Theator

There is no product without the model. Theator's input is raw intraoperative video, and everything downstream, the step identification, the safety milestone tagging, the operative narrative, the departmental analytics, is produced by computer vision and large vision language models operating on that feed. No human transcribes, annotates or reviews the video to make the product work. This is the purest AI centrality case in the category and it stands in direct contrast to C-SATS, which takes the same input, surgical video, and routes the assessment to human expert and crowd reviewers. Same raw material, opposite mechanism, and the two records should be read together.

The documentation path is well governed and the analytics path is not. For Surgery-to-Text the oversight design is explicit and appropriate: the system produces a draft operative report which the surgeon reviews and electronically signs before it becomes the record, so a human remains the author of a legal clinical document. The quality and safety layer carries no equivalent published control. Nothing states whether an automatically tagged critical event or complication is reviewed before it enters departmental benchmarking, whether a surgeon is notified when their case is flagged, or whether they can dispute the tag. An automated adverse event label attached to a named surgeon is consequential in a way a draft report is not, because it is the sort of finding that reaches credentialing and peer review. Publishing the review path for automatically detected safety events would move this to A.

More technically specific than any peer in this category. Theator names computer vision, deep learning and large vision language models rather than gesturing at AI, describes real time inference on the live high definition video feed, and specifies HL7 messaging for the EHR link, which lets a technical buyer reason about what is actually running. The published warm ischemia time validation goes further, describing the concrete mechanism, computer vision detecting clamp placement and removal, and naming expert reviewed video as ground truth. What is absent is a model card or any general accuracy disclosure: no published sensitivity or specificity for step identification or safety event detection across procedures, and no statement of which procedure types the models have been trained and validated on versus which are inferred. Publishing detection performance by procedure type would move this to A.

The strongest evidence posture of any AI native vendor in this category, on two grounds. First, a peer reviewed validation study exists: platform derived warm ischemia times during partial nephrectomy were compared against expert reviewed video as ground truth and against the times surgeons recorded in their own operative reports, using paired sample tests on cases performed between October 2023 and April 2024. That is a narrow use case, but it is a real external measurement of the computer vision against a defined truth standard, which almost nothing else in this lane has. Second, Theator anchors its documentation claim to published external research rather than to its own numbers, citing a Journal of the American College of Surgeons finding that memory written operative reports are only 72.8 percent accurate. Publishing the baseline human error rate is the Octozi pattern and it is the honest way to frame an automation claim. What holds this at B is the absence of any outcome study: nothing published shows that using the platform changes surgical quality, complication rates or documentation driven revenue, which is what the marketing implies.

The data held here is among the most sensitive in the whole index: continuous video from inside a patient's body, captured automatically for every case, linked over HL7 to that patient's record and to the named surgical team operating. Anonymization is described as built into the pipeline, but that claim was located only in a third party directory rather than in vendor published material, so it is recorded as secondary sourced and unconfirmed. The Oracle collaboration places the workload on Oracle Cloud Infrastructure under existing enterprise security and compliance controls, which is meaningful and is also an inherited posture rather than a disclosure by Theator about its own handling. For a vendor whose model is automatic capture of every case, the specific unanswered questions are consent and retention: who consents to the recording, how long video persists, and whether it trains future models. Publishing a data handling statement covering those three would move this to B immediately.

No HIPAA statement, BAA template or scope description was located in vendor published material. The axis is left unrated on absence of evidence rather than graded on inference, following the same treatment the index applied to Anterior. That abstention should not be read as neutral: unlike the workforce vendors elsewhere in this category, where HIPAA genuinely does not apply, Theator plainly processes protected health information, since it holds intraoperative video cross referenced against EHR records. A vendor writing into the Oracle Health EHR under a formal collaboration will have executed BAAs; the gap is that none of it is publicly documented, so a buyer cannot assess scope, subprocessors or video handling terms before entering a sales conversation.

No SOC 2 Type II, HITRUST or ISO 27001 attestation was located, and no trust center was found. Deployment on Oracle Cloud Infrastructure under Oracle's enterprise security and compliance controls provides real infrastructure assurance and is credited, but infrastructure certification is not application certification, and the distinction matters for a platform that ingests and stores operating room video. The comparison inside this category is unflattering: C-SATS handles the same class of data and is HITRUST CSF Certified. A published attestation would move this to B, and given the enterprise partnership the likelier explanation is that one exists without being published.

