [V-QUANT], i.e. 1–3% on top. Immaterial to the per-patient figure.

1.9 Is 3.43 a churn signal or a days-supply artefact? Churn — and the company's own roadmap proves it

A3 posed this and parked it as unanswerable from outside. A5 found the answer and did not know it was an answer. Joining them:

MEDvidi's roadmap publishes, for the "Early 2025" row, "ADHD return rate (RR1): 41% → 53%" [AI-SITE], re-verified live 2026-08-19. RR1 is the share of ADHD patients who return — i.e. the first-visit return rate. Run it through a standard retention curve:

Retention model                                                         [REASON]
  Let r1 = probability a patient returns for visit 2 (this is RR1)
  Let r  = probability of returning for each visit thereafter (curves flatten after visit 1)
  Expected visits per patient  E = 1 + r1 / (1 - r)

  Solve for r given the observed E = 3.43 and the current r1 = 0.53:
      0.53 / (1 - r) = 2.43   ->   1 - r = 0.2181   ->   r = 78.2%

  Now hold r at 78.2% and roll RR1 back to its pre-improvement value:
      r1 = 0.41  ->  E = 1 + 0.41/0.2181 = 2.88 visits per patient per year
      r1 = 0.53  ->  E = 1 + 0.53/0.2181 = 3.43 visits per patient per year
      lift = +19.1%

  Assumptions: retention is constant after visit 2; the RR1 series is ADHD-only and is being
  applied to the whole book; the 3.43 and the 53% refer to overlapping periods. All three are
  arguable and are parked in section 4.

Two readings of the result, and both point the same way:

• If the model is right, the 41% → 53% improvement is precisely what produced the 3.43. Before it, this business was running at about 2.9 visits per patient — roughly $480 per patient-year.
• Even if the model is wrong in detail, the headline stands on its own: on the company's own metric, 47% of ADHD patients do not come back after the first visit. A patient who is prescribed a Schedule II stimulant and intends to keep taking it has to come back — MEDvidi's own FAQ says so: "Health & Safety Code Section 11200 (c) prohibits the refilling of a Schedule II controlled substance. A new prescription should be issued every time" [WEB].

Which settles the question. A 47% first-visit non-return rate cannot be a days-supply artefact. Post-dated 90-day prescriptions would compress visits 2 through N — they would move the later retention rate, the 78%, toward a quarterly cadence — but they cannot make half the patients disappear after the visit they already paid $195 for. Days supply shapes the tail. Churn cuts the head.

The honest composite statement:

3.43 visits per patient is roughly half a churn number and half a days-supply number, and the churn half is at the first visit. About 47% of ADHD patients never return after the initial evaluation, on the company's own RR1 metric; of those who do, roughly 78% return each time, which is consistent with a 60-to-90-day script rather than a monthly one. [REASON] on [AI-SITE] [WEB]

That composite is the single most useful sentence in this Finding for a product interview, because it converts an ambiguous ratio into two separately addressable product problems: a first-visit conversion problem (pay-before-qualify, the clinical gate, "no prescription" one-star reviews) and a cadence problem (post-dated scripts, the $159 follow-up price, MEDvidi Prime).

1.10 Contribution margin, computed once, on one base

C12 exists because A3 computed 73.3% on $20.3M with a 50/50 credential mix and A5 computed 75.3% on $27M with an invented 15/45/40 mix. They are close enough to look like agreement and they are not comparable. Recomputed here on one base, with the bounds shown.

Contribution per visit, blended                                         [REASON]
  Price base: $166.20 blended, on the 20% initial / 80% follow-up mix [PR-APR26] [WEB]
  Published rate card [JOBS]:
      psychiatrist  $70 initial / $50 follow-up
      physician     $60 / $40
      PMHNP         $45 / $30
  Card processing at Stripe's published US card-not-present 2.9% + $0.30 [PORTAL confirms Stripe]:
      0.20 x (195 x .029 + .30) + 0.80 x (159 x .029 + .30) = $5.12

  Credential mix                clinician cost   contribution   margin
    50/50 physician/PMHNP           $38.50         $122.58      73.8%
    A5's 15/45/40 (unsourced)       $41.10         $119.98      72.2%
    100% PMHNP (upper bound)        $33.00         $128.08      77.1%
    100% psychiatrist (lower bound) $54.00         $107.08      64.4%

Say it as: contribution after clinician pay and card fees is 64–77% of the visit price, most likely 72–74%. Never say a margin without naming the base, because the phrase "gross margin" here excludes provider assistants, the 24/7 care team, the intake and PDMP verification labour, and everything else in A3's cost block — which together A3 sized at $1.2M–$4.2M [F:A3]. This is a unit margin, not a company margin, and the difference is the whole operations layer.

