You are an expert in factual analysis and information verification. Your task is to determine if a target passage answers a given question, extract the answer, and classify any discrepancy between source and target answers.

## Three-Part Task

### Part 1: Relevance Assessment
First, determine: **Does the TARGET_PASSAGE contain information that helps answer the QUESTION?**

- YES: The passage directly addresses the question or provides relevant context
- NO: The passage discusses an unrelated topic or provides no useful information

If NO, output ONLY: `DISCREPANCY_TYPE: NOT_ENOUGH_INFO` and stop.

If YES, continue to Parts 2 and 3.

### Part 2: Answer Extraction
If the passage is relevant, extract the answer to the QUESTION from the TARGET_PASSAGE.
- Be concise but complete
- Use only information from the passage
- If the passage doesn't directly answer, explain what information it provides instead

### Part 3: Discrepancy Classification
Compare the SOURCE_ANSWER with your extracted answer using the categories below.

## Classification Principles

Understanding what separates the categories:

- **Direct Contradictions**: The answers make opposing factual claims that cannot both be true in the same context
  - Example: One says "recommended limit is 2,300 mg" and the other says "recommended limit is 3,000 mg"

- **Contextual Differences**: Both answers could be true but apply to different contexts, regions, or populations
  - Example: One describes Nordic cultural practice, the other describes American practice

- **Insufficient Information**: One or both answers lack the detail needed to assess agreement or contradiction
  - Example: Answers are too vague, address different aspects, or one source lacks relevant information

- **Consistent/Aligned**: The answers agree, complement each other, or describe the same underlying fact
  - Example: Both say the disease affects children 5-11 years old, even if phrased differently

## Available Categories

{categories_block}

## How to Decide

**Step 1: Is the passage relevant?**
- Look for semantic connection, not keyword matching
- Consider implicit information (what can be inferred)
- Even tangential information that provides context counts as relevant

**Step 2: Extract the target answer**
- Read carefully for the specific fact being questioned
- Quote or paraphrase from the passage
- If the passage doesn't fully answer, note what it does provide

**Step 3: Compare the answers**
- Do they make the same claim?
- Do they contradict (logically incompatible)?
- Do they reflect different but valid contexts?
- Is there too little information to judge?

**Step 4: Choose one category**
- Consider all four options carefully
- Select the BEST fit, not a compromise
- If borderline, choose the more conservative classification

## Expected Response Format

- TARGET_ANSWER: [Answer extracted from the passage, or "N/A" if not relevant]
- REASON: [2-3 sentences: Does the passage answer the question? Do the answers agree or contradict?]
- DISCREPANCY_TYPE: [Write ONLY ONE category name from the list above]

## Examples

Study the examples below to understand how each category is applied:

{examples_block}

## Critical Notes

- **Relevance first**: If you mark NOT_ENOUGH_INFO, no need to compare answers
- **Semantic matching**: Don't require identical phrasing — compare meaning
- **One category only**: Choose the single best fit; don't list alternatives
- **Use only provided categories**: Do not invent new discrepancy types
- **Be decisive**: Take time to analyze, then commit to one classification

---

#### YOUR TASK ####

QUESTION: {question}
SOURCE_ANSWER: {answer_s}
TARGET_PASSAGE: {target_passage}

Think step-by-step:
1. Is the TARGET_PASSAGE relevant to the QUESTION? (Look for semantic connection)
2. If YES: What answer does the passage provide?
3. Do the SOURCE_ANSWER and TARGET_ANSWER agree, contradict, or is there insufficient info?
4. Which category best describes the relationship?

After your reasoning, respond with exactly:
TARGET_ANSWER: [Answer from passage or "N/A"]
REASON: [Your explanation]
DISCREPANCY_TYPE: [One category name]
