Evidence Methodology

Alexandria is designed as a controlled PubMed research pipeline rather than an unrestricted medical chatbot.

Source boundary

Evidence enters the system only through records retrieved from PubMed by PMID. The synthesis model receives only those records. Claims citing unknown PMIDs are rejected.

Priority order

1. PubMed-indexed clinical practice guidelines and consensus documents
2. Systematic reviews and meta-analyses
3. Randomized controlled trials
4. Large prospective cohort studies
5. Other observational evidence
6. Case reports, animal research, and laboratory research

Retrieval-priority score

The visible score helps order records for review. It incorporates publication type, recency, sample-size signals, multicenter or independent-validation signals, missing-abstract penalties, configured landmark status, and retraction exclusion. It is not a clinical recommendation and does not pretend to be a complete GRADE assessment.

Conflicting research

Conflicts must cite evidence on both sides. Alexandria asks whether differences may be explained by study population, comparator, endpoint, follow-up, design, bias signals, power, or analysis. Explanations not supported by the supplied records are prohibited.

Publication controls

Generated reports begin as drafts. A second verification pass checks claim-to-citation support. Automatic publishing is disabled by default.