Data Mask

Problem It Solves

Replaces real amounts, order counts, conversion rates, and other numeric values in report.docx, Q2_data.xlsx, screenshots, or pasted text with masked values. This avoids manual editing before sharing documents or sending data to LLMs while keeping files in a usable form.

How It Works

• Format-aware entry points: mask.py for plain text, dataframe_mask.py for Excel/CSV, and file_mask.py as the unified entry for Word, PPT, PDF, HTML, Markdown, and images.
• Numeric detection and filtering: recognizes values with units such as 元, 万, 亿, and %, skips IDs, sequence numbers, and years, and can preserve phone numbers with --skip-phone.
• Strategy-based masking: hash creates deterministic alternate numbers and saves a mapping table; placeholder replaces values with [A], [B]; blur and range preserve magnitude or ranges for external use but are not reversible.
• Layout preservation: Word/PPT keeps fonts, tables, headers/footers, and grouped shapes where possible; Excel/CSV preserves column structure; PDF/HTML/images often output as text or simplified PDF.
• Restore and batch processing: -m saves a mapping table, --unmask restores using the same table, and a shared mapping table can keep multiple files consistent.

Boundaries and Caveats

• Masks numbers only; it does not detect person names, places, company names, or other semantic PII, so manual review is still required before external sharing.
• Legacy .doc/.xls, scanned PDFs, and JS-rendered HTML need conversion, OCR, or browser extraction first.
• The mapping table acts like a restore key; keep it separate from masked files and out of public repositories.
• Masking is not encryption; for compliance-heavy workflows, combine placeholder with encryption, approval, and audit controls.