{
  "context": "Tests whether the agent systematically identifies the specific categories of failure modes enumerated in the skill: wrong happy-path behavior, hidden regressions, environment-specific breakage, incomplete edge-case handling, and stale assumptions in tests or manual verification.",
  "type": "weighted_checklist",
  "checklist": [
    {
      "name": "Identifies race condition risk",
      "description": "The report explicitly identifies the window between the redis.exists() check and the redis.setex() call as a potential failure mode (hidden regression / incomplete fix for the original race condition)",
      "max_score": 14
    },
    {
      "name": "Addresses environment-specific breakage",
      "description": "The report identifies at least one environment-specific risk (e.g. Redis connection failure, Redis being unavailable, network partition between check and set) that could cause incorrect behavior",
      "max_score": 12
    },
    {
      "name": "Addresses edge cases",
      "description": "The report considers at least one edge case beyond the happy path (e.g. empty recipient string, TTL expiring mid-campaign, same email sent across campaign boundary)",
      "max_score": 10
    },
    {
      "name": "Addresses stale assumptions",
      "description": "The report notes that the absence of visible tests does not confirm correctness, or explicitly calls out what assumptions the fix relies on that could be wrong",
      "max_score": 10
    },
    {
      "name": "Happy-path behavior examined",
      "description": "The report also addresses whether the basic happy-path claim holds (first send goes through, duplicate is blocked)",
      "max_score": 8
    },
    {
      "name": "Failure modes framed adversarially",
      "description": "The Attempts To Break It section is organized around ways the claim could fail, not a list of positive confirmations",
      "max_score": 10
    },
    {
      "name": "Valid verdict assigned",
      "description": "The verdict is exactly PASS, PARTIAL, or FAIL",
      "max_score": 8
    },
    {
      "name": "Verdict matches evidence",
      "description": "If the race condition risk is identified (check-then-act pattern), the verdict is NOT PASS — it is at minimum PARTIAL",
      "max_score": 14
    },
    {
      "name": "Correct output sections",
      "description": "Report has Claim, Attempts To Break It, Evidence, and Verdict sections",
      "max_score": 8
    },
    {
      "name": "Claim graded not style",
      "description": "The verdict discussion focuses on whether duplicates are prevented, not on code style, naming, or engineering choices",
      "max_score": 6
    }
  ]
}