Treewalker inspects the top alternative candidates at each position in an ungrounded response, converting raw log probabilities into percentage confidence scores. By pruning any alternative candidate whose score falls below the configured threshold percentage, the engine isolates only statistically significant association divergences before completing the alternative paths into full product or concept names.