Legal Disclaimer

1. Read this first

DebriefMD supports medical-training documentation and the human review of assessments. It turns a feedback conversation you had with your preceptor into a draft you then correct. Residents review their coaching notes. Authorized faculty review and sign rotation reports.

It does not practise medicine. It does not know anything about your patients. The software does not decide whether anyone is competent, ready to progress, or passing or failing; people make those decisions. It does not file anything into another system automatically.

Everything below follows from that.

2. DebriefMD is not medical advice

Nothing produced by DebriefMD is medical advice, clinical guidance, diagnosis, or a treatment recommendation. That covers transcripts, summaries, coaching text, competency tags, structured fields, and exported documents.

Never use DebriefMD output to make a clinical decision. Not for a patient discussed in a recording, not for any other patient, not ever. If a transcript appears to contain clinical guidance, it is one of two things. Either it is a garbled rendering of something your preceptor said in a specific context you were present for, or it is a language model producing plausible text. Neither is safe to act on.

Clinical decisions come from your own assessment, your supervision, your program's protocols, and the sources you are trained to use.

3. DebriefMD is not intended as a medical device

DebriefMD is a documentation and educational tool. We do not intend or offer it as a medical device, and it has not been reviewed, licensed, cleared, or approved as one by Health Canada, the U.S. Food and Drug Administration, or any other regulator. How a regulator classifies software depends on what it actually does and what is claimed for it; this notice is not a regulatory determination.

It is not intended to be used for the diagnosis, cure, mitigation, treatment, or prevention of disease, or to affect the structure or function of the body. It must not be used for any of those purposes.

If you are considering a use that would place DebriefMD in that category, stop, and contact us at hello@debriefmd.ca first.

4. DebriefMD does not replace your program's requirements

DebriefMD helps you draft assessment content. It does not:

• enter assessments into your program's official assessment system for you;
• satisfy any requirement your program imposes;
• track whether you have met your EPA, milestone, or evaluation obligations;
• guarantee that anything it produces will be accepted by anyone.

DebriefMD does share reviewed work inside the product and by email, and you should know exactly when:

• Personal download does not newly grant program access. Saving a draft, downloading a reviewed note in any format, or exporting it to a portfolio gives your program's leadership no access in DebriefMD.
• Share with program is a separate action. It shows you who will gain access and asks you to confirm; only then can authorized program and home-site leadership read a copy of the reviewed notes in DebriefMD. It sends no email and submits nothing to another system.
• Email copies go only to the recipients shown to you before you send, and do not grant in-app access. No coaching-note email carries a transcript or audio.

Who receives what, and when, is set out once, in section 11 of the Privacy Policy.

Your program's official assessment system remains the system of record. Your program's requirements remain your responsibility. If a submission deadline is missed, if a required assessment is not completed, or if an export fails to reach your program's system, that is between you and your program.

Check that anything you export actually arrived where it needed to go.

5. No guarantee of acceptance by any program, school, or college

We make no representation that any residency program, medical school, faculty, competency committee, licensing college, or regulatory body will accept an assessment produced with DebriefMD.

Institutions differ. Some require assessments to be entered directly by the preceptor. Some prohibit third-party tools for assessment material. Some restrict recording in clinical settings. Some require privacy-office approval before any tool touches clinical data. Some will simply say no. A deployment for a program needs that institution's own agreement or approval; your use of DebriefMD does not supply it.

DebriefMD has no affiliation with, endorsement from, or partnership with any university, residency program, college, or regulator, unless we have said so explicitly and in writing. References in the product to a program's assessment forms, including form templates named after a specific program, describe field mappings we built. They are not an endorsement by that program.

Before you rely on DebriefMD for anything that counts, confirm with your program that they will accept it.

6. AI transcription and extraction can be wrong

DebriefMD uses automated speech-to-text and a large language model. Both make mistakes. This is not a theoretical caveat. It is the expected behaviour of the technology.

