{
  "schemaVersion": "leaderboard.v1",
  "schema": "schema/leaderboard.v1.json",
  "generatedAt": "2026-08-06T14:30:36.699Z",
  "benchmark": "VigilSAR Defense LLM Benchmark",
  "version": "0.1.0",
  "facts": {
    "benchmark": "VigilSAR Defense LLM Benchmark",
    "version": "0.1.0",
    "generatedAt": "2026-08-06T14:30:36.699Z",
    "sourceGeneratedAt": "2026-08-06T14:24:36.727Z",
    "taskCount": 300,
    "categories": [
      "Communication",
      "Compliance",
      "Domain comprehension",
      "Geospatial reasoning",
      "ISR reasoning",
      "Legal and ethical reasoning",
      "Planning support",
      "Robustness and calibration",
      "Safety",
      "Source evaluation",
      "Sovereign deployability",
      "Structured reporting"
    ],
    "pgTaskCount": 217,
    "taskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
    "scorerGitRev": "c8b266f29c56ab53faf31c64eee2f4d337e9c1a7",
    "scorerDirty": false,
    "heldoutGap": "-18.11",
    "heldoutNote": "34 scored; 0 red flag(s); threshold 5 points"
  },
  "freshness": {
    "status": "fresh",
    "checkedAt": "2026-08-06T14:30:36.721Z",
    "currentTaskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
    "staleModels": [],
    "warnings": []
  },
  "columns": [
    "model",
    "tier",
    "score.mean",
    "ci.lo",
    "ci.hi",
    "rankBand",
    "capability",
    "reliabilitySafety",
    "deployability",
    "ece",
    "eceCoverage",
    "safetyRiskIndex",
    "criticalCount",
    "heldoutGap",
    "costUsd",
    "costPerCorrect",
    "avgLatencyMs",
    "taskSetHash",
    "scorerGitRev",
    "pinned"
  ],
  "rows": [
    {
      "model": "claude-code-cli:claude-fable-5",
      "source": null,
      "tier": "hosted-api",
      "score.mean": 67.77,
      "ci.lo": 64.93,
      "ci.hi": 70.79,
      "rankBand": "A",
      "capability": 67,
      "reliabilitySafety": 66.95,
      "deployability": 53.64,
      "ece": 0.241,
      "eceCoverage": 0.437,
      "safetyRiskIndex": 66,
      "criticalCount": 2,
      "heldoutGap": "pending",
      "costUsd": 25.73128,
      "costPerCorrect": 0.126568,
      "avgLatencyMs": 13805.66,
      "taskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
      "scorerGitRev": "c8b266f29c56ab53faf31c64eee2f4d337e9c1a7",
      "pinned": true,
      "heldoutStatus": "pending",
      "heldoutCanaryHits": 0,
      "heldoutFlagged": false,
      "heldoutFlagReasons": [],
      "heldoutThreshold": 5
    },
    {
      "model": "claude-code-cli:claude-fable-5-early-july",
      "source": null,
      "tier": "hosted-api",
      "score.mean": 65.37,
      "ci.lo": 62.18,
      "ci.hi": 68.54,
      "rankBand": "B",
      "capability": 66.3,
      "reliabilitySafety": 60.72,
      "deployability": 54.04,
      "ece": 0.235,
      "eceCoverage": 0.41,
      "safetyRiskIndex": 283,
      "criticalCount": 11,
      "heldoutGap": -5.81,
      "costUsd": 53.389567,
      "costPerCorrect": 0.27225,
      "avgLatencyMs": 15778.03,
      "taskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
      "scorerGitRev": "c8b266f29c56ab53faf31c64eee2f4d337e9c1a7",
      "pinned": false,
      "heldoutStatus": "scored",
      "heldoutCanaryHits": 0,
      "heldoutFlagged": false,
      "heldoutFlagReasons": [],
      "heldoutThreshold": 5
    },
    {
      "model": "kimi-cli:kimi-k3",
      "source": null,
      "tier": "hosted-api",
      "score.mean": 64.65,
      "ci.lo": 61.63,
      "ci.hi": 67.7,
      "rankBand": "B",
      "capability": 66.63,
      "reliabilitySafety": 57.45,
      "deployability": 53.37,
      "ece": 0.22,
      "eceCoverage": 0.453,
      "safetyRiskIndex": 244,
      "criticalCount": 9,
      "heldoutGap": -10.97,
      "costUsd": "",
      "costPerCorrect": "",
      "avgLatencyMs": 21643.25,
      "taskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
      "scorerGitRev": "c8b266f29c56ab53faf31c64eee2f4d337e9c1a7",
      "pinned": false,
      "heldoutStatus": "scored",
      "heldoutCanaryHits": 0,
      "heldoutFlagged": false,
      "heldoutFlagReasons": [],
      "heldoutThreshold": 5
    },
    {
      "model": "codex-cli:gpt-5.5-medium",
      "source": null,
      "tier": "hosted-api",
      "score.mean": 62.17,
      "ci.lo": 59.15,
      "ci.hi": 65.23,
      "rankBand": "C",
      "capability": 60.83,
      "reliabilitySafety": 61.16,
      "deployability": 50.97,
      "ece": 0.175,
      "eceCoverage": 0.27,
      "safetyRiskIndex": 250,
      "criticalCount": 10,
      "heldoutGap": -9.12,
      "costUsd": "",
      "costPerCorrect": "",
      "avgLatencyMs": 17649.11,
      "taskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
      "scorerGitRev": "c8b266f29c56ab53faf31c64eee2f4d337e9c1a7",
      "pinned": false,
      "heldoutStatus": "scored",
      "heldoutCanaryHits": 0,
      "heldoutFlagged": false,
      "heldoutFlagReasons": [],
      "heldoutThreshold": 5
    },
    {
      "model": "claude-code-cli:claude-opus-5-high",
      "source": null,
      "tier": "hosted-api",
      "score.mean": 61.7,
      "ci.lo": 58.44,
      "ci.hi": 64.87,
      "rankBand": "C",
      "capability": 62.53,
      "reliabilitySafety": 55.84,
      "deployability": 53.37,
      "ece": 0.231,
      "eceCoverage": 0.547,
      "safetyRiskIndex": 160,
      "criticalCount": 5,
      "heldoutGap": -13.02,
      "costUsd": 33.672073,
      "costPerCorrect": 0.181902,
      "avgLatencyMs": 18166.68,
      "taskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
      "scorerGitRev": "c8b266f29c56ab53faf31c64eee2f4d337e9c1a7",
      "pinned": false,
      "heldoutStatus": "scored",
      "heldoutCanaryHits": 0,
      "heldoutFlagged": false,
      "heldoutFlagReasons": [],
