The assessment prompt supplies the search query alongside the attributed grounded segments, instructing the model to evaluate the live page content via url_context and output a strict JSON schema dividing facts into survived, lost, and distortions arrays.
Referenced by
