AI provenance tracking for AI-classified documents

Context paperless-ngx uses AI (OCR, classification, auto-tagging) to process incoming documents. As AI-powered document management grows, there's no metadata trail indicating: • Which documents were AI-classified vs manually sorted • What confidence level the AI had in its classification • Whether AI-generated summaries or tags were human-reviewed Why this matters 1. Audit trail — regulated industries (finance, healthcare, legal) need to know which classifications were AI-driven vs human-confirmed 2. Trust levels — an AI classification with 0.95 confidence is different from 0.50 3. EU AI Act (August 2, 2026) — AI-processed content may need transparency metadata depending on the use case Possible approach When paperless-ngx processes a document with AI: • Store a provenance record: which AI model classified it, confidence score, timestamp • Distinguish AI-classified vs human-confirmed tags in the UI • Optionally embed provenance into the d…