CORD: Bridging the Audio-Text Reasoning Gap via Weighted On-policy Cross-modal Distillation | ScienceToStartup | ScienceToStartup