Reading or Reasoning? Format Decoupled Reinforcement Learning for Document OCR Paper • 2601.08834 • Published Dec 11, 2025 • 1