Known and expected failure modes:

• Mishearing. Accents, background noise, overlapping speech, hallway acoustics, and clinical terminology all degrade transcription. Drug names and dosages are particularly easy to get wrong.
• Misattribution. With one recording covering a two-person conversation, a statement can be attributed to the wrong speaker. Speaker labels are anonymous and do not identify your preceptor.
• Fabrication. A language model can produce fluent, confident text that the preceptor never said. It can look exactly like a real quotation.
• Omission. Important feedback can be dropped, especially if it was brief, hedged, or said at the end of a recording.
• Miscategorisation. Mapping feedback to a competency framework or assessment field is an interpretation, and interpretations can be wrong.
• Confidence scores are not accuracy. Where DebriefMD shows a confidence value, it is a model's self-report. It is not a measurement of correctness. A high score on a wrong answer is entirely possible.
• French and English. DebriefMD handles both. Transcription and extraction of Canadian French medical terminology carries additional risk, which we have documented internally and accepted for production. Review French output with the same care as English, arguably more.

We publish no accuracy figure, and you should not assume one. Treat every output as unverified until you have verified it.

Human review means different things for different people, and none of them stands in for another:

• A note you have reviewed is your review. It is not your preceptor's signature.
• A preceptor's response is optional and applies only to the version that was sent to them. No response is not approval.
• Faculty who sign a rotation report review and approve its ratings, comments and conclusions themselves.

7. You are the quality gate

This is how DebriefMD is designed to work, and it only works if you do your part. You cannot export a coaching note until you have marked it reviewed.

Before you save, download, send, or share anything DebriefMD produced, read it, and confirm it reflects what your preceptor actually said. Personal download does not newly grant program access; Share with program and email copies are separate actions (section 4).

Concretely:

1. Read the full transcript, not just the structured summary.
2. Check that every statement attributed to your preceptor is one they made.
3. Correct anything wrong. The review screen lets you edit every field.
4. Delete anything the model invented.
5. Confirm no patient identifier survived into the text you are about to export.
6. Only then, export.

If you are not confident the output is accurate, do not export it. Write the assessment yourself, or ask your preceptor to enter it directly.

A note you approve and export carries your name and can enter your training record. You are responsible for the content you approve. That does not remove any obligation the law places on us, and the Terms of Service's allocation of risk (sections 15 and 16) applies subject to mandatory law and to any agreement with your institution.

8. De-identification is a safety net, not a guarantee

Recordings are processed before any identifier is removed. The speech-to-text service (Amazon Transcribe) receives the original audio, and the stored audio and the raw transcript it produces can contain whatever was said, including identifiers. The raw transcript is stored as a separate record from the de-identified one, is kept for 30 days, and is not shown in the product or included in any email or export.

DebriefMD then removes patient identifiers in three passes before the transcript is displayed or used for extraction. First, deterministic pattern matching covering all 18 HIPAA identifier categories plus Canadian identifiers: SIN, provincial health numbers (BC PHN, OHIP, RAMQ, Alberta PHN), postal codes, hospital identifiers, and room and bed numbers. The exact pattern table is in the code, and we publish no count of it, because it grows as gaps are found. Then a contextual language-model pass, for identifiers that have no fixed pattern; that model can receive any identifier the first pass missed. Then the deterministic pass again, to catch anything the model reintroduced. Redactions are left as visible inline markers such as [REDACTED-NAME], so a human reviewer can see what was caught. If the contextual pass did not complete, the debrief cannot be exported until it does.

This is a layered control. It is not a guarantee, and we do not certify any transcript as de-identified. Redaction also does not make an assessment anonymous: a coaching note about you is still personal information about you, and about the preceptor whose feedback it records.

Known residual risks, documented internally and repeated here so you are not surprised by them:

• Bare names. A name without a title in front of it, such as "Sarah in bed four", has no deterministic pattern. The language-model pass is what catches it, and a language model can miss.
• Novel formats. A health card or hospital code format we have not yet added to the pattern list will not be matched deterministically.
• Unusual phrasing. Identifiers spelled out, spaced oddly, or embedded in an unusual sentence structure are harder to detect.
• Indirect identifiers. A combination of details (a rare condition, a date, a unit) can identify a patient even when no single item is an identifier.

Do not record patient encounters, and do not deliberately introduce identifiers; the Acceptable Use Policy prohibits both. Read the transcript before you export it. If you find an identifier, remove it and tell us at hello@debriefmd.ca.

