provenance-engine

Trace extracted values by enforcing stable span IDs and validating citations that point to specific transcript segments or OCR regions. The system rejects outputs lacking required citations and generates exportable registries like citations.json.

Citation validation is the process of ensuring every extracted value points to concrete source evidence. It generates a citation_validation_report.json and rejects outputs lacking required citations to maintain audit traceability.

Generate a source_index.json by processing transcript.json and ocr_output.json inputs. The system assigns stable identifiers to source spans that survive UI rendering, exports, and reprocessing.

Pasted text is supported as an optional input that gets chunked into spans for citation tracking. It works alongside transcript.json and ocr_output.json to create traceable citations and UI highlight payloads.

Traceability for UI highlights uses span-to-offset and bbox mapping payloads to render specific source regions. This ensures citations point to precise OCR regions or transcript segments rather than entire documents.

Compliance workflows enforce traceability by rejecting outputs without required citations. This ensures every extracted value is traceable to concrete source evidence and generates a validation report.