provenance-engine

What problem does it solve?

Make every extracted value traceable to concrete source evidence (transcript segments and OCR regions). Enable auditors and staff to answer: “Where did this come from?” quickly and reliably.

Core Features & Use Cases

• Stable IDs: Source spans must have stable identifiers that survive UI rendering, exports, and reprocessing (when possible).
• Minimal ambiguity: A citation should point to a specific segment/region, not an entire document.
• Enforceability: The system must be able to reject outputs that lack required citations.
• Outputs: Generates source_index.json, citations.json, citation_validation_report.json, and UI highlight payloads (span -> offsets/bbox mapping).

Quick Start

Run the provenance-engine to generate traceable citations for your transcripts and OCR outputs. Provide transcript.json and ocr_output.json as inputs, with optional pasted_text that is chunked into spans.