Document Review Comment Exporter

How to use Document Review Comment Exporter

Read actual review comments from a small DOCX package without opening Word or rendering the document. The result retains literal author/date fields, comment text and bounded source-range excerpts, while separating unanchored comments, point references and ambiguous marker evidence. It does not infer modern thread resolution or final visible document content.

1. Choose a supported small DOCX file or load the generated original comment fixture.
2. Run the canonical main-story/classic-comment source inspection.
3. Review author/date privacy and anchor limitations before downloading the separate JSON or formula-protected CSV.

Good to know

• One DOCX ZIP at most 4 MiB, using the existing strict archive directory/expansion/traversal/duplicate-name preflight before selected extraction. Only the canonical macro-free main content type, internal root word/document.xml and internal canonical word/comments.xml relationship are supported.
• Selected parts must be UTF-8 XML, at most 200,000 bytes per part/1 MiB aggregate, with existing DTD/entity rejection and XML node/depth bounds. Selected payloads receive CRC checks. Processing checks a 5-second cooperative budget; native XML parsing cannot be forcibly interrupted.
• At most 200 classic comments, 5,000 source text characters per comment/200,000 together, 1,000 anchor/reference markers and 20,000 projected source events. IDs use canonical nonnegative decimal text 0–999,999. Duplicate selected comment IDs and nested comment anchors are refused.
• Source text includes XML-normalized w:t/delText, tabs/breaks and paragraph newlines. Hidden/deleted source can appear. Range excerpts stop at a safe Unicode boundary within 2,000 UTF-16 characters and carry a truncation flag. Offsets describe this projection, not XML bytes, pages or rendered Word positions.
• Only the main story and classic comments are selected. Threaded replies/resolved state, headers/footers, notes, textboxes, drawings, external/embedded content, field instructions, style visibility and accepted/rejected changes are not reconstructed. This is not full Office rendering, anonymity, sanitization or a malware scanner; raw authors/dates/comments may be private.
• Processing is local. Output is limited to 1,000,000 text characters unless a stricter file limit is stated.