9. Consent to record is yours to obtain

Your preceptor has no DebriefMD account and never agrees to anything inside the product. They agree out loud, to you, before you start recording. Before every recording, DebriefMD requires two separate confirmations before it uses the microphone: "I understand how this recording will be processed and shared, and I will avoid patient identifiers." and "I confirm my preceptor approved this recording and the AI processing and recipients I explained to them." Together they record your understanding and your preceptor's permission to record and process. We store both, with the time you gave them and the version and language of the disclosure you were shown. A recording made before the two confirmations existed keeps the single confirmation it was made under. Faculty recording a closing conversation with a resident confirm separate statements written for that role, stored the same way.

The confirmations are the record of the conversation. They are not a substitute for having it. They are permission to record and process; they do not verify another person's identity, and they are not approval of assessment text that does not exist yet.

We cannot verify those confirmations. We were not there. We have no way to know whether you asked, what your preceptor said, or who they are.

If you record someone who did not agree, that is your act, and the legal, professional, and institutional consequences are yours. See sections 5 and 16 of the Terms of Service. We, you, and your institution each keep our own legal duties; this section does not move ours to you.

10. Where data is held, and which law applies

DebriefMD's own hosting, database, storage, queues, transcription and AI processing are configured in AWS's Canada (Central) region, ca-central-1, in Montréal, Québec. Data in transit uses TLS 1.3 at our load balancer. Section 8 of the Terms of Service lists the services and the encryption, and the Privacy Policy's sections 7 and 12 are the full statement.

That statement covers the systems we control. It does not cover:

• copies in your browser or on your device, which are wherever that device is, including a device used outside Canada;
• recipients' mailboxes: sending from Montréal does not decide where a recipient's email provider stores a message;
• files you download, and anything you or a recipient put into another system.

Which privacy law applies depends on the information, who holds it and in what role, and where you practise, and we cannot determine that for you. PIPEDA, Ontario's PHIPA, British Columbia's PIPA and FIPPA, Alberta's HIA and PIPA, and Quebec's Law 25 each have their own scope and requirements. Section 14 of the Privacy Policy explains this further.

Your institution may also have obligations of its own, and may have a position on whether a tool like DebriefMD may be used at all. Where it does, that position governs.

What DebriefMD shows to whom in the product is not the limit of your rights. It does not restrict a request under privacy law to access or correct information about you, a complaint to us or a regulator, or reporting the law protects.

We do not claim to satisfy every provincial regime for every user. We hold no security or privacy certification, and no independent third-party audit or penetration test has been completed. Our accessibility statement likewise states plainly that no third-party accessibility audit has been completed.

11. Availability, failure, and data loss

DebriefMD is provided "as is" and "as available". Recordings can fail to upload. Processing can fail. A session that stalls is automatically marked failed. We offer no uptime guarantee.

Cloud audio is scheduled for deletion 30 days after recording. The de-identified transcript is retained until you delete your account or ask us to delete it; exporting a debrief removes nothing. Once audio is gone it cannot be recovered, not by us and not by anyone. Section 9 of the Privacy Policy sets out retention, including its limits.

When your browser allows it, a recording you finish is also saved in that browser, before any identifier is removed, and stays there until the debrief has been processed, you discard it, or the site's data is cleared. That copy has no automatic maximum age, and the browser can lose it; the Cookies and Tracking page explains it.

Deleting your account removes what we hold. It cannot recall copies already delivered: emails, downloaded files, or anything entered into a program's own systems.

Do not treat DebriefMD as the sole record of a feedback conversation you cannot afford to lose. If a conversation matters, take your own notes as well.

12. Relationship to other documents

This disclaimer is incorporated by reference into the Terms of Service. The Acceptable Use Policy also applies. The limitation of liability in section 15 of the Terms and the indemnity in section 16 apply to everything described here, subject to mandatory law and to any agreement with your institution.

Where this disclaimer and the Terms of Service conflict, the Terms of Service prevail.

13. Contact

Questions about this disclaimer: hello@debriefmd.ca Privacy questions or identifier leakage reports, for the Privacy Officer role: hello@debriefmd.ca Security reports: hello@debriefmd.ca

1581455 B.C. Ltd. dba WhiteCoat Prep Governed by the laws of British Columbia, Canada.