Toward Template-Free Explainability for Monte Carlo Tree Search

In three linesFramework enabling LLMs to generate evidence-grounded explanations of MCTS decisions from search traces end-to-end, without hand-crafted formal logic constraints. Maps natural-language questions to intent categories, triggers targeted tree expansion when needed, and generates explanations using visit counts, value estimates, and risk information.

Summary generated by Claude — human-verified