{
 "what": "METR's Time Horizon 1.1 re-decided as certificates: for every agent, in the raw runs of the analysis repository and in the per-task file behind the live chart, the penalised logistic fit proved to hold exactly one zero of its score in a box (Krawczyk, outward-rounded interval arithmetic, certified exp), the 50 % and 80 % horizons as enclosures over the box, the printed coefficients and horizons read against them, and the doubling time as an exact least-squares line through the certified horizons.",
 "generated": "2026-09-25",
 "pins": {
  "corpus/metr-horizon/runs-1-1.jsonl": "609f904f4b6ae32129388da89d036e00bac511ad94223d2be1e58fc2b45b55cd",
  "corpus/metr-horizon/release_dates.yaml": "317b92915df5bf935908567a857116bcf6f0c7686ef8b0840897ed04a29232bc",
  "corpus/metr-horizon/figs-1-1.yaml": "6830ffd2b5dc9922313c2c0120b113cdf31b2334058f0b79bf6d7eebf85de8d1",
  "corpus/metr-horizon/params-1-1.yaml": "2ad30555e1b329ae01897e9f16fab09ecbd2d2f31a320711677860e2cc1656af",
  "corpus/metr-horizon/blog-time-horizon-1-1.html": "a039669837a3bf2c7e39110fbfc795a0b47d5b87154aa6cdd367da6c2630ad2c",
  "corpus/metr-horizon/benchmark_results_1_1.yaml": "aae31902b0519a4da73e16643915e5e8aca13cd3315c3aac893ce3d6dfe92ad9",
  "corpus/metr-horizon/task_results_1_1.yaml": "4f85fbf03cbf8b43af102ff17db6bbb4963b3c98784e90f81fb71a146cf26b07"
 },
 "estimator": {
  "weighting": "invsqrt_task_weight (re-derived per agent from the task families, normalised to 1)",
  "regularization": "1/100000",
  "x": "log2(human minutes), the literal read as the rational it denotes and enclosed",
  "y": "successes / runs per task, exact — the same score as METR's per-run form",
  "aggregation": "per task, exact: Σ_runs (w/n)(y_r − p) = w(ȳ − p)",
  "weightsNote": "1/√k for a family of k tasks is the correctly rounded double, taken as a rational; the file's own weights agree to 2.9e-09"
 },
 "agents": {
  "Claude 3 Opus (Inspect)": {
   "alias": "Claude 3 Opus (Inspect)",
   "key": "claude_3_opus_inspect",
   "release_date": "2024-03-04",
   "runsFile": {
    "n_tasks": 228,
    "candidate": {
     "slope": -0.5273557319006651,
     "intercept": 1.0455911105331506,
     "newtonIterations": 6
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.527355731901674,
      -0.5273557318996562
     ],
     "intercept": [
      1.045591110532066,
      1.045591110534235
     ],
     "maxRad": 1.9872992140790302e-14
    },
    "horizons": {
     "0.5": {
      "minutes": [
       3.9523357283233,
       3.9523357283553597
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       0.6390232674029156,
       0.6390232674058335
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 1038
   },
   "live": {
    "n_tasks": 228,
    "candidate": {
     "slope": -0.5273463928274337,
     "intercept": 1.0455560398714971,
     "newtonIterations": 6
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.5273463928284428,
      -0.5273463928264246
     ],
     "intercept": [
      1.045556039870412,
      1.0455560398725823
     ],
     "maxRad": 2.0095036745715333e-14
    },
    "horizons": {
     "0.5": {
      "minutes": [
       3.952249731518153,
       3.952249731550223
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       0.6389887433913306,
       0.6389887433942498
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 1038,
    "printed": {
     "coefficient": -0.527,
     "intercept": 1.046
    },
    "coefficientsVerdict": "REPRODUCED",
    "coefficientsCertifiedRounded": {
     "coefficient": -0.527,
     "intercept": 1.046
    }
   },
   "site": {
    "is_sota": false,
    "release_date": "2024-03-04",
    "p50": {
     "estimate": 3.952262,
     "ci": [
      1.706313,
      8.76484
     ],
     "certified": [
      3.952249731518153,
      3.952249731550223
     ],
     "inside": false,
     "relativeGap": 3.1041767805014565e-06,
     "insideBootstrapCI": true
    },
    "p80": {
     "estimate": 0.638973,
     "ci": [
      0.190491,
      2.118276
     ],
     "certified": [
      0.6389887433913306,
      0.6389887433942498
     ],
     "inside": false,
     "relativeGap": 2.463798370884661e-05,
     "insideBootstrapCI": true
    }
   }
  },
  "Claude 3.5 Sonnet (New) (Inspect)": {
   "alias": "Claude 3.5 Sonnet (New) (Inspect)",
   "key": "claude_3_5_sonnet_20241022_inspect",
   "release_date": "2024-10-22",
   "runsFile": {
    "n_tasks": 226,
    "candidate": {
     "slope": -0.4647122507110154,
     "intercept": 2.0257644463651605,
     "newtonIterations": 7
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.46471225071202454,
      -0.46471225071000627
     ],
     "intercept": [
      2.0257644463630817,
      2.0257644463672393
     ],
     "maxRad": 2.731148640577885e-14
    },
    "horizons": {
     "0.5": {
      "minutes": [
       20.523149064049587,
       20.52314906444628
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       2.5955782994650187,
       2.5955782994918732
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 1016
   },
   "live": {
    "n_tasks": 226,
    "candidate": {
     "slope": -0.46470671630945665,
     "intercept": 2.0257353064751067,
     "newtonIterations": 7
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.46470671631046584,
      -0.46470671630844745
     ],
     "intercept": [
      2.025735306473028,
      2.0257353064771855
     ],
     "maxRad": 2.731148640577885e-14
    },
    "horizons": {
     "0.5": {
      "minutes": [
       20.522995563121167,
       20.522995563517878
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       2.595494969428549,
       2.595494969455404
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 1016,
    "printed": {
     "coefficient": -0.465,
     "intercept": 2.026
    },
    "coefficientsVerdict": "REPRODUCED",
    "coefficientsCertifiedRounded": {
     "coefficient": -0.465,
     "intercept": 2.026
    }
   },
   "site": {
    "is_sota": true,
    "release_date": "2024-10-22",
    "p50": {
     "estimate": 20.522872,
     "ci": [
      10.144026,
      40.82028
     ],
     "certified": [
      20.522995563121167,
      20.522995563517878
     ],
     "inside": false,
     "relativeGap": 6.0207350090768005e-06,
     "insideBootstrapCI": true
    },
    "p80": {
     "estimate": 2.595677,
     "ci": [
      0.89379,
      7.363864
     ],
     "certified": [
      2.595494969428549,
      2.595494969455404
     ],
     "inside": false,
     "relativeGap": 7.013327846707733e-05,
     "insideBootstrapCI": true
    }
   }
  },
  "Claude 3.5 Sonnet (Old) (Inspect)": {
   "alias": "Claude 3.5 Sonnet (Old) (Inspect)",
   "key": "claude_3_5_sonnet_20240620_inspect",
   "release_date": "2024-06-20",
   "runsFile": {
    "n_tasks": 223,
    "candidate": {
     "slope": -0.5006564120427703,
     "intercept": 1.7574199152617256,
     "newtonIterations": 6
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.5006564120437816,
      -0.500656412041759
     ],
     "intercept": [
      1.7574199152599104,
      1.7574199152635408
     ],
     "maxRad": 2.942091015256665e-14
    },
    "horizons": {
     "0.5": {
      "minutes": [
