You are a helpful assistant that creates simple, readable bridge questions.

--TASK--
You are given a list of claims about a shared entity. Each claim is a factual statement from the source documents.

Your task is to create BRIDGE QUESTIONS. A bridge question combines two claims by using a fact from one claim as an indirect reference to ask about a fact from another claim.

--GOOD EXAMPLES--

Example 1:
  Shared entity: Christopher Nolan
  Claim 1: "Christopher Nolan directed the movie Inception"
  Claim 2: "Christopher Nolan was born in London"

  Bridge question: "Where was the director of Inception born?"
  
  This works because:
  - Claim 1 provides the indirect reference ("the director of Inception")
  - Claim 2 provides what we're asking about (birthplace: London)
  - The question replaces "Christopher Nolan" with "the director of Inception"
  - Both claims are about the same person's biography (thematically related)

Example 2:
  Shared entity: CDC
  Claim 1: "The CDC issued new COVID-19 isolation guidelines in 2024"
  Claim 2: "The CDC is headquartered in Atlanta"

  Bridge question: "Where is the headquarters of the agency that issued COVID-19 isolation guidelines?"
  
  This works because:
  - Claim 1 provides the indirect reference ("the agency that issued COVID-19 isolation guidelines")
  - Claim 2 provides what we're asking about (headquarters: Atlanta)
  - Both claims are about the same health agency

Example 3:
  Shared entity: Marie Curie
  Claim 1: "Marie Curie discovered radium in 1898"
  Claim 2: "Marie Curie won two Nobel Prizes"

  Bridge question: "How many Nobel Prizes did the scientist who discovered radium win?"
  
  This works because:
  - Claim 1 provides the indirect reference ("the scientist who discovered radium")
  - Claim 2 provides what we're asking about (Nobel Prize count: two)
  - Both claims are about the same scientist's achievements

--CORE RULES--
1. REPLACE one entity name with a short indirect reference (3-8 words)
2. The reference must describe the entity using a DIFFERENT FACT than what the question asks
3. Reference and question must be THEMATICALLY RELATED (same story/domain)
4. Must be ANSWERABLE ENTIRELY from the provided claims - don't assume external knowledge
5. Must be STANDALONE - the question should make sense without seeing the claims
6. Must be concise and sound NATURAL - like something a curious person would actually ask
7. Keep questions under 20 words, readable in one pass
8. No nested clauses or convoluted structure
9. NO SELF-REFERENCES: NEVER use words like "claims", "both claims", "these cases", "the above", "this context"
10. GRAMMATICALLY CORRECT: The question must be a proper English sentence
11. UNIQUE REFERENCE: The indirect reference must uniquely identify the entity IN THE REAL WORLD (not just within the claims)
   - GOOD: "the director of Inception" (only Christopher Nolan directed Inception)
   - GOOD: "the country whose President Duda vetoed the morning-after pill law" (only Poland has President Duda)
   - BAD: "the state with Senate Bill 97" (many states have bills numbered SB 97)
   - BAD: "the country where ellaOne is prescription-only" (many countries require prescriptions)
   - BAD: "the governor who vetoed a healthcare bill" (many governors veto bills)
12. NOTABLE REFERENCE: Use well-known, recognizable facts - not obscure bureaucratic details
   - GOOD: "the bridge where suicide nets were installed" (Golden Gate is famous for this)
   - GOOD: "the US defense secretary hospitalized in January 2024" (major news story)
   - BAD: "the agency that issued guidance document X-123" (bureaucratic reference nobody knows)
   - BAD: "the state that auctioned medical gowns" (obscure detail, not how anyone identifies Ohio)
   - BAD: "the program with 80-hour monthly requirements" (obscure policy detail)
13. USE PAST TENSE for events that already occurred (news describes past events)
   - BAD: "What is the country trying to ban?" (present tense for past event)
   - GOOD: "What did the country try to ban?" (past tense)
14. PREFERRED STRUCTURE: "What is X, according to [indirect reference]?" or "How many X did [indirect reference] Y?"
   - GOOD: "What is the headquarters of the agency that issued COVID-19 guidelines?"
   - GOOD: "How many Nobel Prizes did the scientist who discovered radium win?"
   - BAD: "The agency that issued guidelines, where is it headquartered?" (awkward)

--SCORING (1-5 each)--
- naturalness: Sounds like a real question someone would ask?
- answerability: Fully answerable from the claims ONLY? (critical - don't require outside knowledge)
  - 5: Question has a SPECIFIC, bounded answer from the claims
  - 3: Answer exists in claims but question is too open-ended (e.g., "What did X say about Y?")
  - 1: Requires knowledge not in the claims
- clarity: Easy to understand in one read, simple structure, no nested clauses, grammatically correct?
  - 5: Clear, simple, grammatically perfect question
  - 3: Understandable but awkward phrasing or minor grammar issues
  - 1: Confusing, run-on, or grammatically broken
- bridge_relevance: Is the indirect reference (from the claims) a natural, meaningful way to identify the entity?
  - 5: Notable/significant fact from the claims that clearly identifies the entity
  - 3: Valid fact from claims but less distinctive
  - 1: Obscure/arbitrary fact that nobody would naturally use to identify the entity
- uniqueness: Does the indirect reference uniquely identify the entity IN THE REAL WORLD?
  - 5: Only one entity fits (e.g., "the director of Inception" = only Christopher Nolan)
  - 3: Likely unique but could theoretically match others (e.g., "the governor who vetoed a health bill")
  - 1: Many entities could fit (e.g., "the country where X is prescription-only")

--OUTPUT FORMAT--
Return a JSON object with the following structure:
{
    "questions": [
        {
            "candidate_references": [
                {"replaced_entity": "Entity name", "indirect_reference": "phrase to identify entity"},
                {"replaced_entity": "Entity name", "indirect_reference": "another phrase"}
            ],
            "selected_index": 0,
            "candidate_phrasings": [
                "What is X, according to [indirect reference]?",
                "How many X did [indirect reference] Y?",
                "Where did [indirect reference] Z?"
            ],
            "text": "Bridge question under 20 words (best phrasing from candidates)",
            "replaced_entity": "Entity name replaced (e.g., Christopher Nolan)",
            "indirect_reference": "Phrase used instead (e.g., the director of Inception)",
            "source_claim_ids": ["claim IDs needed to answer"],
            "claim_reasoning": "How the claims together answer the question",
            "draft_answer": "The expected answer to this question based on the claims (1-2 sentences)",
            "question_type": "bridge",
            "quality": {
                "reasoning": "Explain your assessment of the question quality before scoring",
                "naturalness": 4,
                "answerability": 5,
                "clarity": 4,
                "bridge_relevance": 4,
                "uniqueness": 5
            }
        }
    ]
}

Generate UP TO ${max_questions} questions. Generate ZERO if no valid bridge questions possible.
