{
  "schema_version": 1,
  "study": {
    "episodes": 8025,
    "tasks": 107,
    "seeds": [
      42,
      43,
      44
    ],
    "paper_revision": "e229573da6b2bb6854bf80affd5e7582da856517",
    "aggregate_sha256": "50d9e35a0cb0f208393bdc8ec0e394c5e5077a6bfd7107987361a185d527d0f9"
  },
  "models": [
    {
      "id": "gemma4",
      "label": "Gemma 4 26B-A4B"
    },
    {
      "id": "qwen",
      "label": "Qwen 3.6 35B-A3B"
    },
    {
      "id": "gptoss",
      "label": "GPT-OSS 120B"
    },
    {
      "id": "lightning",
      "label": "Nemotron 3.5 30B-A3B"
    },
    {
      "id": "glm",
      "label": "GLM-4.7-Flash"
    }
  ],
  "harnesses": [
    {
      "id": "openclaw",
      "label": "OpenClaw"
    },
    {
      "id": "hermes",
      "label": "Hermes"
    },
    {
      "id": "zeroclaw",
      "label": "ZeroClaw"
    },
    {
      "id": "codex",
      "label": "Codex"
    },
    {
      "id": "claudecode",
      "label": "Claude Code"
    }
  ],
  "benchmarks": [
    {
      "id": "calc",
      "label": "MedTool",
      "tasks": 32,
      "diagnostic_label": "Calculator-selection F1",
      "diagnostic_tasks": 32
    },
    {
      "id": "memory",
      "label": "MedMemory",
      "tasks": 23,
      "diagnostic_label": "Trace-judge score",
      "diagnostic_tasks": 23
    },
    {
      "id": "healthagent",
      "label": "MedPlanning",
      "tasks": 28,
      "diagnostic_label": "Reference-set recall",
      "diagnostic_tasks": 17
    },
    {
      "id": "healthadmin",
      "label": "MedWeb",
      "tasks": 24,
      "diagnostic_label": "Process-check pass fraction",
      "diagnostic_tasks": 24
    }
  ],
  "results": [
    {
      "model": "gemma4",
      "harness": "openclaw",
      "benchmark": "calc",
      "episodes": 96,
      "seeds": 3,
      "outcome_mean": 0.09647916666666667,
      "outcome_sd": 0.014375181158278788,
      "diagnostic_mean": 0.4533125,
      "diagnostic_sd": 0.035306857956351786,
      "tokens_millions_mean": 0.5174987395833334,
      "tokens_millions_sd": 0.3848229114909143
    },
    {
      "model": "gemma4",
      "harness": "openclaw",
      "benchmark": "memory",
      "episodes": 69,
      "seeds": 3,
      "outcome_mean": 0.5652173913043479,
      "outcome_sd": 0.07530655685082077,
      "diagnostic_mean": 0.35121449275362315,
      "diagnostic_sd": 0.05012918987355274,
      "tokens_millions_mean": 1.0374063768115942,
      "tokens_millions_sd": 0.13128564042484775
    },
    {
      "model": "gemma4",
      "harness": "openclaw",
      "benchmark": "healthagent",
      "episodes": 84,
      "seeds": 3,
      "outcome_mean": 0.0,
      "outcome_sd": 0.0,
      "diagnostic_mean": 0.19954211153060258,
      "diagnostic_sd": 0.012374413141838708,
      "tokens_millions_mean": 4.506343130952382,
      "tokens_millions_sd": 3.490678008922645
    },
    {
      "model": "gemma4",
      "harness": "openclaw",
      "benchmark": "healthadmin",
      "episodes": 72,
      "seeds": 3,
      "outcome_mean": 0.041666666666666664,
      "outcome_sd": 0.041666666666666664,
      "diagnostic_mean": 0.3834325396825397,
      "diagnostic_sd": 0.05721901959754033,
      "tokens_millions_mean": 11.953287388888889,
      "tokens_millions_sd": 1.5740398496936863
    },
    {
      "model": "gemma4",
      "harness": "hermes",
      "benchmark": "calc",
      "episodes": 96,
      "seeds": 3,
      "outcome_mean": 0.13291666666666666,
      "outcome_sd": 0.04254915968128317,
      "diagnostic_mean": 0.46148958333333334,
      "diagnostic_sd": 0.04582145561056342,
      "tokens_millions_mean": 0.10577270833333334,
      "tokens_millions_sd": 0.010510423505082222
    },
    {
      "model": "gemma4",
      "harness": "hermes",
      "benchmark": "memory",
      "episodes": 69,
      "seeds": 3,
      "outcome_mean": 0.5652173913043478,
      "outcome_sd": 0.11503266569846045,
      "diagnostic_mean": 0.2299536231884058,
      "diagnostic_sd": 0.06676128570557281,
      "tokens_millions_mean": 0.7756229710144927,
      "tokens_millions_sd": 0.20010631859010114
    },
    {
      "model": "gemma4",
      "harness": "hermes",
      "benchmark": "healthagent",
      "episodes": 84,
      "seeds": 3,
      "outcome_mean": 0.0,
      "outcome_sd": 0.0,
      "diagnostic_mean": 0.31651637708223385,
      "diagnostic_sd": 0.03583976240388811,
      "tokens_millions_mean": 1.0396553928571428,
      "tokens_millions_sd": 0.042364224092661315
    },
    {
      "model": "gemma4",
      "harness": "hermes",
      "benchmark": "healthadmin",
      "episodes": 72,
      "seeds": 3,
      "outcome_mean": 0.013888888888888888,
      "outcome_sd": 0.024056261216234404,
      "diagnostic_mean": 0.14503968253968252,
      "diagnostic_sd": 0.12853778127477372,
      "tokens_millions_mean": 4.024798361111111,
      "tokens_millions_sd": 2.8884886861103913
    },
    {
      "model": "gemma4",
      "harness": "zeroclaw",
      "benchmark": "calc",
      "episodes": 96,
      "seeds": 3,
      "outcome_mean": 0.08241666666666667,
      "outcome_sd": 0.009391773189650255,
      "diagnostic_mean": 0.37146875,
      "diagnostic_sd": 0.010218511082833918,
      "tokens_millions_mean": 0.10731092708333333,
      "tokens_millions_sd": 0.006689497715368413
    },
    {
      "model": "gemma4",
      "harness": "zeroclaw",
      "benchmark": "memory",
      "episodes": 69,
      "seeds": 3,
      "outcome_mean": 0.4927536231884058,
      "outcome_sd": 0.025102185616940234,
      "diagnostic_mean": 0.3193304347826087,
      "diagnostic_sd": 0.012987858224035151,
      "tokens_millions_mean": 0.483599115942029,
      "tokens_millions_sd": 0.036962770423886605
    },
    {
      "model": "gemma4",
      "harness": "zeroclaw",
      "benchmark": "healthagent",
