HS

Product architecture

Deterministic controls, grounded models, and human authority.

EvidenceGate strategy baseline v2.0
Operating model

Prepare evidence. Apply governed rules.
Keep the human in control.

The platform is an overlay across hospital, TPA, insurer, and NHCX workflows. It does not replace systems of record and cannot approve, reject, diagnose, or settle.

!
Hard boundary

No model or agent has a tool that can publish policy, disposition a case, or communicate an adverse decision.

Intake
Hospital and partner channelsPortal, API, email, DMS, NHCX
Secure intake boundaryIdentity, scan, hash, immutable storage
Evidence engine
Bounded model workersClassify and extract facts with confidence
Evidence graph builderNormalize and link page, region, snippet, and hash
Policy and quality
Published policy as codeClause lineage, golden tests, activation, rollback
Deterministic quality gateCompleteness, contradictions, timelines, rule results
Review
Review packetFindings, evidence, unknowns, rule version
Human actionVerify, correct, override, request evidence
01

Evidence graph

Every material fact opens to an immutable document version, page, region, and snippet.

02

Deterministic truth boundary

Required documents, mandatory fields, comparisons, and insurer rules execute in tested code.

03

Retrieval is not policy

RAG finds approved clauses. Human-authored, tested code owns every policy evaluation.

04

Replayable review

Document, schema, model, prompt, rule, and knowledge versions are pinned to each run.

The startup thesis

What others usually stop before

Extracted JSONEvidence-linked facts

Policy chatbotClause-linked rules

Faster queriesQuery prevention

Black-box automationHuman-control architecture