| name | sci-literature-vetting |
| description | Use when adding, checking, repairing, or interpreting references, citations, related work, official documents, datasets, DOI links, or literature-supported claims in a SCI manuscript |
SCI Literature Vetting
Overview
Use this branch before any source becomes a citation or related-work claim.
Core principle: a reference is not usable until its identity and claim support are verified.
Required Checks
For every candidate source record:
- title
- authors or institution
- venue/report/source type
- year/date
- DOI/arXiv/URL
- source class: peer-reviewed, preprint, official doc, dataset, blog/news, user measurement
- exact manuscript claim it supports
Citation Risk Labels
| Label | Meaning |
|---|
| Direct citation | Verified source directly supports the sentence. |
| Candidate citation | Real source but support is indirect or narrow. |
| Background only | Useful context but not claim support. |
| Unsafe | Hallucination risk, missing source identity, or unverifiable. |
Workflow
- Build or update
LITERATURE_MATRIX.md.
- Put each source into a risk label.
- Rewrite manuscript claims to match what the source actually supports.
- Do not cite search snippets, AI summaries, or memory alone.
Output
- literature matrix rows
- citation risk notes
- safe replacement wording for manuscript sentences
Common Mistakes
- Using secondary news as if it were an official report.
- Treating a dataset/blog as proof of method performance.
- Adding a citation that is real but does not support the sentence.
