Review-First Redaction Detection

THotPDF.DetectLoadedRedactionFindings combines built-in PII patterns with optional application-selected entity and face detectors, returning confidence-scored page-space findings without modifying the loaded document

Built-in matching covers email addresses, telephone numbers, US Social Security numbers, Luhn-valid payment-card numbers, checksum-valid IBANs, and IPv4 addresses

Mandatory review boundary

Every detected finding starts as rrdPending with an empty reviewer identity

THotPDF.StageApprovedLoadedRedactionFindings rejects approved findings without a bounded reviewer identity, validates all approved geometry before mutation, and atomically creates only reviewed /Redact annotations

Staging never removes content and never calls ApplyLoadedRedactions, so burn-in remains a separate explicit application decision

Bounds and coordinates

Text spans are mapped through extracted character quadrilaterals while face pixel rectangles are reversed through the rotation-aware render matrix into PDF page user space

Page, finding, UTF-16 code-unit, raster-pixel, render-slice, confidence, and cancellation limits keep detection bounded and publish no partial array on failure

Related API