LIMITATIONS.md

Scientific and technical limitations

1. No truth inference: link relation labels are user judgments. Mixed supportive and challenging links are not a formal contradiction test.
2. No automatic evidence extraction: bibliographic API records are not article texts. The user must supply checked excerpts and locators.
3. No causal inference: connecting indicators with hypotheses does not estimate a policy effect. No confounding controls or counterfactual assumptions are tested.
4. No meta-analysis: link counts cannot be pooled as effect sizes or votes across independent studies.
5. No semantic DOI deduplication: only exact normalized DOI matches are detected when imported.
6. No persistent storage: download JSON before closing the page. No cross-device sync.
7. Small networks: SVG display grows cluttered with very large graphs; restrict to concise research questions and export GraphML for dedicated network software.
8. External integrations: Crossref and World Bank request availability, errors and CORS depend on live network conditions. The app has no secret API key support.
9. Synthetic data: initial illustration contains entirely fictional research records; never describe them as published evidence.
10. Publication reliability: a locally passing test suite does not guarantee the same behavior on every browser or that all research conclusions are valid.