Final privacy scan for clinical review packets¶

openmed.clinical.review_packet_privacy applies a configured leakage detector to the exact rendered packet immediately before export or local persistence. The detector is dependency-injected, so the gate performs no model download or network call.

from openmed.clinical.review_packet_privacy import persist_review_packet

report = persist_review_packet(
    "review-packet.json",
    rendered_packet,
    configured_local_detector,
)

Detector results can expose findings, entities, or pii_entities. Each item must provide character start and end offsets plus an entity class using entity_class, canonical_label, label, or entity_type. A detector can mark a finding critical with critical=True or a severity/risk_level of critical, high, or blocking. Otherwise OpenMed's direct-identifier and high-risk label set determines whether the finding blocks output; callers can provide an explicit critical_entity_classes set.

Reports contain only schema metadata, counts, content hashes, and finding entity classes, offsets, and hashes. Detector-provided text is ignored. On a critical finding, ReviewPacketPrivacyBlocked is raised before the exporter callback is called or the destination path is opened. Detector failures and malformed offsets also fail closed with value-free errors.

This gate is a final local safety check for a human-review artifact. It is not a compliance certification or an autonomous clinical decision mechanism.