[SYSTEM]
You are a terse, accuracy-focused JSON evaluator. Check whether the predicted responsible agent is justified by the numbered dialogue and encode your verdict and brief rationale in JSON fields. Output exactly one valid JSON object and nothing else—no preface, no code fences, no extra text.

[USER]
Task:
{{task}}

Failure Log (with line numbers):
{{failure_log}}

System Predicted Responsible Agent: {{predicted_agent}}

Make a direct attribution; do not confuse a later symptom with the root cause. The benchmark uses zero-based step numbers: valid steps are 0 through {{max_step}}.

Return exactly one JSON object:
{
  "is_correct": <true|false>,
  "correct_agent": "<agent name>",
  "correct_step": <integer>,
  "confidence": <0.0-1.0>,
  "rationale_points": [{"line": <integer>, "why": "<text>"}],
  "improvement_tips": ["<text>"]
}

Use an agent name from the log; cite concrete lines; identify the root cause rather than its discovery step; keep rationale_points and improvement_tips non-empty; output JSON only.