      "heldoutThreshold": 5
    },
    {
      "model": "codex-cli:gpt-5.5",
      "source": null,
      "tier": "hosted-api",
      "score.mean": 61.66,
      "ci.lo": 58.54,
      "ci.hi": 64.72,
      "rankBand": "C",
      "capability": 60.92,
      "reliabilitySafety": 60.38,
      "deployability": 49.1,
      "ece": 0.199,
      "eceCoverage": 0.283,
      "safetyRiskIndex": 283,
      "criticalCount": 11,
      "heldoutGap": "pending",
      "costUsd": "",
      "costPerCorrect": "",
      "avgLatencyMs": 11839.06,
      "taskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
      "scorerGitRev": "c8b266f29c56ab53faf31c64eee2f4d337e9c1a7",
      "pinned": false,
      "heldoutStatus": "pending",
      "heldoutCanaryHits": 0,
      "heldoutFlagged": false,
      "heldoutFlagReasons": [],
      "heldoutThreshold": 5
    },
    {
      "model": "codex-cli:gpt-5.4-mini",
      "source": null,
      "tier": "hosted-api",
      "score.mean": 60.22,
      "ci.lo": 57.31,
      "ci.hi": 63.12,
      "rankBand": "C",
      "capability": 60.6,
      "reliabilitySafety": 54.95,
      "deployability": 52.17,
      "ece": 0.186,
      "eceCoverage": 0.337,
      "safetyRiskIndex": 116,
      "criticalCount": 4,
      "heldoutGap": -5.61,
      "costUsd": "",
      "costPerCorrect": "",
      "avgLatencyMs": 22229.95,
      "taskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
      "scorerGitRev": "c8b266f29c56ab53faf31c64eee2f4d337e9c1a7",
      "pinned": false,
      "heldoutStatus": "scored",
      "heldoutCanaryHits": 0,
      "heldoutFlagged": false,
      "heldoutFlagReasons": [],
      "heldoutThreshold": 5
    },
    {
      "model": "codex-cli:gpt-5.6-terra",
      "source": null,
      "tier": "hosted-api",
      "score.mean": 58.61,
      "ci.lo": 55.65,
      "ci.hi": 61.5,
      "rankBand": "D",
      "capability": 59.79,
      "reliabilitySafety": 53.29,
      "deployability": 50.84,
      "ece": 0.155,
      "eceCoverage": 0.313,
      "safetyRiskIndex": 183,
      "criticalCount": 7,
      "heldoutGap": -6.56,
      "costUsd": "",
      "costPerCorrect": "",
      "avgLatencyMs": 6572.61,
      "taskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
      "scorerGitRev": "c8b266f29c56ab53faf31c64eee2f4d337e9c1a7",
      "pinned": false,
      "heldoutStatus": "scored",
      "heldoutCanaryHits": 0,
      "heldoutFlagged": false,
      "heldoutFlagReasons": [],
      "heldoutThreshold": 5
    },
    {
      "model": "codex-cli:gpt-5.6-sol",
      "source": null,
      "tier": "hosted-api",
      "score.mean": 57.86,
      "ci.lo": 54.92,
      "ci.hi": 60.87,
      "rankBand": "D",
      "capability": 56.27,
      "reliabilitySafety": 56.53,
      "deployability": 51.64,
      "ece": 0.159,
      "eceCoverage": 0.297,
      "safetyRiskIndex": 233,
      "criticalCount": 9,
      "heldoutGap": -4.86,
      "costUsd": "",
      "costPerCorrect": "",
      "avgLatencyMs": 7373.9,
      "taskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
      "scorerGitRev": "c8b266f29c56ab53faf31c64eee2f4d337e9c1a7",
      "pinned": false,
      "heldoutStatus": "scored",
      "heldoutCanaryHits": 0,
      "heldoutFlagged": false,
      "heldoutFlagReasons": [],
      "heldoutThreshold": 5
    },
    {
      "model": "codex-cli:gpt-5.6-terra-xhigh",
      "source": null,
      "tier": "hosted-api",
      "score.mean": 57.69,
      "ci.lo": 54.88,
      "ci.hi": 60.65,
      "rankBand": "D",
      "capability": 58.44,
      "reliabilitySafety": 51.89,
      "deployability": 50.84,
      "ece": 0.162,
      "eceCoverage": 0.31,
      "safetyRiskIndex": 158,
      "criticalCount": 6,
      "heldoutGap": -6.15,
      "costUsd": "",
      "costPerCorrect": "",
      "avgLatencyMs": 9838.02,
      "taskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
      "scorerGitRev": "c8b266f29c56ab53faf31c64eee2f4d337e9c1a7",
      "pinned": false,
      "heldoutStatus": "scored",
      "heldoutCanaryHits": 0,
      "heldoutFlagged": false,
      "heldoutFlagReasons": [],
      "heldoutThreshold": 5
    },
    {
      "model": "codex-cli:gpt-5.3-codex-spark",
      "source": null,
      "tier": "hosted-api",
      "score.mean": 57.23,
      "ci.lo": 54.22,
      "ci.hi": 60.13,
      "rankBand": "D",
      "capability": 58.94,
      "reliabilitySafety": 50.08,
      "deployability": 50.7,
      "ece": 0.161,
      "eceCoverage": 0.37,
      "safetyRiskIndex": 225,
      "criticalCount": 9,
      "heldoutGap": -3.5,
      "costUsd": "",
      "costPerCorrect": "",
      "avgLatencyMs": 5748.69,
      "taskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
      "scorerGitRev": "c8b266f29c56ab53faf31c64eee2f4d337e9c1a7",
      "pinned": false,
      "heldoutStatus": "scored",
      "heldoutCanaryHits": 0,
      "heldoutFlagged": false,
      "heldoutFlagReasons": [],
      "heldoutThreshold": 5
    },
    {
      "model": "nvidia-build:thinkingmachines/inkling@default",
      "source": null,
      "tier": "hosted-api",
      "score.mean": 55.95,
      "ci.lo": 52.45,
      "ci.hi": 59.32,
      "rankBand": "D",
      "capability": 57.27,
      "reliabilitySafety": 52.72,
      "deployability": 44.04,
      "ece": 0.154,
      "eceCoverage": 0.323,
      "safetyRiskIndex": 116,
      "criticalCount": 4,
      "heldoutGap": -4.02,
      "costUsd": "",
      "costPerCorrect": "",
      "avgLatencyMs": 27198.14,
      "taskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
      "scorerGitRev": "c8b266f29c56ab53faf31c64eee2f4d337e9c1a7",
      "pinned": false,
      "heldoutStatus": "scored",
      "heldoutCanaryHits": 0,
      "heldoutFlagged": false,
