In the Veracity Engine's assessment prompt, survived claims represent specific facts from a cited source page that the model accurately reproduced within its grounded response segments. Lost claims capture significant, query-relevant information from the source page that the model omitted, while distortions identify statements in the output that were altered, misrepresented, or taken out of context relative to the source text. These three categorized claim counts directly feed the engine's per-URL fidelity score formula, measuring the exact proportion of source information that survived intact without degradation.