The model comes to the data.
Patient records stay inside the health system. The environment is built there, runs there, and sends out only results.
Brings a model to be tested or trained.
Tasks · Tools · Graders
Read in place. Never exported.
Scores and failure analysis.
Model builders pay for access. Revenue is shared with the health system.
The environment runs in the health system's own infrastructure.
Patient records are read there and stay there. The health system's technical team controls the infrastructure, and its clinical team reviews the tasks.
Task definitions, scores and aggregate results.
What may leave is agreed with each health system in advance and follows the law of the country it operates in. Patient records are not part of it.
Every grade can be traced to its reason.
Each task has a defined correct outcome. Where clinical rules and guidelines exist, they set it. Where they do not, reviewed clinician judgement does.
-
Write
A clinician defines the task and its correct outcome from real practice.
-
Review
A second clinician reviews it independently.
-
Resolve
Disagreements are settled and recorded.
-
Verify
The task is run through the grader to confirm the expected outcome.
The health system stays in charge.
- The health system approves every use.
- Every run is logged and auditable.
- Ethics and regulatory approvals are obtained locally before any work starts.
- We publish our method, and we say where each environment came from.
What an environment cannot tell you.
An environment shows how a model handles the work we captured. It does not replace monitoring on live patients or clinical oversight after launch. We say so in every report.