Treewalker aggregates observed concepts across runs by normalizing text, clustering variations, and calculating appearance rates and confidence scores from parametric logprobs and rank distributions. Operators can review these discovered brand associations in the Treewalker interface and promote vetted items directly into the property's canonical entities catalog with full provenance tracking and a source:treewalker tag.