{
  "kind": "reliability-curves",
  "title": "Does success survive repeated independent use?",
  "description": "Plot pass^k and pass@k at k=1,2,5 for the two synthetic policies. Label task-resampling intervals and preserve the twenty task success counts in the accompanying table.",
  "task_success_counts": {
    "uniform": [
      6,
      10,
      7,
      8,
      7,
      10,
      6,
      9,
      8,
      5,
      8,
      9,
      6,
      8,
      6,
      9,
      7,
      7,
      10,
      7
    ],
    "split": [
      10,
      10,
      10,
      10,
      10,
      10,
      10,
      10,
      10,
      10,
      5,
      5,
      7,
      4,
      5,
      5,
      5,
      6,
      3,
      7
    ]
  },
  "data": {
    "columns": [
      "Policy",
      "k",
      "pass^k",
      "pass@k",
      "Bootstrap low",
      "Bootstrap high"
    ],
    "rows": [
      [
        "uniform",
        1,
        0.765,
        0.765,
        0.705,
        0.825
      ],
      [
        "uniform",
        2,
        0.589,
        0.941,
        0.491,
        0.694
      ],
      [
        "uniform",
        5,
        0.295,
        1.0,
        0.153,
        0.443
      ],
      [
        "split",
        1,
        0.76,
        0.76,
        0.65,
        0.87
      ],
      [
        "split",
        2,
        0.629,
        0.891,
        0.471,
        0.802
      ],
      [
        "split",
        5,
        0.511,
        0.994,
        0.309,
        0.713
      ]
    ],
    "provenance": "Validated offline scripted demonstration; not LLM or published benchmark measurements."
  },
  "nodes": [
    {
      "id": "n",
      "label": "10 trials: 8 successes, 2 failures"
    },
    {
      "id": "all",
      "label": "All-two: C(8,2)/C(10,2)=28/45"
    },
    {
      "id": "one",
      "label": "At-least-one: 1-C(2,2)/C(10,2)=44/45"
    }
  ],
  "edges": [
    {
      "from": "n",
      "to": "all",
      "label": "reliability"
    },
    {
      "from": "n",
      "to": "one",
      "label": "best-of-two"
    }
  ],
  "scope": "Original conceptual diagram and/or validated synthetic demo data; not paper figures or measured LLM performance."
}