Document-to-Markdown Shoot-Off
Real-world parsers tested on messy spreadsheets, decks with speaker notes, complex tables, and image scans.
mdforge
90.6%
87 / 96 pts
pandoc
80.5%
29 / 36 pts · 3 formats
AnythingMD
45.2%
38 / 84 pts · 3 failures
GetMarkdown
19.0%
16 / 84 pts · 5 failures
| Test Case | Format | mdforge | AnythingMD | GetMarkdown | pandoc |
|---|---|---|---|---|---|
| S1Messy Finance Hidden confidential sheet, merged cells, currency formatting | XLSX | 11/12PASS Clean GFM table, hidden sheet filtered | 0/12FAIL FATAL: Leaked hidden sheet | 0/12FAIL FATAL: Leaked hidden sheet | —— Unsupported format |
| S2Wide Inventory 22 columns x 41 rows, SKU 01000 leading zero, €/$ symbols | XLSX | 12/12PASS All 22 cols kept, currency symbols preserved | 10/12WARN Stripped currency symbols ($323 → 323) | 10/12WARN Stripped currency symbols | —— Unsupported format |
| S3Mixed Report Rowspan/colspan tables, inline formulas, nested lists | DOCX | 9/12WARN Clean HTML table for complex merges | 11/12PASS Rowspan expanded, image to CDN | 0/12FAIL Browser crashed, zero output | 10/12PASS HTML table fallback, clean text |
| S4Dense Deck 8 slides, two-column flow, 5 slides with speaker notes | PPTX | 12/12PASS 100% speaker notes extracted to blockquotes | 6/12FAIL Discarded all 5 speaker notes | 0/12FAIL Corrupted/truncated ZIP download | —— Unsupported format |
| S5Academic Two-Col 2-column layout, cross-column table, $LaTeX$ math equations | 10/12PASS Correct reading order, table intact | 11/12PASS Correct reading order, GFM table | 5/12FAIL Table rows broke into ### headings | —— Unsupported format | |
| S6Scanned Memo 150 DPI image scan, no text layer, small printed table | 9/12PASS Vision VLM OCR extracted text & table | 0/12FAIL Silent failure, reset to dropzone | 0/12FAIL Hung indefinitely on upload | —— Unsupported format | |
| S7Dirty Blog Ad noise, tracking scripts, nested tables, cookie banner | HTML | 12/12PASS Main content extracted, noise stripped | 0/12FAIL Conversion failed to parse file | 1/12FAIL Dumped raw HTML & scripts | 11/12PASS Good extraction, preserved nested tables |
| S8Mini Book 3 chapters, TOC, 4 paired footnotes [^1], blockquotes | EPUB | 12/12PASS All 4 footnote pairs matched cleanly | —— Unsupported format | —— Unsupported format | 8/12WARN Valid markdown with raw HTML tags |
📐 Scoring Methodology (6 Dimensions · Max 12 Pts)
- • Structure: Proper heading levels, lists, and reading order
- • Tables: Clean GFM pipes, preserved columns, escaped characters
- • Completeness: No dropped numbers, preserved leading zeros
- • Noise Suppression: Stripped ad scripts, tracking code, and hidden sheets
- • Compliance: Valid Markdown renderable by Obsidian / GitHub
- • Media & Notes: Preserved footnotes, diagrams, and speaker notes
⚠️ Fatal Rule: Leaking confidential sheets or crashing = 0 pts.
⚖️ Honest Limitations & Quotas
Because GitHub-Flavored Markdown has no native rowspan support, complex Word tables fall back to clean HTML <table> tags to prevent misaligned columns.
PDF & DOCX run on high-accuracy cloud infrastructure, with five lifetime conversions per verified email account. Browser-local tools (Excel, PPT, HTML, EPUB) remain unlimited.