The Veracity Engine provides the model with the query, target URL, and its attributed response segments via the url_context tool, prompting it to compare the live page text against the grounded text and return a schema-constrained JSON object categorising claims into survived, lost, and distortions arrays.
flowchart LR
A[Grounded Query & Attributed Segments] --> C[url_context Model Call]
B[Live Source Page URL] --> C
C --> D[Model Evaluates Content Alignment]
D --> E[JSON Output]
E --> F[survived: Accurate Facts]
E --> G[lost: Omitted Source Facts]
E --> H[distortions: Misrepresented Claims]