Treewalker aggregates branch results by grouping identical normalized concepts across test runs, computing their appearance rates, mean rank positions, and token confidence scores to surface which latent associations consistently compete with primary recall.