      "episodes": 84,
      "seeds": 3,
      "outcome_mean": 0.011904761904761904,
      "outcome_sd": 0.020619652471058063,
      "diagnostic_mean": 0.2035339618939363,
      "diagnostic_sd": 0.018231064706324414,
      "tokens_millions_mean": 1.7855557142857144,
      "tokens_millions_sd": 0.6972223530789974
    },
    {
      "model": "gemma4",
      "harness": "zeroclaw",
      "benchmark": "healthadmin",
      "episodes": 72,
      "seeds": 3,
      "outcome_mean": 0.027777777777777776,
      "outcome_sd": 0.024056261216234404,
      "diagnostic_mean": 0.19910714285714284,
      "diagnostic_sd": 0.0460322805456104,
      "tokens_millions_mean": 6.674875791666667,
      "tokens_millions_sd": 0.7691005315551686
    },
    {
      "model": "gemma4",
      "harness": "codex",
      "benchmark": "calc",
      "episodes": 96,
      "seeds": 3,
      "outcome_mean": 0.09386458333333333,
      "outcome_sd": 0.005821593469217283,
      "diagnostic_mean": 0.4741458333333334,
      "diagnostic_sd": 0.0396999775333165,
      "tokens_millions_mean": 0.05659528125,
      "tokens_millions_sd": 0.009836155567826901
    },
    {
      "model": "gemma4",
      "harness": "codex",
      "benchmark": "memory",
      "episodes": 69,
      "seeds": 3,
      "outcome_mean": 0.7681159420289855,
      "outcome_sd": 0.05020437123388053,
      "diagnostic_mean": 0.47633333333333333,
      "diagnostic_sd": 0.06604253660681032,
      "tokens_millions_mean": 0.7316941739130435,
      "tokens_millions_sd": 0.38070476658173197
    },
    {
      "model": "gemma4",
      "harness": "codex",
      "benchmark": "healthagent",
      "episodes": 84,
      "seeds": 3,
      "outcome_mean": 0.011904761904761904,
      "outcome_sd": 0.020619652471058063,
      "diagnostic_mean": 0.31416527963842544,
      "diagnostic_sd": 0.04034424988195444,
      "tokens_millions_mean": 2.3048095714285712,
      "tokens_millions_sd": 0.2424954549306133
    },
    {
      "model": "gemma4",
      "harness": "codex",
      "benchmark": "healthadmin",
      "episodes": 72,
      "seeds": 3,
      "outcome_mean": 0.1111111111111111,
      "outcome_sd": 0.02405626121623441,
      "diagnostic_mean": 0.29279100529100527,
      "diagnostic_sd": 0.05251927658641471,
      "tokens_millions_mean": 20.044801833333334,
      "tokens_millions_sd": 6.931775797443185
    },
    {
      "model": "gemma4",
      "harness": "claudecode",
      "benchmark": "calc",
      "episodes": 96,
      "seeds": 3,
      "outcome_mean": 0.06435416666666667,
      "outcome_sd": 0.01949600152648828,
      "diagnostic_mean": 0.27538541666666666,
      "diagnostic_sd": 0.013131298092090259,
      "tokens_millions_mean": 0.29162580208333333,
      "tokens_millions_sd": 0.11286990850679536
    },
    {
      "model": "gemma4",
      "harness": "claudecode",
      "benchmark": "memory",
      "episodes": 69,
      "seeds": 3,
      "outcome_mean": 0.782608695652174,
      "outcome_sd": 0.043478260869565244,
      "diagnostic_mean": 0.35845410628019325,
      "diagnostic_sd": 0.14227524741410602,
      "tokens_millions_mean": 1.2267840144927538,
      "tokens_millions_sd": 0.15476753951224256
    },
    {
      "model": "gemma4",
      "harness": "claudecode",
      "benchmark": "healthagent",
      "episodes": 84,
      "seeds": 3,
      "outcome_mean": 0.03571428571428571,
      "outcome_sd": 0.03571428571428571,
      "diagnostic_mean": 0.28521732364443364,
      "diagnostic_sd": 0.06784032768830367,
      "tokens_millions_mean": 4.75427969047619,
      "tokens_millions_sd": 0.6627111451663298
    },
    {
      "model": "gemma4",
      "harness": "claudecode",
      "benchmark": "healthadmin",
      "episodes": 72,
      "seeds": 3,
      "outcome_mean": 0.09722222222222222,
      "outcome_sd": 0.09622504486493763,
      "diagnostic_mean": 0.29933862433862435,
      "diagnostic_sd": 0.017494198878915547,
      "tokens_millions_mean": 9.664612555555555,
      "tokens_millions_sd": 0.3552113015915578
    },
    {
      "model": "qwen",
      "harness": "openclaw",
      "benchmark": "calc",
      "episodes": 96,
      "seeds": 3,
      "outcome_mean": 0.10140625,
      "outcome_sd": 0.005362995315120085,
      "diagnostic_mean": 0.42175,
      "diagnostic_sd": 0.02904275202718399,
      "tokens_millions_mean": 0.46655939583333333,
      "tokens_millions_sd": 0.04808827542112382
    },
    {
      "model": "qwen",
      "harness": "openclaw",
      "benchmark": "memory",
      "episodes": 69,
      "seeds": 3,
      "outcome_mean": 0.8260869565217391,
      "outcome_sd": 0.08695652173913043,
      "diagnostic_mean": 0.44107391304347826,
      "diagnostic_sd": 0.04309645575978168,
      "tokens_millions_mean": 0.8257554057971015,
      "tokens_millions_sd": 0.119257072582191
    },
    {
      "model": "qwen",
      "harness": "openclaw",
      "benchmark": "healthagent",
      "episodes": 84,
      "seeds": 3,
      "outcome_mean": 0.09523809523809523,
      "outcome_sd": 0.020619652471058063,
      "diagnostic_mean": 0.4711981894206958,
      "diagnostic_sd": 0.02501661026564873,
      "tokens_millions_mean": 2.8385386666666665,
      "tokens_millions_sd": 0.46729324618538237
    },
    {
      "model": "qwen",
      "harness": "openclaw",
      "benchmark": "healthadmin",
      "episodes": 72,
      "seeds": 3,
      "outcome_mean": 0.4583333333333333,
      "outcome_sd": 0.04166666666666666,
      "diagnostic_mean": 0.4671626984126984,
      "diagnostic_sd": 0.03700702676836275,
      "tokens_millions_mean": 7.928591763888889,
      "tokens_millions_sd": 0.6182918827092666
    },
    {
      "model": "qwen",
      "harness": "hermes",
      "benchmark": "calc",
      "episodes": 96,
      "seeds": 3,
      "outcome_mean": 0.13430208333333332,
