The Association engine evaluates an AI's ungrounded parametric memory across three distinct probe directions, using a Matryoshka prefix-matching algorithm to resolve multi-tiered brand hierarchies and compute competitive share of association.
flowchart LR
subgraph Probes["Probe Vectors"]
E2B["E2B: Entity to Brands"]
B2E["B2E: Brand to Entities"]
Q2B["Q2B: Query to Brands"]
end
subgraph LLM["Parametric Memory"]
Model["Target LLM<br/>Gemini, GPT, Claude"]
end
subgraph Normalization["Resolution Layer"]
Matryoshka["Matryoshka Hierarchy<br/>Prefix Resolution"]
end
subgraph Metrics["Outputs"]
SoA["Share of Association<br/>Exclusive and Inclusive"]
Centrality["Competitor Centrality<br/>and Topic Gaps"]
end
Probes --> Model
Model --> Matryoshka
Matryoshka --> MetricsReferenced by