       11.394229887754372,
       11.39422988792369
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       1.6716559377640443,
       1.67165593777592
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 1005
   },
   "live": {
    "n_tasks": 223,
    "candidate": {
     "slope": -0.5006548734127971,
     "intercept": 1.7574113357009682,
     "newtonIterations": 6
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.5006548734138084,
      -0.5006548734117858
     ],
     "intercept": [
      1.7574113356991523,
      1.757411335702784
     ],
     "maxRad": 2.9531932455029164e-14
    },
    "horizons": {
     "0.5": {
      "minutes": [
       11.394179744936505,
       11.394179745105841
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       1.6716387211907742,
       1.6716387212026524
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 1005,
    "printed": {
     "coefficient": -0.501,
     "intercept": 1.757
    },
    "coefficientsVerdict": "REPRODUCED",
    "coefficientsCertifiedRounded": {
     "coefficient": -0.501,
     "intercept": 1.757
    }
   },
   "site": {
    "is_sota": true,
    "release_date": "2024-06-20",
    "p50": {
     "estimate": 11.395377,
     "ci": [
      5.489734,
      22.384214
     ],
     "certified": [
      11.394179744936505,
      11.394179745105841
     ],
     "inside": false,
     "relativeGap": 0.00010507602041534917,
     "insideBootstrapCI": true
    },
    "p80": {
     "estimate": 1.671757,
     "ci": [
      0.57927,
      4.573519
     ],
     "certified": [
      1.6716387211907742,
      1.6716387212026524
     ],
     "inside": false,
     "relativeGap": 7.075620331498095e-05,
     "insideBootstrapCI": true
    }
   }
  },
  "Claude 3.7 Sonnet (Inspect)": {
   "alias": "Claude 3.7 Sonnet (Inspect)",
   "key": "claude_3_7_sonnet_inspect",
   "release_date": "2025-02-24",
   "runsFile": {
    "n_tasks": 228,
    "candidate": {
     "slope": -0.5974673077547866,
     "intercept": 3.534725851326118,
     "newtonIterations": 7
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.5974673077558059,
      -0.5974673077537673
     ],
     "intercept": [
      3.534725851322458,
      3.5347258513297777
     ],
     "maxRad": 6.328271240363392e-14
    },
    "horizons": {
     "0.5": {
      "minutes": [
       60.38770205065293,
       60.387702052011086
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       12.091304433142042,
       12.091304433347599
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 1043
   },
   "live": {
    "n_tasks": 228,
    "candidate": {
     "slope": -0.5974549266690904,
     "intercept": 3.5346505939822412,
     "newtonIterations": 7
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.5974549266701099,
      -0.5974549266680709
     ],
     "intercept": [
      3.5346505939785797,
      3.5346505939859028
     ],
     "maxRad": 6.417089082333405e-14
    },
    "horizons": {
     "0.5": {
      "minutes": [
       60.387561330354124,
       60.387561331712654
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       12.090873275367516,
       12.09087327557314
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 1043,
    "printed": {
     "coefficient": -0.597,
     "intercept": 3.535
    },
    "coefficientsVerdict": "REPRODUCED",
    "coefficientsCertifiedRounded": {
     "coefficient": -0.597,
     "intercept": 3.535
    }
   },
   "site": {
    "is_sota": true,
    "release_date": "2025-02-24",
    "p50": {
     "estimate": 60.388937,
     "ci": [
      33.006168,
      104.226017
     ],
     "certified": [
      60.387561330354124,
      60.387561331712654
     ],
     "inside": false,
     "relativeGap": 2.27806789273791e-05,
     "insideBootstrapCI": true
    },
    "p80": {
     "estimate": 12.09179,
     "ci": [
      4.575608,
      28.889879
     ],
     "certified": [
      12.090873275367516,
      12.09087327557314
     ],
     "inside": false,
     "relativeGap": 7.58195550979397e-05,
     "insideBootstrapCI": true
    }
   },
   "post": {
    "th11": [
     60,
     32,
     106
    ],
    "certifiedRounded": 60.0,
    "verdict": "REPRODUCED",
    "relativeGap": 0.0064179000193243406
   }
  },
  "Claude 4 Opus (Inspect)": {
   "alias": "Claude 4 Opus (Inspect)",
   "key": "claude_4_opus_inspect",
   "release_date": "2025-05-22",
   "runsFile": {
    "n_tasks": 228,
    "candidate": {
     "slope": -0.6036223028133756,
     "intercept": 4.013617494723423,
     "newtonIterations": 7
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.6036223028144106,
      -0.6036223028123405
     ],
     "intercept": [
      4.013617494719158,
      4.013617494727688
     ],
     "maxRad": 8.615330671091215e-14
    },
    "horizons": {
     "0.5": {
      "minutes": [
       100.37248373111936,
       100.3724837336897
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       20.42967757569275,
       20.42967757610434
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 1368
   },
   "live": {
    "n_tasks": 228,
    "candidate": {
     "slope": -0.6036241351955829,
     "intercept": 4.013629387286679,
     "newtonIterations": 8
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.603624135196607,
      -0.6036241351945588
     ],
     "intercept": [
      4.013629387282496,
      4.013629387290862
     ],
     "maxRad": 8.615330671091215e-14
    },
    "horizons": {
     "0.5": {
      "minutes": [
       100.37245015062868,
       100.37245015316327
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       20.429769465852456,
       20.429769466257977
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 1368,
    "printed": {
     "coefficient": -0.604,
     "intercept": 4.014
    },
    "coefficientsVerdict": "REPRODUCED",
    "coefficientsCertifiedRounded": {
     "coefficient": -0.604,
     "intercept": 4.014
    }
   },
   "site": {
    "is_sota": false,
    "release_date": "2025-05-22",
    "p50": {
     "estimate": 100.366123,
     "ci": [
      59.994073,
      163.454074
     ],
     "certified": [
      100.37245015062868,
      100.37245015316327
     ],
     "inside": false,
     "relativeGap": 6.303675115805413e-05,
     "insideBootstrapCI": true
    },
    "p80": {
     "estimate": 20.429752,
     "ci": [
      8.483125,
      41.937268
     ],
     "certified": [
      20.429769465852456,
      20.429769466257977
     ],
     "inside": false,
     "relativeGap": 8.549415109717258e-07,
     "insideBootstrapCI": true
    }
   },
   "post": {
    "th11": [
     101,
     58,
     170
    ],
    "certifiedRounded": 100.0,
    "verdict": "DIFFERS",
    "relativeGap": 0.006252212107548821
   }
  },
  "Claude 4.1 Opus (Inspect)": {
   "alias": "Claude 4.1 Opus (Inspect)",
   "key": "claude_4_1_opus_inspect",
   "release_date": "2025-08-05",
   "runsFile": {
    "n_tasks": 228,
    "candidate": {
     "slope": -0.6605183465544667,
     "intercept": 4.392874287481577,
     "newtonIterations": 8
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.6605183465554946,
      -0.6605183465534389
     ],
     "intercept": [
      4.392874287476989,