      "outcome_sd": 0.03272484323237215,
      "diagnostic_mean": 0.44989583333333333,
      "diagnostic_sd": 0.04559172945031627,
      "tokens_millions_mean": 0.24251264583333335,
      "tokens_millions_sd": 0.036042543535821335
    },
    {
      "model": "qwen",
      "harness": "hermes",
      "benchmark": "memory",
      "episodes": 69,
      "seeds": 3,
      "outcome_mean": 0.782608695652174,
      "outcome_sd": 0.07530655685082077,
      "diagnostic_mean": 0.5468608695652174,
      "diagnostic_sd": 0.03562614426245774,
      "tokens_millions_mean": 0.413604884057971,
      "tokens_millions_sd": 0.05566417991644185
    },
    {
      "model": "qwen",
      "harness": "hermes",
      "benchmark": "healthagent",
      "episodes": 84,
      "seeds": 3,
      "outcome_mean": 0.08333333333333333,
      "outcome_sd": 0.05455447255899809,
      "diagnostic_mean": 0.5013878739901758,
      "diagnostic_sd": 0.06112887723479179,
      "tokens_millions_mean": 1.6918409047619047,
      "tokens_millions_sd": 0.1289825322996501
    },
    {
      "model": "qwen",
      "harness": "hermes",
      "benchmark": "healthadmin",
      "episodes": 72,
      "seeds": 3,
      "outcome_mean": 0.33333333333333337,
      "outcome_sd": 0.07216878364870322,
      "diagnostic_mean": 0.4160714285714286,
      "diagnostic_sd": 0.06800644081308398,
      "tokens_millions_mean": 7.833452486111111,
      "tokens_millions_sd": 1.0220178423882655
    },
    {
      "model": "qwen",
      "harness": "zeroclaw",
      "benchmark": "calc",
      "episodes": 96,
      "seeds": 3,
      "outcome_mean": 0.07570833333333334,
      "outcome_sd": 0.02581772672062421,
      "diagnostic_mean": 0.25233333333333335,
      "diagnostic_sd": 0.06307647681254347,
      "tokens_millions_mean": 0.40605962500000004,
      "tokens_millions_sd": 0.036694381282135154
    },
    {
      "model": "qwen",
      "harness": "zeroclaw",
      "benchmark": "memory",
      "episodes": 69,
      "seeds": 3,
      "outcome_mean": 0.6666666666666666,
      "outcome_sd": 0.06641414050660634,
      "diagnostic_mean": 0.3043492753623188,
      "diagnostic_sd": 0.06765235404767196,
      "tokens_millions_mean": 0.47939091304347825,
      "tokens_millions_sd": 0.09727997644832354
    },
    {
      "model": "qwen",
      "harness": "zeroclaw",
      "benchmark": "healthagent",
      "episodes": 84,
      "seeds": 3,
      "outcome_mean": 0.023809523809523808,
      "outcome_sd": 0.020619652471058063,
      "diagnostic_mean": 0.19461418625421184,
      "diagnostic_sd": 0.04715918021700183,
      "tokens_millions_mean": 4.02977425,
      "tokens_millions_sd": 0.9833701039530315
    },
    {
      "model": "qwen",
      "harness": "zeroclaw",
      "benchmark": "healthadmin",
      "episodes": 72,
      "seeds": 3,
      "outcome_mean": 0.3194444444444445,
      "outcome_sd": 0.08673608331108887,
      "diagnostic_mean": 0.4499338624338624,
      "diagnostic_sd": 0.020882484851027006,
      "tokens_millions_mean": 6.72515325,
      "tokens_millions_sd": 0.866223381050503
    },
    {
      "model": "qwen",
      "harness": "codex",
      "benchmark": "calc",
      "episodes": 96,
      "seeds": 3,
      "outcome_mean": 0.11719791666666667,
      "outcome_sd": 0.01036748284400959,
      "diagnostic_mean": 0.49689583333333337,
      "diagnostic_sd": 0.008031270265862905,
      "tokens_millions_mean": 0.22972803125,
      "tokens_millions_sd": 0.010324094335847035
    },
    {
      "model": "qwen",
      "harness": "codex",
      "benchmark": "memory",
      "episodes": 69,
      "seeds": 3,
      "outcome_mean": 0.855072463768116,
      "outcome_sd": 0.025102185616940234,
      "diagnostic_mean": 0.649272463768116,
      "diagnostic_sd": 0.04027165700474715,
      "tokens_millions_mean": 0.6193036666666667,
      "tokens_millions_sd": 0.18271016284170774
    },
    {
      "model": "qwen",
      "harness": "codex",
      "benchmark": "healthagent",
      "episodes": 84,
      "seeds": 3,
      "outcome_mean": 0.05952380952380952,
      "outcome_sd": 0.05455447255899809,
      "diagnostic_mean": 0.4459497929606625,
      "diagnostic_sd": 0.011030189846702659,
      "tokens_millions_mean": 2.125207773809524,
      "tokens_millions_sd": 0.18178917426964877
    },
    {
      "model": "qwen",
      "harness": "codex",
      "benchmark": "healthadmin",
      "episodes": 72,
      "seeds": 3,
      "outcome_mean": 0.4166666666666667,
      "outcome_sd": 0.04166666666666666,
      "diagnostic_mean": 0.4703373015873016,
      "diagnostic_sd": 0.02843832670756009,
      "tokens_millions_mean": 5.079714597222222,
      "tokens_millions_sd": 0.5199575795602558
    },
    {
      "model": "qwen",
      "harness": "claudecode",
      "benchmark": "calc",
      "episodes": 96,
      "seeds": 3,
      "outcome_mean": 0.10185416666666666,
      "outcome_sd": 0.023009685743634864,
      "diagnostic_mean": 0.46069791666666665,
      "diagnostic_sd": 0.032567977786214085,
      "tokens_millions_mean": 0.2531481354166667,
      "tokens_millions_sd": 0.006410797723032435
    },
    {
      "model": "qwen",
      "harness": "claudecode",
      "benchmark": "memory",
      "episodes": 69,
      "seeds": 3,
      "outcome_mean": 0.8550724637681159,
      "outcome_sd": 0.100408742467761,
      "diagnostic_mean": 0.41884057971014493,
      "diagnostic_sd": 0.03342771766571244,
      "tokens_millions_mean": 0.6198206231884058,
      "tokens_millions_sd": 0.16923623005339852
    },
    {
      "model": "qwen",
      "harness": "claudecode",
      "benchmark": "healthagent",
      "episodes": 84,
      "seeds": 3,
      "outcome_mean": 0.09523809523809523,
      "outcome_sd": 0.041239304942116126,
      "diagnostic_mean": 0.45444990798251667,
