The Veracity engine evaluates grounding fidelity through a two-phase assessment that contrasts an AI model's grounded statements against original source pages. First, it executes a grounded search to capture the generated response segments and cited URLs, then uses Gemini's url_context tool to compare each page's actual text against those claims. By tracking which facts survived intact, which were omitted, and which were altered, it calculates an exact mathematical fidelity score both per source and as an overall run aggregate.