      "heldoutFlagReasons": [],
      "heldoutThreshold": 5
    },
    {
      "model": "codex-cli:gpt-5.6-luna",
      "source": null,
      "tier": "hosted-api",
      "score.mean": 55.49,
      "ci.lo": 52.53,
      "ci.hi": 58.65,
      "rankBand": "E",
      "capability": 53.61,
      "reliabilitySafety": 53.73,
      "deployability": 51.1,
      "ece": 0.143,
      "eceCoverage": 0.29,
      "safetyRiskIndex": 258,
      "criticalCount": 10,
      "heldoutGap": -7.54,
      "costUsd": "",
      "costPerCorrect": "",
      "avgLatencyMs": 7895.3,
      "taskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
      "scorerGitRev": "c8b266f29c56ab53faf31c64eee2f4d337e9c1a7",
      "pinned": false,
      "heldoutStatus": "scored",
      "heldoutCanaryHits": 0,
      "heldoutFlagged": false,
      "heldoutFlagReasons": [],
      "heldoutThreshold": 5
    },
    {
      "model": "nvidia-build:z-ai/glm-5.2@default",
      "source": null,
      "tier": "hosted-api",
      "score.mean": 54.48,
      "ci.lo": 51.19,
      "ci.hi": 57.82,
      "rankBand": "E",
      "capability": 55.45,
      "reliabilitySafety": 50.47,
      "deployability": 44.7,
      "ece": 0.134,
      "eceCoverage": 0.307,
      "safetyRiskIndex": 266,
      "criticalCount": 10,
      "heldoutGap": -16.04,
      "costUsd": "",
      "costPerCorrect": "",
      "avgLatencyMs": 151065.39,
      "taskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
      "scorerGitRev": "c8b266f29c56ab53faf31c64eee2f4d337e9c1a7",
      "pinned": false,
      "heldoutStatus": "scored",
      "heldoutCanaryHits": 0,
      "heldoutFlagged": false,
      "heldoutFlagReasons": [],
      "heldoutThreshold": 5
    },
    {
      "model": "nvidia-build:nvidia/nemotron-3-ultra-550b-a55b@default",
      "source": null,
      "tier": "hosted-api",
      "score.mean": 54.39,
      "ci.lo": 51.25,
      "ci.hi": 57.58,
      "rankBand": "E",
      "capability": 58.03,
      "reliabilitySafety": 43.51,
      "deployability": 48.7,
      "ece": 0.134,
      "eceCoverage": 0.357,
      "safetyRiskIndex": 233,
      "criticalCount": 9,
      "heldoutGap": -6.86,
      "costUsd": "",
      "costPerCorrect": "",
      "avgLatencyMs": 12220.78,
      "taskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
      "scorerGitRev": "c8b266f29c56ab53faf31c64eee2f4d337e9c1a7",
      "pinned": false,
      "heldoutStatus": "scored",
      "heldoutCanaryHits": 0,
      "heldoutFlagged": false,
      "heldoutFlagReasons": [],
      "heldoutThreshold": 5
    },
    {
      "model": "nvidia-build:minimaxai/minimax-m3@default",
      "source": null,
      "tier": "hosted-api",
      "score.mean": 53.6,
      "ci.lo": 50.44,
      "ci.hi": 56.92,
      "rankBand": "E",
      "capability": 53.52,
      "reliabilitySafety": 49.27,
      "deployability": 46.44,
      "ece": 0.125,
      "eceCoverage": 0.357,
      "safetyRiskIndex": 233,
      "criticalCount": 9,
      "heldoutGap": -17.66,
      "costUsd": "",
      "costPerCorrect": "",
      "avgLatencyMs": 37999.25,
      "taskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
      "scorerGitRev": "c8b266f29c56ab53faf31c64eee2f4d337e9c1a7",
      "pinned": false,
      "heldoutStatus": "scored",
      "heldoutCanaryHits": 0,
      "heldoutFlagged": false,
      "heldoutFlagReasons": [],
      "heldoutThreshold": 5
    },
    {
      "model": "gemini-cli:gemini-3.1-pro-preview",
      "source": null,
      "tier": "hosted-api",
      "score.mean": 53.03,
      "ci.lo": 49.92,
      "ci.hi": 56.22,
      "rankBand": "E",
      "capability": 55.42,
      "reliabilitySafety": 45.28,
      "deployability": 45.5,
      "ece": 0.133,
      "eceCoverage": 0.257,
      "safetyRiskIndex": 383,
      "criticalCount": 15,
      "heldoutGap": "pending",
      "costUsd": "",
      "costPerCorrect": "",
      "avgLatencyMs": 15791.07,
      "taskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
      "scorerGitRev": "c8b266f29c56ab53faf31c64eee2f4d337e9c1a7",
      "pinned": false,
      "heldoutStatus": "pending",
      "heldoutCanaryHits": 0,
      "heldoutFlagged": false,
      "heldoutFlagReasons": [],
      "heldoutThreshold": 5
    },
    {
      "model": "mlx-openai:qwen/qwen3.6-35b-a3b",
      "source": null,
      "tier": "sovereign-deployable",
      "score.mean": 52.78,
      "ci.lo": 49.71,
      "ci.hi": 55.85,
      "rankBand": "E",
      "capability": 55.06,
      "reliabilitySafety": 47.18,
      "deployability": 68.64,
      "ece": 0.124,
      "eceCoverage": 0.32,
      "safetyRiskIndex": 100,
      "criticalCount": 4,
      "heldoutGap": "pending",
      "costUsd": "",
      "costPerCorrect": "",
      "avgLatencyMs": 20177.4,
      "taskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
      "scorerGitRev": "c8b266f29c56ab53faf31c64eee2f4d337e9c1a7",
      "pinned": false,
      "heldoutStatus": "pending",
      "heldoutCanaryHits": 0,
      "heldoutFlagged": false,
      "heldoutFlagReasons": [],
      "heldoutThreshold": 5
    },
    {
      "model": "nvidia-build:deepseek-ai/deepseek-v4-flash@default",
      "source": null,
      "tier": "hosted-api",
      "score.mean": 52.45,
      "ci.lo": 49.08,
      "ci.hi": 55.72,
      "rankBand": "E",
      "capability": 55.46,
      "reliabilitySafety": 42.02,
      "deployability": 48.3,
      "ece": 0.099,
      "eceCoverage": 0.29,
      "safetyRiskIndex": 341,
      "criticalCount": 13,
      "heldoutGap": -5.01,
      "costUsd": "",
      "costPerCorrect": "",
      "avgLatencyMs": 27385.04,
      "taskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
      "scorerGitRev": "c8b266f29c56ab53faf31c64eee2f4d337e9c1a7",
      "pinned": false,