      "diagnostic_sd": 0.02718707314090079,
      "tokens_millions_mean": 2.676642392857143,
      "tokens_millions_sd": 0.08803325049936928
    },
    {
      "model": "qwen",
      "harness": "claudecode",
      "benchmark": "healthadmin",
      "episodes": 72,
      "seeds": 3,
      "outcome_mean": 0.4722222222222222,
      "outcome_sd": 0.0481125224324688,
      "diagnostic_mean": 0.47559523809523807,
      "diagnostic_sd": 0.013583302032433379,
      "tokens_millions_mean": 8.107142680555555,
      "tokens_millions_sd": 1.002008954262317
    },
    {
      "model": "gptoss",
      "harness": "openclaw",
      "benchmark": "calc",
      "episodes": 96,
      "seeds": 3,
      "outcome_mean": 0.11834375,
      "outcome_sd": 0.0059552366661619685,
      "diagnostic_mean": 0.43153125000000003,
      "diagnostic_sd": 0.02369378627008566,
      "tokens_millions_mean": 0.23949164583333335,
      "tokens_millions_sd": 0.0698242628908765
    },
    {
      "model": "gptoss",
      "harness": "openclaw",
      "benchmark": "memory",
      "episodes": 69,
      "seeds": 3,
      "outcome_mean": 0.5072463768115942,
      "outcome_sd": 0.05020437123388047,
      "diagnostic_mean": 0.32029710144927537,
      "diagnostic_sd": 0.02909284477130433,
      "tokens_millions_mean": 0.5148938985507246,
      "tokens_millions_sd": 0.10583228003643975
    },
    {
      "model": "gptoss",
      "harness": "openclaw",
      "benchmark": "healthagent",
      "episodes": 84,
      "seeds": 3,
      "outcome_mean": 0.023809523809523808,
      "outcome_sd": 0.020619652471058063,
      "diagnostic_mean": 0.19515487354362035,
      "diagnostic_sd": 0.03549413382560162,
      "tokens_millions_mean": 0.938718,
      "tokens_millions_sd": 0.09586446433176414
    },
    {
      "model": "gptoss",
      "harness": "openclaw",
      "benchmark": "healthadmin",
      "episodes": 72,
      "seeds": 3,
      "outcome_mean": 0.027777777777777776,
      "outcome_sd": 0.024056261216234404,
      "diagnostic_mean": 0.21736111111111112,
      "diagnostic_sd": 0.03105142882458672,
      "tokens_millions_mean": 1.7704017777777779,
      "tokens_millions_sd": 0.5029769858449151
    },
    {
      "model": "gptoss",
      "harness": "hermes",
      "benchmark": "calc",
      "episodes": 96,
      "seeds": 3,
      "outcome_mean": 0.11959375,
      "outcome_sd": 0.014783496129552714,
      "diagnostic_mean": 0.4916770833333333,
      "diagnostic_sd": 0.028006736308356848,
      "tokens_millions_mean": 0.10189292708333333,
      "tokens_millions_sd": 0.010039050524814519
    },
    {
      "model": "gptoss",
      "harness": "hermes",
      "benchmark": "memory",
      "episodes": 69,
      "seeds": 3,
      "outcome_mean": 0.6811594202898551,
      "outcome_sd": 0.06641414050660634,
      "diagnostic_mean": 0.22222608695652174,
      "diagnostic_sd": 0.00964434563080891,
      "tokens_millions_mean": 0.3746790144927536,
      "tokens_millions_sd": 0.015570700596105507
    },
    {
      "model": "gptoss",
      "harness": "hermes",
      "benchmark": "healthagent",
      "episodes": 84,
      "seeds": 3,
      "outcome_mean": 0.03571428571428571,
      "outcome_sd": 0.03571428571428571,
      "diagnostic_mean": 0.3804306384389505,
      "diagnostic_sd": 0.03271155369352184,
      "tokens_millions_mean": 0.6088028571428571,
      "tokens_millions_sd": 0.05055025765376714
    },
    {
      "model": "gptoss",
      "harness": "hermes",
      "benchmark": "healthadmin",
      "episodes": 72,
      "seeds": 3,
      "outcome_mean": 0.0,
      "outcome_sd": 0.0,
      "diagnostic_mean": 0.05648148148148148,
      "diagnostic_sd": 0.031262017929304844,
      "tokens_millions_mean": 0.2196304861111111,
      "tokens_millions_sd": 0.09389189682486117
    },
    {
      "model": "gptoss",
      "harness": "zeroclaw",
      "benchmark": "calc",
      "episodes": 96,
      "seeds": 3,
      "outcome_mean": 0.12117708333333334,
      "outcome_sd": 0.018555668285481757,
      "diagnostic_mean": 0.46123958333333337,
      "diagnostic_sd": 0.025127416187718795,
      "tokens_millions_mean": 0.10342508333333333,
      "tokens_millions_sd": 0.01812366250955063
    },
    {
      "model": "gptoss",
      "harness": "zeroclaw",
      "benchmark": "memory",
      "episodes": 69,
      "seeds": 3,
      "outcome_mean": 0.36231884057971014,
      "outcome_sd": 0.025102185616940265,
      "diagnostic_mean": 0.2584608695652174,
      "diagnostic_sd": 0.06936159315438864,
      "tokens_millions_mean": 0.29279195652173917,
      "tokens_millions_sd": 0.02323967380326242
    },
    {
      "model": "gptoss",
      "harness": "zeroclaw",
      "benchmark": "healthagent",
      "episodes": 84,
      "seeds": 3,
      "outcome_mean": 0.011904761904761904,
      "outcome_sd": 0.020619652471058063,
      "diagnostic_mean": 0.11123898834896277,
      "diagnostic_sd": 0.05581496540318747,
      "tokens_millions_mean": 0.6300289761904763,
      "tokens_millions_sd": 0.057395689040636244
    },
    {
      "model": "gptoss",
      "harness": "zeroclaw",
      "benchmark": "healthadmin",
      "episodes": 72,
      "seeds": 3,
      "outcome_mean": 0.0,
      "outcome_sd": 0.0,
      "diagnostic_mean": 0.2745039682539683,
      "diagnostic_sd": 0.018546812049875412,
      "tokens_millions_mean": 1.236296875,
      "tokens_millions_sd": 0.061536472626052774
    },
    {
      "model": "gptoss",
      "harness": "codex",
      "benchmark": "calc",
      "episodes": 96,
      "seeds": 3,
      "outcome_mean": 0.13408333333333333,
      "outcome_sd": 0.021654476238086043,
      "diagnostic_mean": 0.43680208333333337,
      "diagnostic_sd": 0.05735668564361818,
      "tokens_millions_mean": 0.21706067708333332,
      "tokens_millions_sd": 0.027188744169400585
    },
