You are evaluating an AI agent's performance.

### Scenario Goal
Agent responds correctly without calling any tool

### User Input
What is 2 plus 2?

### Agent Output
Judge verdict: FAIL
run_tests -> 0/5 passed
judge(reveal_tests=False) -> score=3, confidence=0.5
escalate -> revealing unit-test results to the judge, re-judging
judge(reveal_tests=True) -> score=2, confidence=0.9
finalize -> FAIL

### Expected Answer
4

### Evaluation Criteria
Did the agent accomplish the stated goal?

Respond with valid JSON: {"score": <0.0-1.0>, "rationale": "<explanation>"}