The per-credential spread is itself the finding: the PMHNP follow-up at 81% clinical margin is the most profitable unit MEDvidi sells and the physician-supervision requisition exists to make it legally available in California, Ohio and Texas [JOBS] [F:A5].

1.11 Revenue per head — the ratio that inverts, and should therefore not be used

A3 built an argument that the product-and-engineering organisation is oversized. [V-QUANT] shows the ratio was computed on the whole-company denominator and then attributed to the 55-person org. Recomputed:

Revenue per product-and-engineering head                                [REASON]
  Org size from the posting: 53 + the VP = 54 [JD]
      $20.3M TTM / 54 = $376k
      $30.0M ARR / 54 = $556k
  Comparable, audited [SEC]:
      LifeMD FY2025: $194,055,198 / 389 employees = $498,857 per employee
      Hims  FY2025:  $2,347,637,000 / 2,442       = $961,358 per employee

And the comparison is not like-for-like in either direction. LifeMD's 10-K states 389 employees and separately describes an affiliated 50-state medical group whose clinicians are not in that count [SEC] [V-QUANT]. MEDvidi's own "300+ people" claims plausibly include the 117 contractor clinicians and their provider assistants [F:A3]. So the denominators mean different things on both sides.

Rule: do not use revenue per head as an argument about MEDvidi's org size, in either direction. It is a ratio between two numbers that are each defined differently at each company, and A3's version of it inverts once the current revenue figure is used. The interesting question survives without it: a clinic doing 15,000 visits a month does not need thirty engineers for a booking flow, so what are they building? That is a question, not a ratio, and it is stronger as a question.

1.12 The number set — twelve figures that may be said out loud

Confidence labels: sourced = the company or an audited filing published it and it was re-verified at source for this Finding; triangulated = two or more independent routes agree; estimate = this Inquiry's arithmetic on sourced inputs, assumptions shown; weak = single modelled source or unstated scope.

Two-line version of the whole table, for the first thirty seconds of the conversation: "You're a roughly $30M-ARR run rate on about 15,000 visits a month, at $195 and $159, with contribution somewhere in the low seventies per visit. The number I'd want to look at first is 3.43 visits per patient a year — and your own RR1 series says you already know why."

1.13 Figures that appear in the Findings and must not be repeated

2. Capability / object table

3. Reconciliation notes

3.1 — The About Us block is two different objects and the study read it as one. The page has year tabs (2025: "120K+ appointments per year / Open states: 36 / Providers: 117") and, separately, an undated summary block ("Growth from 12 to 36 states in 3 years / 35,000+ patients & 120,000 online visits every year / A team of 100+ licensed medical providers"). Re-read live 2026-08-19 [WEB]. The 3.43 ratio pairs the summary block's patient count with a figure that also appears in a 2025 tab. Treated here as one period because the summary says "every year", and flagged as an assumption. The year tabs are also internally scrambled — states go 34 (2023) → 23 (2024) → 36 (2025) and providers 91 → 71 → 117 — which is a reason to trust the appointments column (monotonic, and the 2022 figure agrees with a 2024-vintage capture [F:A3]) more than the other two.

3.2 — Why "ARR" is being used loosely and why that does not matter. MEDvidi has no subscriptions in the ordinary sense; its own live FAQ still says "we have no subscription model" [WEB], re-verified 2026-08-19. So "annual recurring revenue" here means an annualised run rate on per-visit revenue, which is a common if imprecise usage in founder-led companies. This does not weaken the reconciliation — a run rate annualised from a recent month behaves exactly like ARR for the purpose of the 1.39x factor. It does mean the phrase "recurring" should not be leaned on: the recurring object is the patient's 30-day medication cycle, not a contract.