      4.3928742874861655
     ],
     "maxRad": 9.992007221626409e-14
    },
    "horizons": {
     "0.5": {
      "minutes": [
       100.47180500067253,
       100.47180500308221
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       23.45541673984003,
       23.45541674029634
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 1043
   },
   "live": {
    "n_tasks": 228,
    "candidate": {
     "slope": -0.6605181419595312,
     "intercept": 4.3928727769629585,
     "newtonIterations": 8
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.660518141960559,
      -0.6605181419585033
     ],
     "intercept": [
      4.39287277695837,
      4.392872776967547
     ],
     "maxRad": 9.947598300641403e-14
    },
    "horizons": {
     "0.5": {
      "minutes": [
       100.47178920334922,
       100.4717892057589
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       23.4554024825407,
       23.45540248299701
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 1043,
    "printed": {
     "coefficient": -0.661,
     "intercept": 4.393
    },
    "coefficientsVerdict": "REPRODUCED",
    "coefficientsCertifiedRounded": {
     "coefficient": -0.661,
     "intercept": 4.393
    }
   },
   "site": {
    "is_sota": false,
    "release_date": "2025-08-05",
    "p50": {
     "estimate": 100.472004,
     "ci": [
      59.272391,
      159.451918
     ],
     "certified": [
      100.47178920334922,
      100.4717892057589
     ],
     "inside": false,
     "relativeGap": 2.137880219748818e-06,
     "insideBootstrapCI": true
    },
    "p80": {
     "estimate": 23.455761,
     "ci": [
      10.538456,
      47.785951
     ],
     "certified": [
      23.4554024825407,
      23.45540248299701
     ],
     "inside": false,
     "relativeGap": 1.528506959398658e-05,
     "insideBootstrapCI": true
    }
   }
  },
  "Claude Mythos Preview (early)": {
   "alias": "Claude Mythos Preview (early)",
   "key": "claude_mythos_preview_early_inspect",
   "release_date": null,
   "live": {
    "n_tasks": 228,
    "candidate": {
     "slope": -0.5566660462156086,
     "intercept": 5.582691030259519,
     "newtonIterations": 11
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.5566660462166951,
      -0.556666046214522
     ],
     "intercept": [
      5.582691030253215,
      5.582691030265823
     ],
     "maxRad": 3.5926817076870066e-13
    },
    "horizons": {
     "0.5": {
      "minutes": [
       1044.645302842901,
       1044.6453028876606
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       185.90806300191488,
       185.90806300862752
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 1149,
    "printed": {
     "coefficient": -0.557,
     "intercept": 5.582
    },
    "coefficientsVerdict": "DIFFERS",
    "coefficientsCertifiedRounded": {
     "coefficient": -0.557,
     "intercept": 5.583
    }
   },
   "site": {
    "is_sota": true,
    "release_date": "2026-04-07",
    "p50": {
     "estimate": 1044.780145,
     "ci": [
      508.876789,
      3304.261235
     ],
     "certified": [
      1044.645302842901,
      1044.6453028876606
     ],
     "inside": false,
     "relativeGap": 0.0001290793695544712,
     "insideBootstrapCI": true
    },
    "p80": {
     "estimate": 185.911829,
     "ci": [
      97.30292,
      398.514614
     ],
     "certified": [
      185.90806300191488,
      185.90806300862752
     ],
     "inside": false,
     "relativeGap": 2.0257314418310062e-05,
     "insideBootstrapCI": true
    }
   }
  },
  "Claude Opus 4.5 (Inspect)": {
   "alias": "Claude Opus 4.5 (Inspect)",
   "key": "claude_opus_4_5_inspect",
   "release_date": "2025-11-24",
   "runsFile": {
    "n_tasks": 228,
    "candidate": {
     "slope": -0.5399734495013105,
     "intercept": 4.4249567953850875,
     "newtonIterations": 9
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.5399734495023518,
      -0.5399734495002693
     ],
     "intercept": [
      4.424956795380344,
      4.424956795389831
     ],
     "maxRad": 1.603162047558726e-13
    },
    "horizons": {
     "0.5": {
      "minutes": [
       293.0021321529904,
       293.00213216298016
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       49.433917764373064,
       49.43391776571908
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 1044
   },
   "live": {
    "n_tasks": 228,
    "candidate": {
     "slope": -0.5399699968540652,
     "intercept": 4.424931108657104,
     "newtonIterations": 9
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.539969996855107,
      -0.5399699968530234
     ],
     "intercept": [
      4.424931108652357,
      4.42493110866185
     ],
     "maxRad": 1.6298074001497298e-13
    },
    "horizons": {
     "0.5": {
      "minutes": [
       293.00311266714255,
       293.0031126771379
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       49.43352070164215,
       49.43352070298888
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 1044,
    "printed": {
     "coefficient": -0.54,
     "intercept": 4.425
    },
    "coefficientsVerdict": "REPRODUCED",
    "coefficientsCertifiedRounded": {
     "coefficient": -0.54,
     "intercept": 4.425
    }
   },
   "site": {
    "is_sota": true,
    "release_date": "2025-11-24",
    "p50": {
     "estimate": 292.994594,
     "ci": [
      161.717714,
      623.704698
     ],
     "certified": [
      293.00311266714255,
      293.0031126771379
     ],
     "inside": false,
     "relativeGap": 2.90736745433936e-05,
     "insideBootstrapCI": true
    },
    "p80": {
     "estimate": 49.430584,
     "ci": [
      20.660485,
      104.641318
     ],
     "certified": [
      49.43352070164215,
      49.43352070298888
     ],
     "inside": false,
     "relativeGap": 5.9407117825991594e-05,
     "insideBootstrapCI": true
    }
   },
   "post": {
    "th11": [
     320,
     170,
     729
    ],
    "certifiedRounded": 293.0,
    "verdict": "DIFFERS",
    "relativeGap": 0.09213856836418087
   }
  },
  "Claude Opus 4.6 (Inspect)": {
   "alias": "Claude Opus 4.6 (Inspect)",
   "key": "claude_opus_4_6_inspect",
   "release_date": "2026-02-05",
   "runsFile": {
    "n_tasks": 228,
    "candidate": {
     "slope": -0.4122131948469003,
     "intercept": 3.9117777822693185,
     "newtonIterations": 10
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.4122131948479397,
      -0.41221319484586083
     ],
     "intercept": [
      3.9117777822650996,
      3.9117777822735373
     ],
     "maxRad": 1.5409895581797173e-13
    },
    "horizons": {
     "0.5": {
      "minutes": [
       718.9241210959478,
       718.9241211300028
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       69.87206223551986,
       69.8720622380082
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 1391
   },
   "live": {
    "n_tasks": 228,
    "candidate": {
     "slope": -0.4122079302734472,
     "intercept": 3.9117384898730125,
     "newtonIterations": 10
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.41220793027448627,
      -0.41220793027240815
     ],
     "intercept": [
      3.9117384898687972,
      3.911738489877228
     ],
     "maxRad": 1.5276668818842154e-13
