What the extraction preserves
Each table retains merged cell spans, column groupings, and nested headers as structured Markdown. Cell content that crosses row or column boundaries is represented with repeated markers so downstream readers can reconstruct the original layout. Formatting such as bold, italic, and inline code within cells survives the conversion.
How it works across formats
PDF tables are parsed using positional analysis to detect cell boundaries even when lines are absent. DOCX tables use the underlying XML grid structure for precise span detection. Image-based tables go through layout-aware OCR that identifies grid lines and cell content separately. All three paths produce consistent Markdown pipe-table output. Feedback: sapsap@qq.com.