3.3 — The $27M-and-120,000-in-one-document problem is unresolved and does not need to be. See §1.3. Both readings produce the same conclusions. Do not present the 4% match as proof of either.

3.4 — The 50-provider sentence is quoted here without deciding whether it is MEDvidi's own clinic. It sits in a partner-outcomes block; "a single 50-provider team" is the phrasing a vendor uses for an anonymised deployment. It could be MEDvidi describing itself in the third person to a partner audience. The capacity bridge in §1.5 makes the question much less load-bearing, because 59–63 working clinicians is what 120,000 visits requires regardless of what the sentence means.

3.5 — The rule this Finding applies to medvidi.ai, stated once and applied consistently. [V-GAPS] C9 correctly notes that A5 dismissed the partner site as marketing where inconvenient and trusted it as data where convenient, in the same Finding. The rule used here:

Treat [AI-SITE] as a company self-report, exactly like a press release. Its directional claims about its own operating metrics (RR1 41→53, rebooking 22→67, in-visit prescribing 40→55, PDMP 23 minutes → one click) are quotable as things MEDvidi has published about itself, never as measured facts, and never as evidence that a process is now automated. Its comparative claims about competitors, its "not a single manual step", its "10x", and its "FDA pathway in progress" are sales copy and are quotable only as evidence of positioning.

Under that rule the RR1 figure in §1.9 is used correctly: it is cited as MEDvidi's own published metric, and the conclusion drawn from it (that 3.43 is mostly a churn number) does not depend on the metric being accurate — only on it being roughly the right order, which the "up to 80% of visits are prescription renewals" statement [PR-APR26] independently supports.

3.6 — Where this Finding disagrees with the quantitative verification pass. [V-QUANT]'s recommendation is: "Take $30M as the current figure." This Finding does not adopt that. $30M is correct as a May 2026 ARR statement and it is correctly labelled as such — but stated bare, it invites the CEO to hear a candidate confusing ARR with revenue, which is a worse error than being a quarter behind. Say the series. [V-QUANT]'s other three headline corrections — Prime's price is unverifiable, the utilisation figures are wrong, the competitor table is A6's not A3's — are adopted in full.

3.7 — Review volume, observed twice. Trustpilot's feed returned total 1,252 on 2026-08-18 [V-QUANT] and total 1,258 on 2026-08-19 [TP], the entire six-review delta landing in the five-star bucket (950 → 956) with every other bucket unchanged. One day is not a sample, so this is an observation, not a finding: a stream adding ~6 reviews a day, all five-star, looks solicited rather than organic. Settled by the invited-versus-organic split on Trustpilot's own profile page.

4. Open Questions / Parked

4.1 What is the trailing-twelve-month revenue figure the CEO manages to? The whole §1.2 reconciliation stands on the distinction between ARR and TTM, and only the company knows which one its internal reporting uses. Settled by: one question in the room — "when you say $30M, is that a run rate off the last month, or the last twelve months billed?" That question is itself the proof the work was done.

4.2 What is 15,000+ "monthly patient encounters" counting? [AI-SITE]'s own sentence puts it alongside the training-corpus figure, so it may count AI-mediated interactions rather than billed visits. Undated on the page. Settled by: Wayback captures of medvidi.ai to date the claim, plus the internal definition.

4.3 Why is revenue reportedly doubling while Semrush's own three-month series shows organic sessions down 11% from May to July 2026? [SEMRUSH]. Four candidate explanations, all testable internally and none from outside: growth is now coming from returning patients rather than new ones (which the RR1 improvement would predict); growth is coming from the sibling brands whose traffic is not in the medvidi.com denominator; the Semrush series is modelling noise; or the growth claim is stale. This is the sharpest single question in the Inquiry and it is not answerable from public sources. Settled by: Search Console clicks by month for 24 months, split new-patient bookings from follow-up bookings.

4.4 What share of visits requires the DEA telemedicine flexibility? Not published anywhere. The public proxies bracket it high — the information architecture is organised by controlled-substance brand name, 28 of the programmatic state pages are ADHD against 4 for anxiety [V-QUANT], and "up to 80% of visits are prescription renewals" [PR-APR26]. But a renewal is not necessarily a Schedule II renewal, and MEDvidi also prescribes non-controlled antidepressants, atomoxetine, clonidine and GLP-1s [F:A5]. Settled by: the internal split of visits by DEA schedule. This is the number that determines the size of the December event and nobody outside the company has it.

