How TryQuerra calculates trust scores

Not all research sources are equal. We assign each source a quality weight based on publication tier, study design, sample size, and institutional reputation before it contributes to a trust score.

Highest-weight sources include: systematic reviews and meta-analyses from Cochrane Library, JAMA, NEJM, Lancet, Nature, and comparable journals; government health body guidelines (WHO, CDC, NHS, FDA); and large-scale randomised controlled trials with replicated results.

Lower-weight sources include: observational studies, small sample research, preprints not yet peer-reviewed, conference presentations without published data, and industry-funded research without independent replication.

Consensus score measures the degree of agreement across all retrieved sources for a given answer. It's not a simple count — each source vote is weighted by its quality score.

When all high-quality sources point in the same direction, consensus is high. When credible sources actively disagree — even if the majority agrees — the consensus score drops and the contradictions section is populated on the answer page.

Science moves. An answer fully supported by 2015 data may no longer reflect the current consensus. Freshness scores penalise answers that rely heavily on older sources and rewards answers verified against recent research.

The recency penalty scales by category: medical and pharmaceutical evidence decays faster than, say, physics or historical fact. The system automatically flags answers for re-verification when significant new literature appears on that topic.

Research decay rates by category: Drug interactions & clinical medicine — re-checked every 6 months. Nutrition & lifestyle — re-checked annually. Physical sciences and established facts — re-checked every 3–5 years.

Our system screens all sources for potential conflicts of interest before assigning them quality weights. Sources published by organisations with financial interests in a particular outcome receive a conflict-of-interest penalty.

Detected patterns include: industry funding without independent replication, citations that trace back to a single self-referencing research group, promotional framing in scientific language, and affiliate link patterns in referenced websites.

Sources flagged for potential conflicts remain in the source list but are clearly labelled and carry reduced weight in the trust score calculation. High conflict-of-interest concentration across an answer's source pool will reduce the total score.

Every answer produced by our AI research system passes through human review before publication. Reviewers check for factual balance, potential for harm, inappropriate framing, and AI hallucination artefacts.

Reviewers do not override scientific consensus — they ensure answers are not misleadingly framed, that important caveats are included, and that sensitive topics (mental health, medications, legal matters) carry appropriate disclaimers.

Answers dealing with contested political, social, or cultural topics are handled with additional care and may include multiple perspectives where scientific consensus is genuinely absent.

Community insights are moderated before appearing publicly. Our review criteria: the insight must be relevant to the verified answer, must add genuine context or nuance, must not contain promotional links, and must not make claims contradicting strong scientific consensus without extraordinary evidence.

Expert badges are awarded when professional credentials are verified. Insights from non-expert contributors are equally welcome — lived experience and practical knowledge are valuable. All insights are labelled clearly by contributor type.

Answers that have not completed human review, or where our quality systems have flagged significant concerns, are held in a pending state. Pending answers are not publicly accessible and carry a noindex directive to prevent search engine indexing until they pass review.

Reasons an answer may be held pending include: AI confidence score below threshold, significant contradictions detected that require manual resolution, sensitive topic requiring specialist reviewer, or initial human review flagged the need for additional sourcing.

We believe intellectual honesty requires stating clearly what our system cannot do.