{
  "n_tasks": 100,
  "n_candidates": 800,
  "generation": {
    "pass@1": 0.771,
    "pass@8_oracle": 0.89
  },
  "verification_intrinsic": {
    "no_think": {
      "auroc": 0.773,
      "balanced_acc": 0.627,
      "raw_acc": 0.814,
      "say_A_rate": 0.912
    },
    "think": {
      "auroc": 0.926,
      "balanced_acc": 0.827,
      "raw_acc": 0.909,
      "say_A_rate": 0.828
    }
  },
  "verifier_selected": {
    "pass@1_random": 0.771,
    "no_think_selected": 0.8,
    "think_selected": 0.86,
    "oracle": 0.89,
    "gap_closed_no_think": 0.242,
    "gap_closed_think": 0.747
  },
  "foreign_control": {
    "mean_pa_nothink": 0.009,
    "reject_rate_nothink": 1.0,
    "mean_pa_think": 0.0,
    "reject_rate_think": 1.0
  },
  "forced_think_frac": 0.374
}