Grounded fidelity scoring calculates the percentage of factual claims from cited source pages that survive accurately in an AI model's output relative to the total volume of extracted facts.
Referenced by