{
  "source": "results/pilot_plans.csv.gz",
  "configuration": "results/configuration.json",
  "target_power": 0.8,
  "alpha": 0.05,
  "definition": "Fraction of evaluable plans with observed_iid >= requested_power; point estimates, not guaranteed true power.",
  "rows": [
    {
      "benchmark": "arc",
      "pilot_n": 128,
      "total": 200,
      "evaluated": 191,
      "reached": 77
    },
    {
      "benchmark": "arc",
      "pilot_n": 256,
      "total": 200,
      "evaluated": 197,
      "reached": 73
    },
    {
      "benchmark": "gsm8k",
      "pilot_n": 128,
      "total": 200,
      "evaluated": 196,
      "reached": 45
    },
    {
      "benchmark": "gsm8k",
      "pilot_n": 256,
      "total": 200,
      "evaluated": 197,
      "reached": 31
    },
    {
      "benchmark": "winogrande",
      "pilot_n": 128,
      "total": 200,
      "evaluated": 188,
      "reached": 75
    },
    {
      "benchmark": "winogrande",
      "pilot_n": 256,
      "total": 200,
      "evaluated": 196,
      "reached": 99
    },
    {
      "benchmark": "hellaswag",
      "pilot_n": 128,
      "total": 200,
      "evaluated": 185,
      "reached": 62
    },
    {
      "benchmark": "hellaswag",
      "pilot_n": 256,
      "total": 200,
      "evaluated": 199,
      "reached": 69
    },
    {
      "benchmark": "mmlu",
      "pilot_n": 128,
      "total": 200,
      "evaluated": 196,
      "reached": 93
    },
    {
      "benchmark": "mmlu",
      "pilot_n": 256,
      "total": 200,
      "evaluated": 197,
      "reached": 59
    }
  ]
}