The per-URL fidelity score measures source claim survival as a percentage, calculated using the formula:

\[\text{Score} = \frac{|\text{survived}|}{|\text{survived}| + |\text{lost}| + |\text{distortions}|} \times 100\]

When the LLM assessment detects no extractable facts across all three categories, empty sets default to a 100.0% score, whereas all other values are rounded to a single decimal place. Once each individual source is evaluated through the two-phase pipeline, an aggregate fidelity score for the entire run is computed as the unweighted arithmetic mean of all valid per-URL scores.