Reviewed Node PDF library docs for table handling and printed-versus-index page numbering to see if the existing runtime could meet the requirements alone.
- What worked
- Clarified that text-stream extraction leaves table structure and OCR gaps the task needed to close.