    },
    "horizons": {
     "0.5": {
      "minutes": [
       718.9370163639824,
       718.9370163980213
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       69.87123529635896,
       69.87123529884589
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 1391,
    "printed": {
     "coefficient": -0.412,
     "intercept": 3.912
    },
    "coefficientsVerdict": "REPRODUCED",
    "coefficientsCertifiedRounded": {
     "coefficient": -0.412,
     "intercept": 3.912
    }
   },
   "site": {
    "is_sota": true,
    "release_date": "2026-02-05",
    "p50": {
     "estimate": 718.80683,
     "ci": [
      316.685725,
      3633.786163
     ],
     "certified": [
      718.9370163639824,
      718.9370163980213
     ],
     "inside": false,
     "relativeGap": 0.00018108178471567326,
     "insideBootstrapCI": true
    },
    "p80": {
     "estimate": 69.874587,
     "ci": [
      27.026521,
      170.437873
     ],
     "certified": [
      69.87123529635896,
      69.87123529884589
     ],
     "inside": false,
     "relativeGap": 4.796972068445859e-05,
     "insideBootstrapCI": true
    }
   }
  },
  "GPT-4 0314": {
   "alias": "GPT-4 0314",
   "key": "gpt_4",
   "release_date": "2023-03-14",
   "runsFile": {
    "n_tasks": 213,
    "candidate": {
     "slope": -0.6405311544122474,
     "intercept": 1.2780931404970903,
     "newtonIterations": 6
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.6405311544132594,
      -0.6405311544112354
     ],
     "intercept": [
      1.2780931404957652,
      1.2780931404984155
     ],
     "maxRad": 2.3425705819590803e-14
    },
    "horizons": {
     "0.5": {
      "minutes": [
       3.987168350391902,
       3.987168350420771
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       0.8895057199373386,
       0.8895057199402199
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 1401
   },
   "live": {
    "n_tasks": 213,
    "candidate": {
     "slope": -0.6405172006053937,
     "intercept": 1.278045901117951,
     "newtonIterations": 6
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.6405172006064055,
      -0.640517200604382
     ],
     "intercept": [
      1.2780459011166267,
      1.2780459011192753
     ],
     "maxRad": 2.3647750424515834e-14
    },
    "horizons": {
     "0.5": {
      "minutes": [
       3.987084659752489,
       3.987084659781346
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       0.8894579798822982,
       0.8894579798851778
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 1401,
    "printed": {
     "coefficient": -0.641,
     "intercept": 1.278
    },
    "coefficientsVerdict": "REPRODUCED",
    "coefficientsCertifiedRounded": {
     "coefficient": -0.641,
     "intercept": 1.278
    }
   },
   "site": {
    "is_sota": true,
    "release_date": "2023-03-14",
    "p50": {
     "estimate": 3.987428,
     "ci": [
      1.93292,
      7.995283
     ],
     "certified": [
      3.987084659752489,
      3.987084659781346
     ],
     "inside": false,
     "relativeGap": 8.611310689659419e-05,
     "insideBootstrapCI": true
    },
    "p80": {
     "estimate": 0.889561,
     "ci": [
      0.342984,
      2.523746
     ],
     "certified": [
      0.8894579798822982,
      0.8894579798851778
     ],
     "inside": false,
     "relativeGap": 0.00011582347905342824,
     "insideBootstrapCI": true
    }
   },
   "post": {
    "th11": [
     3.5,
     1.6,
     6.9
    ],
    "certifiedRounded": 4.0,
    "verdict": "DIFFERS",
    "relativeGap": 0.12216561757065678
   }
  },
  "GPT-4 1106 (Inspect)": {
   "alias": "GPT-4 1106 (Inspect)",
   "key": "gpt_4_1106_inspect",
   "release_date": "2023-11-06",
   "runsFile": {
    "n_tasks": 228,
    "candidate": {
     "slope": -0.5851899175106726,
     "intercept": 1.1797933324211511,
     "newtonIterations": 6
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.5851899175116843,
      -0.585189917509661
     ],
     "intercept": [
      1.179793332419924,
      1.1797933324223784
     ],
     "maxRad": 2.3869795029440866e-14
    },
    "horizons": {
     "0.5": {
      "minutes": [
       4.044850066609324,
       4.044850066640637
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       0.7830201554114583,
       0.7830201554143982
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 1029
   },
   "live": {
    "n_tasks": 228,
    "candidate": {
     "slope": -0.5851825772024222,
     "intercept": 1.1797659740439024,
     "newtonIterations": 6
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.5851825772034338,
      -0.5851825772014105
     ],
     "intercept": [
      1.1797659740426756,
      1.1797659740451292
     ],
     "maxRad": 2.375877272697835e-14
    },
    "horizons": {
     "0.5": {
      "minutes": [
       4.044789892184809,
       4.044789892216119
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       0.7829923790049514,
       0.7829923790078904
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 1029,
    "printed": {
     "coefficient": -0.585,
     "intercept": 1.18
    },
    "coefficientsVerdict": "REPRODUCED",
    "coefficientsCertifiedRounded": {
     "coefficient": -0.585,
     "intercept": 1.18
    }
   },
   "site": {
    "is_sota": true,
    "release_date": "2023-11-06",
    "p50": {
     "estimate": 4.044959,
     "ci": [
      1.866859,
      8.443226
     ],
     "certified": [
      4.044789892184809,
      4.044789892216119
     ],
     "inside": false,
     "relativeGap": 4.180880087685809e-05,
     "insideBootstrapCI": true
    },
    "p80": {
     "estimate": 0.783032,
     "ci": [
      0.276599,
      2.358414
     ],
     "certified": [
      0.7829923790049514,
      0.7829923790078904
     ],
     "inside": false,
     "relativeGap": 5.0602018756373456e-05,
     "insideBootstrapCI": true
    }
   },
   "post": {
    "th11": [
     3.6,
     1.6,
     7.5
    ],
    "certifiedRounded": 4.0,
    "verdict": "DIFFERS",
    "relativeGap": 0.10996613026999227
   }
  },
  "GPT-4 Turbo (Inspect)": {
   "alias": "GPT-4 Turbo (Inspect)",
   "key": "gpt_4_turbo_inspect",
   "release_date": "2024-04-09",
   "runsFile": {
    "n_tasks": 228,
    "candidate": {
     "slope": -0.6903304254564744,
     "intercept": 1.3117902820505152,
     "newtonIterations": 6
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.6903304254574885,
      -0.6903304254554603
     ],
     "intercept": [
      1.3117902820491516,
      1.3117902820518788
     ],
     "maxRad": 2.609024107869118e-14
    },
    "horizons": {
     "0.5": {
      "minutes": [
       3.732740863211591,
       3.732740863236265
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       0.9279215577544574,
       0.9279215577572033
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 1032
   },
   "live": {
    "n_tasks": 228,
    "candidate": {
     "slope": -0.6903239774027946,
     "intercept": 1.3117673325913817,
     "newtonIterations": 6
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.6903239774038085,
      -0.6903239774017808
     ],
     "intercept": [
      1.3117673325900185,
      1.3117673325927448
