{
  "context": "Tests whether the agent grades the behavioral claim (does the retry logic work?) rather than the code quality (is it well-written?), and whether it runs functional tests rather than relying on code inspection.",
  "type": "weighted_checklist",
  "checklist": [
    {
      "name": "No code quality criticism in verdict",
      "description": "The verdict and the evidence section do NOT base the conclusion on code style, verbosity, naming conventions, or readability — they focus only on whether the retry behavior is correct",
      "max_score": 14
    },
    {
      "name": "Claim-based verdict",
      "description": "The verdict explicitly addresses whether the claimed behaviors hold (retries on 503, exponential backoff, gives up after N attempts) — not whether the code is good",
      "max_score": 12
    },
    {
      "name": "Functional proof executed",
      "description": "The agent actually runs the code (e.g. runs the test file or writes and executes additional tests) rather than only reasoning about it by reading",
      "max_score": 12
    },
    {
      "name": "Backoff calculation verified",
      "description": "The agent checks whether the exponential backoff delays are computed correctly (the implementation uses attempt_number after increment, so delays are 2x, 4x, 8x rather than 1x, 2x, 4x)",
      "max_score": 12
    },
    {
      "name": "Verdict matches actual behavior",
      "description": "If the agent discovers the backoff calculation is off-by-one (or confirms it is correct), the verdict reflects this finding accurately",
      "max_score": 10
    },
    {
      "name": "Attempts are falsification-oriented",
      "description": "Attempts include at least one adversarial test beyond the provided test suite (e.g. verifying exact delay values, testing with max_attempts=1, testing non-503 error codes)",
      "max_score": 10
    },
    {
      "name": "Valid verdict value",
      "description": "Verdict is exactly PASS, PARTIAL, or FAIL",
      "max_score": 8
    },
    {
      "name": "Correct output format",
      "description": "Report contains Claim, Attempts To Break It, Evidence, and Verdict sections",
      "max_score": 8
    },
    {
      "name": "One-sentence claim",
      "description": "The Claim section contains a single sentence summarizing what is being verified",
      "max_score": 8
    },
    {
      "name": "Evidence contains execution output",
      "description": "The Evidence section shows actual output from running the code (test results, printed values) rather than paraphrased descriptions",
      "max_score": 6
    }
  ]
}