The Fact & Gap Checker acts as an automated reconciliation engine that evaluates fresh live-web research against a brand's historical memory claims using Gemini and a structured JSON schema. When a research finding aligns with an unprompted parametric claim, the system updates the record with a correct verdict and saves backing URLs into the claim_sources table with positive support flags. Conversely, conflicting evidence marks the claim as incorrect while linking counter-evidence URLs to document the model's hallucination.

Crucially, the reconciliation logic enforces specific safeguards against circular validation and accidental data overwrites:

• Self-Grading Prevention: Findings that match previously generated research claims—where parametric memory was never involved—are classified as regrounding updates rather than genuine confirmations, preserving their status as knowledge gaps.
• Knowledge Gap Discovery: Entirely new findings that models never recalled unprompted are inserted into the claims ledger with a gap verdict, flagging brand facts that search can establish but AI memory completely overlooks.
• Human Override Protection: If an operator has manually ruled on a claim's veracity, automated evaluation passes will append the newly discovered source citations for auditing without overriding the locked human verdict.