Skip to content

Document-intake quarantine policy

openmed.interop.document_quarantine provides a local, deterministic intake gate for attachments that must be classified before a parser sees them. It does not make a clinical decision, certify compliance, or contact a remote validator.

Classify an attachment

Pass the bytes, the MIME value supplied by the transport boundary, and the filename used for the extension check:

from openmed.interop.document_quarantine import classify_document

result = classify_document(
    b"%PDF-1.7\nsynthetic fixture",
    declared_mime="application/pdf",
    filename="synthetic-report.pdf",
)

safe_report = result.to_dict()

The default policy accepts common local document, image, text, JSON, XML, and archive types up to 25,000,000 bytes. A top-level archive has depth one; one archive layer is allowed, while nested archives beyond the configured limit are rejected. ZIP and tar members are checked for path traversal, symlinks, entry count, declared uncompressed size, and recursively inspectable nested archives.

Use DocumentQuarantinePolicy to change the limits or allowlists for a local deployment:

from openmed.interop.document_quarantine import (
    DocumentQuarantinePolicy,
    classify_document,
)

policy = DocumentQuarantinePolicy(
    max_size_bytes=10_000_000,
    max_archive_depth=1,
)
result = classify_document(
    payload,
    declared_mime="application/pdf",
    filename="synthetic-report.pdf",
    policy=policy,
)

Dispositions and reason codes

An attachment is accepted only when its declared MIME, extension, and local content sniff agree, the MIME and extension are allowlisted, and archive limits pass. Missing or conflicting signals are quarantined for an explicit local review. Empty or oversized payloads, unsupported types, malformed archives, unsafe archive members, and archive-depth violations are rejected.

QuarantineResult exposes only:

  • disposition: accepted, quarantined, or rejected;
  • reason_codes: stable machine-readable values such as declared_mime_sniff_mismatch or archive_depth_exceeded; and
  • sha256: a digest of the supplied bytes.

to_dict() is safe to place in a log, report, or audit record. It does not include the filename, declared MIME, sniffed type, archive member names, payload bytes, or exception details. Callers should discard or separately protect the original payload after routing it according to the disposition.

The classifier uses only Python's local standard-library inspection helpers. It performs no mandatory network call and has no telemetry or remote fallback.