      "heldoutStatus": "scored",
      "heldoutCanaryHits": 0,
      "heldoutFlagged": false,
      "heldoutFlagReasons": [],
      "heldoutThreshold": 5
    },
    {
      "model": "gemini-cli:gemini-3.1-flash-lite",
      "source": null,
      "tier": "hosted-api",
      "score.mean": 51.15,
      "ci.lo": 47.9,
      "ci.hi": 54.43,
      "rankBand": "F",
      "capability": 53.76,
      "reliabilitySafety": 41.67,
      "deployability": 48.97,
      "ece": 0.119,
      "eceCoverage": 0.3,
      "safetyRiskIndex": 400,
      "criticalCount": 16,
      "heldoutGap": "pending",
      "costUsd": "",
      "costPerCorrect": "",
      "avgLatencyMs": 7764.57,
      "taskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
      "scorerGitRev": "c8b266f29c56ab53faf31c64eee2f4d337e9c1a7",
      "pinned": false,
      "heldoutStatus": "pending",
      "heldoutCanaryHits": 0,
      "heldoutFlagged": false,
      "heldoutFlagReasons": [],
      "heldoutThreshold": 5
    },
    {
      "model": "nvidia-build:qwen/qwen3.5-397b-a17b@default",
      "source": null,
      "tier": "hosted-api",
      "score.mean": 48.95,
      "ci.lo": 45.82,
      "ci.hi": 52.17,
      "rankBand": "F",
      "capability": 50.04,
      "reliabilitySafety": 47.01,
      "deployability": 36.3,
      "ece": 0.112,
      "eceCoverage": 0.273,
      "safetyRiskIndex": 33,
      "criticalCount": 1,
      "heldoutGap": -17.78,
      "costUsd": "",
      "costPerCorrect": "",
      "avgLatencyMs": 195031.35,
      "taskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
      "scorerGitRev": "c8b266f29c56ab53faf31c64eee2f4d337e9c1a7",
      "pinned": false,
      "heldoutStatus": "scored",
      "heldoutCanaryHits": 0,
      "heldoutFlagged": false,
      "heldoutFlagReasons": [],
      "heldoutThreshold": 5
    },
    {
      "model": "nvidia-build:qwen/qwen3-next-80b-a3b-instruct@default",
      "source": null,
      "tier": "hosted-api",
      "score.mean": 48.13,
      "ci.lo": 45.11,
      "ci.hi": 51.42,
      "rankBand": "G",
      "capability": 49.46,
      "reliabilitySafety": 42.82,
      "deployability": 43.64,
      "ece": 0.117,
      "eceCoverage": 0.277,
      "safetyRiskIndex": 183,
      "criticalCount": 7,
      "heldoutGap": -2.11,
      "costUsd": "",
      "costPerCorrect": "",
      "avgLatencyMs": 254810.55,
      "taskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
      "scorerGitRev": "c8b266f29c56ab53faf31c64eee2f4d337e9c1a7",
      "pinned": false,
      "heldoutStatus": "scored",
      "heldoutCanaryHits": 0,
      "heldoutFlagged": false,
      "heldoutFlagReasons": [],
      "heldoutThreshold": 5
    },
    {
      "model": "nvidia-build:mistralai/mistral-medium-3.5-128b@default",
      "source": null,
      "tier": "hosted-api",
      "score.mean": 47.53,
      "ci.lo": 44.57,
      "ci.hi": 50.62,
      "rankBand": "G",
      "capability": 49.39,
      "reliabilitySafety": 38.94,
      "deployability": 45.64,
      "ece": 0.116,
      "eceCoverage": 0.26,
      "safetyRiskIndex": 425,
      "criticalCount": 17,
      "heldoutGap": -7.14,
      "costUsd": "",
      "costPerCorrect": "",
      "avgLatencyMs": 214062.35,
      "taskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
      "scorerGitRev": "c8b266f29c56ab53faf31c64eee2f4d337e9c1a7",
      "pinned": false,
      "heldoutStatus": "scored",
      "heldoutCanaryHits": 0,
      "heldoutFlagged": false,
      "heldoutFlagReasons": [],
      "heldoutThreshold": 5
    },
    {
      "model": "nvidia-build:google/gemma-4-31b-it@default",
      "source": null,
      "tier": "hosted-api",
      "score.mean": 47.13,
      "ci.lo": 43.8,
      "ci.hi": 50.64,
      "rankBand": "G",
      "capability": 48.38,
      "reliabilitySafety": 40.45,
      "deployability": 44.3,
      "ece": 0.099,
      "eceCoverage": 0.257,
      "safetyRiskIndex": 100,
      "criticalCount": 4,
      "heldoutGap": -13.89,
      "costUsd": "",
      "costPerCorrect": "",
      "avgLatencyMs": 42388.12,
      "taskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
      "scorerGitRev": "c8b266f29c56ab53faf31c64eee2f4d337e9c1a7",
      "pinned": false,
      "heldoutStatus": "scored",
      "heldoutCanaryHits": 0,
      "heldoutFlagged": false,
      "heldoutFlagReasons": [],
      "heldoutThreshold": 5
    },
    {
      "model": "nvidia-build:nvidia/nemotron-3-nano-omni-30b-a3b-reasoning@default",
      "source": null,
      "tier": "hosted-api",
      "score.mean": 46.88,
      "ci.lo": 43.61,
      "ci.hi": 50.1,
      "rankBand": "G",
      "capability": 49.99,
      "reliabilitySafety": 36.49,
      "deployability": 45.1,
      "ece": 0.127,
      "eceCoverage": 0.29,
      "safetyRiskIndex": 500,
      "criticalCount": 20,
      "heldoutGap": -5.05,
      "costUsd": "",
      "costPerCorrect": "",
      "avgLatencyMs": 11472.2,
      "taskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
      "scorerGitRev": "c8b266f29c56ab53faf31c64eee2f4d337e9c1a7",
      "pinned": false,
      "heldoutStatus": "scored",
      "heldoutCanaryHits": 0,
      "heldoutFlagged": false,
      "heldoutFlagReasons": [],
      "heldoutThreshold": 5
    },
    {
      "model": "nvidia-build:openai/gpt-oss-20b@default",
      "source": null,
      "tier": "hosted-api",
      "score.mean": 46.72,
      "ci.lo": 43.5,
      "ci.hi": 50.02,
      "rankBand": "G",
      "capability": 51.89,
      "reliabilitySafety": 34.45,
      "deployability": 44.84,
      "ece": 0.143,
      "eceCoverage": 0.343,
      "safetyRiskIndex": 316,
      "criticalCount": 12,
      "heldoutGap": -15.34,
      "costUsd": "",
      "costPerCorrect": "",
      "avgLatencyMs": 2699.75,
