{
  "context": "Tests whether the agent correctly issues a FAIL verdict when a concrete counterexample is found, and whether it uses functional execution (running code with adversarial inputs) rather than code inspection to find the failure.",
  "type": "weighted_checklist",
  "checklist": [
    {
      "name": "FAIL verdict issued",
      "description": "The verdict is FAIL (the function has real gaps — e.g. a number with international dial prefix '00' instead of '+' raises ValueError rather than normalizing correctly)",
      "max_score": 14
    },
    {
      "name": "Specific counterexample documented",
      "description": "The Evidence section includes at least one specific input value and the actual output or error produced (not just a description of what might fail)",
      "max_score": 14
    },
    {
      "name": "Counterexample found by running code",
      "description": "The counterexample was discovered by actually executing the function with adversarial inputs, not only by reasoning about it from code inspection",
      "max_score": 12
    },
    {
      "name": "Shortest path used",
      "description": "The agent does not build elaborate test infrastructure before trying simple adversarial inputs — it tries direct inputs quickly to find failure",
      "max_score": 10
    },
    {
      "name": "International formats tested",
      "description": "Attempts include at least one non-US international number format (e.g. using '00' instead of '+' prefix, numbers from non-US/UK countries)",
      "max_score": 10
    },
    {
      "name": "Correct output sections",
      "description": "Report contains Claim, Attempts To Break It, Evidence, and Verdict sections",
      "max_score": 8
    },
    {
      "name": "One-sentence claim",
      "description": "The Claim section is a single sentence",
      "max_score": 6
    },
    {
      "name": "Verdict not based on code quality",
      "description": "The FAIL verdict references a behavioral failure (wrong output or unhandled case), not code style or structural concerns",
      "max_score": 8
    },
    {
      "name": "Remaining Risks or gaps described",
      "description": "Given a FAIL verdict, the report describes what specifically fails or what remaining gaps were found",
      "max_score": 8
    },
    {
      "name": "Falsification-oriented attempts",
      "description": "The Attempts To Break It section lists inputs designed to break the normalizer, not to confirm it works",
      "max_score": 10
    }
  ]
}