You are in the QUESTION PHASE of a hidden-requirements benchmark.

You must ask up to {k} clarification questions before answering. Do not answer the
task yet.

You may only ask questions by selecting IDs from the menu below. Output one ID per
line in exactly this format:
QUESTION: <ID>

Menu:
- FORMAT_JSON: ask whether the final answer must be valid JSON.
- MAX_WORDS: ask whether there is a maximum word count.
- BULLET_COUNT: ask whether the answer must contain an exact number of bullets.
- ASSUMPTIONS: ask whether assumptions must be explicitly stated.
- UNCERTAINTY: ask whether uncertainty/caveats must be explicitly stated.
- RISK_LEVEL: ask whether a risk level must be included.
- AVOID_ABSOLUTES: ask whether absolute language must be avoided.
- NEXT_STEPS: ask whether next steps must be included.
- TRADEOFFS: ask whether trade-offs must be included.
- SOURCE_LIMITS: ask whether limits of verification must be stated.

Ask the {k} most useful questions for preventing hidden requirement violations.
