Prove every clinical AI is reliable.
Without a single chart ever leaving the health system.
Your health system runs a stack of AI documentation, decision-support, and triage tools. The Med Board measures every one of them — how distinct each tool actually is, whether its answers are drifting, whether its outputs stay clinically grounded — then does two things with that measurement. It shows you which tools are behaviorally identical, so you can stop paying for the same capability twice. And it emits one signed, tamper-evident record per output documenting how the tool performed against its FDA PCCP change-control gate and your §1557 nondiscrimination duty — carrying none of the chart. The model is measured inside your network; only the proof crosses the boundary.
You can’t govern what you can’t measure.
A hosted clinical AI tool is a black box. It documents, it triages, it recommends — and every one of those outputs runs through rules that put the responsibility on the health system, not the vendor. The FDA’s PCCP framework demands you control and monitor change in a SaMD tool. HHS §1557 demands the tool not discriminate across protected classes. ONC HTI-1 demands decision-support transparency, and CMS scrutinizes AI in utilization-management coverage decisions. The measurement is designed so the chart never enters the audit surface; only the signed record crosses the boundary.
None of those questions can be answered by a vendor’s 510(k) summary or a marketing claim. They require a measurement of how the tool actually behaves — taken on your fleet, repeatable by an adversary, and produced without shipping a single chart to anyone.
What the Med Board measures.
The Med Board points the Astrolabe at your fleet of clinical AI tools and grades each one on three axes. The same measurement saves money and survives a review.
The smaller bill. Health systems rarely run one AI tool — they run several, often with overlapping mandates bought by different service lines. When the Med Board clusters them by behavior, the tools that produce the same outputs collapse into the same group. Every duplicate is a license you can drop without losing a capability. The measurement names which ones.
The defensible record. The tools that survive get graded against the doctrine — and every grade is signed. When a surveyor, a payer, or a plaintiff’s expert asks how you knew the tool was reliable, you have a record, not an assurance.
Proof for every output.
Every time a clinical tool produces an output, the Med Board emits one signed, tamper-evident record of the output's behavior against your reliability bar — cryptographically tied to a behavioral measurement of the model, but carrying none of the underlying chart. For a CMIO, that is how you evidence a tool's behavior against its PCCP and §1557 gates without the proof ever touching PHI: the record crosses the boundary, the chart never does. When a regulator, a payer, or your own quality committee asks how the health system supervised its AI, this is the record you hand them.
The doctrine it maps to.
The Med Board is not a separate model. It is the Astrolabe reading shaped for clinical practice: the contracts it asserts at runtime, the doctrinal anchors it scores against, and the substrate primitives that make the record byte-for-byte reproducible by a third party.
The audit happens in code. audit_medicine(responses, clinical_band=…) checks all five mapped review areas — HHS §1557, HIPAA data-minimization posture, FDA SaMD/PCCP, ONC HTI-1, and CMS Utilization-Management — before it returns; then it scores every doctrinal axis and emits the record through the Port — the same boundary the Law Board, the School Board, and the War Board ship through. Want the full corpus? The complete pin-and-state reference lives at the Medicine corpus surface.
Next step.
The first engagement is a scoping conversation. Tool inventory, hosting topology, the clinical band (decision-support, ambient-documentation, utilization-management, samd, or predictive-dsi), the doctrine anchor (HHS §1557 / FDA SaMD-PCCP / ONC HTI-1 / CMS UM / system-specific), and the target review date. We return a Phase 0 scope and an evidence-record delivery date. No PHI changes hands at any stage.
Medicine AI workflow.
69 compliance pins. 50 state constituencies. Every regulatory anchor and every state buyer has a canonical Planisphere page framed for clinical AI. Pick the pin the evaluator names; the bundle lands the cite-anchor next to every SHA. This is the corpus colossum for health-system AI governance.