Nirog Core OCR Ops

Nirog Core OCR Ops is an unauthenticated, private Python subsystem for operating a bounded prescription-evidence pipeline. It is not a patient-facing application and it is not a clinical authority.

The subsystem creates synthetic demo jobs, produces provisional review packets, records human review decisions, stores model-governance approvals, and can submit an identifier-only receipt to Nirog Core after a confirmed decision. Nirog Core remains the source of truth for clinical data and workflow authority.

Demo-only runtime

Every current result is explicitly marked resultSource: demo and demoFixtureId: demo.prescription-v1. No model adapter is installed or runnable.

What belongs here

Capability	Boundary
Name normalization and candidate ranking	Provisional and deterministic; safe abstention is required.
Human review workflow	Requires explicit actor, timestamp, decision, and reason.
Model governance	Records an approval boundary but never activates a missing model adapter.
Core handoff	Accepts only confirmed outcomes and sends no document text, candidate list, or diagnosis hypothesis.

What does not belong here

The service must not diagnose, prescribe, recommend treatment, mutate Core medication or regimen records, or expose the Core worker identity. The absence of in-application authentication is intentional: this is a private network subsystem, not a public user product.

Explore

Start with Architecture to understand the clean module boundaries, then use HTTP API to integrate operational tooling or inspect the service’s generated OpenAPI view at /api/docs.