On the ATT&CK extraction benchmark, ARBITRA achieves a stronger precision-oriented operating point than sentence-level pipelines, recent LLM-based extraction systems, and frontier models including GPT-5.5.