     ],
     "maxRad": 2.6201263381153694e-14
    },
    "horizons": {
     "0.5": {
      "minutes": [
       3.732700772351538,
       3.732700772376206
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       0.927899527238477,
       0.927899527241222
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 1032,
    "printed": {
     "coefficient": -0.69,
     "intercept": 1.312
    },
    "coefficientsVerdict": "REPRODUCED",
    "coefficientsCertifiedRounded": {
     "coefficient": -0.69,
     "intercept": 1.312
    }
   },
   "site": {
    "is_sota": false,
    "release_date": "2024-04-09",
    "p50": {
     "estimate": 3.732787,
     "ci": [
      1.980046,
      6.736613
     ],
     "certified": [
      3.732700772351538,
      3.732700772376206
     ],
     "inside": false,
     "relativeGap": 2.3100605625981537e-05,
     "insideBootstrapCI": true
    },
    "p80": {
     "estimate": 0.927933,
     "ci": [
      0.428277,
      2.196806
     ],
     "certified": [
      0.927899527238477,
      0.927899527241222
     ],
     "inside": false,
     "relativeGap": 3.6073691752606557e-05,
     "insideBootstrapCI": true
    }
   }
  },
  "GPT-4o (Inspect)": {
   "alias": "GPT-4o (Inspect)",
   "key": "gpt_4o_inspect",
   "release_date": "2024-05-13",
   "runsFile": {
    "n_tasks": 228,
    "candidate": {
     "slope": -0.5625928568729721,
     "intercept": 1.5783152365583941,
     "newtonIterations": 6
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.562592856873984,
      -0.5625928568719603
     ],
     "intercept": [
      1.5783152365567625,
      1.5783152365600257
     ],
     "maxRad": 2.708944180085382e-14
    },
    "horizons": {
     "0.5": {
      "minutes": [
       6.990669532226873,
       6.990669532303899
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       1.2669100535156914,
       1.266910053521865
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 1044
   },
   "live": {
    "n_tasks": 228,
    "candidate": {
     "slope": -0.5625882694741298,
     "intercept": 1.5782952845034333,
     "newtonIterations": 6
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.5625882694751418,
      -0.5625882694731177
     ],
     "intercept": [
      1.5782952845018006,
      1.578295284505066
     ],
     "maxRad": 2.7200464103316335e-14
    },
    "horizons": {
     "0.5": {
      "minutes": [
       6.990608531684881,
       6.990608531761936
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       1.2668813542850172,
       1.2668813542911945
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 1044,
    "printed": {
     "coefficient": -0.563,
     "intercept": 1.578
    },
    "coefficientsVerdict": "REPRODUCED",
    "coefficientsCertifiedRounded": {
     "coefficient": -0.563,
     "intercept": 1.578
    }
   },
   "site": {
    "is_sota": true,
    "release_date": "2024-05-13",
    "p50": {
     "estimate": 6.991195,
     "ci": [
      4.001482,
      12.905741
     ],
     "certified": [
      6.990608531684881,
      6.990608531761936
     ],
     "inside": false,
     "relativeGap": 8.389374293542012e-05,
     "insideBootstrapCI": true
    },
    "p80": {
     "estimate": 1.267009,
     "ci": [
      0.562997,
      3.01517
     ],
     "certified": [
      1.2668813542850172,
      1.2668813542911945
     ],
     "inside": false,
     "relativeGap": 0.00010075585574849939,
     "insideBootstrapCI": true
    }
   }
  },
  "GPT-5 (Inspect)": {
   "alias": "GPT-5 (Inspect)",
   "key": "gpt_5_2025_08_07_inspect",
   "release_date": "2025-08-07",
   "runsFile": {
    "n_tasks": 228,
    "candidate": {
     "slope": -0.5762752008794506,
     "intercept": 4.417325437800079,
     "newtonIterations": 8
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.5762752008804888,
      -0.5762752008784124
     ],
     "intercept": [
      4.417325437795374,
      4.4173254378047835
     ],
     "maxRad": 1.4432899320127035e-13
    },
    "horizons": {
     "0.5": {
      "minutes": [
       202.99571298966111,
       202.99571299584602
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       38.31118532629268,
       38.31118532722973
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 1056
   },
   "live": {
    "n_tasks": 228,
    "candidate": {
     "slope": -0.5762749706471274,
     "intercept": 4.417323640519252,
     "newtonIterations": 8
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.5762749706481657,
      -0.5762749706460891
     ],
     "intercept": [
      4.417323640514546,
      4.417323640523959
     ],
     "maxRad": 1.4521717162097048e-13
    },
    "horizons": {
     "0.5": {
      "minutes": [
       202.99570505925905,
       202.9957050654454
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       38.311158307698484,
       38.311158308635754
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 1056,
    "printed": {
     "coefficient": -0.576,
     "intercept": 4.417
    },
    "coefficientsVerdict": "REPRODUCED",
    "coefficientsCertifiedRounded": {
     "coefficient": -0.576,
     "intercept": 4.417
    }
   },
   "site": {
    "is_sota": true,
    "release_date": "2025-08-07",
    "p50": {
     "estimate": 203.012577,
     "ci": [
      112.641357,
      405.551565
     ],
     "certified": [
      202.99570505925905,
      202.9957050654454
     ],
     "inside": false,
     "relativeGap": 8.311476706374613e-05,
     "insideBootstrapCI": true
    },
    "p80": {
     "estimate": 38.312431,
     "ci": [
      19.184874,
      70.093125
     ],
     "certified": [
      38.311158307698484,
      38.311158308635754
     ],
     "inside": false,
     "relativeGap": 3.3219885739046854e-05,
     "insideBootstrapCI": true
    }
   },
   "post": {
    "th11": [
     214,
     117,
     480
    ],
    "certifiedRounded": 203.0,
    "verdict": "DIFFERS",
    "relativeGap": 0.05420949637465336
   }
  },
  "GPT-5.1-Codex-Max (Inspect)": {
   "alias": "GPT-5.1-Codex-Max (Inspect)",
   "key": "gpt_5_1_codex_max_inspect",
   "release_date": "2025-11-19",
   "runsFile": {
    "n_tasks": 228,
    "candidate": {
     "slope": -0.6467002577105011,
     "intercept": 5.047879894838223,
     "newtonIterations": 8
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.6467002577115397,
      -0.6467002577094625
     ],
     "intercept": [
      5.047879894832875,
      5.047879894843571
     ],
     "maxRad": 1.5010215292932116e-13
    },
    "horizons": {
     "0.5": {
      "minutes": [
       223.7268156592879,
       223.726815665742
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       50.631087294010165,
       50.631087295229115
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 1369
   },
   "live": {
    "n_tasks": 228,
    "candidate": {
     "slope": -0.646700112023795,
     "intercept": 5.047878733855931,
     "newtonIterations": 8
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.6467001120248339,
      -0.646700112022756
     ],
     "intercept": [
      5.047878733850583,
      5.04787873386128
     ],
     "maxRad": 1.5143442055887135e-13
    },
    "horizons": {
     "0.5": {
      "minutes": [
