Logprobs represent the log probabilities of generated tokens, which the Treewalker engine converts to percentage confidence scores to identify weak associations and trigger branching completions along alternative paths.

flowchart LR
    A["Target Token<br>85% Confidence"] -->|Above Threshold| B["Primary Recall Output"]
    A -.->|Alternative Token<br>12% Confidence| C["Branch Prompt Formulation"]
    C --> D["Alternative Association Path"]