Live demo — a seeded registry, fully interactive to read, sealed to write. Every number is computed by the real engine. Get this for your institution

Clinical AI GovernanceReadmission risk stratification

Readmission risk stratification
CareSignal (demo vendor) · v3.0 · Inpatient · Discharge planning support · deployed 2026-08-25
moderate Suspended — equity

Performance against the locked baseline

baseline 0.88 · sens 0.81 · spec 0.86 · floor at −10% · locked 2026-08-25
QuarterAUROCDecaySens/SpecParity gapVerdictAttested by
2026-Q3 0.874 0.68% 0.79 / 0.85 0.09 ⚠ equity [email protected]
2026-08-25

Quarterly attestation

The number, on the record, once a quarter. Breach the threshold on an active model and suspension is automatic — continuation is what needs a signature.

Lock a baseline

Insert-only: a new lock supersedes the old on the record, never edits it. The threshold you choose is the automatic suspension floor.

Lock an equity threshold and a quarterly parity gap above it suspends the model automatically — the same enforcement as decay. The headline metric holding while a subgroup slides is exactly the failure AUROC alone cannot see.

Accountable sign-offs

status changes need an org admin

You sign as who you are signed in as — email from the session, your stated role, your rationale, and a hash anchored in the audit chain. Deployment, suspension, reactivation and retirement require an organisation administrator.

ActionSigned byRationaleSignature
acknowledge drift
2026-08-25
 [email protected]
CMIO
 Equity suspension acknowledged. Subgroup analysis commissioned; model remains suspended until the gap is explained and a revalidated baseline is locked. 09d335fc26d3…
approve deployment
2026-08-25
 [email protected]
CMIO
 Local validation on 41,250 discharges complete; baseline and equity threshold locked; discharge-planning use only, care team retains full discretion. 0ed4e6ee9dd7…

The fifteen-control checklist

6/15

The Institute's published RUAIH / CHAI / NIST crosswalk, as a working checklist for this model. Change a status and it saves — every change lands in the audit chain. 'Not started' is the honest default; the binder shows the gaps.

RUAIH 1 — Governance
The AI governance committee
CHAI: Organizational structure · NIST: GOVERN · evidence: AI governance committee charter
The organizational AI policy
CHAI: Organizational AI policy · NIST: GOVERN · evidence: Organizational AI policy
The AI inventory and product registry
CHAI: Responsible AI lifecycle management · NIST: MAP · evidence: AI use case registry
Board and executive reporting
CHAI: Organizational resources · NIST: GOVERN · evidence: Board reporting pack
Resourcing the governance programme
CHAI: Organizational resources · NIST: GOVERN · evidence: Programme resourcing plan
RUAIH 2 — Effective data management
Data governance and minimum necessary
CHAI: Responsible data management and use · NIST: MAP · evidence: Data use agreement and access policy
Data security controls for AI systems
CHAI: Responsible data management and use · NIST: GOVERN · evidence: Access control and audit log review record
RUAIH 3 — Risk and bias reduction
The risk-tiering method
CHAI: Risk and impact assessments · NIST: MAP · evidence: Intake and risk-tiering procedure
Bias and equity assessment
CHAI: Risk and impact assessments · NIST: MEASURE · evidence: Local bias assessment
Third-party and vendor due diligence
CHAI: Third-party management · NIST: GOVERN · evidence: Vendor AI disclosure request and completed questionnaire
RUAIH 4 — Monitoring, evaluating and validating safety performance, effectiveness and responsible use
Local validation before deployment
CHAI: Responsible AI lifecycle management · NIST: MEASURE · evidence: Local validation study protocol and result
Post-deployment monitoring
CHAI: Responsible AI lifecycle management · NIST: MEASURE · evidence: Monitoring plan with named thresholds
AI safety event reporting
CHAI: Responsible AI lifecycle management · NIST: MANAGE · evidence: AI safety event reporting route
RUAIH 5 — Transparency, education and training
Patient disclosure and consent
CHAI: Education, training and feedback · NIST: GOVERN · evidence: Patient disclosure standard
Role-specific workforce training
CHAI: Education, training and feedback · NIST: GOVERN · evidence: Training curriculum and completion records

Survey binder

The full governance record as one download — assembled live, with the audit chain's verification verdict inside. This is the document you hand the surveyor.

Export survey binder

Audit trail — this model

acknowledge drift — Readmission risk stratification
[email protected] · 2026-08-25T22:14 UTC · signed as cmio, sig 09d335fc26d35048d4a6e88d1d90bcf99c9276f36c26019067cd095269381906: Equi
suspended drift — Readmission risk stratification
system · 2026-08-25T22:14 UTC · active → suspended_drift: automatic: 2026-Q3 demographic parity gap 0.09 exceeded locked e
drift attested — Readmission risk stratification
[email protected] · 2026-08-25T22:14 UTC · 2026-Q3: AUROC 0.874 vs baseline 0.88, decay 0.68% — EQUITY THRESHOLD BREACHED (gap 0.09 >
active — Readmission risk stratification
[email protected] · 2026-08-25T22:14 UTC · under_review → active: approve_deployment signed by [email protected] as cmio
approve deployment — Readmission risk stratification
[email protected] · 2026-08-25T22:14 UTC · signed as cmio, sig 0ed4e6ee9dd77b2ca41f00bf1c959c93a48195c0829c7f866e2fa5c49bec75fd: Loca
under review — Readmission risk stratification
[email protected] · 2026-08-25T22:14 UTC · proposed → under_review: baseline locked, awaiting deployment sign-off
baseline locked — Readmission risk stratification
[email protected] · 2026-08-25T22:14 UTC · AUROC 0.88, threshold 10%, equity threshold gap 0.05, evaluated 2026-08-25
model registered — Readmission risk stratification
[email protected] · 2025-11-03T01:00 UTC · Readmission risk stratification (CareSignal (demo vendor)) — moderate risk, Inpatient