AKPS — an open standard for coverage policy criteria, in a form that AI agents can actually consume.

Coverage policies live in PDFs. AI agents need structured inputs. AKPS is the bridge.

Every payer maintains coverage policies — what services are covered, under what conditions, for which populations. Today, those policies live in human-readable documents: PDFs, Word files, intranet pages. Engineering teams ingest them by writing code that encodes each rule. When the policy changes, the code changes. When a new payer onboards, the code changes again. The result is a policy substrate that engineers maintain rather than clinicians govern.

AKPS inverts the relationship. Coverage criteria are authored by clinical policy teams in structured markdown — a format engineers can parse and reviewers can read. When a criterion changes, the AKPS pack changes. When a new payer onboards, a new pack is added. The agent platform reads the pack at request time; no code changes required, no engineering bottleneck, no version drift between the policy document and the rule that executes against it.

An AKPS pack is structured markdown plus YAML front matter.

Every pack is a single markdown file with YAML front matter declaring identity, scope, version, and authority. The body contains criteria, each conformant to a schema that the reference parser validates.

Pack types

Example: a minimal Criterion Pack

---
id: cms-ncd-220-3-mri-spine-2026
type: criterion-pack
version: 1.0.0
authority: CMS NCD 220.3
effective_date: 2026-01-01
jurisdiction: [medicare-ffs]
service_category: advanced-imaging
authored_by: [email protected]
reviewed_by: [email protected]
---

# MRI of the Lumbar Spine — Coverage Criteria

## Criterion C-1: Clinical Indication

**Requirement:** Documented neurological deficit OR
intractable pain unresponsive to at least 6 weeks
of conservative management.

**Evidence sources accepted:**
- Physician progress note documenting deficit findings
- Pain management consultation documenting failed therapy
- PT/OT notes documenting conservative treatment trial

**Evidence binding rule:** The supporting passage must
be in a document dated within 90 days of the
authorization request.

Every criterion in the pack carries a stable identifier, an explicit requirement statement, the evidence sources accepted as proof, and the binding rule that defines what constitutes valid evidence. At request time, the AI agent maps each submitted document against each criterion independently — producing the per-criterion citation chain that anchors the audit trail.

Extension Pack mechanism

AKPS supports an Extension Pack mechanism for compliant additive customization. A base pack defines the canonical criteria; an Extension Pack adds plan-specific overrides, tighter SLAs, or jurisdiction-specific clarifications without modifying the base. The compliance envelope ensures that extensions cannot weaken validation rules, bypass safety gates, or remove required fields — only add to or refine what the base pack already specifies. This makes pack reuse safe across multi-payer deployments.

AKPS is open. The architecture that consumes it is patented.

The Agentic Knowledge Pack Specification itself is open. The architecture that ingests AKPS packs, routes them across deployments, and propagates lessons learned back into pack evolution is protected by Genzeon's patent portfolio.

This bifurcation is intentional. AKPS adoption produces a network of interoperable coverage policy artifacts; the more vendors and payers adopt the format, the more reusable the resulting library becomes. The patented architecture provides commercial differentiation on the consumption layer — how packs are routed, how agents reason over them, how feedback flows back into pack evolution — without locking the underlying specification behind proprietary licenses.