SCS + SCP: CHAI Alignment¶

Coalition for Health AI — Blueprint for Trustworthy AI¶

Supervisory Control Plane — Ohana AI Strategy & Systems Architecture Contact: info@ohana-tech.com | supervisoryplane.com

What CHAI Is¶

The Coalition for Health AI (CHAI) is the leading industry body for responsible AI in healthcare, co-founded by major health systems, academic medical centers, and technology companies. CHAI publishes governance frameworks that inform healthcare AI procurement, deployment standards, and emerging regulation.

Two CHAI frameworks are directly relevant to AI-driven prior authorization:

CHAI Blueprint for Trustworthy AI (2023) — General principles for healthcare AI: safety and effectiveness, equity and fairness, transparency and explainability, privacy and security, accountability and oversight.
CHAI T&E Framework for Prior Authorization AI-Supported Criteria Matching (December 2025) — Prior-authorization-specific requirements across four principles: Transparency, Fairness, Safety, and Usefulness.

SCS + SCP do not merely satisfy CHAI requirements — CHAI requirements are implemented as SCS structured context documents. SCS ships CHAI as machine-readable, versioned governance context that any AI agent can receive at runtime.

The Key Differentiator¶

Most vendor compliance mappings are documentation exercises — a table showing which features correspond to which requirements.

SCS inverts this. The CHAI requirements are encoded as SCS SCDs (Structured Context Documents): versioned YAML files that define the requirement, its benchmark, implementation guidance, and evidence requirements in a structured format. Load them into any project with one command:

/scs-team:use chai

The result: CHAI governance requirements are delivered to your AI agents as structured context at runtime via SCP — not referenced in a document, but active in the agent's operating context.

CHAI Blueprint for Trustworthy AI (2023)¶

General principles, applicable to all healthcare AI deployments.

CHAI Principle	SCS / SCP Coverage
Safety and Effectiveness — AI must demonstrably improve outcomes without introducing new harms	SCS safety concern bundle documents pre-deployment safety requirements. SCP enforces scope boundaries at runtime — agents cannot exceed their defined intent.
Equity and Fairness — AI must not perpetuate or amplify health disparities	SCS fairness requirements are encoded as standards SCDs with specific benchmarks (e.g., PPV parity thresholds). SCP audit trails provide the monitoring data needed to detect demographic disparities post-deployment.
Transparency and Explainability — Stakeholders must understand what the AI does and why	SCS bundles are the AI's documented operating context — every policy, constraint, and requirement is explicit, versioned, and human-readable. SCP audit logs record what context was active for every decision.
Privacy and Security — PHI must be protected	SCS PHI handling policies encode HIPAA minimum necessary requirements as structured context. SCP role-based delivery ensures agents receive only the PHI-relevant context their role permits.
Accountability and Oversight — Clear accountability chains, human oversight for clinical decisions	SCS provenance fields record who authored and updated every governance artifact. SCP policy layer is human-controlled: humans define what agents receive; agents cannot override governance context.

CHAI T&E Framework — Prior Authorization (December 2025)¶

The CHAI T&E Framework is the most detailed healthcare AI governance standard specifically for prior authorization. All four principles are implemented as SCS SCDs in the CHAI Prior Authorization bundle.

Transparency¶

Requirement	ID	SCS / SCP Coverage
CHAI Applied Model Card — document model purpose, training, evaluation, and limitations before production	CHAI-PA-TRANS-001	SCS project bundle is the model card: structured documentation of what the agent does, what policies it operates under, and what its constraints are.
Decision Traceability — every criteria matching assessment traceable to specific policy criteria and evidence	CHAI-PA-TRANS-002	SCP logs policy version, SCDs delivered, and intent for every context request. The reasoning chain from policy to decision is reconstructable.
Context Version Auditability — must be possible to determine exactly which governance context was active at time of any specific decision	CHAI-PA-TRANS-003	This is SCP's core capability. Every context request logs the bundle version and SCD versions active at that moment. Any past decision can be reconstructed: what the agent was told, under which policy version, at what exact timestamp.

CHAI-PA-TRANS-003 is the requirement most auditors will focus on. SCP is the only purpose-built infrastructure that satisfies it — not through logging alone, but through immutable versioned bundles that preserve the exact governance context as a replayable artifact.

