PrecisionOS

The product is a validated VR simulation library with a measured score; AI is a layer on it. What PrecisionOS sells is immersive orthopedic and podiatric procedure simulation on Meta Quest headsets, an expert-authored library of hundreds of cases and skills, and the Precision Score, a composite of procedural parameters that the company's randomized trial validated against masked expert ratings. Since the build, the company has added Delphi, which it calls an AI Attending: an assistant in the mobile app, currently in beta, that takes questions by text or speech, recommends VR and video practice, and gives personalized feedback, with AI-identified learning gaps and transcript-based insights listed in the subscription. The privacy policy says Delphi generates responses using third-party AI systems.

Generating language and recognizing speech are learned functions, so a learned component is found; the Precision Score itself is described as a measurement, not a learned model. The simulation and the score carry the product's value without Delphi.

The system scores and recommends, and faculty review sits in the institutional tier; the review of individual scores is not described. The Precision Score is computed from procedural parameters and validated against masked expert ratings; Delphi, the beta AI assistant, recommends practice and gives feedback; and the institutional subscription adds program-level reports on resident progress, simulation use and training gaps for Clinical Competency Committee meetings, so faculty see the data before competency decisions. Individual users can also add their analytics to residency or fellowship applications.

What is not published is whether faculty can override or annotate a score, whether a resident can contest one, and how Delphi is kept from giving wrong clinical guidance.

The scoring approach is described in published work and the AI supplier is described by type. The Precision Score is a composite of procedural parameters relevant to safe implantation, validated in the 2020 JAMA Network Open trial against masked expert ratings. Delphi, the beta AI assistant, uses third-party AI systems, speech-to-text transcription and AI infrastructure providers, per the privacy policy, which says only de-identified text is sent to them.

The score's parameter weights are proprietary, the third-party AI provider and model are not named, and no version history or recalibration practice is published for either.

The architecture and the terms with the AI provider are described; no provider is named. The privacy policy lists service provider categories (cloud hosting, data storage, AI or machine learning infrastructure, speech to text transcription, customer support, database management, analytics and payment processing). It says Delphi text is deidentified before reaching any third party AI system and is not used to train or fine tune models. It also refers to a subprocessors page without linking to it, and that page may name the providers.

A peer reviewed randomized trial of the product is published with its method available to read. Lohre et al., JAMA Network Open 2020, block randomized 18 senior (PGY-4 and PGY-5) orthopedic residents from nine Canadian institutions at a national course to immersive VR training or technical video for a reverse shoulder arthroplasty step, then had masked assessors rate them on cadavers. The VR group scored higher on OSATS (15.9 vs 9.4; difference 6.9, 95% CI 3.3 to 9.7; P < .001), completed training faster, and made fewer critical errors in the early steps (15 vs 65 percent error rate); the transfer effectiveness ratio was 0.79. The assessment was on cadavers, not on patients.

The company co-funded the trial with the Canadian Shoulder and Elbow Society; its founder and chief executive and another author held equity, which the paper discloses, and assessors were masked. The sample is small and covers one procedure, and no independent replication is published. The company's Resource Hub lists further studies, and a program case study describes 30 residents over 21 months generating 88,000 data points.

Categorical commitments on the AI assistant are published; retention periods are not. The privacy policy (updated 4 September 2026) says text from Delphi conversations is de-identified by removing or masking information that directly identifies the user before it goes to a third-party AI system, that identifiable transcripts are never provided to those systems, and that Delphi chats are not used to train or fine-tune AI models; users are told not to submit information identifying a patient. The terms prohibit uploading patient information and require x-rays and video to be anonymized.

Retention is as long as needed for the purposes described, with residual backup copies for a limited period, and no period is stated for performance records or transcripts. Safety engineering, meaning how a clinically wrong Delphi answer is caught and handled, is not published beyond the beta label. The sensitive record is a quantified history of an identified surgeon's performance, which institutions and possibly device companies can see.

The company states that patient information does not belong in the service and explains how, with a service privacy policy behind it. The terms require users not to upload any media containing health-protected or patient information and say all x-rays and video must be anonymized; the privacy policy (updated 4 September 2026), which covers the platform, the mobile app and Delphi, tells users not to submit information that directly identifies a patient to Delphi. Training runs on simulated cases.

