The engine gathers active property claims alongside newly extracted research findings, feeding both sets into a Gemini model using a structured JSON schema to classify relationships into confirms, contradicts, or new. Confirmed or contradicted parametric claims are updated with definitive verification verdicts and citation source links, while unprompted findings are logged as gap claims without overwriting existing human rulings.