The Veracity Engine operates via a two-phase pipeline that first executes a grounded search to map attributed response segments to source URLs, then uses url_context calls to assess page content directly against those claims. It audits source representation by classifying extracted facts into survived, lost, and distorted sets, calculating an individual URL fidelity percentage—the ratio of survived facts over total observed claims—before averaging them into an aggregate score.