Entities and claims
Entities
Section titled “Entities”Each analysed page yields the real-world things it is about: the organisation, products, people, places, technologies and concepts it defines or mentions, with a role (main, supporting, passing) and a salience. Entities are resolved across pages — aliases merged, mention counts summed — and the page that most consistently defines each one becomes its defining page, or entity home.
Names are linked to Wikidata by exact label or alias match through the public search API. The model never guesses an identifier: a link exists only when the label matches, which keeps the graph honest and lets you see which of your entities have no public record at all.
An entity mentioned often but with no defining page has no canonical home for engines to resolve it to. That is a finding.
Claims
Section titled “Claims”Factual, citable statements — the sentences an answer engine would quote — with:
- evidence type — statistic, citation, quote, first-party data, or none;
- specificity — from 0 (marketing) to 1 (concrete, quantified, dated);
- verifiability — whether a reader could check it against an external source;
- freshness — the date or period it is bound to, if any.
Claims are matched back to the passage that carries them, so coverage verdicts can report the specificity of what was cited.
Contradictions
Section titled “Contradictions”The same subject and predicate with different objects across pages — two founding years, two prices, two headcounts — lower a model’s confidence in every statement about that entity. They are surfaced in the dossier and, when material, as findings with both sources.