    {
      "model": "gptoss",
      "harness": "codex",
      "benchmark": "memory",
      "episodes": 69,
      "seeds": 3,
      "outcome_mean": 0.6811594202898551,
      "outcome_sd": 0.06641414050660634,
      "diagnostic_mean": 0.3753666666666667,
      "diagnostic_sd": 0.03907013671412917,
      "tokens_millions_mean": 0.5094249420289855,
      "tokens_millions_sd": 0.06456150050581282
    },
    {
      "model": "gptoss",
      "harness": "codex",
      "benchmark": "healthagent",
      "episodes": 84,
      "seeds": 3,
      "outcome_mean": 0.03571428571428571,
      "outcome_sd": 0.0,
      "diagnostic_mean": 0.4200980392156863,
      "diagnostic_sd": 0.08022323885328624,
      "tokens_millions_mean": 0.631245130952381,
      "tokens_millions_sd": 0.14599941730349342
    },
    {
      "model": "gptoss",
      "harness": "codex",
      "benchmark": "healthadmin",
      "episodes": 72,
      "seeds": 3,
      "outcome_mean": 0.0,
      "outcome_sd": 0.0,
      "diagnostic_mean": 0.0,
      "diagnostic_sd": 0.0,
      "tokens_millions_mean": 0.10367793055555556,
      "tokens_millions_sd": 0.029286295386034722
    },
    {
      "model": "gptoss",
      "harness": "claudecode",
      "benchmark": "calc",
      "episodes": 96,
      "seeds": 3,
      "outcome_mean": 0.15572916666666667,
      "outcome_sd": 0.02224520079327973,
      "diagnostic_mean": 0.52025,
      "diagnostic_sd": 0.020629899986124522,
      "tokens_millions_mean": 0.29579656249999997,
      "tokens_millions_sd": 0.10204318158711964
    },
    {
      "model": "gptoss",
      "harness": "claudecode",
      "benchmark": "memory",
      "episodes": 69,
      "seeds": 3,
      "outcome_mean": 0.7971014492753623,
      "outcome_sd": 0.09050721736809272,
      "diagnostic_mean": 0.3521739130434783,
      "diagnostic_sd": 0.06052351599368436,
      "tokens_millions_mean": 0.467263652173913,
      "tokens_millions_sd": 0.07197451392445538
    },
    {
      "model": "gptoss",
      "harness": "claudecode",
      "benchmark": "healthagent",
      "episodes": 84,
      "seeds": 3,
      "outcome_mean": 0.047619047619047616,
      "outcome_sd": 0.041239304942116126,
      "diagnostic_mean": 0.27884655069757375,
      "diagnostic_sd": 0.034990061647253747,
      "tokens_millions_mean": 1.5530451785714285,
      "tokens_millions_sd": 0.24989339027159135
    },
    {
      "model": "gptoss",
      "harness": "claudecode",
      "benchmark": "healthadmin",
      "episodes": 72,
      "seeds": 3,
      "outcome_mean": 0.2638888888888889,
      "outcome_sd": 0.08673608331108887,
      "diagnostic_mean": 0.33323412698412697,
      "diagnostic_sd": 0.08562657411477706,
      "tokens_millions_mean": 4.085164694444444,
      "tokens_millions_sd": 0.5928220735558312
    },
    {
      "model": "lightning",
      "harness": "openclaw",
      "benchmark": "calc",
      "episodes": 96,
      "seeds": 3,
      "outcome_mean": 0.05651041666666667,
      "outcome_sd": 0.009528756504436102,
      "diagnostic_mean": 0.2838541666666667,
      "diagnostic_sd": 0.07469788979337591,
      "tokens_millions_mean": 0.6540043125,
      "tokens_millions_sd": 0.01702850243952791
    },
    {
      "model": "lightning",
      "harness": "openclaw",
      "benchmark": "memory",
      "episodes": 69,
      "seeds": 3,
      "outcome_mean": 0.5942028985507246,
      "outcome_sd": 0.05020437123388053,
      "diagnostic_mean": 0.2386463768115942,
      "diagnostic_sd": 0.05004158466053123,
      "tokens_millions_mean": 1.3547960579710145,
      "tokens_millions_sd": 0.08559641297125162
    },
    {
      "model": "lightning",
      "harness": "openclaw",
      "benchmark": "healthagent",
      "episodes": 84,
      "seeds": 3,
      "outcome_mean": 0.0,
      "outcome_sd": 0.0,
      "diagnostic_mean": 0.1451717039472794,
      "diagnostic_sd": 0.009557091703059307,
      "tokens_millions_mean": 4.995744238095238,
      "tokens_millions_sd": 0.33307208788391823
    },
    {
      "model": "lightning",
      "harness": "openclaw",
      "benchmark": "healthadmin",
      "episodes": 72,
      "seeds": 3,
      "outcome_mean": 0.0,
      "outcome_sd": 0.0,
      "diagnostic_mean": 0.08148148148148147,
      "diagnostic_sd": 0.015143300306594162,
      "tokens_millions_mean": 12.222517472222222,
      "tokens_millions_sd": 0.8017615144042272
    },
    {
      "model": "lightning",
      "harness": "hermes",
      "benchmark": "calc",
      "episodes": 96,
      "seeds": 3,
      "outcome_mean": 0.06727083333333334,
      "outcome_sd": 0.004707492089301194,
      "diagnostic_mean": 0.3644375,
      "diagnostic_sd": 0.017775482368630663,
      "tokens_millions_mean": 0.22937557291666666,
      "tokens_millions_sd": 0.007511088863173033
    },
    {
      "model": "lightning",
      "harness": "hermes",
      "benchmark": "memory",
      "episodes": 69,
      "seeds": 3,
      "outcome_mean": 0.6231884057971014,
      "outcome_sd": 0.05020437123388053,
      "diagnostic_mean": 0.31159565217391305,
      "diagnostic_sd": 0.05273446859212447,
      "tokens_millions_mean": 0.8544001739130435,
      "tokens_millions_sd": 0.1464128740881948
    },
    {
      "model": "lightning",
      "harness": "hermes",
      "benchmark": "healthagent",
      "episodes": 84,
      "seeds": 3,
      "outcome_mean": 0.023809523809523808,
      "outcome_sd": 0.041239304942116126,
      "diagnostic_mean": 0.3961073559858726,
      "diagnostic_sd": 0.0808618615897866,
      "tokens_millions_mean": 1.8682710119047619,
      "tokens_millions_sd": 0.4830207998985771
    },
    {
      "model": "lightning",
      "harness": "hermes",
      "benchmark": "healthadmin",
      "episodes": 72,
      "seeds": 3,
      "outcome_mean": 0.0,
      "outcome_sd": 0.0,
