OCR Text Verification

Problem Being Solved

OCR text often contains clause skips, bottom-of-page truncation, missing sub-clauses, 1 misread as l, incomplete formulas, and mixed version content. Manual page-by-page review is slow and hard to keep consistent. This skill targets OCR output after PDF-to-text conversion and provides an executable check-and-repair workflow for structured documents such as quantity lists, standard clauses, appendices, tables, and formulas.

How It Works

• Four-stage check: extract X.Y.Z clause numbers with regex and verify continuity → compare PDF pages against OCR text page by page for truncation, misalignment, cross-references, and version differences → run script-based fallback checks → sample-validate formulas, amounts, core definitions, and repaired areas.
• Six-dimension review: beyond clause numbering, it covers completeness, page mapping, character accuracy, cross-reference consistency, and version consistency, reducing the blind spots of manual checks that only look for obvious omissions.
• Repair loop: prioritize issues from P0 to P3, restore missing or truncated content from PDF pages, then rerun checks and sample validation to avoid introducing new problems.

Boundaries And Notes

It fits documents with section numbers, clause numbers, table indexes, and appendix references. For blurry scans, complex layouts, or image-only non-clause content, base OCR quality must still be ensured first; formulas, cross-page tables, dual page-number systems, and version-difference regions remain high-risk and require manual review against the original pages.