Reviewing detected PII before you export

What Lex Redactus detects

Ten entity types: people, organizations, locations, addresses, SSNs, phone numbers, emails, case numbers, account numbers, and parcel IDs. Detection is multi-pass — a neural NER model plus pattern libraries tuned for legal documents — and runs entirely on your machine.

Why review matters

No detector catches everything, especially on scanned or all-caps text. A missed name is the failure this tool exists to prevent, so the app puts you in the loop before anything is exported. Human review is the safety gate — treat the suggestions as a high-recall starting point, not a final answer.

Confirm, edit, or add

• Confirm a suggested entity to include it in the redaction.
• Add a missed entity by selecting any text in the document and marking it as the appropriate type — useful for names in captions or scanned pages the model may under-detect.
• Relabel an entity if its type is wrong (e.g. an organization tagged as a person).

Removing a false positive

Click the ×next to an entity you don't want redacted. You'll be asked whether to suppress it just on this document or across the entire matter. The Don't-Tag list is matter-scoped and you can revisit it any time from the matter settings.

If a document was scanned (OCR)

When the text came from OCR, the app flags it — review those pages carefully, because scanner errors and unusual formatting (all-caps captions, names split across lines) can lower detection quality. When in doubt, add the entity manually.

A quick pre-export checklist

1. Every party in the caption is detected (plaintiff, defendants, counsel).
2. Identifiers are caught: SSNs, account numbers, case numbers, addresses.
3. No real name remains visible in the tokenized preview.
4. Anything you excluded was excluded on purpose.