Position EvidenceGate as Evidence Intelligence infrastructure
OCR, RAG, and agents remain implementation capabilities instead of the product category.
HS
Decisions and risks
Record ownership, trade-offs, mitigation, and review dates.
EvidenceGate strategy baseline v2.0
Architecture decision records
Why we chose this path
Use planned cashless pre-authorization as the beachhead
The first measurable value is preventing packet defects and avoidable queries before payer review.
Make the evidence graph the system of record for findings
Every fact, comparison, rule input, and human action preserves page-level lineage.
Keep policy evaluation outside models
RAG locates approved clauses while human-authored tested code evaluates policy.
Prohibit autonomous case disposition
EvidenceGate can prepare and validate a packet, but a human remains responsible for every material decision.
Risk register
What can break the plan
CriticalHigh
A lawful representative packet set is unavailable
Mitigation
Make the data agreement and sample audit the first commercial gate; stop model work until access is real.
HighMedium
The hospital workflow baseline cannot be measured
Mitigation
Agree query-cycle and handling-time definitions with the partner before shadow mode.
CriticalHigh
Insurer rule onboarding becomes bespoke consulting
Mitigation
Use a common typed bundle, reusable primitives, import validation, golden tests, and separately priced onboarding.
CriticalMedium
Model output is mistaken for policy or a final decision
Mitigation
Enforce service and tool boundaries; label observed facts, retrieved context, rule results, and human actions separately.
HighMedium
Extraction accuracy hides critical-field failures
Mitigation
Report by field and confidence band; use abstention and held-out critical-field evaluation instead of blended accuracy.
HighMedium
Hospital operators do not change their workflow
Mitigation
Begin in read-only shadow mode, integrate with existing intake, and use weekly operator feedback.