During token-level analysis, whenever an alternative candidate's log-probability meets or exceeds the user's confidence threshold, Treewalker splices that token into the preceding text to construct a divergent string fragment. It then dispatches a focused branch completion prompt through Vertex AI to complete the alternative concept, recording the replaced token probability alongside the newly discovered association.