Grounded Memory research findings are generated via an iterative search-and-extract loop using Google Search and URL context tools that uncovers fresh brand facts and their source URLs until research yields no new information.

sequenceDiagram
    participant E as Grounded Memory Engine
    participant G as Gemini + Google Search
    participant DB as Postgres DB (grounded_memory_facts)
    participant C as Fact & Gap Checker

    loop Research Rounds (1 to 8)
        E->>G: Prompt with established facts + brand
        G-->>E: Extract fresh facts + source URLs
        E->>DB: Store facts & sources
        Note over E,G: Stop when no fresh facts found
    end
    E->>C: Trigger Fact & Gap Checker (run_check)