Open engineering evidence

Judge PDF-to-Word quality from the files, not a slogan

Download one rights-cleared PDF and the three DOCX files PDFCore 8.0 produced from it. Review the rendered pages, inspect the Word structure, and read exactly what our current test evidence does—and does not—prove.

Same input, three modesPublic sample downloadsRepeatable checks
Verified historical checkpoint

What passed in the public PDFCore 8.0.0 sample build

The figures below remain tied to the downloadable 8.0.0 sample outputs. The current 8.0.1 release suite has 47 tests; these historical figures are kept separate so the evidence is not silently relabeled.

42 / 428.0.0 release tests passed
210 / 210Five repeated 8.0.0 suite runs passed
0Open XML validation errors in all three sample DOCX files
3 modesEditable, precise and flowing outputs published below
Public evidence available

Structural checks go beyond “the file opened”

The test pipeline creates PDFs, calls the production conversion service, opens the resulting DOCX package, runs the Open XML validator and checks editable text, images, tables, headings, headers, footers, page geometry and page-range behavior. The editable and flowing samples contain one real Word table, 17 heading paragraphs, two header parts and three footer parts.

Important limit: this controlled corpus is mainly synthetic. It does not establish Top‑5 status or parity with Adobe, Foxit, Nitro, PDFelement or PDF‑XChange. A 30+ document, same-input blind comparison with independent review is still required for that claim.

Compare the output models

One PDF, three different editing goals

A fixed PDF does not contain all the paragraph, table and style information Word expects. PDFCore exposes three reconstruction strategies instead of claiming that one output can maximize both visual fidelity and free-flow editing.

First page of PDFCore editable-layout Word output
Balanced default

Editable layout

Reconstructs editable blocks, page geometry, Word-native tables, headings and repeated headers or footers while retaining non-text visual regions.

Download editable DOCX · 44,375 bytes

First page of PDFCore precise-layout Word output
Visual placement

Precise layout

Places editable text lines over cleaned page regions. It is useful when a form, flyer or diagram must stay near its original coordinates.

Download precise DOCX · 44,074 bytes

First page of PDFCore flowing-text Word output
Rewrite first

Flowing text

Prioritizes headings, paragraphs, reading order and tables without positioned shapes. It is the smallest, cleanest sample for substantial rewriting.

Download flowing DOCX · 6,801 bytes

The previews are Word-rendered pages from the downloadable files. Different Word versions, fonts and printer metrics can change pagination.

Download the evidence set

Reproduce the inspection

The corpus includes portrait and landscape pages, multi-column text, aligned data, graphics and repeated page-edge content. Start with the PDF, open each DOCX in a current desktop version of Word, turn on formatting marks, and inspect the table, headings, header/footer regions and page orientation.

What to inspect

Six practical checks

  1. Select and edit text in the body.
  2. Click the aligned data and confirm whether it is a real Word table.
  3. Use Word’s Navigation pane to inspect headings.
  4. Open the header and footer regions.
  5. Check the landscape section and column reading order.
  6. Save the DOCX back to PDF and compare the rendered result.

See the full scoring and limitations →

Current boundaries

What remains unproven

Real-document breadth

The published sample is a controlled engineering corpus. The next benchmark must cover at least 30 rights-cleared born-digital, scanned, mixed-script, table-heavy, rotated and malformed PDFs.

Competitor parity

We have not published a same-input blind score against installed commercial leaders. Until that exists, PDFCore describes the target and its evidence—not a Top‑5 conclusion.

Human edit cost

Validation proves structure and package correctness, but a useful benchmark must also time a standard edit script and record manual repairs required after conversion.

Try the same document locally

Download PDFCore 8.0, convert the published corpus with each mode, and compare your output to the files above. PDF processing stays on the Windows PC.