Grounded memory auditing conducts multi-round web research using tools such as Google Search and URL context to extract source-backed facts about a brand, continuing iteratively until a round yields no novel information or reaches a configured round cap. In each iteration, established facts feed into continuation prompts to prevent duplicate coverage, and all discovered statements are bound directly to their source URLs and titles via a strict JSON schema.

Downstream, the Fact & Gap Checker reconciles these discoveries against the property's parametric claims ledger. Recalled claims supported by live sources are flagged as confirmed (correct), while conflicting assertions are marked contradicted (incorrect). Novel discoveries absent from model memory are appended as gap entries without self-verifying, ensuring research never artificially validates previous research findings while preserving human rulings.