{
  "schemaVersion": 1,
  "lessonId": "rl-02",
  "contentHash": "2b067eae4b74d07513fcdf4a0311cb087d4b88923073fdc1669c330e647038ed",
  "rendererHash": "a1bd5b648f4731ae5d5f0a24811c2540d1a0a132149ce589d17c4113509cbedf",
  "kind": "executed",
  "generatedAt": "2026-10-20T10:46:24.417916+00:00",
  "provenance": "recorded_cpu_computation",
  "environment": {
    "backend": "cpu",
    "deviceCount": 1,
    "jax": "0.9.2",
    "numpy": "2.4.4",
    "matplotlib": "3.10.8"
  },
  "stdout": "return mean: 0.5848437547683716\nactive environments by transition: [64 64 58 53 47 44 39 29]\n",
  "data": {
    "kind": "line",
    "x": [
      1,
      2,
      3,
      4,
      5,
      6,
      7,
      8
    ],
    "xlabel": "transition index",
    "ylabel": "active environments before transition",
    "series": [
      {
        "label": "uniform random policy",
        "y": [
          64,
          64,
          58,
          53,
          47,
          44,
          39,
          29
        ]
      }
    ]
  },
  "diagram": {
    "title": "A terminal transition still counts",
    "prediction": "Should every cell after the first done flag be treated identically to the transition that caused done?",
    "reading": "Each row is one environment. A terminating transition can still earn reward; later cells are padding under this absorbing-state contract. Tensor length stays fixed while activity changes. This is a conceptual rollout example, not an auto-reset implementation.",
    "alt": "Each row is one environment. A terminating transition can still earn reward; later cells are padding under this absorbing-state contract. Tensor length stays fixed while activity changes. This is a conceptual rollout example, not an auto-reset implementation.",
    "scope": "Architecture and dataflow mechanism diagram.",
    "height": 4.1,
    "panel": {
      "kind": "table",
      "headers": [
        "Environment",
        "Step 0",
        "Step 1",
        "Step 2",
        "Step 3"
      ],
      "rows": [
        [
          "A",
          "active",
          "terminal\n+ reward",
          "padding",
          "padding"
        ],
        [
          "B",
          "active",
          "active",
          "terminal\n+ reward",
          "padding"
        ],
        [
          "C",
          "active",
          "active",
          "active",
          "active"
        ]
      ]
    }
  }
}