      "taskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
      "scorerGitRev": "c8b266f29c56ab53faf31c64eee2f4d337e9c1a7",
      "pinned": false,
      "heldoutStatus": "scored",
      "heldoutCanaryHits": 0,
      "heldoutFlagged": false,
      "heldoutFlagReasons": [],
      "heldoutThreshold": 5
    },
    {
      "model": "nvidia-build:mistralai/mistral-nemotron@default",
      "source": null,
      "tier": "hosted-api",
      "score.mean": 46.56,
      "ci.lo": 43.6,
      "ci.hi": 49.66,
      "rankBand": "G",
      "capability": 48.19,
      "reliabilitySafety": 38.2,
      "deployability": 47.36,
      "ece": 0.115,
      "eceCoverage": 0.193,
      "safetyRiskIndex": 300,
      "criticalCount": 12,
      "heldoutGap": -8.83,
      "costUsd": "",
      "costPerCorrect": "",
      "avgLatencyMs": 8584.13,
      "taskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
      "scorerGitRev": "c8b266f29c56ab53faf31c64eee2f4d337e9c1a7",
      "pinned": false,
      "heldoutStatus": "scored",
      "heldoutCanaryHits": 0,
      "heldoutFlagged": false,
      "heldoutFlagReasons": [],
      "heldoutThreshold": 5
    },
    {
      "model": "nvidia-build:nvidia/nemotron-3-super-120b-a12b@default",
      "source": null,
      "tier": "hosted-api",
      "score.mean": 46.27,
      "ci.lo": 42.92,
      "ci.hi": 49.93,
      "rankBand": "G",
      "capability": 49.57,
      "reliabilitySafety": 34.83,
      "deployability": 46.84,
      "ece": 0.14,
      "eceCoverage": 0.283,
      "safetyRiskIndex": 333,
      "criticalCount": 13,
      "heldoutGap": -7.09,
      "costUsd": "",
      "costPerCorrect": "",
      "avgLatencyMs": 8713.13,
      "taskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
      "scorerGitRev": "c8b266f29c56ab53faf31c64eee2f4d337e9c1a7",
      "pinned": false,
      "heldoutStatus": "scored",
      "heldoutCanaryHits": 0,
      "heldoutFlagged": false,
      "heldoutFlagReasons": [],
      "heldoutThreshold": 5
    },
    {
      "model": "nvidia-build:nvidia/llama-3.3-nemotron-super-49b-v1.5@default",
      "source": null,
      "tier": "hosted-api",
      "score.mean": 46.09,
      "ci.lo": 43,
      "ci.hi": 49.54,
      "rankBand": "G",
      "capability": 48.9,
      "reliabilitySafety": 37.18,
      "deployability": 46.57,
      "ece": 0.124,
      "eceCoverage": 0.263,
      "safetyRiskIndex": 408,
      "criticalCount": 16,
      "heldoutGap": -7.78,
      "costUsd": "",
      "costPerCorrect": "",
      "avgLatencyMs": 43007.05,
      "taskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
      "scorerGitRev": "c8b266f29c56ab53faf31c64eee2f4d337e9c1a7",
      "pinned": false,
      "heldoutStatus": "scored",
      "heldoutCanaryHits": 0,
      "heldoutFlagged": false,
      "heldoutFlagReasons": [],
      "heldoutThreshold": 5
    },
    {
      "model": "nvidia-build:meta/llama-3.3-70b-instruct@default",
      "source": null,
      "tier": "hosted-api",
      "score.mean": 43.69,
      "ci.lo": 40.56,
      "ci.hi": 46.83,
      "rankBand": "H",
      "capability": 47.95,
      "reliabilitySafety": 39.02,
      "deployability": 30.3,
      "ece": 0.159,
      "eceCoverage": 0.3,
      "safetyRiskIndex": 75,
      "criticalCount": 3,
      "heldoutGap": -9.17,
      "costUsd": "",
      "costPerCorrect": "",
      "avgLatencyMs": 37492.27,
      "taskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
      "scorerGitRev": "c8b266f29c56ab53faf31c64eee2f4d337e9c1a7",
      "pinned": false,
      "heldoutStatus": "scored",
      "heldoutCanaryHits": 0,
      "heldoutFlagged": false,
      "heldoutFlagReasons": [],
      "heldoutThreshold": 5
    },
    {
      "model": "nvidia-build:mistralai/mistral-large-3-675b-instruct-2512@default",
      "source": null,
      "tier": "hosted-api",
      "score.mean": 43.64,
      "ci.lo": 40.64,
      "ci.hi": 46.96,
      "rankBand": "H",
      "capability": 46.18,
      "reliabilitySafety": 37.72,
      "deployability": 37.77,
      "ece": 0.124,
      "eceCoverage": 0.227,
      "safetyRiskIndex": 350,
      "criticalCount": 14,
      "heldoutGap": -11.29,
      "costUsd": "",
      "costPerCorrect": "",
      "avgLatencyMs": 66204.47,
      "taskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
      "scorerGitRev": "c8b266f29c56ab53faf31c64eee2f4d337e9c1a7",
      "pinned": false,
      "heldoutStatus": "scored",
      "heldoutCanaryHits": 0,
      "heldoutFlagged": false,
      "heldoutFlagReasons": [],
      "heldoutThreshold": 5
    },
    {
      "model": "nvidia-build:mistralai/mistral-small-4-119b-2603@default",
      "source": null,
      "tier": "hosted-api",
      "score.mean": 43.36,
      "ci.lo": 40.01,
      "ci.hi": 46.67,
      "rankBand": "H",
      "capability": 46.94,
      "reliabilitySafety": 34.83,
      "deployability": 39.1,
      "ece": 0.16,
      "eceCoverage": 0.233,
      "safetyRiskIndex": 625,
      "criticalCount": 25,
      "heldoutGap": -18.11,
      "costUsd": "",
      "costPerCorrect": "",
      "avgLatencyMs": 1435.01,
      "taskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
      "scorerGitRev": "c8b266f29c56ab53faf31c64eee2f4d337e9c1a7",
      "pinned": false,
      "heldoutStatus": "scored",
      "heldoutCanaryHits": 0,
      "heldoutFlagged": false,
      "heldoutFlagReasons": [],
      "heldoutThreshold": 5
    },
    {
      "model": "nvidia-build:nvidia/nemotron-3-nano-30b-a3b@default",
      "source": null,
      "tier": "hosted-api",
      "score.mean": 43.19,
      "ci.lo": 40.08,
      "ci.hi": 46.53,
      "rankBand": "H",
      "capability": 45.1,
      "reliabilitySafety": 35.71,
      "deployability": 43.77,
      "ece": 0.177,
      "eceCoverage": 0.32,
