How Resar scores products.

The research pipeline searches indexed scholarly metadata providers, deduplicates candidates by DOI and title, and sends only the provided candidate metadata and abstracts to the AI curation step.

The AI curation prompt is constrained to strict JSON and forbids external knowledge, invented studies, invented citations, and unsupported clinical claims.

Each product aims to use five peer-reviewed or clinically relevant papers. When fewer than five usable papers are found, Resar marks transparent research-gap slots instead of pretending stronger evidence exists.

The five selected papers are ranked by intended-use relevance, human applicability, study design, safety relevance, recency, and metadata quality.

Evidence Score combines study quality, product-match confidence, outcome relevance, recency, citation signal, and research completeness.

Safety Score reflects reported safety context from the selected papers and is cautious when safety evidence is thin.

Effectiveness Score reflects how directly the selected evidence supports the intended shopper use case.

Research Confidence Score reflects completeness and quality of the selected evidence set.

Product Fit Score combines evidence-backed benefit, product quality signals, price context when available, and practical-use alignment.

AI-generated summaries are stored with a review status. Admins can approve or reject summaries before treating them as reviewed research content.

Scores are explainable and historical score changes are stored when refreshed research changes a product score.

Resar does not diagnose, treat, cure, or prevent disease. Product scores are research-navigation tools, not medical recommendations.

Many products have indirect evidence. Resar surfaces that uncertainty instead of presenting brand-specific proof where only category-level or ingredient-level research exists.