What problem does it solve? Research workflows often require manually reading dozens of papers and pulling out claims, statistics, and methods with citations. This Skill automates that extraction, producing a structured evidence package with confidence and strength labels so downstream synthesis steps can integrate results without re-reading sources. ## Core Features & Use Cases - Structured Evidence Extraction: Pulls claims, findings, statistics, methods, mechanisms, and limitations from a literature_sources.json package or embedded source list, each with source citations. - Evidence Classification: Labels every item by evidence type (empirical, statistical, theoretical, mechanistic, opinion), confidence (HIGH/MEDIUM/LOW), strength, and domain theme across biomedicine, chemistry, materials, finance, and computer science. - Quality and Traceability Checks: Runs fabrication checks, preserves cross-source contradictions, records metadata gaps, and returns a SUFFICIENT, PARTIAL, or INSUFFICIENT verdict. - Use Case: After a literature retrieval step produces a deduplicated source package, run this Skill to extract traceable evidence items from each paper's abstract or full text, then hand the embedded JSON evidence package to a report-writing step. ## Quick Start Extract evidence from the sources in literature_sources.json for my research objective on transformer architectures and return the evidence package with confidence ratings.