      "safetyRiskIndex": 383,
      "criticalCount": 15,
      "heldoutGap": -5.56,
      "costUsd": "",
      "costPerCorrect": "",
      "avgLatencyMs": 4132.44,
      "taskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
      "scorerGitRev": "c8b266f29c56ab53faf31c64eee2f4d337e9c1a7",
      "pinned": false,
      "heldoutStatus": "scored",
      "heldoutCanaryHits": 0,
      "heldoutFlagged": false,
      "heldoutFlagReasons": [],
      "heldoutThreshold": 5
    },
    {
      "model": "nvidia-build:google/diffusiongemma-26b-a4b-it@default",
      "source": null,
      "tier": "hosted-api",
      "score.mean": 42.68,
      "ci.lo": 39.45,
      "ci.hi": 45.92,
      "rankBand": "H",
      "capability": 43.7,
      "reliabilitySafety": 38.48,
      "deployability": 37.64,
      "ece": 0.167,
      "eceCoverage": 0.253,
      "safetyRiskIndex": 308,
      "criticalCount": 12,
      "heldoutGap": -9.95,
      "costUsd": "",
      "costPerCorrect": "",
      "avgLatencyMs": 12388.39,
      "taskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
      "scorerGitRev": "c8b266f29c56ab53faf31c64eee2f4d337e9c1a7",
      "pinned": false,
      "heldoutStatus": "scored",
      "heldoutCanaryHits": 0,
      "heldoutFlagged": false,
      "heldoutFlagReasons": [],
      "heldoutThreshold": 5
    },
    {
      "model": "nvidia-build:nvidia/ising-calibration-1.5-31b@default",
      "source": null,
      "tier": "hosted-api",
      "score.mean": 41.96,
      "ci.lo": 38.66,
      "ci.hi": 45.5,
      "rankBand": "H",
      "capability": 46.65,
      "reliabilitySafety": 34.39,
      "deployability": 35.24,
      "ece": 0.167,
      "eceCoverage": 0.267,
      "safetyRiskIndex": 583,
      "criticalCount": 23,
      "heldoutGap": -11.58,
      "costUsd": "",
      "costPerCorrect": "",
      "avgLatencyMs": 3017.89,
      "taskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
      "scorerGitRev": "c8b266f29c56ab53faf31c64eee2f4d337e9c1a7",
      "pinned": false,
      "heldoutStatus": "scored",
      "heldoutCanaryHits": 0,
      "heldoutFlagged": false,
      "heldoutFlagReasons": [],
      "heldoutThreshold": 5
    },
    {
      "model": "nvidia-build:openai/gpt-oss-120b@high",
      "source": null,
      "tier": "hosted-api",
      "score.mean": 41.59,
      "ci.lo": 38.27,
      "ci.hi": 45.25,
      "rankBand": "H",
      "capability": 45.11,
      "reliabilitySafety": 28.84,
      "deployability": 47.1,
      "ece": 0.177,
      "eceCoverage": 0.353,
      "safetyRiskIndex": 258,
      "criticalCount": 10,
      "heldoutGap": -11.19,
      "costUsd": "",
      "costPerCorrect": "",
      "avgLatencyMs": 300061.15,
      "taskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
      "scorerGitRev": "c8b266f29c56ab53faf31c64eee2f4d337e9c1a7",
      "pinned": false,
      "heldoutStatus": "scored",
      "heldoutCanaryHits": 0,
      "heldoutFlagged": false,
      "heldoutFlagReasons": [],
      "heldoutThreshold": 5
    },
    {
      "model": "nvidia-build:nvidia/nvidia-nemotron-nano-9b-v2@default",
      "source": null,
      "tier": "hosted-api",
      "score.mean": 41.57,
      "ci.lo": 38.26,
      "ci.hi": 44.89,
      "rankBand": "H",
      "capability": 44.57,
      "reliabilitySafety": 32.49,
      "deployability": 41.9,
      "ece": 0.173,
      "eceCoverage": 0.267,
      "safetyRiskIndex": 608,
      "criticalCount": 24,
      "heldoutGap": -6.88,
      "costUsd": "",
      "costPerCorrect": "",
      "avgLatencyMs": 9395.03,
      "taskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
      "scorerGitRev": "c8b266f29c56ab53faf31c64eee2f4d337e9c1a7",
      "pinned": false,
      "heldoutStatus": "scored",
      "heldoutCanaryHits": 0,
      "heldoutFlagged": false,
      "heldoutFlagReasons": [],
      "heldoutThreshold": 5
    },
    {
      "model": "nvidia-build:meta/llama-4-maverick-17b-128e-instruct@default",
      "source": null,
      "tier": "hosted-api",
      "score.mean": 41.15,
      "ci.lo": 38.05,
      "ci.hi": 44.46,
      "rankBand": "I",
      "capability": 45.06,
      "reliabilitySafety": 36.94,
      "deployability": 25.24,
      "ece": 0.179,
      "eceCoverage": 0.287,
      "safetyRiskIndex": 208,
      "criticalCount": 8,
      "heldoutGap": -9.43,
      "costUsd": "",
      "costPerCorrect": "",
      "avgLatencyMs": 1814.62,
      "taskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
      "scorerGitRev": "c8b266f29c56ab53faf31c64eee2f4d337e9c1a7",
      "pinned": false,
      "heldoutStatus": "scored",
      "heldoutCanaryHits": 0,
      "heldoutFlagged": false,
      "heldoutFlagReasons": [],
      "heldoutThreshold": 5
    },
    {
      "model": "nvidia-build:poolside/laguna-xs-2.1@default",
      "source": null,
      "tier": "hosted-api",
      "score.mean": 40.92,
      "ci.lo": 37.35,
      "ci.hi": 44.6,
      "rankBand": "I",
      "capability": 44.06,
      "reliabilitySafety": 33.85,
      "deployability": 35.5,
      "ece": 0.176,
      "eceCoverage": 0.243,
      "safetyRiskIndex": 500,
      "criticalCount": 20,
      "heldoutGap": -3.1,
      "costUsd": "",
      "costPerCorrect": "",
      "avgLatencyMs": 12305.77,
      "taskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
      "scorerGitRev": "c8b266f29c56ab53faf31c64eee2f4d337e9c1a7",
      "pinned": false,
      "heldoutStatus": "scored",
      "heldoutCanaryHits": 0,
      "heldoutFlagged": false,
      "heldoutFlagReasons": [],
      "heldoutThreshold": 5
    }
  ],
  "cards": {
    "generatedAt": "2026-08-06T14:30:36.699Z",
    "benchmark": "VigilSAR Defense LLM Benchmark",
    "version": "0.1.0",
    "taskCount": 300,
    "provenance": {