       223.72680994953257,
       223.72680995598867
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       50.63106905409927,
       50.631069055318534
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 1369,
    "printed": {
     "coefficient": -0.647,
     "intercept": 5.048
    },
    "coefficientsVerdict": "REPRODUCED",
    "coefficientsCertifiedRounded": {
     "coefficient": -0.647,
     "intercept": 5.048
    }
   },
   "site": {
    "is_sota": false,
    "release_date": "2025-11-19",
    "p50": {
     "estimate": 223.714694,
     "ci": [
      134.31077,
      396.206723
     ],
     "certified": [
      223.72680994953257,
      223.72680995598867
     ],
     "inside": false,
     "relativeGap": 5.415513675538117e-05,
     "insideBootstrapCI": true
    },
    "p80": {
     "estimate": 50.632499,
     "ci": [
      27.30204,
      88.365435
     ],
     "certified": [
      50.63106905409927,
      50.631069055318534
     ],
     "inside": false,
     "relativeGap": 2.8242459174701003e-05,
     "insideBootstrapCI": true
    }
   }
  },
  "GPT-5.2": {
   "alias": "GPT-5.2",
   "key": "gpt_5_2",
   "release_date": "2025-12-11",
   "runsFile": {
    "n_tasks": 228,
    "candidate": {
     "slope": -0.5738142494250806,
     "intercept": 4.85472175420836,
     "newtonIterations": 9
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.5738142494261493,
      -0.5738142494240119
     ],
     "intercept": [
      4.854721754202964,
      4.854721754213757
     ],
     "maxRad": 2.686739719592879e-13
    },
    "horizons": {
     "0.5": {
      "minutes": [
       352.24642758450796,
       352.2464275967983
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       66.00541504855258,
       66.0054150504437
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 961
   },
   "live": {
    "n_tasks": 228,
    "candidate": {
     "slope": -0.5738182717487014,
     "intercept": 4.854752293515392,
     "newtonIterations": 9
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.5738182717497707,
      -0.573818271747632
     ],
     "intercept": [
      4.85475229350999,
      4.854752293520794
     ],
     "maxRad": 2.7000623958883807e-13
    },
    "horizons": {
     "0.5": {
      "minutes": [
       352.2449420349746,
       352.2449420472737
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       66.00591148520977,
       66.00591148710244
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 961,
    "printed": {
     "coefficient": -0.574,
     "intercept": 4.855
    },
    "coefficientsVerdict": "REPRODUCED",
    "coefficientsCertifiedRounded": {
     "coefficient": -0.574,
     "intercept": 4.855
    }
   },
   "site": {
    "is_sota": true,
    "release_date": "2025-12-11",
    "p50": {
     "estimate": 352.249302,
     "ci": [
      198.067494,
      815.177445
     ],
     "certified": [
      352.2449420349746,
      352.2449420472737
     ],
     "inside": false,
     "relativeGap": 1.2377651188439268e-05,
     "insideBootstrapCI": true
    },
    "p80": {
     "estimate": 66.002649,
     "ci": [
      31.665111,
      131.730719
     ],
     "certified": [
      66.00591148520977,
      66.00591148710244
     ],
     "inside": false,
     "relativeGap": 4.942719567115632e-05,
     "insideBootstrapCI": true
    }
   }
  },
  "GPT-5.3-Codex": {
   "alias": "GPT-5.3-Codex",
   "key": "gpt_5_3_codex",
   "release_date": "2026-02-05",
   "runsFile": {
    "n_tasks": 228,
    "candidate": {
     "slope": -0.5182779332585183,
     "intercept": 4.378998759747057,
     "newtonIterations": 9
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.5182779332595832,
      -0.5182779332574535
     ],
     "intercept": [
      4.378998759742176,
      4.378998759751939
     ],
     "maxRad": 2.4646951146678475e-13
    },
    "horizons": {
     "0.5": {
      "minutes": [
       349.49600917317747,
       349.49600918615477
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       54.732407100019316,
       54.73240710163451
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 1602
   },
   "live": {
    "n_tasks": 228,
    "candidate": {
     "slope": -0.5182783877646185,
     "intercept": 4.379001997858679,
     "newtonIterations": 9
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.5182783877656831,
      -0.518278387763554
     ],
     "intercept": [
      4.379001997853798,
      4.3790019978635595
     ],
     "maxRad": 2.5135449277513544e-13
    },
    "horizons": {
     "0.5": {
      "minutes": [
       349.49572775525786,
       349.4957277682323
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       54.73245201852386,
       54.73245202013871
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 1602,
    "printed": {
     "coefficient": -0.518,
     "intercept": 4.379
    },
    "coefficientsVerdict": "REPRODUCED",
    "coefficientsCertifiedRounded": {
     "coefficient": -0.518,
     "intercept": 4.379
    }
   },
   "site": {
    "is_sota": false,
    "release_date": "2026-02-05",
    "p50": {
     "estimate": 349.530732,
     "ci": [
      194.913134,
      816.351994
     ],
     "certified": [
      349.49572775525786,
      349.4957277682323
     ],
     "inside": false,
     "relativeGap": 0.00010015643100121854,
     "insideBootstrapCI": true
    },
    "p80": {
     "estimate": 54.739407,
     "ci": [
      22.361504,
      122.058963
     ],
     "certified": [
      54.73245201852386,
      54.73245202013871
     ],
     "inside": false,
     "relativeGap": 0.00012707235323182171,
     "insideBootstrapCI": true
    }
   }
  },
  "GPT-5.4": {
   "alias": "GPT-5.4",
   "key": "gpt_5_4",
   "release_date": null,
   "live": {
    "n_tasks": 228,
    "candidate": {
     "slope": -0.5201552033103755,
     "intercept": 4.378021334776262,
     "newtonIterations": 9
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.5201552033114364,
      -0.5201552033093145
     ],
     "intercept": [
      4.378021334771412,
      4.3780213347811126
     ],
     "maxRad": 2.375877272697835e-13
    },
    "horizons": {
     "0.5": {
      "minutes": [
       341.74103110805237,
       341.741031120606
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       53.877254639911506,
       53.8772546414845
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 1371,
    "printed": {
     "coefficient": -0.52,
     "intercept": 4.378
    },
    "coefficientsVerdict": "REPRODUCED",
    "coefficientsCertifiedRounded": {
     "coefficient": -0.52,
     "intercept": 4.378
    }
   },
   "site": {
    "is_sota": false,
    "release_date": "2026-03-05",
    "p50": {
     "estimate": 341.735276,
     "ci": [
      186.581591,
      768.779526
     ],
     "certified": [
      341.74103110805237,
      341.741031120606
     ],
     "inside": false,
     "relativeGap": 1.684059004372661e-05,
     "insideBootstrapCI": true
    },
    "p80": {
     "estimate": 53.877851,
     "ci": [
      23.957027,
      108.679232
     ],
     "certified": [
      53.877254639911506,
      53.8772546414845
     ],
     "inside": false,
     "relativeGap": 1.106886556265206e-05,
