The Veracity Engine measures how accurately an AI represents cited web pages by executing a two-phase pipeline that extracts attributed response segments and evaluates whether source claims were survived, lost, or distorted to calculate an overall fidelity score.

flowchart TD
    A[Query + Grounded Search] --> B[Phase 1: Extract Segments & URLs]
    B --> C[Resolve Redirects & Clean URLs]
    C --> D[Phase 2: Evaluate URL Context]
    D --> E[Classify Fact Status]
    E --> E1[Survived Facts]
    E --> E2[Lost Facts]
    E --> E3[Distortions]
    E1 --> F[Compute Fidelity Scores]
    E2 --> F
    E3 --> F
    F --> G[Per-URL & Aggregate Fidelity]