Fairness and Bias Management¶

Requirement	ID	SCS / SCP Coverage
Predictive Parity — PPV must not vary by more than 0.05 across demographic subgroups	CHAI-PA-FAIR-001	SCS fairness SCD encodes the 0.05 PPV threshold as a structured benchmark. SCP audit trails provide the demographic-stratified outcome data needed to measure parity.
Toxicity Language Scoring — generative outputs screened for harmful language	CHAI-PA-FAIR-002	SCS toxicity requirements encoded with benchmark (<0.1 score), including clinical term allowlisting guidance.
Auto-Approval vs. Manual Review Outcome Disparity — address the Symptoms Paradox	CHAI-PA-FAIR-003	SCS fairness SCD documents the Symptoms Paradox monitoring obligation. SCP routing logs enable the disparity analysis the standard requires.
Demographic Appeal Delay Monitoring	CHAI-PA-FAIR-004	SCP audit trails provide the per-request demographic and timing data needed for appeal delay analysis within clinical categories.

Safety and Reliability¶

Requirement	ID	SCS / SCP Coverage
Pre-Deployment Failure Analysis — FMEA/FTA with RPN <100 for critical failure modes	CHAI-PA-SAFE-001	SCS safety SCD encodes the four AI-specific risk categories (data drift, model bias, hallucination, policy change impact) with RPN <100 benchmark. Hallucination risk specifically calls for grounding criteria matching against structured policy documents — the SCS bundle is that structure.
System Reliability — ≥99.9% uptime, fail-safe to human review	CHAI-PA-SAFE-002	SCP deployment on Docker Compose / AWS with documented fallback pathways. Fail-safe is a SCP design principle: when context cannot be delivered, requests route to human review.

On hallucination (RISK-003): The CHAI safety SCD specifies: "Ground all criteria matching against structured policy documents. Require citation of specific policy sections for each criterion evaluated." An SCS bundle containing your coverage criteria, clinical guidelines, and escalation paths is the structured policy document CHAI requires. This is the clearest statement in the CHAI framework of the SCS value proposition.

Usefulness, Usability, and Efficacy¶

Requirement	ID	SCS / SCP Coverage
Turnaround Time Compliance — 72 hrs expedited, 7 days standard	CHAI-PA-USE-001	SCS usefulness SCD encodes CMS TAT benchmarks. SCP context delivery is sub-50ms — not a bottleneck in TAT compliance.
Policy Coverage Completeness — ≥90% of criteria represented	CHAI-PA-USE-002	SCS structured policy bundles provide the criteria inventory. Policy diffs between versions surface missing or changed criteria before production deployment.
Scope Boundary Enforcement — ≥95% correct rejection of out-of-scope queries	CHAI-PA-USE-003	SCP intent validation enforces scope boundaries at runtime. Agents are registered with explicit allowed intents; requests outside that scope are blocked, not just logged. SCS usefulness SCD encodes the in-scope/out-of-scope boundary explicitly.
Policy Update Resilience — monitor performance after policy updates	CHAI-PA-USE-005	SCS semantic versioning: every policy update produces a new bundle version. SCP delivers the specific version to agents. Test against staging before production deployment is built into the versioned bundle workflow.
Appeal Overturn Rate Monitoring	CHAI-PA-USE-006	SCP audit trails provide the pre/post AI implementation comparison data the standard requires.

The CHAI Governance Note¶

The CHAI T&E Framework includes an explicit governance note that defines the boundary of AI authority in prior authorization:

"The AI model does not render actual approvals or denials. It informs organizational routing logic. The agent's output is a criteria matching assessment, not a coverage determination. Final decisions remain with authorized human reviewers."

SCS encodes this boundary as a structured governance constraint delivered to the agent at runtime. SCP enforces it: the agent's allowed intents are defined and registered; exceeding them triggers routing to human review. Governance is not aspirational policy — it is the agent's operating context.

Loading CHAI in One Command¶

For organizations using the scs-team Claude Code plugin:

/scs-team:use chai     # General CHAI Blueprint (accountability + transparency)

For prior authorization deployments, the full CHAI T&E Framework bundle is available as SCS SCDs covering all four principles with benchmarks, implementation guidance, and evidence requirements.

CHAI requirements become the structured context delivered to your prior authorization agents at runtime via SCP — not referenced in documentation, but active in every decision the agent makes.

Summary¶

Capability	SCS	SCP
CHAI requirements encoded as versioned structured context	✓
CHAI delivered to agents at runtime		✓
Context version auditability (CHAI-PA-TRANS-003)	✓ (versioned bundles)	✓ (audit log)
Scope boundary enforcement (CHAI-PA-USE-003)	✓ (intent definitions)	✓ (runtime enforcement)
Policy update resilience (CHAI-PA-USE-005)	✓ (semantic versioning)	✓ (version-locked delivery)
Hallucination mitigation (RISK-003)	✓ (structured policy documents)	✓ (grounded context delivery)
Demographic audit data (CHAI-PA-FAIR-001 through -004)		✓ (audit trails)