LibreOffice-class reconstruction. Layout will drift. Inspect every section.
If you need a round-trip that looks like the PDF you printed, stop. This path rebuilds an office document from a PDF. That is a guess with structure. It is useful. It is not Acrobat.
What the engine actually does
PDF→DOCX, PDF→XLSX, and PDF→PPTX are reconstruction. LibreOffice-class conversion. Tables shift. Headers collapse. Floating figures wander. Multi-column magazine layouts become a stack of boxes.
Word→PDF, Excel→PDF, PPT→PDF go the other way through LibreOffice headless. Fonts substitute. Tracked changes, macros, OLE, animations, and speaker notes are not a fidelity promise. Print a page and compare.
These slugs can sit in the catalog while the worker binary is missing. Queue may return ENGINE_UNAVAILABLE. That is an honest error, not a “try a different browser.”
Never promised
Not Adobe Acrobat export. Not InDesign. Not a print RIP. Not a legal-page clone. Forms and signatures do not survive as live fields. Treat output as a draft you restyle.
Scans are pictures of text. Run OCR PDF first if you need a text layer. Even then you get searchable/selectable text, not a designed document. Handwriting is out.
When the draft is still worth it
Born-digital PDFs with simple headings: you get something you can edit in Word or LibreOffice instead of fighting a frozen layout. Then you fix the wrecked bits. That is the job.
Tables: PDF→XLSX is heuristic. Ruled grids sometimes land in cells. Multi-column invoices often do not. Merged headers lie. Formulas are never reverse-engineered — values only.
Decks: PDF→PPTX is one page ≈ one slide. You do not get a rebuilt master. Use it to get pages onto a stage, then restyle.
A sane workflow
Keep the PDF. Export once. Open the office file. Inspect every section, every table, every figure. If it is garbage, raster pages to JPEG/PNG for a reading copy instead of pretending the DOCX is the source.
Going back (DOCX→PDF) is a second reconstruction. Font substitution will move titles. Do not loop PDF→DOCX→PDF to “clean” a file. You will only stack errors.
Steps
- Check the slug status. If the tool is stubbed or the worker is down, you will get ENGINE_UNAVAILABLE. Use a live raster instead if you only need stills.
- Prefer born-digital input. Scans need OCR first. OCR does not restore layout.
- Export once. Queue PDF→DOCX, XLSX, or PPTX. Download. Keep the PDF.
- Inspect in an office app. Walk headings, tables, figures, and page breaks. Restyle. Do not ship the first download unread.
FAQ
Is this Adobe Acrobat export?
No. Do not expect Acrobat fidelity. Open the file and inspect every section.
Why is the tool in the catalog if it fails?
The slug is real in the API catalog. Missing binaries return ENGINE_UNAVAILABLE. Coming-soon AI is a different story: dropzone off, noindex.
Can I get a pixel-perfect Word file?
Not on this product. If you need pixels, raster. If you need editable text, accept drift.