{
  "harness": "sj-harness 0.8.3",
  "job": "job:2ee082733d03071290288a4cc05c5cde",
  "kind": "adversarial_review",
  "bundle": "sha256:496cde080901add9e0038c2f28330a12f18556c63ae67e5698024ce5cf88e0b5",
  "claims": [
    {
      "local_id": "C1",
      "claim_id": "claim:7709661604ef7df46b5e13ae1dd4a49dec461e938355bfc85144f12dd6cdddc0",
      "verdict": "minor_issues",
      "significance": "moderate",
      "reason": "Arithmetic and two confirmed source substitutions support the claim as scoped; residual issues are fragile p=0.048 crossing (4/6 LOO include null; Evolut-omit and long-horizon sensitivities do not), ~59% weight in the changed rows, unresolved inputs held fixed, count-model vs unavailable survival reconstruction, and single-agent forest-plot extraction—none an observed numeric contradiction after re-run."
    }
  ]
}
