Based on your analysis above, provide your decomposition plan as a JSON object only:
{"sub_questions": [{"question": "<sub-question>", "evidence_needed": "<what evidence would answer it>", "risk": "low|medium|high"}]}
- Cover the question's full scope — split it at its natural seams, neither padding with near-duplicates nor leaving parts under-covered.
Output ONLY the JSON object. Do NOT call any tools.