The Veracity Engine submits a structured assessment prompt to Gemini along with the source page content, instructing the model to output a strict JSON schema containing arrays for survived, lost, and distortions to calculate the per-URL fidelity score.
flowchart LR
A[Grounded Segments + Query] --> C[Assessment Prompt]
B[Page URL Context] --> C
C --> D[Gemini Evaluation]
D --> E[JSON Output]
E --> F["survived: string[]"]
E --> G["lost: string[]"]
E --> H["distortions: string[]"]
F --> I[Fidelity Score %]
G --> I
H --> I