Stop DeepSeek-R1 Switching Languages Mid-Answer

While DeepSeek-R1 was trained via multi-stage SFT and GRPO to address the poor readability and language-switching of DeepSeek-R1-Zero, studies note it can still provide different responses displaying different cultural values across languages. Explicitly state the required output language and analytical framework in your user prompt.

User Prompt: Analyze the corporate governance requirements for dual-class share structures under EU market regulations. Write your complete response strictly in English and ground all evaluations in statutory EU law.