Documents, measurements and observations retain their provenance and time.
Clinical oversight
AI can propose a hypothesis. The doctor makes the decision.
BIOS retains sources, identifies missing data and separates machine-generated conclusions from clinical decisions.
A use case is launched only after its rules and responsibilities have been agreed.
A possible association does not conceal uncertainty or missing data.
Confirmation or rejection is recorded separately from the system's conclusion.
The patient is not given a system response as a final medical assessment.
Decision model
Four independent stages
The path from fact to release can be traced.
Review policy
What must not be released automatically
Higher risk requires stricter stop rules.
Clinically significant conclusion
Requires a doctor's review.
Potential urgency signal
Referred to a separate review procedure.
Insufficient data
The system stops.
Sources contradict one another
The discrepancy remains visible.
Permission absent
Release is blocked.
The conclusion cannot be explained
It is not assigned clinical status.
Change control
Every change has a version
An AI update must not silently change medical meaning.
Source
The rule's provenance.
Version
Revision and owner.
Review
Accuracy and safety.
Approval for use
Use case and audience.
Monitoring
Deviations and incidents.
Oversight is tested alongside the product
Activation depends on the use case, clinic and completed review.
Accountability
Each use case has a clinical and a technical owner.
Traceability
The fact, hypothesis and decision are recorded separately.
Incident
An event can be stopped and linked to a version.
Rejection
A rejected conclusion does not become a fact.
Oversight begins with a specific use case.
Owners, rules and stop conditions are defined before launch.