Unicode text cleaner

Cleanup is not normalization by force

Unicode has legitimate distinctions that look invisible. A production-safe cleaner should separate routine prose cleanup from changes that may alter language, emoji, code, or bidirectional display. Text Harmonizer selects ordinary space runs, stray BOMs, and trailing whitespace by default while leaving joiners and bidi controls untouched.

The reconstruction model

1. Tokenize the source while retaining UTF-16 positions.
2. Classify each finding as safe, review recommended, or potentially semantic.
3. Preview selected replacements and removals.
4. Build a new sanitized copy only after confirmation.
5. Keep the source independently available for restoration and diffing.

Multilingual and technical text

The default policy protects Arabic, Persian, Indic scripts, variation selectors, emoji sequences, URLs, Markdown, and code. Tabs are presented for review rather than removed because they may be indentation or table structure. Unicode normalization is not performed automatically.

Offline operation

The cleaner uses browser APIs and bundled rules. Once the PWA shell has loaded, inspection, reconstruction, diffing, copying, and export remain available without a network connection.