The Veracity Engine tracks claim accuracy by extracting attributed response segments from grounding metadata, evaluating the live source page via a structured evaluation prompt to categorize facts into survived, lost, and distorted arrays, and computing an aggregate fidelity score.

flowchart LR
    A[Grounded Search Response] --> B[Extract Segments & URLs]
    B --> C[Fetch Source via url_context]
    C --> D[Run Assessment Prompt]
    D --> E1[Survived Facts]
    D --> E2[Lost Facts]
    D --> E3[Distorted Facts]
    E1 --> F[Calculate Fidelity Score]
    E2 --> F
    E3 --> F