Evidence & consensus

The AllNutrition API ships every answer with three labels — evidence_strength, consensus_level, and per-source evidence_level — so calling agents can decide how confidently to use the answer.

evidence_strength

Overall strength of the evidence backing the answer. Set on the response.

Value	What it means	How to treat it
`strong`	Top-tier sources (clinical guidelines, systematic reviews) with high composite trust.	Safe to surface as a recommendation.
`moderate`	Solid evidence (RCTs, large observational studies) with some gaps.	Surface as guidance, but encourage user judgement.
`limited`	Sparse or lower-tier evidence (small studies, expert opinion).	Surface as preliminary; pair with a "more research needed" note.
`insufficient`	Not enough trusted sources to answer confidently.	The `answer` field will say so explicitly. Suggest the user consult a clinician.

consensus_level

How well the cited sources agree.

Value	What it means
`high`	Sources align — the answer reflects the mainstream view.
`moderate`	Mostly aligned, with minor variation in detail.
`mixed`	Notable disagreement — the answer presents the mainstream view but acknowledges the disagreement.
`low`	Sources conflict; treat the topic as contested.

evidence_level (per source)

Each cited source carries an evidence tier. Higher tiers carry more weight in ranking and labelling.

Tier	What it is
`guideline`	Clinical or public health guideline (e.g. WHO, NICE, AHA, NIH).
`systematic_review`	Systematic review or meta-analysis of primary studies.
`rct`	Randomised controlled trial.
`observational`	Cohort, case–control, or cross-sectional study.
`expert_opinion`	Position statement or expert consensus document.
`review`	Narrative review (lower tier than systematic reviews).

trust_score

A composite trust score in [0, 1] attached to each source. It blends:

Publisher authority — the credibility of the journal, body, or organisation.
Recency — newer sources score higher; sources >10 years old decay sharply.
Evidence tier — guidelines and systematic reviews outrank narrative reviews and expert opinion.
Consensus signal — how aligned the source is with the wider literature.

Use trust_score when you want to do your own ranking — e.g. show only sources with trust_score >= 0.7 in a clinician-facing UI.

Worked example

A response with evidence_strength: "strong", consensus_level: "high", and a top source at evidence_level: "guideline" with trust_score: 0.92 is the gold standard — equivalent to a current public-health recommendation. Surface it directly.

A response with evidence_strength: "limited" and consensus_level: "mixed" means the topic is genuinely uncertain. Don't render it as a recommendation — show the cited sources alongside the answer and let the user decide.

Always pair answers with sources AllNutrition is designed so the citations are the value-add. If you only display the answer string and drop the sources array, you lose the thing that distinguishes this from a generic LLM. At minimum, show a "Sources (N)" expandable section linking to source.url.