By requesting the top five log probabilities for each generated token through the Vertex AI API, Treewalker extracts chosen tokens alongside their alternative candidates and transforms logarithmic values into percentage-based confidence scores. Whenever an unchosen candidate exceeds a designated probability threshold, the system branches off to generate completions for that sub-path, systematically mapping out latent brand associations and competitive concepts lurking just beneath the surface in the model's parametric memory.