Verify Med CodesClinical Truth Layer
Back to Impact

PHI containment is architectural.

Default is block. Allow requires all 3 layers.

We don't scrub PHI. We extract clinical facts as codes — leaving no room for PHI to exist.

Every outbound payload passes a three-layer egress gate: field allowlist, value pattern scan, and structural integrity check. No field leaves the system without clearing all three. This is not a filter — it's a fail-closed architecture.

3Egress gate layers
BlockDefault posture
0 fieldsPHI in output
Fail-closedArchitecture
Three-Layer Egress Gate
Layer 1: Field-Level Allowlist

Only pre-approved field names pass through. Unknown fields are stripped.

All fields in allowlist
Layer 2: Value-Level Pattern Scan

Regex + NER scan on all output values. Catches names, DOBs, MRNs, SSNs, addresses, phone numbers.

No PHI patterns detected in values
Layer 3: Payload-Level Hash Check

Final payload hash compared against known-safe schema. Any structural deviation triggers block.

Payload hash matches known-safe schema
Payload Inspector — Select a scenario
Egress-Safe PayloadPASSED
{
"fact_id": "f_demo_s001",
"icd10_code": "I50.22",
"icd10_display": "Chronic systolic heart failure",
"snomed_code": "441481004",
"hcc_category": "226",
"raf_weight": "0.36",
"assertion": "present",
"section": "PMH",
"phrase": "chronic systolic heart failure",
"confidence": "0.846"
}
Gate Audit Log
Layer 1: Field-Level Allowlist
All fields in allowlist
Layer 2: Value-Level Pattern Scan
No PHI patterns detected in values
Layer 3: Payload-Level Hash Check
Payload hash matches known-safe schema
Cleared — All 3 Layers

This payload contains only structured clinical codes, standardized values, and pre-approved field names. No PHI. No free text. No identifiers. Safe for downstream AI, analytics, or external transmission.

Stays Local
Clinical note text
Patient names
DOB, SSN, MRN
Addresses, phones
Provider names
Original sentences
Egress-Safe
ICD-10 / SNOMED codes
HCC categories + RAF
Lab values (no dates)
Assertion status
Decision trace rules
Confidence scores

Without architectural containment, AI cannot safely reason over clinical state. We have structured clinical data with no PHI. That means any AI — GPT-4, Claude, open-source — can safely process our output without HIPAA exposure. This is not a feature. It's the reason the rest of the system is possible.

Deployment
Customer VPC
Dedicated tenant
On-premise available
Certifications
SOC 2 Type II (planned)
HITRUST CSF (planned)
TLS 1.3 encryption
Data Ownership
All data customer-owned
No model training on your data
Full deletion on request

Synthetic payloads. Fail-closed architecture. Every output auditable. Zero PHI in transit.