Auditable method

What the package audit checks—and what it cannot know.

Relationship graph

The parser interprets namespaced image references and image-type relationships in supported Word XML parts, including footnotes, endnotes and comments as well as the main document, headers and footers. It resolves package target paths and checks entry membership. XML comments are not active drawings. Invalid, ambiguous or unsupported coverage is reported instead of an all-clear, and can suppress an unreferenced-entry conclusion.

A present entry is not proof of nonempty or decodable image data. Missing relationships and NULL targets have an unknown intended target; only a known internal target with no matching entry is reported absent. An unrelated surviving entry is never assigned to a broken rId.

No macro, object, hyperlink or external image is executed or fetched.

Microsoft LearnDocuments the image-part error caused by a relationship target set to NULL and a manual package inspection workflow.

Read Microsoft’s workflow

OOXML package boundaryImages are related package parts. A relationship can be internal or external; byte presence is separate from Word display behavior.

Current-supply reviewRaw DOCX inspectors and broad repair checks exist. This prototype tests whether an image-specific reference/target/recoverability ledger reduces the remaining work.

Known limits

It cannot diagnose Word display options, GPU rendering, permissions, cloud-version history or files not inside the selected package.
It does not validate every OOXML extension or vendor-specific drawing model.
It cannot reconstruct image bytes that are absent, infer the original layout from an orphan alone or guarantee that a manual repair is safe.
Input is capped at 50 MB, 2,000 entries and 64 MB of declared expansion. XML is streamed with actual-output caps of 2 MB per part and 8 MB total; its size and CRC are checked before parsing. Binary image payloads are not expanded or decoded.
Encrypted, multi-volume, ZIP64, unsafe/duplicate-path, malformed or over-budget archives are rejected. This reader supports UTF-8 XML and UTF-8 or ASCII entry names, not every historical ZIP/XML encoding.