Theator is positioned as documentation, analytics and quality software rather than a diagnostic or therapeutic device, so no FDA pathway was located and the axis is rated accordingly rather than penalized. Two regulatory exposures apply instead and both are sharper than the FDA question. The first is discoverability, as with any scored surgical video: automatically detected complications and safety events tied to a named surgeon and a specific case are exactly what a malpractice plaintiff seeks, and state peer review privilege statutes vary and were not drafted with continuous automated adverse event detection in mind. The second is documentation integrity, because Surgery-to-Text produces reports optimized for billing as well as clinical accuracy; the index treats coding drift as a False Claims Act exposure wherever an automated system shapes what gets billed, and the same scrutiny applies here. Both belong in counsel review before deployment.

The governance risk here is not demographic bias in the usual sense, it is unadjusted comparison between surgeons. Theator automatically tags complications and critical events and feeds them into system wide quality benchmarking and comparative safety analysis across multi site networks. Nothing published states whether those comparisons are risk adjusted for case mix, patient acuity or anatomical difficulty. Without adjustment, the surgeon who accepts the hardest referrals produces the worst dashboard, and the rational response to being measured that way is to stop taking difficult cases, which harms exactly the patients with the fewest alternatives. No false positive rate for automated event detection is published either, so a buyer cannot tell how often a flagged complication is not one. Publishing the risk adjustment method and the event detection false positive rate would move this to B, and they are the two disclosures a surgical department should demand before any benchmarking goes live.

The deepest clinical integration in this category by a wide margin, and the only vendor here for which the EHR axis is assessed on its literal terms rather than on a workforce systems analogue. Theator links surgical video to the electronic health record over HL7 messaging, cross references intraoperative observations against EHR data to build the operative narrative, and under the June 2026 Oracle Health collaboration writes structured reports directly into the Oracle Health EHR where clinical, quality and billing staff access them immediately, with the compute running on Oracle Cloud Infrastructure. That is native write back into a major EHR rather than an asserted integration. The residual limitation is concentration: the documented depth is with one EHR vendor, and equivalent integration with Epic, the dominant system in US surgical departments, was not located.

More concrete than most of this category. The platform ingests from cameras already installed in the operating room rather than requiring new capture hardware, processes video in real time, and under the Oracle collaboration the compute intensive computer vision workload runs on Oracle Cloud Infrastructure. Naming the cloud is a genuine disclosure and it lets a buyer reason about the security and compliance envelope. What is not published is residency: no regions are named, nothing states whether operating room video crosses borders, and given research and development operations in Tel Aviv alongside a US commercial footprint that is a question a European or Canadian institution would need answered before recording a single case. Naming the storage regions would move this to A.

No pricing, rate, unit or charging mechanism was located in vendor material. The clearest independent signal comes from an unexpected place: the published warm ischemia time validation study notes that because the platform analyses many parameters across laparoscopic, robotic and endoscopic procedures, determining the cost attributable to any single use case is difficult. When academic authors evaluating the technology say the pricing cannot be apportioned, buyers should expect the same problem in procurement. The commercial framing also spans two budgets, quality improvement and revenue integrity, since Surgery-to-Text is sold partly on billing accuracy, and a system charged against recovered revenue warrants the savings linked scrutiny the index applies elsewhere. Publishing whether pricing is per operating room, per case or per surgeon would move this to B.

Coverage follows the camera. Any minimally invasive or robotic procedure is already visualized, which is the structural insight the company is built on, and the platform is described as covering laparoscopic, robotic and increasingly open surgery across hospital surgical departments and academic medical centers internationally. Documented specialty depth is narrower than the general claim: the peer reviewed validation is in urology, specifically partial nephrectomy, and the broader evidence base for other specialties was not located. The practical limit is that open and non visualized procedures fall outside the model's reach without new capture hardware, so a department's coverage is determined by its case mix rather than by the contract. Buyers should ask which specific procedures have validated step and event models rather than assuming the platform generalizes across their service line.