Machine-readable risk control map for SCRIMED enterprise diligence.

Keep public and demo surfaces synthetic/no-PHI; require BAA/DPA, data classification, retention, deletion, and incident controls before live data.

Require human review, TrustOS gates, Clinical Robustness Lab scorecards, and no autonomous clinical authority.

Require citations, evidence cards, unsupported-claim refusals, and hallucination-risk scorecards.

Use model registry placeholders, eval snapshots, regression checks, and canary/rollback gates before provider activation.

Track demographic bias risk in synthetic robustness scenarios and require reviewer disposition before any clinical claim expands.

Preserve AAL2, least privilege, token redaction, fail-closed APIs, rate limits, and provider-call kill switch.

Use provider-neutral model routing with synthetic fallback and future adapters for OpenAI, Anthropic, Google, NVIDIA, Azure, AWS, and open models.

Keep provider calls disabled by default, use request thresholds, rate limits, safe errors, and cost guardrail contract checks.

Bind execution attempts to policy decisions, hashes, model route metadata, reviewer gates, and durable-store readiness.

Block HIPAA/SOC/HITRUST/FDA/ONC/security/accessibility certification and clinical validation claims until qualified approval exists.

Keep EHR writeback and production connectors blocked; use synthetic FHIR/HL7/DICOM readiness only.

Allow synthetic policy synthesis and reviewer-held packets only; block payer submission, final coding, claims, appeals, and reimbursement guarantees.

Use build/smoke validation, fail-closed protected routes, DNS readiness, rate limits, and no-secret tests before promotion.