      "scorerGitRev": "c8b266f29c56ab53faf31c64eee2f4d337e9c1a7",
      "scorerDirty": true,
      "taskSetHash": "7b252ca27acaac32f59c40078a6f01a934036944c2330f78ebe927f03164796d",
      "taskCount": 300,
      "generatedAt": "2026-08-06T14:30:36.682Z",
      "stalenessWindowDays": 14,
      "perRun": [
        {
          "model": "claude-code-cli:claude-fable-5",
          "runDir": "results/live-claude-fable-5-claude-code-cli-claude-fable-5",
          "responsesMtime": "2026-06-10T02:02:20.257Z",
          "mtime": "2026-06-10T02:02:20.257Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-06-10T02:02:20.257Z"
        },
        {
          "model": "claude-code-cli:claude-fable-5-early-july",
          "runDir": "results/claude-fable-5-early-july",
          "responsesMtime": "2026-07-16T21:45:17.077Z",
          "mtime": "2026-07-16T21:45:17.077Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-16T21:45:17.077Z"
        },
        {
          "model": "kimi-cli:kimi-k3",
          "runDir": "results/kimi-k3",
          "responsesMtime": "2026-07-17T09:41:06.088Z",
          "mtime": "2026-07-17T09:41:06.088Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-17T09:41:06.088Z"
        },
        {
          "model": "codex-cli:gpt-5.5-medium",
          "runDir": "results/codex-gpt-5.5-medium",
          "responsesMtime": "2026-07-16T09:02:30.711Z",
          "mtime": "2026-07-16T09:02:30.711Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-16T09:02:30.711Z"
        },
        {
          "model": "claude-code-cli:claude-opus-5-high",
          "runDir": "results/claude-opus-5-high",
          "responsesMtime": "2026-07-26T20:06:38.014Z",
          "mtime": "2026-07-26T20:06:38.014Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-26T20:06:38.014Z"
        },
        {
          "model": "codex-cli:gpt-5.5",
          "runDir": "results/live-gpt-5-5-codex-cli-gpt-5-5",
          "responsesMtime": "2026-06-10T15:45:35.609Z",
          "mtime": "2026-06-10T15:45:35.609Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-06-10T15:45:35.609Z"
        },
        {
          "model": "codex-cli:gpt-5.4-mini",
          "runDir": "results/live-codex-gpt-5-4-mini-gpt-5-4-mini",
          "responsesMtime": "2026-06-10T02:46:29.192Z",
          "mtime": "2026-06-10T02:46:29.192Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-06-10T02:46:29.192Z"
        },
        {
          "model": "codex-cli:gpt-5.6-terra",
          "runDir": "results/codex-gpt-5.6-terra",
          "responsesMtime": "2026-07-12T18:59:51.328Z",
          "mtime": "2026-07-12T18:59:51.328Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-12T18:59:51.328Z"
        },
        {
          "model": "codex-cli:gpt-5.6-sol",
          "runDir": "results/codex-gpt-5.6-sol",
          "responsesMtime": "2026-07-12T18:51:09.912Z",
          "mtime": "2026-07-12T18:51:09.912Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-12T18:51:09.912Z"
        },
        {
          "model": "codex-cli:gpt-5.6-terra-xhigh",
          "runDir": "results/codex-gpt-5.6-terra-xhigh",
          "responsesMtime": "2026-07-16T07:33:53.104Z",
          "mtime": "2026-07-16T07:33:53.104Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-16T07:33:53.104Z"
        },
        {
          "model": "codex-cli:gpt-5.3-codex-spark",
          "runDir": "results/live-gpt-5-3-codex-spark-codex-cli-gpt-5-3-codex-spark",
          "responsesMtime": "2026-06-11T06:14:08.762Z",
          "mtime": "2026-06-11T06:14:08.762Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-06-11T06:14:08.762Z"
        },
        {
          "model": "codex-cli:gpt-5.6-luna",
          "runDir": "results/repeat/gpt-5.6-luna",
          "responsesMtime": "2026-07-16T09:52:10.115Z",
          "mtime": "2026-07-16T09:52:10.115Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-16T09:52:10.115Z"
        },
        {
          "model": "nvidia-build:thinkingmachines/inkling@default",
          "runDir": "results/nvidia-thinkingmachines-inkling",
          "responsesMtime": "2026-08-06T14:01:50.050Z",
          "mtime": "2026-08-06T14:01:50.050Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-08-06T14:01:50.050Z"
        },
        {
          "model": "nvidia-build:z-ai/glm-5.2@default",
          "runDir": "results/nvidia-z-ai-glm-5-2",
          "responsesMtime": "2026-08-06T12:41:39.662Z",
          "mtime": "2026-08-06T12:41:39.662Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-08-06T12:41:39.662Z"
        },
        {
          "model": "nvidia-build:nvidia/nemotron-3-ultra-550b-a55b@default",
          "runDir": "results/nvidia-nvidia-nemotron-3-ultra-550b-a55b",
          "responsesMtime": "2026-07-21T18:39:20.459Z",
          "mtime": "2026-07-21T18:39:20.459Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-21T18:39:20.459Z"
        },
        {
          "model": "nvidia-build:minimaxai/minimax-m3@default",
          "runDir": "results/nvidia-minimaxai-minimax-m3",
          "responsesMtime": "2026-07-22T00:57:19.316Z",
          "mtime": "2026-07-22T00:57:19.316Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-22T00:57:19.316Z"
        },
        {
          "model": "mlx-openai:qwen/qwen3.6-35b-a3b",
          "runDir": "results/live-qwen-qwen3-6-35b-a3b-lm-studio-mlx-qwen-qwen3-6-35b-a3b",
          "responsesMtime": "2026-06-10T17:18:39.935Z",
          "mtime": "2026-06-10T17:18:39.935Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-06-10T17:18:39.935Z"
        },
        {
          "model": "gemini-cli:gemini-3.1-pro-preview",
          "runDir": "results/live-gemini-3-1-pro-preview-gemini-cli-gemini-3-1-pro-preview",
