A per-URL fidelity score is derived by fetching the source page via url_context, extracting the survived, lost, and distorted claims against the attributed answer segments, and computing the proportion of intact facts as a percentage.