Parametric memory auditing queries models across multiple ungrounded iterations to evaluate the stability, confidence, and accuracy of what is stored directly in their internal weights from training. In contrast, grounded memory auditing deploys recursive web search probes to capture live, citation-backed facts currently discoverable about the brand.

By feeding both output streams into an automated Fact & Gap Checker, the system compares unprompted recall against real-time research to systematically categorize findings:

• Confirms: Verifies that recalled parametric statements align with live web evidence.
• Contradicts: Flags outdated assumptions, misattributions, or active model hallucinations.
• Gaps: Highlights authoritative brand facts present on the live web that models completely fail to recall unprompted.