Systematic literature search across PubMed, preprint servers, and curated genomics databases. Candidate studies are identified by gene, variant, phenotype, and health domain.

Candidate studies are screened for relevance to actionable genotype-phenotype relationships. Studies outside scope — non-human models, non-germline variants, insufficient sample size — are excluded at this stage.

Study design, methodology, and statistical robustness are assessed. Replicated findings from independent cohorts are weighted more heavily than single-study results.

Each finding receives a confidence grade based on replication, effect size, and study quality. Grades are surfaced in every API response and user-facing output — nothing is presented without an evidence level.

Beyond single-variant associations, epistatic interactions between variants are mapped. This layer captures combinatorial effects that single-SNP analysis misses and is a core source of differentiation in the knowledge base.

Curated findings are encoded into a structured schema with strict validation rules. Every entry is linked to its source citations and must pass schema integrity checks before entering the production substrate.

As new research is published, the substrate is continuously reviewed and expanded. Existing entries are re-graded when contradictory or reinforcing evidence emerges. The knowledge base strengthens over time — not at release milestones.