Ranking Scanned PDFs While Preserving OCR Uncertainty

Combine lexical and semantic retrieval, then use region evidence and page metadata to rank results without concealing weak extraction. A scanned PDF presents two separate search problems: finding the