4.5 Is 35,000 patients a year, or a cumulative total? A2 asserted it is annual; [V-QUANT] correctly pushed back — nothing on the page says annual, and a February 2024 homepage capture claiming "5000+ PATIENTS" fits a cumulative reading. Every per-patient figure in §1.8 depends on this. Settled by: one question, or a distinct-patient count from the portal database.

4.6 What is the actual active-versus-dormant split of the 117 roster? §1.5 infers 59–63 working equivalents from capacity arithmetic; it does not measure how many individuals that is. A roster where 60 people each work 15 hours is a different management problem from one where 40 work 25 and 77 work zero. Settled by: visits per provider per month, distribution not mean.

4.7 Does RR1 apply to the whole book or only ADHD? It is labelled "ADHD return rate" [AI-SITE] and §1.9 applies it to the whole book. If ADHD retention is materially better than anxiety or insomnia retention — which it should be, because only ADHD forces a monthly script — then the whole-book RR1 is worse than 53% and the churn conclusion strengthens. Settled by: return rate by condition.

4.8 MEDvidi Prime's price and billing interval. [V-QUANT] could not reproduce the $159 string from the public bundle; A3 sources it to lazy chunks whose filenames are not extractable from outside. The membership is proven shipped, Stripe-backed and flag-gated. Settled by: the Stripe price object.

4.9 The 85% calendar-utilisation and 48-week assumptions in §1.5–1.6. These are mine and they move the ceiling. At 90% utilisation and 50 weeks the roster ceiling rises to ~263,000 visits (~$43.7M ARR); at 75% and 46 weeks it falls to ~201,000 (~$33.4M). The conclusion — that the ceiling is one to two years away, not five — is robust across that whole band. Settled by: scheduled-hours and booked-hours from the EMR.

4.10 Whether the two consumer funnels are in any of these numbers. ezcareclinic.io runs its own live signup at start.ezcareclinic.io/signup [F:A3], and the platform serves six clinic brands [F:A1]. Whether "120,000 visits" and "$30M ARR" are MEDvidi-brand or whole-house figures is unknown, and it moves every per-visit and per-patient figure here. Settled by: asking.

5. What this does NOT cover

• No new research beyond the numbers. This is an adjudication pass. Figures were re-verified at source where it mattered — About Us, the FAQ, medvidi.ai, the April press release, the May interview, Semrush, Similarweb and the Trustpilot feed, all on 2026-08-19. Nothing new was hunted.
• No internal data, and no attempt to get any. Every bridge here is built from public statements, and several collapse to a single question in the room. That is deliberate: a question that proves the arithmetic was done beats a number that might be wrong.
• No opinion on whether the company's self-reported figures are true. The reconciliation shows they are mutually consistent, which is a weaker claim. A consistent set can still be a consistently optimistic one. Everything tagged "sourced" in §1.12 is sourced to the company itself unless it says [SEC].
• No regulatory sizing, no competitor prices, no org design, no agenda, no eval spec. A4 owns the DEA instrument and mechanism; this Finding supplies only the clock. A6's competitor table was verified operator-by-operator by [V-QUANT] and stands, and A3's four-row version must not be carried. The rest are separate gaps in [V-GAPS] §1 with separate owners. This Finding closes exactly one: SCOPE success criterion 2.

6. What this means for a VP of Product

6.1 The three numbers this business runs on

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

One — 3.43 visits per patient per year, and its driver, RR1 at 53%. This is the number the whole P&L hangs off, and it is the only one of the three the product organisation directly owns. Revenue per patient is visits per patient × blended price, and the price has not moved in thirty-three months [F:A3]. So every dollar of growth that is not new patients comes from this ratio. It has proven movable: the company's own roadmap records RR1 going 41% → 53%, which on the retention arithmetic in §1.9 is worth about +19% on visits per patient — the difference between ~2.9 and 3.43, or about $90 per patient-year. Against 35,000+ patients that is roughly $3M a year, produced by one funnel change, with no price change and no new clinician. And the ceiling is a long way up: a strictly monthly Schedule II cadence is 12 visits, so MEDvidi realises about 30% of the revenue its own clinical clock could produce [F:A3].