There is no business associate agreement to publish, and the position rests on users following the rule.

Safeguards are asserted with nothing independent behind them. The privacy policy says the company uses reasonable administrative, technical and organizational safeguards; no SOC 2 report, ISO 27001 certificate, penetration testing statement, trust page or security page is published on the site, in the privacy policy or in the terms.

The data is identified performance history for surgeons and trainees, Delphi transcripts and session recordings, generated partly on personal headsets at home and shared with institutions.

No device claim is made and none applies: this is surgical training software, and the terms say the offerings are provided for informational purposes and are not a substitute for professional medical advice, diagnosis or treatment. That clause allocates liability rather than stating a regulatory position, and no statement of intended purpose or device position is published.

The exposure that applies is legal: a quantified performance score tied to a named surgeon is the kind of record a plaintiff may seek, and whether it is protected depends on state peer review privilege.

The scoring instrument has been validated overall and not across groups, and nothing is published on testing Delphi. The company's 2020 randomized trial (JAMA Network Open, 18 senior orthopedic residents from nine Canadian institutions) reports that the Precision Score correlated with masked expert OSATS ratings (r = 0.74) and final implant position (r = 0.73); the groups were balanced on hand dominance and gender, but no results are reported by group. Delphi is labeled beta, and no page describes how its recommendations and feedback are checked.

Scores in VR are sensitive to factors unrelated to surgical judgment, such as handedness, visual acuity, prior gaming or VR exposure and simulator sickness, and institutional reports feed Clinical Competency Committee meetings.

The accuracy of the score the product produces is published and cited by the company. Lohre et al., JAMA Network Open 2020 (a randomized trial the company co-funded and its chief executive co-authored, listed in the company's Resource Hub), reports that the Precision Score agreed with masked assessors' OSATS ratings of cadaveric reverse shoulder arthroplasty with r = 0.74 and with final implant position with r = 0.73, in 18 senior residents; it is a test-set result in one procedure, not performance in routine use. That is a published measure of how well the product's output tracks a reference standard, with its method and case count.

Nothing comparable is published for Delphi, the beta AI assistant that gives feedback and recommendations. The terms say the offerings are provided for informational purposes only and that the company accepts no liability for decisions made in reliance on them, and no route to contest a score is described.

No clinical record integration exists: this is surgical simulation, and the terms forbid uploading patient information. For a residency program the systems of record that matter are learning management, evaluation and competency tracking systems, and no integration with any of them is claimed.

The institutional subscription provides program-level reports for Clinical Competency Committee meetings and an ACGME-aligned curriculum, and data is reached through the company's dashboard; no export, standard or connection to a residency management system is described.

A single hosted service with location implied rather than committed. The privacy policy says personal information may be processed in Canada, the United States or other countries where the company or its service providers operate, and names cloud hosting and data storage among its provider categories without naming a host or region. Delivery runs on Meta Quest 3 and 3S headsets, the mobile app and a web dashboard; personal headsets are used for at-home training, and multi-user sessions and recording are supported.

Individual pricing is published with a free trial. The home, individuals and pricing pages list the Individual subscription at $199 a month or $1,999 a year, covering the library of hundreds of cases and skills, Delphi feedback and recommendations and transcript-based insights, with a free trial started online; the headset is not included (Meta Quest 3 and 3S are supported), and the mobile app is free. The Institutions subscription, which adds program-level reporting for Clinical Competency Committees and an ACGME-aligned curriculum, is quoted on request.

A surgeon can start without a sales conversation. A program can compare the per person price with its quote. How institutional pricing scales with residents, and whether headsets are supplied, is not stated.

Coverage is named by specialty, with validation behind part of it. The home page lists the library as available today for sports arthroscopy, shoulder and elbow, orthopedic trauma, spine, pediatrics, podiatry and foot and ankle, and arthroplasty, with physical medicine and rehabilitation, pain medicine, oral and maxillofacial surgery, dental surgery, physiotherapy and anesthesiology coming. Buyers are individual surgeons and trainees, residency programs and institutions.

The randomized evidence covers one shoulder arthroplasty step in senior residents; other specialties rest on case studies (arthroscopy, trauma and hip approaches in named residents' stories).