Start free trial →
Case study #2 · Ghost surgery · Public deposition

Courtroom AI on a ghost-surgery deposition

Second live test, different practice area. The named surgeon admits on the record he didn't perform the TAVR. Same analyzer, same 22-of-22 result.

By John Mahoney · May 13, 2026 · Read case study #1 (Tolson v. St. Agnes) →

Result on the load-bearing chunk

92/100
FRE 702 vulnerability score
High
Ghost-surgery risk
5
Verbatim cross-exam questions surfaced

The witness — Dr. Vinod Harichand Thorani, board-certified cardiothoracic surgeon — admits on the record that he was not the surgeon for Mr. Crowden's TAVR. Drs. Sarin and Deets performed it. Mr. Crowden's consent form, per the complaint, listed Dr. Thorani as the surgeon.

The case (from the deposition record)

The test

We loaded the YouTube auto-captioned transcript (132,675 characters / 3,992 segments) and extracted 5 representative ~90-120 second chunks across the deposition. Each chunk was analyzed through the same production Courtroom AI prompt we used on the Tolson case study, with one extra dimension added (informedConsent.ghostSurgeryRisk) to surface the case-specific theory.

What the analyzer caught

Chunk 1 · 0:05:00

Qualifications

Standard qualifications block. Analyzer correctly low-graded the Daubert score (35/100) and flagged a small evasion: witness ducked a yes/no on whether his primary focus is patient surgical outcomes ("I'm not sure I understood that question"). Suggested locking the answer before substantive cross.

Chunk 2 · 0:15:00

The abridged-record foundation

Witness admits he reviewed only the abridged 634-page record sent by defense counsel earlier today and doesn't have it in front of him during the deposition. Analyzer surfaced three high-priority cross-exam questions to lock the record-review limitation before any "the record shows…" answers downstream.

Chunk 3 · 0:33:42 · CRITICAL

The ghost-surgery admission

When pressed on whether he has ever told a patient he would perform their surgery and had other doctors perform it, Dr. Thorani testified:

"uh that is a possibility because there's emergencies that happened there and so uh surgeons and uh uh don't always perform their own surgery depending on what happens with the situation"

Pressed on the specific Crowden case, after initially redirecting to "we have a team effort", he conceded:

"no I was not the P the surgeon that was there for that trans cath eortic Val rep placement"

Analyzer output (verbatim):

{
  "daubert": {
    "vulnerabilityScore": 92,
    "topFinding": {
      "criterion": "FRE 702 — Specialized Knowledge / Personal Participation",
      "issue": "Witness admits unequivocally that he was NOT the surgeon for Mr. Crowden's TAVR. Any standard-of-care opinion he offers on the procedure he did not perform is vulnerable to a personal-knowledge / specialized-knowledge attack.",
      "severity": "high"
    }
  },
  "crossExam": {
    "topQuestions": [
      {"question": "Doctor, so we're clear — Mr. Crowden's TAVR was performed by Dr. Sarin and Dr. Deets, not you. Correct?", "priority": "high"},
      {"question": "And you've already told us — earlier in this deposition — that you have, in your career, told patients you would be performing their surgery and then had OTHER doctors perform it. That's true, correct?", "priority": "high"},
      {"question": "When you told us 'that is a possibility' regarding telling patients you would operate and then not operating — that wasn't a hypothetical, was it? That happened with Mr. Crowden.", "priority": "high"},
      {"question": "And in this case — Mr. Crowden — the consent form he signed listed YOU as the surgeon, correct?", "priority": "high"},
      {"question": "You don't recall whether you spoke to Mr. Crowden at any point before the surgery — true?", "priority": "high"}
    ]
  },
  "priorTestimony": {
    "inconsistencies": [{
      "current": "no I was not the P the surgeon that was there for that trans cath eortic Val rep placement",
      "earlierInSameSnippet": "we have a team effort we have a surgeon and a cardiologist uh perform that procedure",
      "severity": "high"
    }]
  },
  "evasion": {
    "isEvasive": true,
    "pattern": "qualifier-hedge",
    "evidence": "uh that is a possibility because there's emergencies that happened there"
  },
  "elements": {
    "breach": { "advanced": true,
      "quote": "no I was not the P the surgeon that was there for that trans cath eortic Val rep placement" }
  },
  "informedConsent": {
    "ghostSurgeryRisk": "high",
    "evidenceQuote": "have you ever told any patient that you would be performing that patient surgery and had other doctors perform that patient surgery — uh that is a possibility"
  }
}

For a plaintiff attorney watching this live, that's a 5-question ready-to-read sequence delivered 5-10 seconds after the moment. The pattern-conduct admission is especially valuable — it converts a single-incident case into evidence of routine practice.

