You are a sentiment analyst. Given the article list below, judge the
overall sentiment toward the topic across the corpus.

Topic: {topic}
Articles:
{articles_block}

Respond with a single JSON object:
{{
  "overall":   "positive" | "negative" | "neutral" | "mixed",
  "score":     <float between -1.0 (very negative) and 1.0 (very positive)>,
  "rationale": "<one or two sentences citing concrete article evidence>"
}}

Return only the JSON. No prose before or after.