      "diagnostic_mean": 0.07202380952380952,
      "diagnostic_sd": 0.03474248275108949,
      "tokens_millions_mean": 11.544927888888889,
      "tokens_millions_sd": 3.701542113914483
    },
    {
      "model": "lightning",
      "harness": "zeroclaw",
      "benchmark": "calc",
      "episodes": 96,
      "seeds": 3,
      "outcome_mean": 0.050645833333333334,
      "outcome_sd": 0.01759717255046484,
      "diagnostic_mean": 0.22501041666666666,
      "diagnostic_sd": 0.04265202970575765,
      "tokens_millions_mean": 0.5338255104166667,
      "tokens_millions_sd": 0.051003040231995755
    },
    {
      "model": "lightning",
      "harness": "zeroclaw",
      "benchmark": "memory",
      "episodes": 69,
      "seeds": 3,
      "outcome_mean": 0.5507246376811594,
      "outcome_sd": 0.09050721736809272,
      "diagnostic_mean": 0.20483478260869564,
      "diagnostic_sd": 0.06754763956339267,
      "tokens_millions_mean": 0.8838199855072464,
      "tokens_millions_sd": 0.1116207634569287
    },
    {
      "model": "lightning",
      "harness": "zeroclaw",
      "benchmark": "healthagent",
      "episodes": 84,
      "seeds": 3,
      "outcome_mean": 0.011904761904761904,
      "outcome_sd": 0.020619652471058063,
      "diagnostic_mean": 0.16829050460764017,
      "diagnostic_sd": 0.1036631481205114,
      "tokens_millions_mean": 3.2265130476190476,
      "tokens_millions_sd": 0.6984112038744142
    },
    {
      "model": "lightning",
      "harness": "zeroclaw",
      "benchmark": "healthadmin",
      "episodes": 72,
      "seeds": 3,
      "outcome_mean": 0.0,
      "outcome_sd": 0.0,
      "diagnostic_mean": 0.08174603174603175,
      "diagnostic_sd": 0.01921148306369596,
      "tokens_millions_mean": 9.263923027777778,
      "tokens_millions_sd": 0.8046342571220227
    },
    {
      "model": "lightning",
      "harness": "codex",
      "benchmark": "calc",
      "episodes": 96,
      "seeds": 3,
      "outcome_mean": 0.07645833333333334,
      "outcome_sd": 0.0066197158565027055,
      "diagnostic_mean": 0.3256354166666667,
      "diagnostic_sd": 0.025789569364780676,
      "tokens_millions_mean": 0.6030483229166667,
      "tokens_millions_sd": 0.009481297649143456
    },
    {
      "model": "lightning",
      "harness": "codex",
      "benchmark": "memory",
      "episodes": 69,
      "seeds": 3,
      "outcome_mean": 0.6811594202898551,
      "outcome_sd": 0.05020437123388047,
      "diagnostic_mean": 0.4106333333333333,
      "diagnostic_sd": 0.05446685840165732,
      "tokens_millions_mean": 0.7211970869565217,
      "tokens_millions_sd": 0.07292999530089038
    },
    {
      "model": "lightning",
      "harness": "codex",
      "benchmark": "healthagent",
      "episodes": 84,
      "seeds": 3,
      "outcome_mean": 0.023809523809523808,
      "outcome_sd": 0.020619652471058063,
      "diagnostic_mean": 0.24898975290599332,
      "diagnostic_sd": 0.09431679543973984,
      "tokens_millions_mean": 2.949551619047619,
      "tokens_millions_sd": 0.3499111532312531
    },
    {
      "model": "lightning",
      "harness": "codex",
      "benchmark": "healthadmin",
      "episodes": 72,
      "seeds": 3,
      "outcome_mean": 0.041666666666666664,
      "outcome_sd": 0.041666666666666664,
      "diagnostic_mean": 0.08386243386243386,
      "diagnostic_sd": 0.013003350255677955,
      "tokens_millions_mean": 8.010737875,
      "tokens_millions_sd": 0.7998603948889721
    },
    {
      "model": "lightning",
      "harness": "claudecode",
      "benchmark": "calc",
      "episodes": 96,
      "seeds": 3,
      "outcome_mean": 0.06308333333333334,
      "outcome_sd": 0.00771370871133551,
      "diagnostic_mean": 0.2929479166666667,
      "diagnostic_sd": 0.025719914412684834,
      "tokens_millions_mean": 0.3453730104166667,
      "tokens_millions_sd": 0.05333479152456749
    },
    {
      "model": "lightning",
      "harness": "claudecode",
      "benchmark": "memory",
      "episodes": 69,
      "seeds": 3,
      "outcome_mean": 0.7101449275362318,
      "outcome_sd": 0.05020437123388047,
      "diagnostic_mean": 0.28695652173913044,
      "diagnostic_sd": 0.048761170554378896,
      "tokens_millions_mean": 1.2005951449275363,
      "tokens_millions_sd": 0.2389232426679113
    },
    {
      "model": "lightning",
      "harness": "claudecode",
      "benchmark": "healthagent",
      "episodes": 84,
      "seeds": 3,
      "outcome_mean": 0.0,
      "outcome_sd": 0.0,
      "diagnostic_mean": 0.21513036373969877,
      "diagnostic_sd": 0.09467125427104116,
      "tokens_millions_mean": 3.528921416666667,
      "tokens_millions_sd": 0.5279575800459736
    },
    {
      "model": "lightning",
      "harness": "claudecode",
      "benchmark": "healthadmin",
      "episodes": 72,
      "seeds": 3,
      "outcome_mean": 0.013888888888888888,
      "outcome_sd": 0.024056261216234404,
      "diagnostic_mean": 0.12152777777777778,
      "diagnostic_sd": 0.04804159008928491,
      "tokens_millions_mean": 9.763401027777778,
      "tokens_millions_sd": 0.4401184839420695
    },
    {
      "model": "glm",
      "harness": "openclaw",
      "benchmark": "calc",
      "episodes": 96,
      "seeds": 3,
      "outcome_mean": 0.024614583333333332,
      "outcome_sd": 0.0009863983834300098,
      "diagnostic_mean": 0.2888645833333333,
      "diagnostic_sd": 0.024844542702238116,
      "tokens_millions_mean": 0.5830493229166667,
      "tokens_millions_sd": 0.1467894679612273
    },
    {
      "model": "glm",
      "harness": "openclaw",
      "benchmark": "memory",
      "episodes": 69,
      "seeds": 3,
      "outcome_mean": 0.43478260869565216,
      "outcome_sd": 0.043478260869565216,
      "diagnostic_mean": 0.3642521739130435,
      "diagnostic_sd": 0.017471473914817376,
