{
  "harness": "sj-harness 0.3.1",
  "job": "job:ac2ae45bad5f554cb431f9a065fa0d3d",
  "kind": "adversarial_review",
  "bundle": "sha256:210a3d2ac6a282b19e20bbd0595fffca8b28088c7cab96d453612f426071c7fb",
  "claims": [
    {
      "local_id": "C1",
      "claim_id": "claim:e453823134841beae44eb0ad213b6027bd145ccb7b35a393e162d410424db809",
      "verdict": "minor_issues",
      "significance": "minor",
      "reason": "Numbers exact (independently confirmed with scipy fisher_exact 0.2156707 and 200k-rep Monte Carlo 0.21565±0.0009). Issues: 75-design grid includes 12 duplicate k=1 rows and 11 of 31 exceedances are trivial perfect-copy controls (only 20/45 beta-binomial designs exceed nominal); headline max 0.754 is the perfect-copy control, unlabeled; oracle bound is tautological; R2 booleans cannot be false; misses classic clustered-binary literature (Hurlbert 1984, Williams 1975, Rao-Scott 1992)."
    }
  ]
}