The engine calculates a per-URL fidelity score as the proportion of verified facts that survived intact, then averages these across all URLs to produce an overall aggregate score:
\[\text{Score}_{\text{URL}} = \frac{|\text{survived}|}{|\text{survived}| + |\text{lost}| + |\text{distortions}|} \times 100\]
flowchart LR
A[Source Page Content] --> B[Gemini Assessment]
C[Grounded Segments] --> B
B --> D[Fact Classification]
D --> E[Survived S]
D --> F[Lost L]
D --> G[Distorted D]
E --> H["Per-URL Score: S / (S + L + D) * 100"]
F --> H
G --> H
H --> I[Aggregate Score: Mean Across URLs]Referenced by