     "insideBootstrapCI": true
    }
   }
  },
  "Gemini 3 Pro": {
   "alias": "Gemini 3 Pro",
   "key": "gemini_3_pro",
   "release_date": "2025-11-18",
   "runsFile": {
    "n_tasks": 228,
    "candidate": {
     "slope": -0.6759398974248514,
     "intercept": 5.278803081526174,
     "newtonIterations": 8
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.6759398974258977,
      -0.6759398974238051
     ],
     "intercept": [
      5.278803081520536,
      5.278803081531812
     ],
     "maxRad": 1.816324868286756e-13
    },
    "horizons": {
     "0.5": {
      "minutes": [
       224.3449098021705,
       224.34490980852567
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       54.141426411849004,
       54.14142641314438
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 1044
   },
   "live": {
    "n_tasks": 228,
    "candidate": {
     "slope": -0.6759461562004612,
     "intercept": 5.2788499398302005,
     "newtonIterations": 8
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.6759461562015069,
      -0.6759461561994154
     ],
     "intercept": [
      5.278849939824567,
      5.278849939835834
     ],
     "maxRad": 1.7896795156957523e-13
    },
    "horizons": {
     "0.5": {
      "minutes": [
       224.34444507740588,
       224.34444508375705
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       54.142026918252185,
       54.142026919546716
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 1044,
    "printed": {
     "coefficient": -0.676,
     "intercept": 5.279
    },
    "coefficientsVerdict": "REPRODUCED",
    "coefficientsCertifiedRounded": {
     "coefficient": -0.676,
     "intercept": 5.279
    }
   },
   "site": {
    "is_sota": true,
    "release_date": "2025-11-18",
    "p50": {
     "estimate": 224.325884,
     "ci": [
      139.565157,
      379.235434
     ],
     "certified": [
      224.34444507740588,
      224.34444508375705
     ],
     "inside": false,
     "relativeGap": 8.273475971578075e-05,
     "insideBootstrapCI": true
    },
    "p80": {
     "estimate": 54.142849,
     "ci": [
      25.650451,
      103.295501
     ],
     "certified": [
      54.142026918252185,
      54.142026919546716
     ],
     "inside": false,
     "relativeGap": 1.5183800729413554e-05,
     "insideBootstrapCI": true
    }
   }
  },
  "Gemini 3.1 Pro": {
   "alias": "Gemini 3.1 Pro",
   "key": "gemini_3_1_pro",
   "release_date": null,
   "live": {
    "n_tasks": 228,
    "candidate": {
     "slope": -0.6610787099245459,
     "intercept": 5.675804895515881,
     "newtonIterations": 10
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.6610787099256115,
      -0.6610787099234803
     ],
     "intercept": [
      5.675804895509674,
      5.675804895522089
     ],
     "maxRad": 2.504663143554353e-13
    },
    "horizons": {
     "0.5": {
      "minutes": [
       384.1888629675755,
       384.1888629799502
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       89.80060579283716,
       89.80060579530898
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 1364,
    "printed": {
     "coefficient": -0.661,
     "intercept": 5.676
    },
    "coefficientsVerdict": "REPRODUCED",
    "coefficientsCertifiedRounded": {
     "coefficient": -0.661,
     "intercept": 5.676
    }
   },
   "site": {
    "is_sota": false,
    "release_date": "2026-02-19",
    "p50": {
     "estimate": 384.147435,
     "ci": [
      233.50073,
      694.750898
     ],
     "certified": [
      384.1888629675755,
      384.1888629799502
     ],
     "inside": false,
     "relativeGap": 0.00010783232920975847,
     "insideBootstrapCI": true
    },
    "p80": {
     "estimate": 89.801503,
     "ci": [
      52.025934,
      158.618017
     ],
     "certified": [
      89.80060579283716,
      89.80060579530898
     ],
     "inside": false,
     "relativeGap": 9.991103678143908e-06,
     "insideBootstrapCI": true
    }
   }
  },
  "human": {
   "alias": "human",
   "key": "human",
   "release_date": null,
   "runsFile": {
    "n_tasks": 188,
    "candidate": {
     "slope": -0.4447876644177658,
     "intercept": 2.9950704535487374,
     "newtonIterations": 7
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.4447876644187848,
      -0.4447876644167468
     ],
     "intercept": [
      2.995070453545617,
      2.9950704535518575
     ],
     "maxRad": 6.328271240363392e-14
    },
    "horizons": {
     "0.5": {
      "minutes": [
       106.42611894327051,
       106.42611894658222
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       12.269061861846065,
       12.269061862106373
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 773
   }
  },
  "o1 (Inspect)": {
   "alias": "o1 (Inspect)",
   "key": "o1_inspect",
   "release_date": "2024-12-05",
   "runsFile": {
    "n_tasks": 228,
    "candidate": {
     "slope": -0.5650541474364478,
     "intercept": 2.982994429800361,
     "newtonIterations": 7
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.5650541474374621,
      -0.5650541474354335
     ],
     "intercept": [
      2.982994429797291,
      2.9829944298034308
     ],
     "maxRad": 4.440892098500626e-14
    },
    "horizons": {
     "0.5": {
      "minutes": [
       38.83081889904477,
       38.830818899847536
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       7.0898102296367425,
       7.089810229740019
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 1044
   },
   "live": {
    "n_tasks": 228,
    "candidate": {
     "slope": -0.5650534120497429,
     "intercept": 2.982989777180757,
     "newtonIterations": 7
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.5650534120507575,
      -0.5650534120487283
     ],
     "intercept": [
      2.982989777177685,
      2.9829897771838287
     ],
     "maxRad": 4.529709940470639e-14
    },
    "horizons": {
     "0.5": {
      "minutes": [
       38.8307822013824,
       38.830782202185524
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       7.089787838294639,
       7.089787838397964
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 1044,
    "printed": {
     "coefficient": -0.565,
     "intercept": 2.983
    },
    "coefficientsVerdict": "REPRODUCED",
    "coefficientsCertifiedRounded": {
     "coefficient": -0.565,
     "intercept": 2.983
    }
   },
   "site": {
    "is_sota": true,
    "release_date": "2024-12-05",
    "p50": {
     "estimate": 38.831588,
     "ci": [
      21.164512,
      64.95249
     ],
     "certified": [
      38.8307822013824,
      38.830782202185524
     ],
     "inside": false,
     "relativeGap": 2.0751542253940472e-05,
     "insideBootstrapCI": true
    },
    "p80": {
     "estimate": 7.090121,
     "ci": [
      3.036033,
      16.640636
     ],
     "certified": [
      7.089787838294639,
      7.089787838397964
     ],
     "inside": false,
     "relativeGap": 4.699177365520601e-05,
     "insideBootstrapCI": true
    }
   }
  },
  "o1-preview": {
   "alias": "o1-preview",
   "key": "o1_preview",
   "release_date": "2024-09-12",
   "runsFile": {
    "n_tasks": 213,
    "candidate": {
     "slope": -0.629820941380409,
     "intercept": 2.7368846740899686,
