What problem does it solve? Document extraction pipelines often fabricate values, lose provenance, or silently drop rows and gaps, making downstream validation and decisions untrustworthy. This Skill provides a structured expert review lens that audits extraction code against strict invariants for immutable source ingest, verified fields, and honest gap reporting. ## Core Features & Use Cases - Provenance and Dependency Auditing: Verifies that every source-derived field closes the chain from immutable source hash through geometry, raw spans, transformations, and normalized values. - Fabrication Detection: Flags authored recovery rows, synthetic anchors, fixed confidence, unlocated AI output, and value-bearing placeholders presented as extracted truth. - Architecture Enforcement: Checks source-neutral extraction, one-way Extraction-to-Interpretation flow, shadow pipeline isolation, and parity/cutover evidence. - Use Case: When reviewing a PR that adds a new XLSX table parser, run this review to confirm table-derived fields carry cell coordinates, rejected candidates are persisted honestly, and no filename-based routing leaked into the generic extraction layer. ## Quick Start Review this pull request touching the PDF extraction pipeline for provenance closure, fabricated values, and honest gap reporting.