You are grading a predicted answer against a gold answer for a factoid question.

Output exactly one JSON object (no prose, no code fences):
{{
  "judgment": "CORRECT" | "INCORRECT" | "NOT_ATTEMPTED",
  "reason": "<one short sentence>"
}}

Grading rubric:

- **CORRECT**: The predicted answer conveys the same information as the gold answer. Semantic equivalence is allowed (e.g. "Dostoevsky" vs "Fyodor Dostoevsky"; "408" vs "17 × 24 = 408"). Extra accurate detail is fine. Ignore formatting, casing, and whitespace.

- **INCORRECT**: The predicted answer disagrees with the gold answer in any factual way, OR includes the correct answer alongside an incorrect alternative, OR hedges so much that it could be read either way.

- **NOT_ATTEMPTED**: The predicted answer refuses, says "I don't know", says the search results were insufficient, or otherwise declines to commit to an answer. Do NOT grade such responses as correct even if the gold answer is somewhere implied.

Be strict. Hallucinated justifications do not count as attempts — they count as INCORRECT.

Question: {question}

Gold answer: {gold_answer}

Predicted answer: {predicted_answer}
