Document Source Text Exporter

How to use Document Source Text Exporter

Retrieve stored source wording from a small local document without uploading or rendering it. A single workflow handles a canonical DOCX main story or selected EPUB 3 XHTML spine items, keeping source structure and omissions visible in JSON evidence and offering a plain-text copy for further editing.

1. Choose a small supported local DOCX or EPUB, or load the generated original DOCX example.
2. Choose JSON evidence or plain text and the explicitly limited revision/spine source policy.
3. Run the local export, review selected and omitted parts in JSON and verify wording in a trusted reader before reusing or sharing it.

Good to know

• A local 4 MiB ZIP, 200 safe entries, stored/DEFLATE and no encryption, ZIP64, symlinks or legacy non-UTF8 names. ZIP directory/local headers and expansion limits are preflighted; selected extraction checks actual size and CRC. Unselected payloads are not inspected for integrity or malware.
• Selected metadata/content: 200,000 bytes per entry and 1 MiB aggregate before decompression, at most 23 XML documents/40,000 total elements, the strict parser’s 20,000 per-document elements/60 XML levels and a checked 5 second processing budget. UTF-8 XML only; DTD/entity declarations and malformed XML are refused. Content is limited to 100,000 token characters, 5,000 tokens, 2,000 blocks and 20 block levels, with separate export allocation limits.
• DOCX: canonical internal word/document.xml relationship and macro-free main content type required. Export main-story paragraphs and direct source table rows/cells, including nested tables. Markup-compatibility AlternateContent, row/cell wrappers and unexpected nested paragraph content are refused. Other stories, textboxes, drawings, embedded data, field instructions, numbering, style inheritance and visual table merges are not reconstructed. Direct hidden markup and stored revisions are source observations, not a visibility decision.
• EPUB: one EPUB 3 package root, 1–20 selected local XHTML spine documents, valid explicit linear flags and no repeated selected paths. Metadata xml:base, selected foreign-content fallbacks, remote/unsafe paths and encryption/font obfuscation declarations are refused. Source whitespace and generated block/cell separators remain explicit; CSS, scripts, image alt attributes, SVG/MathML and embedded resources are not converted or loaded.
• Source output can contain names, deleted wording, hidden text and stale field results. The deletion filter is not tracked-change acceptance, document sanitization or anonymization. JSON carries structure and omission evidence; TEXT provides the declared source mapping only. Use a trusted reader for final display and a dedicated privacy review before sharing.
• Processing is local. Output is limited to 1,000,000 text characters unless a stricter file limit is stated.