      "tokens_millions_mean": 0.5901048260869565,
      "tokens_millions_sd": 0.00988488843490459
    },
    {
      "model": "glm",
      "harness": "openclaw",
      "benchmark": "healthagent",
      "episodes": 84,
      "seeds": 3,
      "outcome_mean": 0.0,
      "outcome_sd": 0.0,
      "diagnostic_mean": 0.11452523038200788,
      "diagnostic_sd": 0.03548932405489222,
      "tokens_millions_mean": 1.6269058452380951,
      "tokens_millions_sd": 0.312058579119203
    },
    {
      "model": "glm",
      "harness": "openclaw",
      "benchmark": "healthadmin",
      "episodes": 72,
      "seeds": 3,
      "outcome_mean": 0.0,
      "outcome_sd": 0.0,
      "diagnostic_mean": 0.27192460317460315,
      "diagnostic_sd": 0.062004206348749195,
      "tokens_millions_mean": 3.7116306666666667,
      "tokens_millions_sd": 0.4419535630834865
    },
    {
      "model": "glm",
      "harness": "hermes",
      "benchmark": "calc",
      "episodes": 96,
      "seeds": 3,
      "outcome_mean": 0.03903125,
      "outcome_sd": 0.015617591993886253,
      "diagnostic_mean": 0.32129166666666664,
      "diagnostic_sd": 0.03727330349789019,
      "tokens_millions_mean": 0.4054481354166667,
      "tokens_millions_sd": 0.08280423088034737
    },
    {
      "model": "glm",
      "harness": "hermes",
      "benchmark": "memory",
      "episodes": 69,
      "seeds": 3,
      "outcome_mean": 0.34782608695652173,
      "outcome_sd": 0.07530655685082073,
      "diagnostic_mean": 0.4802028985507247,
      "diagnostic_sd": 0.060388970019888674,
      "tokens_millions_mean": 0.27538040579710144,
      "tokens_millions_sd": 0.06170274677179289
    },
    {
      "model": "glm",
      "harness": "hermes",
      "benchmark": "healthagent",
      "episodes": 84,
      "seeds": 3,
      "outcome_mean": 0.011904761904761904,
      "outcome_sd": 0.020619652471058063,
      "diagnostic_mean": 0.32971302047389006,
      "diagnostic_sd": 0.04367528757469858,
      "tokens_millions_mean": 1.0288993095238095,
      "tokens_millions_sd": 0.1611462548203468
    },
    {
      "model": "glm",
      "harness": "hermes",
      "benchmark": "healthadmin",
      "episodes": 72,
      "seeds": 3,
      "outcome_mean": 0.05555555555555555,
      "outcome_sd": 0.04811252243246881,
      "diagnostic_mean": 0.1425595238095238,
      "diagnostic_sd": 0.03378861065138025,
      "tokens_millions_mean": 1.1026568472222222,
      "tokens_millions_sd": 0.38577170308949793
    },
    {
      "model": "glm",
      "harness": "zeroclaw",
      "benchmark": "calc",
      "episodes": 96,
      "seeds": 3,
      "outcome_mean": 0.03244791666666667,
      "outcome_sd": 0.00631113330914768,
      "diagnostic_mean": 0.3052916666666667,
      "diagnostic_sd": 0.0032971835361158414,
      "tokens_millions_mean": 0.41280790624999997,
      "tokens_millions_sd": 0.15169167730997077
    },
    {
      "model": "glm",
      "harness": "zeroclaw",
      "benchmark": "memory",
      "episodes": 69,
      "seeds": 3,
      "outcome_mean": 0.4782608695652174,
      "outcome_sd": 0.11503266569846048,
      "diagnostic_mean": 0.4376942028985507,
      "diagnostic_sd": 0.04521214232354581,
      "tokens_millions_mean": 0.2320424347826087,
      "tokens_millions_sd": 0.0703907342137128
    },
    {
      "model": "glm",
      "harness": "zeroclaw",
      "benchmark": "healthagent",
      "episodes": 84,
      "seeds": 3,
      "outcome_mean": 0.0,
      "outcome_sd": 0.0,
      "diagnostic_mean": 0.09745057443267162,
      "diagnostic_sd": 0.061700843011434074,
      "tokens_millions_mean": 1.551051880952381,
      "tokens_millions_sd": 0.5064902418661947
    },
    {
      "model": "glm",
      "harness": "zeroclaw",
      "benchmark": "healthadmin",
      "episodes": 72,
      "seeds": 3,
      "outcome_mean": 0.15277777777777776,
      "outcome_sd": 0.10485881160098263,
      "diagnostic_mean": 0.31173941798941796,
      "diagnostic_sd": 0.0626436316757511,
      "tokens_millions_mean": 2.648781638888889,
      "tokens_millions_sd": 0.7817134265384624
    },
    {
      "model": "glm",
      "harness": "codex",
      "benchmark": "calc",
      "episodes": 96,
      "seeds": 3,
      "outcome_mean": 0.035916666666666666,
      "outcome_sd": 0.0037816804162744033,
      "diagnostic_mean": 0.284,
      "diagnostic_sd": 0.025188120339755427,
      "tokens_millions_mean": 0.6114646770833333,
      "tokens_millions_sd": 0.18990657051575388
    },
    {
      "model": "glm",
      "harness": "codex",
      "benchmark": "memory",
      "episodes": 69,
      "seeds": 3,
      "outcome_mean": 0.536231884057971,
      "outcome_sd": 0.05020437123388047,
      "diagnostic_mean": 0.384063768115942,
      "diagnostic_sd": 0.08699281854780307,
      "tokens_millions_mean": 1.5504156086956522,
      "tokens_millions_sd": 0.7436613109209865
    },
    {
      "model": "glm",
      "harness": "codex",
      "benchmark": "healthagent",
      "episodes": 84,
      "seeds": 3,
      "outcome_mean": 0.011904761904761904,
      "outcome_sd": 0.020619652471058063,
      "diagnostic_mean": 0.28252555853259176,
      "diagnostic_sd": 0.06290530888594316,
      "tokens_millions_mean": 2.6386781904761905,
      "tokens_millions_sd": 1.5404502455918136
    },
    {
      "model": "glm",
      "harness": "codex",
      "benchmark": "healthadmin",
      "episodes": 72,
      "seeds": 3,
      "outcome_mean": 0.20833333333333331,
      "outcome_sd": 0.1502313031443329,
      "diagnostic_mean": 0.33779761904761907,
      "diagnostic_sd": 0.10715773706198214,
      "tokens_millions_mean": 8.304172958333334,
      "tokens_millions_sd": 3.519159132443811
    },
    {
      "model": "glm",
      "harness": "claudecode",
      "benchmark": "calc",
      "episodes": 96,
