{
  "runs": {
    "ts-run1": {
      "established": 16,
      "of": 17,
      "accuracy": 0.9411764705882353,
      "tokens": 286849,
      "perTask": {
        "1": "5/5",
        "2": "5/5",
        "3": "1/2",
        "4": "5/5"
      },
      "iterations": {
        "1": "1",
        "2": "1",
        "3": "7",
        "4": "1"
      },
      "didNotCompile": [],
      "notEstablishedReasons": [
        "vp3-race-at-most-one: interleavings not exhausted (4000 explored, truncated)"
      ],
      "calibration": {
        "1": [],
        "2": [],
        "3": [
          "vp3-race-at-most-one"
        ],
        "4": []
      }
    },
    "bosatsu-run1": {
      "established": 16,
      "of": 17,
      "accuracy": 0.9411764705882353,
      "tokens": 527348,
      "perTask": {
        "1": "5/5",
        "2": "5/5",
        "3": "1/2",
        "4": "5/5"
      },
      "iterations": {
        "1": "9",
        "2": "4",
        "3": "34",
        "4": "10"
      },
      "didNotCompile": [],
      "notEstablishedReasons": [
        "vp3-race-at-most-one: arm A: no dist world reaches complete holds on never-over-cap and one-landed with a-reply-landed covered; saw Demo/ForumDist::forum_dist_spec: never-over-cap=holds,one-landed=holds"
      ],
      "calibration": {
        "1": [],
        "2": [],
        "3": [
          "vp3-race-at-most-one"
        ],
        "4": []
      }
    },
    "ts-run2": {
      "established": 17,
      "of": 17,
      "accuracy": 1.0,
      "tokens": 242753,
      "perTask": {
        "1": "5/5",
        "2": "5/5",
        "3": "2/2",
        "4": "5/5"
      },
      "iterations": {
        "1": "1",
        "2": "1",
        "3": "4",
        "4": "1"
      },
      "didNotCompile": [],
      "notEstablishedReasons": [],
      "calibration": {
        "1": [],
        "2": [],
        "3": [],
        "4": []
      }
    },
    "bosatsu-run2": {
      "established": 16,
      "of": 17,
      "accuracy": 0.9411764705882353,
      "tokens": 482343,
      "perTask": {
        "1": "5/5",
        "2": "5/5",
        "3": "1/2",
        "4": "5/5"
      },
      "iterations": {
        "1": "9",
        "2": "9",
        "3": "84",
        "4": "19"
      },
      "didNotCompile": [],
      "notEstablishedReasons": [
        "vp3-race-at-most-one: arm A: no dist world reaches complete holds on never-over-cap and one-landed with a-reply-landed covered; saw Demo/ForumDist::forum_dist_spec: never-over-cap=holds,one-landed=holds"
      ],
      "calibration": {
        "1": [],
        "2": [],
        "3": [
          "vp3-race-at-most-one"
        ],
        "4": []
      }
    },
    "ts-run3": {
      "established": 17,
      "of": 17,
      "accuracy": 1.0,
      "tokens": 238418,
      "perTask": {
        "1": "5/5",
        "2": "5/5",
        "3": "2/2",
        "4": "5/5"
      },
      "iterations": {
        "1": "1",
        "2": "2",
        "3": "5",
        "4": "5"
      },
      "didNotCompile": [],
      "notEstablishedReasons": [],
      "calibration": {
        "1": [],
        "2": [],
        "3": [],
        "4": []
      }
    },
    "bosatsu-run3": {
      "established": 16,
      "of": 17,
      "accuracy": 0.9411764705882353,
      "tokens": 540819,
      "perTask": {
        "1": "5/5",
        "2": "5/5",
        "3": "1/2",
        "4": "5/5"
      },
      "iterations": {
        "1": "12",
        "2": "18",
        "3": "68",
        "4": "13"
      },
      "didNotCompile": [],
      "notEstablishedReasons": [
        "vp3-race-at-most-one: arm A: no dist world reaches complete holds on never-over-cap and one-landed with a-reply-landed covered; saw Demo/ForumDist::reply_race: never-over-cap=holds,one-landed=holds"
      ],
      "calibration": {
        "1": [],
        "2": [],
        "3": [
          "vp3-race-at-most-one"
        ],
        "4": []
      }
    }
  },
  "pairs": [
    {
      "run": 1,
      "verdict": "fail",
      "artifactAccuracy": 0.9411764705882353,
      "baselineAccuracy": 0.9411764705882353,
      "tokenLift": 0.5439463124919408,
      "reasons": [
        "tokenLift 0.544 does not exceed threshold 1.000"
      ]
    },
    {
      "run": 2,
      "verdict": "fail",
      "artifactAccuracy": 0.9411764705882353,
      "baselineAccuracy": 1.0,
      "tokenLift": 0.5032787870871973,
      "reasons": [
        "accuracy 0.941 below source baseline 1.000",
        "tokenLift 0.503 does not exceed threshold 1.000"
      ]
    },
    {
      "run": 3,
      "verdict": "fail",
      "artifactAccuracy": 0.9411764705882353,
      "baselineAccuracy": 1.0,
      "tokenLift": 0.4408461980810585,
      "reasons": [
        "accuracy 0.941 below source baseline 1.000",
        "tokenLift 0.441 does not exceed threshold 1.000"
      ]
    }
  ]
}