Reasoning strategy that explores multiple solution branches with reflection or scoring, guiding the model to better final answers than single‑pass decoding.