What problem does it solve? Document intelligence pipelines often produce mismatches between what a source document says and what appears in the UI, and finding where the divergence happens is slow and error-prone. This Skill provides a bounded investigation workflow that pinpoints the first divergence between source truth, extracted artifacts, canonical assembly, and rendered surfaces. ## Core Features & Use Cases - End-to-End Pipeline Tracing: Walks the full path from source file through parser output, normalized models, persisted facts, view model assembly, and UI render conditions. - Sibling Surface Comparison: Compares Extraction, Evidence, Fact Ledger, and Validator surfaces that should agree, to isolate which layer consumes the wrong data shape. - Domain-Specific Playbooks: Covers contract pricing and rate schedule assembly, invoice totals and line items, transaction_data spreadsheet summaries, evidence anchors, and OCR/table noise. - Use Case: An invoice total renders correctly in the Extraction tab but is missing in the Fact Ledger. Use this Skill to trace whether persistence stored the right data, the view model built the wrong shape, or a render condition blocked the section, then apply the smallest fix at the first true divergence. ## Quick Start Use the eightforge-extraction-trace-debugger skill to trace why the rate schedule rows shown in the Evidence tab do not match the rows rendered in the Extraction tab.
