Frequently asked questions — for skeptics

Where's the peer-reviewed paper for the 87.4% accuracy claim?

It doesn't exist yet. The 87.4% figure comes from an internal retrospective validation conducted at Integrative Psychiatry Manhattan (N = 124 charts, five diagnostic categories) and is being written up as a manuscript. An NIH R21 (PAR-25-310), with a companion R01 (PAR-25-283), has been submitted through the Sultan Lab for Mental Health Informatics to fund prospective validation across outpatient settings with Guideline Concordance Score as the primary outcome. A preprint will go up when the writeup is finalized — it will be linked here and on the Evidence-based AI page.

Is Sigmund HIPAA-compliant? Is there a BAA?

Yes, and yes. A Business Associate Agreement is signed with every clinic before any patient data flows. More than that: Sigmund runs inside your own HIPAA-compliant environment, so identifiable PHI never has to leave it — movement is minimized, and anything that must move is de-identified first. The architecture is HIPAA-aligned from the audit log out, not bolted on: AES-256 at rest, TLS 1.3 in transit, role-scoped access, audit log on every read/write. Full detail on the Security & HIPAA page.

Do you train your models on my patients' data?

No. Sigmund runs on its own infrastructure and all model inference is local — PHI is never sent to a third-party LLM or external AI service. There is no OpenAI in the loop and no external inference call. Your patients' data is not used to train any model, ours or anyone else's. Full architecture detail, key management, and audit-log description on the Security & HIPAA page.

How is this different from Abridge, Freed, Suki, or Heidi?

Those are general-medical ambient scribes — strong on transcription breadth across specialties, further along in deployment scale, and a reasonable choice if all you need is the conversation captured. Sigmund is a psychiatry-specific presence with an evidence engine on top: he hears a psychiatric note structure (MSE, risk assessment, controlled-substance documentation) as the default rather than a template you have to coax out, and the medication recommendations he surfaces carry citation chains traceable to PMIDs and guideline sections. If you're a generalist, the general-medical scribes are likely the better fit. If you're a psychiatrist documenting MSEs every day, Sigmund is built for that work. Full side-by-side on the Compare page.

Who built this? Is it actually a psychiatrist?

Yes. Sigmund is built by Dr. Ryan Sultan, Assistant Professor of Clinical Psychiatry at Columbia University Irving Medical Center, Director of the Sultan Lab for Mental Health Informatics, and founder of Integrative Psychiatry Manhattan. Double board-certified in adult and child/adolescent psychiatry, NIH NIDA K12 awardee, with peer-reviewed publications in JAMA Psychiatry, JAMA Network Open, and Pediatrics. He's the one writing the notes the product is supposed to write — the design choices come from a working clinic, not a whiteboard.

What's the status of the R21?

Submitted under PAR-25-310 (June 5, 2026 deadline). Not yet funded — review is pending. The academic home is the Sultan Lab for Mental Health Informatics at Columbia and the New York State Psychiatric Institute. Primary outcome of the prospective study is Guideline Concordance Score across two outpatient psychiatry settings. If you want to follow the validation work as it publishes, the lab page is the canonical source; results will also be linked on the Decisions page as they land.

What does it cost?

For most clinicians, nothing. The core tools — initial patient analysis, symptom tracking, and the AI scribe — are free, and free forever for verified trainees (.edu email). The full decision-support engine becomes a paid Practice tier in Q3 2026. We don't run ads and we don't sell your data; the free tier is funded by grant-backed research infrastructure and the paid tier. Specifics on the pricing page — no surprise enterprise quotes, no usage-based gotchas.

Will it work with the systems I already use?

Yes. Sigmund connects to the clinical informatics infrastructure your practice already runs — including Epic, athenahealth, Practice Fusion, Valant, AdvancedMD, ICANotes, and Tebra — reading the chart and writing the note back into it. We've begun connecting these at our design-partner clinic. If the system you use isn't on that list, tell us in the waitlist form and we'll prioritize based on demand.

What happens if I sign up and don't like it?

Your notes are yours. Export to PDF or structured JSON at any time, in bulk, including the full citation chain on each note. There is no data lock-in by design — the export format is documented so you can ingest it into the next system without manual rework. A standard notice period applies for cancellation, but your charts and audit trail leave with you.

Is Sigmund FDA-cleared or a medical device?

No. Sigmund is investigational decision support — not an FDA-cleared medical device. Decisions about diagnosis, prescribing, and treatment remain with the licensed clinician. The signature line stays empty until you sign it.

What does "evidence-based" actually mean in your context?

Multi-vector retrieval over a curated psychiatric literature corpus, split into five section-level vectors per paper (overview, population, intervention, outcomes, timing). Evidence is weighted by study quality (RCT > meta-analysis > cohort), effect size, and recency. Every recommendation surfaces a citation chain traceable to PMIDs and guideline sections — you can click any line and read the studies underneath. No black box and no "trust us" — the mechanism is documented on the Evidence-based AI page.

How does it capture the MSE — does it eavesdrop the whole session?

No. Sigmund sits in only on the recorded encounter that the clinician explicitly initiates for each visit — he does not run in the background between visits, between rooms, or before you press record. He watches the observed behavioral content of the visit (speech, thought process, affect as conveyed in the audio), reads the clinician's own annotations, and drafts the MSE as an editable note. The signature line stays empty until you sign; every field can be edited, struck, or rewritten before that signature lands.

What if a patient declines AI recording?

Sigmund doesn't run for that encounter. Clinical AI use is opt-in per patient. The clinician informs, the patient decides, and the documentation pathway falls back to whatever you would have done before Sigmund existed.

Why should I trust an early product from a single clinic?

You shouldn't trust it beyond what the data supports. The deployment evidence is from real practice (N = 124 charts, retrospective). Prospective, multi-site validation is exactly what the R21 is for, and we don't claim more than the validation data shows. If "validated at one clinic" isn't enough for your standards yet, that's a reasonable position — wait for the prospective data, then revisit. Either way the core tools are free, so trying it costs you nothing but the time to set it up.

How do I report a bug or a privacy concern?

Submit through the waitlist with "Privacy" or "Security" in the Additional info field. It goes to the founder. The commitment is to acknowledge within 2 business days and, for any privacy concern, to start a documented review the same business day it lands. Notable issues get logged on the Decisions page.

Does it handle telehealth, and does it work for adult as well as child psychiatry?

Yes to telehealth — same MSE precision, same citation chain whether the patient is in the room or on video. Sigmund covers the core of outpatient mental health: depression, anxiety, attention, mood, neurocognitive, and trauma. The clinical knowledge nodes behind it hold a dynamic understanding of the scholarly literature in each of these areas and update as that literature evolves — feeding straight into the note to raise its quality.

Who owns the note, and who is liable for its clinical content?

You own the note. You are responsible for its clinical content. Sigmund drafts; he stays present; the clinician signs. The product is decision support — never a decision-maker — and that line is drawn intentionally in both the UI and the BAA.