Method

How Text Harmonizer works

Four independent systems, explicit approval gates, and a source document that never becomes disposable.

Open the local text workspace

1. Character inspection

The first pass identifies Unicode and whitespace issues. It does not apply writing-pattern rules or infer authorship. Selected actions reconstruct a new sanitized string after confirmation.

2. Versioned pattern analysis

The approved local ruleset scans the sanitized copy. Each detector is configured as data rather than prose inside a component, so the same text and ruleset version always produce the same result.

3. Deterministic harmonization

The local provider proposes only configured, reversible edits. Names, dates, numbers, currencies, URLs, email addresses, quotations, citations, identifiers, acronyms, and code are treated as protected atoms.

4. Validation and scoring

Every candidate is checked for protected-fact changes and rescanned. Candidates that increase the score are rejected by default. The user chooses which valid edits to apply, then runs a final scan and exports the result.

Rule governance

The protected admin console compares MediaWiki revision IDs, stores source snapshots, stages changes, runs fixtures, and requires explicit publication. A source update never silently replaces production rules.