A working AI platform should be transparent about the methodology underneath the metrics. This page is our running statement of where Referral Review stands today — operating data, validation approach, and the documents available to diligence reviewers.
Observed against pre-deployment routing patterns; the engine routes guideline-concordant primary-care-manageable conditions back to PCP-LED management. The figure is a directional operating metric, not yet a validated outcome. It is the primary signal that the engine is doing the work it was designed to do.
Of the engine's recommendations reviewed by deploying clinicians, 88%+ concord with the engine's pathway. Coverage spans cardiology, endocrinology, GI, dermatology, orthopedics, and twenty-plus other specialties; the remainder are documented overrides with reason captured in the audit log.
Production batch mode. Sub-10-second decision latency under typical load. Failover and audit log integrity preserved through deploy windows.
Across the production window. The 39 hardcoded patient-safety rules fire before any other engine logic and cannot be bypassed by any user. Closed-loop accountability requires immutable audit records — the safety rule layer fires upstream of every recommendation, and the audit log proves it.
Operating data tells us the engine is doing what it was designed to do. Validation tells the world. Here's the framework — and our honest position on each piece.
Production-validated cohorts are pre-specified — patient population, time window, exclusion criteria, denominator basis. Status: documented for the operating dataset; comparator-arm definitions in development.
Operating measurement uses pre-deployment routing patterns at the operating sites. A retrospective comparator with matched-cohort design is the next step. Status: pre-deployment baseline measured; matched-cohort design in development.
Effect size, CI, sensitivity to cohort definition, robustness across specialties and sites. Status: descriptive statistics complete on the operating dataset; inferential analysis pending matched-cohort design.
The current measurement cohort is eight production sites, with multiple additional sites expected to come online. Status: multi-site production deployment complete and expanding.
Buyers, plan diligence teams, and clinical advisors regularly ask to see the underlying work. Most of it exists in defensible form today and is available under a standard mutual NDA.
The validation methodology — authoring by board-certified physicians, stress-testing against the production cohort, ongoing pilot review by 20+ practicing primary care clinicians, and twice-yearly minimum update cadence — is documented and available under NDA.
Request documents under NDA →The right diligence question for any healthcare AI platform isn't whether we'll show you the metrics — it's whether we'll show you the methodology underneath them. Ask us. The answer is yes.
Request validation documents →