Evidence over assertion
Every material conclusion must be traceable to evidence and method.
Real Intelligence Standard
A defined standardfor a defined clinical use.
The Standard brings the principles, evidence requirements, assessment method, and publication rules for medical AI into one risk-sensitive framework.
Working framework / 2026
Governing principle
Working framework publishedEvery conclusion is limited to the conditions actually reviewed.
A finding applies only to the product version, use, population, setting, and period of validity under review. It is not universal approval and does not imply an absence of risk.
Human consequence first
Evaluation begins with the decision a product may influence and the person who may experience the result.
Rigor proportional to risk
Higher-consequence uses require stronger evidence, more demanding tests, and closer review.
Uncertainty made visible
Incomplete or contradictory evidence is reported rather than converted into false certainty.
Evaluation dimensions
Eight dimensions inform each assessment.
The applicable questions and evidence thresholds vary by product and use. The dimensions remain connected: strong performance in one area does not erase a consequential weakness in another.
Clinical validity and utility
Whether outputs are supported and meaningfully useful for the stated purpose.
Safety and failure behavior
What happens when a product is wrong, uncertain, incomplete, or used outside ideal conditions.
Reliability and calibration
Whether performance is consistent and stated confidence corresponds to observed accuracy.
Generalizability and equity
Whether performance holds across relevant populations, settings, languages, and subgroups.
Evidence quality
Whether claims are supported by appropriate studies, representative data, and independent validation.
Human use and workflow
How the product changes comprehension, decisions, escalation, and responsibility.
Transparency and communication
Whether users can understand intended use, limitations, uncertainty, and appropriate next steps.
Privacy, security, and governance
How sensitive information is collected, processed, retained, protected, and monitored over time.
Risk proportionality
Evidence requirements rise with clinical consequence.
The same evidence threshold should not apply to a low-consequence administrative aid and a product that could materially alter diagnosis or treatment.
01
Possible harm
How severe could a plausible error be?
02
Product autonomy
How directly does the output shape or initiate action?
03
Error visibility
How likely is a wrong result to be detected?
04
Population vulnerability
Who bears the risk and what protections are needed?
05
Reach and immediacy
How widely and quickly could a failure affect decisions?
How assessment works
From a defined question to a documented finding.
The method is specific enough for consequential decisions, adaptable across product categories, and transparent enough for qualified scrutiny.
01Scope
Statement of Scope
A structured record of intended use, population, users, inputs, outputs, material claims, and version.
02Review
Evidence Review
A map connecting every material claim to supporting studies, datasets, contradictory evidence, and limitations.
03Test
Clinical Test Protocol
Specialty-designed representative cases, edge cases, safety-critical scenarios, and subgroup analyses.
04Measure
Reliability and Calibration Assessment
Quantitative testing of accuracy, calibration, consistency, robustness, and consequential failure patterns.
05Observe
Human Factors Assessment
Evaluation of how people and clinical teams understand, question, and act on product outputs.
06Monitor
Continuing Review
Ongoing tracking of product changes, new evidence, safety reports, performance drift, and incidents.
Public by design
Principles, evidence expectations, conflict rules, and determination requirements are open to scrutiny.
Protected where necessary
Selected cases and security-sensitive procedures may remain confidential to preserve test integrity.
Reviewed over time
Methods and findings are revisited as products, evidence, and evaluation practices change.
Public determinations
Findings will be stated with scope and limitations.
No RIR product determinations have been issued.
A public record will explain what an evaluated product can be trusted to do, for whom, under what conditions, and with what uncertainty. It will not reduce materially different findings to a universal score or “best product” list.
01
The product version, intended use, users, population, and setting
02
The evidence reviewed and methods applied
03
The finding, limitations, and material uncertainty
04
The review date, validity period, and any later status change
Depending on the evidence, a scoped finding may state that the applicable requirements are met, are not met, or cannot yet be determined. Later product changes or new evidence may amend, suspend, supersede, or expire that finding.
Independence
Payment funds assessment work. It does not determine the finding.
Fees are fixed in advance and are not contingent on outcome.
Commercial relationships do not alter methods or findings.
Any permitted remediation is documented and retested.
Published conclusions include their evidence, limitations, and period of validity.