     "newtonIterations": 7
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.6298209413814263,
      -0.6298209413793917
     ],
     "intercept": [
      2.7368846740871398,
      2.7368846740927975
     ],
     "maxRad": 4.6629367034256575e-14
    },
    "horizons": {
     "0.5": {
      "minutes": [
       20.329410335109635,
       20.329410335434115
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       4.4210946780272495,
       4.421094678076019
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 1661
   },
   "live": {
    "n_tasks": 213,
    "candidate": {
     "slope": -0.6298201729754966,
     "intercept": 2.736880161863878,
     "newtonIterations": 7
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.6298201729765143,
      -0.629820172974479
     ],
     "intercept": [
      2.736880161861048,
      2.7368801618667082
     ],
     "maxRad": 4.75175454539567e-14
    },
    "horizons": {
     "0.5": {
      "minutes": [
       20.32938408822374,
       20.32938408854835
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       4.421080740679002,
       4.421080740727794
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 1661,
    "printed": {
     "coefficient": -0.63,
     "intercept": 2.737
    },
    "coefficientsVerdict": "REPRODUCED",
    "coefficientsCertifiedRounded": {
     "coefficient": -0.63,
     "intercept": 2.737
    }
   },
   "site": {
    "is_sota": true,
    "release_date": "2024-09-12",
    "p50": {
     "estimate": 20.326586,
     "ci": [
      11.716193,
      33.379877
     ],
     "certified": [
      20.32938408822374,
      20.32938408854835
     ],
     "inside": false,
     "relativeGap": 0.0001376376449087926,
     "insideBootstrapCI": true
    },
    "p80": {
     "estimate": 4.420545,
     "ci": [
      2.012646,
      8.912444
     ],
     "certified": [
      4.421080740679002,
      4.421080740727794
     ],
     "inside": false,
     "relativeGap": 0.00012117867987899774,
     "insideBootstrapCI": true
    }
   }
  },
  "o3 (Inspect)": {
   "alias": "o3 (Inspect)",
   "key": "o3_inspect",
   "release_date": "2025-04-16",
   "runsFile": {
    "n_tasks": 228,
    "candidate": {
     "slope": -0.6939566785705277,
     "intercept": 4.790848845494354,
     "newtonIterations": 8
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.6939566785715627,
      -0.6939566785694926
     ],
     "intercept": [
      4.790848845489311,
      4.790848845499397
     ],
     "maxRad": 1.2745360322696797e-13
    },
    "horizons": {
     "0.5": {
      "minutes": [
       119.73252986457136,
       119.73252986748754
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       29.98157671748073,
       29.981576718087062
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 1044
   },
   "live": {
    "n_tasks": 228,
    "candidate": {
     "slope": -0.6939559701183827,
     "intercept": 4.790843862642025,
     "newtonIterations": 8
    },
    "certified": true,
    "rounds": 1,
    "box": {
     "slope": [
      -0.6939559701194178,
      -0.6939559701173477
     ],
     "intercept": [
      4.790843862636982,
      4.790843862647068
     ],
     "maxRad": 1.2789769243681803e-13
    },
    "horizons": {
     "0.5": {
      "minutes": [
       119.73251887020378,
       119.73251887311996
      ],
      "verdict": "ENCLOSED"
     },
     "0.8": {
      "minutes": [
       29.981531582456494,
       29.98153158306284
      ],
      "verdict": "ENCLOSED"
     }
    },
    "runs": 1044,
    "printed": {
     "coefficient": -0.694,
     "intercept": 4.791
    },
    "coefficientsVerdict": "REPRODUCED",
    "coefficientsCertifiedRounded": {
     "coefficient": -0.694,
     "intercept": 4.791
    }
   },
   "site": {
    "is_sota": true,
    "release_date": "2025-04-16",
    "p50": {
     "estimate": 119.732634,
     "ci": [
      74.615398,
      190.943818
     ],
     "certified": [
      119.73251887020378,
      119.73251887311996
     ],
     "inside": false,
     "relativeGap": 9.61558291051086e-07,
     "insideBootstrapCI": true
    },
    "p80": {
     "estimate": 29.981603,
     "ci": [
      15.127348,
      57.823511
     ],
     "certified": [
      29.981531582456494,
      29.98153158306284
     ],
     "inside": false,
     "relativeGap": 2.3820512074003744e-06,
     "insideBootstrapCI": true
    }
   },
   "post": {
    "th11": [
     121,
     74,
     201
    ],
    "certifiedRounded": 120.0,
    "verdict": "DIFFERS",
    "relativeGap": 0.010585938893482336
   }
  }
 },
 "trend": {
  "certified": {
   "from_2023_on": {
    "agents": [
     "Claude 3.5 Sonnet (New) (Inspect)",
     "Claude 3.5 Sonnet (Old) (Inspect)",
     "Claude 3.7 Sonnet (Inspect)",
     "Claude Opus 4.5 (Inspect)",
     "Claude Opus 4.6 (Inspect)",
     "GPT-4 0314",
     "GPT-4 1106 (Inspect)",
     "GPT-4o (Inspect)",
     "GPT-5 (Inspect)",
     "GPT-5.2",
     "Gemini 3 Pro",
     "o1 (Inspect)",
     "o1-preview",
     "o3 (Inspect)"
    ],
    "n": 14,
    "verdict": "ENCLOSED",
    "slopeBitsPerDay": [
     0.007767572975212326,
     0.007767572975301208
    ],
    "doublingDays": [
     128.74034182617024,
     128.74034182764345
    ],
    "since": "2023-01-01",
    "excludedOverMinutes": 960
   },
   "from_2024_on": {
    "agents": [
     "Claude 3.5 Sonnet (New) (Inspect)",
     "Claude 3.5 Sonnet (Old) (Inspect)",
     "Claude 3.7 Sonnet (Inspect)",
     "Claude Opus 4.5 (Inspect)",
     "Claude Opus 4.6 (Inspect)",
     "GPT-4o (Inspect)",
     "GPT-5 (Inspect)",
     "GPT-5.2",
     "Gemini 3 Pro",
     "o1 (Inspect)",
     "o1-preview",
     "o3 (Inspect)"
    ],
    "n": 12,
    "verdict": "ENCLOSED",
    "slopeBitsPerDay": [
     0.009550767524028347,
     0.009550767524183181
    ],
    "doublingDays": [
     104.70362695646534,
     104.7036269581628
    ],
    "since": "2024-01-01",
    "excludedOverMinutes": 960
   }
  },
  "printed": {
   "all_time_stitched": {
    "point_estimate": 187.778
   },
   "from_2023_on": {
    "ci_high": 158.012,
    "ci_low": 104.428,
    "point_estimate": 128.744
   }
  },
  "post": {
   "all": {
    "th1": "195.8 [162, 223]",
    "th11": "196.5",
    "th11Note": "the stitched trend, TH1.0 estimates for pre-2023 models"
   },
   "from2023": {
    "th1": "165.3 [129, 211]",
    "th11": 130.8,
    "th11ci": [
     107,
     161
    ]
   },
   "from2024": {
    "th1": 108.9,
    "th11": 88.6
   }
  }
 },
 "own": {
  "missing": {
   "baselineFiles": 1,
   "baselineAttempts": 0,
   "attemptsSolved": 0,
   "tasksWithoutHumanTime": 18,
   "frontierLogs": 15,
   "frontierRollouts": 180,
   "rolloutsOnTasksWithoutHumanTime": 180
  },
  "tasksListed": 18,
  "tasksWithHumanTime": 0,
  "models": {},
  "verdict": "NO DATA",
  "rule": "a task's human time is the median wall-clock minutes over the baseline attempts that SOLVED it; a model's outcome is successes/runs over its non-control Inspect logs; equal task weights; λ = 1e-5; the same certified fit as above"
 },
 "counts": {
  "runsFileAgents": 21,
  "liveAgents": 23,
  "siteResults": 26,
  "certified": 44,
  "fits": 44
 },
 "seconds": 120.2
}