/claim #16

Adds a self-contained citation-retraction-evidence-guard package for the AI-Powered Research Assistant Suite.

This slice prevents AI peer-review and research-assistant output from being released when manuscript claims rely on unsafe citation evidence. It validates citation retraction status, expressions of concern, superseded preprints, stale evidence windows, claim-to-citation support direction, missing citation-context quotes, and assistant recommendations that point at unsafe sources.

Why this is distinct:

• not a broad assistant suite
• not a reproducibility dependency checker
• not reviewer calibration or inter-rater agreement
• not model-assumption diagnostics
• not sample chain-of-custody
• not external-validity transfer
• not structured abstract consistency
• not systematic-review screening drift

Artifacts included:

• deterministic synthetic citation packets
• JSON results
• Markdown reviewer report
• SVG summary
• short MP4 demo artifact

Validation run locally:

• npm test
• npm run demo
• node –check src/index.js
• node –check scripts/demo.js
• node –check test/citationRetractionEvidenceGuard.test.js
• git diff –check

Boundaries:

• synthetic citation packets only
• no live publisher, Crossref, PubMed, OpenAlex, or Retraction Watch calls
• no private manuscripts
• no credentials
• no external AI APIs
• no production medical or legal review claims

AI-assisted with OpenAI Codex; I reviewed the implementation and kept the module isolated.