Dictionary, exceptions and detection

Automatic redaction and pseudonymisation are controlled through four building blocks that apply per project:

The dictionary

The central list of terms and patterns that are redacted or pseudonymised automatically. Every entry has a type — word, word pair (“Mona Muster”, for instance), pattern (a regular expression) or pattern (line) — plus an action (redact a word, line, sentence, paragraph or document, or pseudonymise), a style, and for words a tolerance (0–5) for matches that tolerate typos and OCR errors (“Musterrnann” is still found). Details: dictionary (tenant administration).

Entries come about in three ways:

• Manually through the dictionary page (including import from JSON or from the directory service).
• From the viewer: Shift-clicking a word puts it into the dictionary project-wide (the click matrix).
• From automatic detection: findings from PII detection are adopted by import.

The exception list

The counterpart to the dictionary: words and expressions listed here are not redacted, even when the dictionary or the tolerance search matches them — useful for your own company name or for technical terms that look like personal names.

Area rules

Fixed areas of the page (the letterhead on every page, for example) are redacted project-wide — regardless of the text content.

Detection profiles (PII detection)

Automatic detection runs with a profile — Fast, Standard (recommended) or Deep (more matches, more false positives) — and with selectable detection methods (names, addresses, phone numbers, IBANs, tax numbers, dates, companies, barcodes and many more). Across the installation, the system administrator maintains the underlying PII word lists and learning suggestions.

Priorities

Where rules overlap, the following applies:

1. Words redacted manually in the viewer always take precedence — they are redacted even against the exception list.
2. Exceptions protect terms from dictionary and tolerance matches.
3. Dictionary entries take effect project-wide in all documents; document-specific markings only in the document concerned.
4. Inactive dictionary entries (the Active switch turned off) are ignored.