literature-search-citation-scout

Citation verification checks three questions per source: existence (real with exact metadata), support (backs the exact claim), and proximity (close enough in domain and operating regime). Each source is tagged verified, manuscript-only, or unverifiable with a safe claim boundary.

Use the escalating search ladder: start manuscript-first, then verify metadata, then run targeted proximity checks on the few sources that affect positioning. Shortlists default to 3-5 sources prioritizing IEEE/ACM Transactions, archival journals, and official IEEE/IEC standards.

Yes, a low-RAG mode classifies each cited source as direct domain support, adjacent methodological support, generic background, or unverifiable from the manuscript alone. It states what claims are safe now and marks gaps requiring later verification instead of inventing certainty.

Yes, it detects Korean input and responds in Korean with a compressed three-line evidence block per source: judgment, evidence title, and why it matters. English titles, venues, and DOIs are preserved exactly on a compact evidence line.

Avoid it for pure code debugging or artifact editing with no literature component, and never use it to fabricate sources or bypass research-integrity requirements. Claims about your own results need internal evidence, not literature support.