A paired ungrounded and grounded analysis runs identical association probes across parametric memory and live search, comparing the results in a holistic analysis to identify training-data discrepancies, hallucinations, and emerging search-only brand signals.

flowchart LR
    A["Brand Entity Query"] --> B["Ungrounded Probe<br/>Parametric Memory"]
    A --> C["Grounded Probe<br/>Live Web Search"]
    B --> D["Appearance Rates &<br/>Token Logprobs"]
    C --> E["Appearance Rates &<br/>Retrieved Sources"]
    D --> F["Holistic Analysis<br/>Cross-Comparison"]
    E --> F
    F --> G["Grounded-Only<br/>Signals"]
    F --> H["Ungrounded-Only<br/>Hallucinations"]
    F --> I["Converged<br/>Core Associations"]