      "seeds": 3,
      "outcome_mean": 0.022770833333333334,
      "outcome_sd": 0.0032656374601038205,
      "diagnostic_mean": 0.27727083333333336,
      "diagnostic_sd": 0.028119310767172685,
      "tokens_millions_mean": 0.87469,
      "tokens_millions_sd": 0.28791833624392477
    },
    {
      "model": "glm",
      "harness": "claudecode",
      "benchmark": "memory",
      "episodes": 69,
      "seeds": 3,
      "outcome_mean": 0.4782608695652174,
      "outcome_sd": 0.07530655685082073,
      "diagnostic_mean": 0.34202898550724636,
      "diagnostic_sd": 0.07506910629837603,
      "tokens_millions_mean": 1.281176304347826,
      "tokens_millions_sd": 0.3412734088198629
    },
    {
      "model": "glm",
      "harness": "claudecode",
      "benchmark": "healthagent",
      "episodes": 84,
      "seeds": 3,
      "outcome_mean": 0.05952380952380952,
      "outcome_sd": 0.05455447255899809,
      "diagnostic_mean": 0.2272811540074967,
      "diagnostic_sd": 0.13656197771110293,
      "tokens_millions_mean": 2.795349666666667,
      "tokens_millions_sd": 0.4396505459738856
    },
    {
      "model": "glm",
      "harness": "claudecode",
      "benchmark": "healthadmin",
      "episodes": 72,
      "seeds": 3,
      "outcome_mean": 0.25,
      "outcome_sd": 0.04166666666666667,
      "diagnostic_mean": 0.3189484126984127,
      "diagnostic_sd": 0.04438450376229575,
      "tokens_millions_mean": 4.4031694027777775,
      "tokens_millions_sd": 0.6549486331166288
    }
  ],
  "variance": {
    "models": [
      "gemma4",
      "qwen",
      "gptoss",
      "lightning",
      "glm"
    ],
    "by_benchmark": {
      "calc": {
        "model": 0.7318876915290591,
        "harness": 0.05252905922994654,
        "interaction": 0.09543109936281814,
        "run_to_run": 0.12015214987817614,
        "harness_plus_interaction": 0.14796015859276468,
        "hv_over_mv": 0.17884298074900323,
        "grand_mean": 0.08472708333333333,
        "total_ss": 0.12184788346354165
      },
      "memory": {
        "model": 0.47411767882699213,
        "harness": 0.2539086760456889,
        "interaction": 0.14370966435341456,
        "run_to_run": 0.1282639807739043,
        "harness_plus_interaction": 0.3976183403991035,
        "hv_over_mv": 0.6435751748251748,
        "grand_mean": 0.624927536231884,
        "total_ss": 1.866817895400126
      },
      "healthagent": {
        "model": 0.3642276422764227,
        "harness": 0.09593495934959345,
        "interaction": 0.14146341463414627,
        "run_to_run": 0.3983739837398375,
        "harness_plus_interaction": 0.23739837398373972,
        "hv_over_mv": 0.4694533762057878,
        "grand_mean": 0.02857142857142857,
        "total_ss": 0.10459183673469391
      },
      "healthadmin": {
        "model": 0.7317656344305679,
        "harness": 0.09454942197873623,
        "interaction": 0.09417800269278981,
        "run_to_run": 0.07950694089790614,
        "harness_plus_interaction": 0.18872742467152603,
        "hv_over_mv": 0.22849915682967953,
        "grand_mean": 0.1322222222222222,
        "total_ss": 1.9943518518518517
      }
    },
    "mean": {
      "model": 0.5754996617657604,
      "harness": 0.12423052915099128,
      "interaction": 0.11869554526079219,
      "run_to_run": 0.18157426382245603,
      "harness_plus_interaction": 0.2429260744117835
    }
  },
  "variance_three_models": {
    "models": [
      "gemma4",
      "qwen",
      "gptoss"
    ],
    "by_benchmark": {
      "calc": {
        "model": 0.25379927623300896,
        "harness": 0.1604329785945932,
        "interaction": 0.25455265791461323,
        "run_to_run": 0.3312150872577847,
        "harness_plus_interaction": 0.41498563650920645,
        "hv_over_mv": 0.8163353154247995,
        "grand_mean": 0.10996180555555556,
        "total_ss": 0.0391559724392361
      },
      "memory": {
        "model": 0.2905917476842992,
        "harness": 0.4687284697236468,
        "interaction": 0.10862742095996329,
        "run_to_run": 0.13205236163209064,
        "harness_plus_interaction": 0.57735589068361,
        "hv_over_mv": 1.4462128475551297,
        "grand_mean": 0.6792270531400966,
        "total_ss": 1.0975005250997687
      },
      "healthagent": {
        "model": 0.3826291079812207,
        "harness": 0.11971830985915494,
        "interaction": 0.09859154929577467,
        "run_to_run": 0.39906103286384964,
        "harness_plus_interaction": 0.21830985915492962,
        "hv_over_mv": 0.4536585365853658,
        "grand_mean": 0.03809523809523809,
        "total_ss": 0.07244897959183673
      },
      "healthadmin": {
        "model": 0.7844143723751751,
        "harness": 0.1061336651630632,
        "interaction": 0.05656659926375277,
        "run_to_run": 0.05288536319800902,
        "harness_plus_interaction": 0.16270026442681598,
        "hv_over_mv": 0.19346485819975334,
        "grand_mean": 0.17222222222222222,
        "total_ss": 1.4881944444444446
      }
    },
    "mean": {
      "model": 0.42785862606842595,
      "harness": 0.21375335583511454,
      "interaction": 0.129584556858526,
      "run_to_run": 0.2288034612379335,
      "harness_plus_interaction": 0.3433379126936405
    }
  },
  "notes": [
    "Outcome and diagnostic means weight the three seed means equally. SD is the sample SD across seeds.",
    "MedPlanning recall covers 17 of its 28 tasks. MedMemory diagnostic is an LLM trace-judge score.",
    "Variance shares describe these selected model-harness configurations. Mean weights the four environments equally.",
    "Token use is input plus output tokens, with repeated context counted on each call. It is not latency or dollars.",
    "The historical identifier lightning denotes Nemotron 3.5 in the displayed results."
  ]
}