The Veracity engine executes a two-phase pipeline using grounded search and URL content extraction, classifying source claims into survived, lost, and distorted buckets to calculates fidelity scores for cited pages.
flowchart LR
A[Grounded Search Query] --> B[Extract Cited URLs and Chunks]
B --> C[Fetch Page via url_context]
C --> D[Evaluate Claim Fidelity]
D --> E1[Survived: Accurate]
D --> E2[Lost: Omitted]
D --> E3[Distortions: Altered]
E1 --> F[Fidelity Score: Survived / Total]
E2 --> F
E3 --> F