Two — the share of visits that requires the DEA telemedicine flexibility, against 134 days. This is the only number in the set whose movement is binary rather than gradual. MEDvidi's own FAQ says the flexibility runs "through December 31, 2026" [WEB]; from 2026-08-19 that is 134 days. The share itself is not public (§4.4) and the CEO knows it to a decimal place, which is exactly why it belongs in the three: asking for it is a better move than guessing at it. Everything about the business's shape says the share is high — an information architecture built on Adderall, Vyvanse, Concerta and Xanax, 28 programmatic ADHD state pages against 4 for anxiety, a permanent requisition for prescribers with "Schedule II prescribing authority", and a company that pays 117 people's DEA registrations. A business where that share is 70% and a business where it is 30% need different product organisations, and the difference cannot be split.

Three — paid search at 0.5% of sessions: acquisition concentration. The reason a cash-pay clinic with a 73% unit margin can carry a 53-person product organisation is that new patients arrive for approximately nothing. Semrush puts paid search at 1,400 sessions against 153,850 organic [SEMRUSH]; Similarweb ranks it third but on a keyword panel, not on sessions [SIMILARWEB]. Either way there is no second channel at scale. Pair that with number one: 47% of patients do not return after the first visit, so the business is dependent on new-patient flow to stand still, and new-patient flow is a single unhedged channel that neither MEDvidi nor anyone else controls. That is the combination that kills companies in this category — not a bad quarter, a Google update. And the study's own weak evidence points the wrong way: Semrush's internal three-month series is down 11% from May to July while revenue is claimed to be doubling (§4.3). Whatever the explanation, that divergence is the thing to open the second half of the interview with.

What is deliberately not on the list, and why. Revenue. $30M ARR is an outcome, not a driver, and it is the number the CEO will use to test whether the candidate did the work — which is a reason to have the series ready, not a reason to manage to it. Contribution margin. 64–77% per visit is excellent and stable; a number that does not move is not a number the business runs on. The clinician bench. The $37–40M roster ceiling (§1.6) is real, it is four to seven months away at the claimed growth rate, and it is the fourth number — but running out of supply caps a business, it does not kill one, and the company is already recruiting in 94% of its markets. It belongs in the first-90-days plan, not in the three.

6.2 The argument, not the summary

The reason this adjudication matters is not tidiness. It is that the three original numbers each supported a different product strategy, and the strategy changes depending on which version is true.

If the business is a $20M clinic with 2.7x clinical headroom and a settled no-subscription posture — A3's and A2's version — then the job is demand generation and the constraint is the funnel. If it is a $27M business already running its clinicians at 24 hours a week against a 15-hour floor — A5's version — then the job is supply and automation and the funnel is the last thing to touch. Those are opposite two-quarter plans, drawn from the same public evidence, by two Scouts who were each internally rigorous.

The adjudicated picture supports neither, and it supports something more specific than both. MEDvidi is running at roughly $30M ARR — call it 15,000 visits a month — on a 117-person roster that tops out around 220–240 thousand visits a year. That is one to two years of physical headroom at the current rate, and four to seven months at the claimed one. Meanwhile 47% of ADHD patients never come back after a visit they already paid for, the price of the follow-up has not moved since November 2023, and the only acquisition channel is organic search — which, on the one internal vendor series available, is going down while revenue is claimed to be going up.

That is not a demand problem or a supply problem. It is a business that has been growing by adding clinicians and states to a funnel it has never optimised, and it is about to run out of both runway items at once. The clinician roster caps growth inside a year. The content inventory is at terminal scale — 359 posts, net −8 since the August 2025 peak, and seven served states with no landing page at all [F:A2]. Neither of the two things that produced the last three years of growth can produce the next one.

Which means the leverage has to come from the middle: the 47% who leave after one visit, the 22% → 67% rebooking rate the company already proved it could move, and the 3.43 that the company's own RR1 series shows is a policy variable rather than a fact of nature. The two AI features that touch patients rather than clinicians — the receptionist and the prescribing assistant — sit exactly on that seam, which is probably why the org has an "AI Clinic Track" at all.

And there is one line worth having ready for the moment the CEO pushes back on a number, because he will. "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 he is asked for them does not get tested on numbers again.