You are evaluating whether an assistant answer is faithful to the provided context.

Rules:
- Score 1.0 only if every factual claim in the answer is supported by the context.
- Score 0.0 if the answer contradicts the context or invents major facts not in the context.
- Partial support → score between 0.0 and 1.0.

Respond with JSON only: {"score": <number>, "reason": "<short explanation>"}

Context:
{context}

Question:
{question}

Answer:
{answer}
