Reference Architecture · Rev 2026.05

JIL on Snowflake.
Six product lines.
One governed runtime.

Settlements, Retroactive, Pre-Clearance, Asset Intel, Wallet Intel, and Utilization Management all execute as containerized services inside the customer's Snowflake account. No data movement. No copies. Verdicts produced where the data already lives.

Runtime Snowpark Container Services Services 273 Checks 175 / 16 categories Read the full TDD → Anchor CourtChain™ L1
Source 01
MCO Claims & Prior Auth Feeds
Source 02
Federal & State Program Data
Source 03
Provider, Member, Eligibility
Source 04
Wallet, Asset, KYC Streams
Snowflake AI Data Cloud
Customer Account · Customer VPC · Customer Region
K8s
Snowpark Container Services
Managed Kubernetes inside Snowflake
SVC
Service Functions & Streamlit
SQL-callable verdicts, in-account UI
IAM
Native RBAC & Network Policy
No keys leave the account
PRODUCT · 01
Pre-Settlement
Block fraud and improper payments before they settle. Sub-second verdicts on inbound claims and wires.
Real-time 175 checks SLA < 800ms
PRODUCT · 02
Retroactive
Forensic sweeps on settled claims. Proves overpayment, ineligibility, and bad actor exposure with court-ready evidence.
Batch CREB™ output FRE 902(14)
PRODUCT · 03
Pre-Clearance
Attestyx grant and disbursement clearance. Approves grantees, vendors, and counterparties before funds move.
Foundations Grants Federal
PRODUCT · 04
Asset Intel
Counterparty and asset risk graph. Shell companies, beneficial ownership, sanctions, and exposure mapping.
Graph OFAC / FinCEN UBO
PRODUCT · 05
Wallet Intel
On-chain plus off-chain wallet attribution. Flags laundering patterns, mule networks, and pre-settlement risk on payouts.
On-chain Attribution Patterns
NEW
PRODUCT · 06
Utilization Management
Proof layer beneath the plan's medical team. Validates medical necessity evidence, flags pattern outliers, never decides care.
Proof only Pattern outliers Audit trail
Verdict Engine
175 checks · 16 categories
Bad Actor Registry
Cross-jurisdiction graph
CREB™ Generator
Court Ready Evidence
Sealed Escrow / PoCS
Proof of Compliant Settlement
Native Data Sharing
Zero-copy customer feeds
Elastic Compute
Per-workload warehouses
Governed Storage
Row, column, masking
Horizon Catalog
Lineage, audit, policy
Anchor · 01
CourtChain™ L1
Every verdict, every CREB™, every settlement event hashed and notarized on JIL's purpose-built Layer 1. Ed25519 plus Dilithium-III hybrid signatures, ML-DSA-65 PQC.
Anchor · 02
Five Jurisdictional Vaults
Sovereign evidence replication across regulated trust zones.
US
United States
CH
Switzerland
UAE
Emirates
SG
Singapore
BR
Brazil

The whole system runs inside the customer's Snowflake account.

Every box rendered inside the framed cloud region is a workload that executes inside the customer's own Snowflake tenant. Not a SaaS endpoint they pipe data to. Not a copy that lives in JIL infrastructure. The customer's claims, prior auth requests, eligibility files, wallets, and counterparty data never move. JIL's services are deployed to the data, the data is not deployed to JIL.

That single architectural decision is what unlocks payer, federal, and sovereign procurement. No data exfiltration risk. No third-party vendor data residency review. No new HIPAA pathway. The customer's existing Snowflake governance posture covers JIL by definition because JIL is running under it.

Snowpark Container Services is the runtime.

The orchestration layer at the top of the cloud is Snowflake's managed Kubernetes, branded Snowpark Container Services. JIL ships its 273 services as containers. Snowflake schedules them, allocates compute through warehouses, and exposes results through SQL-callable Service Functions and Streamlit apps that customer analysts and counsel can use without leaving Snowsight. RBAC, network policy, and key management are all native. JIL inherits them.

Six product lines. One verdict spine.

The six product cards in the middle are not separate platforms. They are differentiated entry points that all converge on the same shared engines below them: the Verdict Engine running 175 checks across 16 categories, the Bad Actor Registry as the cross-product knowledge graph, the CREB™ generator producing FRE 902(14) court ready evidence packages, and the Sealed Escrow PoCS layer. A claim seen by Pre-Settlement and a wallet seen by Wallet Intel hit the same registry. A grantee cleared by Attestyx and a provider scrutinized by Utilization Management share the same evidence schema.

"Six surfaces. One brain. One sealed evidence record. The customer chooses which surfaces they switch on."

Utilization Management is the new sixth surface.

Utilization Management is highlighted in gold because it is the newest addition and because it carries the most carefully scoped legal posture. It is a proof layer beneath the plan's medical team, never above it. It does not approve or deny care. It validates that the medical necessity evidence on file is internally consistent, that the documentation pattern matches peer norms, and that any flagged outliers are surfaced for the human reviewer with a sealed audit trail. That deliberate scoping is what avoids the NaviHealth pattern of litigation. The plan's clinicians make the call. JIL proves what the plan's clinicians had in front of them when they made it.

The anchor layer is the only thing outside Snowflake.

CourtChain™ and the five jurisdictional vaults sit beneath the cloud because they have to live somewhere with sovereign neutrality. Hashes leave Snowflake. Data does not. Every verdict, every CREB™, every PoCS receipt is hashed inside the customer's account, written to CourtChain™, and replicated across the US, Switzerland, UAE, Singapore, and Brazil vaults. The customer can prove a verdict was rendered at a specific time on specific evidence without ever exposing the evidence itself. That is the FRE 902(14) admissibility bridge and it is the reason an MCO general counsel, a state Medicaid director, and a federal IG can all use the same artifact.

Why Snowflake instead of building it ourselves.

Three reasons, in order of weight to the buyer. First, the largest payers, the largest banks, and a growing share of federal civilian agencies have already standardized on Snowflake. Meeting them inside their existing data plane removes the longest line item in any procurement cycle. Second, Snowpark Container Services gives JIL a managed Kubernetes substrate with native governance, native sharing, and native compute elasticity that would take a multi-year platform investment to replicate. Third, the commercial story to investors is cleaner. JIL is a detection and proof IP company riding on top of the most defensible data infrastructure layer in the enterprise. JIL does not have to be an infrastructure company to win an infrastructure-grade procurement.

What the customer turns on, in plain language.

An MCO could activate Pre-Settlement and Utilization Management on day one and add Retroactive in quarter two. A foundation activates Pre-Clearance only. A federal program office activates Pre-Settlement and Asset Intel together. Each of those decisions is a SKU. None of them require a separate stack, a separate integration, or a separate audit. They are all the same engine, lit up against different pieces of the customer's existing Snowflake account.