The Veracity Engine operates via a two-phase execution pipeline that first executes a grounded search to capture raw grounding metadata, resolving redirect URIs to identify the exact source URLs cited by the LLM. It then evaluates each cited page in parallel using Gemini's url_context tool, directly comparing the extracted webpage text against the specific response segments attributed to that source. By categorizing facts into survived, lost, and distorted claims, the engine calculates a per-URL score as survived / (survived + lost + distortions) alongside an overall aggregate score, verifying whether AI models faithfully represent source content or introduce hallucinations and factual omissions.