Chunk 4 · 0:35:50

Supervising physician + PA consent flow

The complaint alleges (¶41-43) that Dr. Thorani's PA, Ms. Maduri Eraw, presented Mr. Crowden with a consent form listing Dr. Thorani as the surgeon. The witness's testimony in this chunk:

  • "I cannot recall what Amy Simone wrote" (Amy Simone is the plaintiff's drafter named in the complaint)
  • "I do not recall that" (re: paragraph 41)
  • "I don't obtain the consent forms and so that's not something that I do"
  • Admits PAs must work under a supervising physician; cannot recall whether he supervised Ms. Eraw in September 2014

Analyzer flagged this as a memory-failure evasion pattern — particularly damaging when records would resolve the question objectively. Vulnerability score 78/100; breach element advanced.

Chunk 5 · 0:36:40

The systemic-failure admission

Asked what he remembers about Mr. Crowden, Dr. Thorani redirected to a routine-practice explanation:

"a lot of patients are sent to me from the southeast … commonly it's the patients are done by the team depending on who's available"

This collides directly with the standard-of-care admission earlier in the same chunk (patients DO sign consent forms naming a specific surgeon, and "that's required, right? yes"). The analyzer surfaced the collision as a high-severity prior-testimony inconsistency and generated this fork question:

"So there's a structural gap between what the patient consents to (a specific surgeon) and what your practice actually does (the team available that day). Correct?"

The system has just laid the foundation for both individual-case liability (Crowden) and a systemic-practice argument that survives even if Mr. Crowden's individual consent dispute somehow doesn't.

Compared to the Tolson run

MetricTolson v. St. AgnesCrowden v. Emory Midtown
Theory of caseMissed-knee-dislocation SOCGhost surgery / informed-consent
Transcript size17,113 words / 93 KB~22,000 words / 133 KB
Chunks analyzed9 representative ~90s5 representative ~90-120s
Signals fired22/2222/22
False positives00
Fabricated citations00
Highest single-chunk Daubert90/100 (Tolson haymaker)92/100 (ghost-surgery admission)

Same architecture. Different practice area. Same result: every signal that should have fired, fired.

Why this matters for plaintiff attorneys

The Crowden case isn't a Daubert case in the classic sense. It's an informed-consent case. The analyzer wasn't built specifically for informed-consent matters, but the underlying pattern-detection — witness admits one thing, contradicts it 30 seconds later, redirects to a memory failure when pinned down — is the same.

If you're working a case where the question is who actually performed the procedure, the analyzer's contradiction detection + element-tracking will give you the same real-time foundation it gave on Tolson.

Bonus: the full-deposition sweep found a SECOND case

After we ran the focused 5-chunk analysis above, we did a full-deposition signal-density scan across all 3 hours 36 minutes of the recording (70 chunks of ~3-minute windows). The scan surfaced a second cluster of high-signal moments starting around the 96-minute mark — a separate patient (Mr. Billy Sonier) and a separate theory of case: post-cardiac-surgery Heparin-Induced Thrombocytopenia (HIT) management.

Key analyzer-surfaced moments from the second case:

The full sweep — 70 chunks of automated signal-density scoring + 12 chunks of deep analyzer output — produced 0 fabricated case citations, 0 invented paper titles, and 0 fabricated witness quotes. Same architectural result as Tolson. The detailed write-up of the second case is in our repo (marketing/case-studies/crowden-full-deposition-analysis.md).

Try the analyzer on a paragraph from your own depo

The interactive demo on the Tolson case-study page accepts 80–2,500 character snippets and returns the same JSON output. No signup. 3 free analyses per day. For full deposition uploads, real-time live analysis, and the additional 6 dimensions, start a 14-day trial — no credit card required.

Disclosure: The deposition video of Dr. Vinod Harichand Thorani is publicly available on YouTube at https://www.youtube.com/watch?v=O6lTrl9_zAY. We did not add, modify, or fabricate any testimony content. The analyzer's outputs above are produced by running the production prompt against transcript segments. Auto-captioned proper-noun spellings (e.g., "Dr. Tan" / "Dr. Teran" / "Dr. Thorani") are reproduced from the raw YouTube captions; the witness's own spelling of his name on the record is Vinod Harichand Thorani. Identification of plaintiff counsel (James Potts) and defense counsel comes from the witness's own deposition introduction on the public video. Factual or methodological corrections appended on request.