          "responsesMtime": "2026-06-11T22:05:47.760Z",
          "mtime": "2026-06-11T22:05:47.760Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-06-11T22:05:47.760Z"
        },
        {
          "model": "nvidia-build:deepseek-ai/deepseek-v4-flash@default",
          "runDir": "results/nvidia-deepseek-ai-deepseek-v4-flash",
          "responsesMtime": "2026-07-21T20:51:51.596Z",
          "mtime": "2026-07-21T20:51:51.596Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-21T20:51:51.596Z"
        },
        {
          "model": "gemini-cli:gemini-3.1-flash-lite",
          "runDir": "results/live-gemini-3-1-flash-lite-gemini-cli-gemini-3-1-flash-lite",
          "responsesMtime": "2026-06-10T20:32:30.917Z",
          "mtime": "2026-06-10T20:32:30.917Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-06-10T20:32:30.917Z"
        },
        {
          "model": "nvidia-build:qwen/qwen3.5-397b-a17b@default",
          "runDir": "results/nvidia-qwen-qwen3-5-397b-a17b",
          "responsesMtime": "2026-07-22T11:01:48.734Z",
          "mtime": "2026-07-22T11:01:48.734Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-22T11:01:48.734Z"
        },
        {
          "model": "nvidia-build:qwen/qwen3-next-80b-a3b-instruct@default",
          "runDir": "results/nvidia-qwen-qwen3-next-80b-a3b-instruct",
          "responsesMtime": "2026-07-23T21:02:27.015Z",
          "mtime": "2026-07-23T21:02:27.015Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-23T21:02:27.015Z"
        },
        {
          "model": "nvidia-build:mistralai/mistral-medium-3.5-128b@default",
          "runDir": "results/nvidia-mistralai-mistral-medium-3-5-128b",
          "responsesMtime": "2026-07-24T16:26:39.626Z",
          "mtime": "2026-07-24T16:26:39.626Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-24T16:26:39.626Z"
        },
        {
          "model": "nvidia-build:openai/gpt-oss-20b@default",
          "runDir": "results/nvidia-openai-gpt-oss-20b",
          "responsesMtime": "2026-07-23T07:16:31.033Z",
          "mtime": "2026-07-23T07:16:31.033Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-23T07:16:31.033Z"
        },
        {
          "model": "nvidia-build:google/gemma-4-31b-it@default",
          "runDir": "results/nvidia-google-gemma-4-31b-it",
          "responsesMtime": "2026-07-22T22:01:21.325Z",
          "mtime": "2026-07-22T22:01:21.325Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-22T22:01:21.325Z"
        },
        {
          "model": "nvidia-build:nvidia/nemotron-3-nano-omni-30b-a3b-reasoning@default",
          "runDir": "results/nvidia-nvidia-nemotron-3-nano-omni-30b-a3b-reasoning",
          "responsesMtime": "2026-08-06T08:56:51.262Z",
          "mtime": "2026-08-06T08:56:51.262Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-08-06T08:56:51.262Z"
        },
        {
          "model": "nvidia-build:mistralai/mistral-nemotron@default",
          "runDir": "results/nvidia-mistralai-mistral-nemotron",
          "responsesMtime": "2026-08-06T10:38:10.078Z",
          "mtime": "2026-08-06T10:38:10.078Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-08-06T10:38:10.078Z"
        },
        {
          "model": "nvidia-build:nvidia/llama-3.3-nemotron-super-49b-v1.5@default",
          "runDir": "results/nvidia-nvidia-llama-3-3-nemotron-super-49b-v1-5",
          "responsesMtime": "2026-07-21T21:35:56.561Z",
          "mtime": "2026-07-21T21:35:56.561Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-21T21:35:56.561Z"
        },
        {
          "model": "nvidia-build:nvidia/nemotron-3-super-120b-a12b@default",
          "runDir": "results/nvidia-nvidia-nemotron-3-super-120b-a12b",
          "responsesMtime": "2026-07-21T18:21:48.120Z",
          "mtime": "2026-07-21T18:21:48.120Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-21T18:21:48.120Z"
        },
        {
          "model": "nvidia-build:mistralai/mistral-large-3-675b-instruct-2512@default",
          "runDir": "results/nvidia-mistralai-mistral-large-3-675b-instruct-2512",
          "responsesMtime": "2026-07-22T03:18:28.321Z",
          "mtime": "2026-07-22T03:18:28.321Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-22T03:18:28.321Z"
        },
        {
          "model": "nvidia-build:meta/llama-3.3-70b-instruct@default",
          "runDir": "results/nvidia-meta-llama-3-3-70b-instruct",
          "responsesMtime": "2026-07-23T01:12:48.612Z",
          "mtime": "2026-07-23T01:12:48.612Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-23T01:12:48.612Z"
        },
        {
          "model": "nvidia-build:nvidia/nemotron-3-nano-30b-a3b@default",
          "runDir": "results/nvidia-nvidia-nemotron-3-nano-30b-a3b",
          "responsesMtime": "2026-07-21T17:58:53.871Z",
          "mtime": "2026-07-21T17:58:53.871Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-21T17:58:53.871Z"
        },
        {
          "model": "nvidia-build:mistralai/mistral-small-4-119b-2603@default",
          "runDir": "results/nvidia-mistralai-mistral-small-4-119b-2603",
          "responsesMtime": "2026-07-22T22:25:43.350Z",
          "mtime": "2026-07-22T22:25:43.350Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-22T22:25:43.350Z"
        },
        {
          "model": "nvidia-build:google/diffusiongemma-26b-a4b-it@default",
          "runDir": "results/nvidia-google-diffusiongemma-26b-a4b-it",
          "responsesMtime": "2026-08-06T09:47:56.710Z",
          "mtime": "2026-08-06T09:47:56.710Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-08-06T09:47:56.710Z"
        },
        {
          "model": "nvidia-build:nvidia/ising-calibration-1.5-31b@default",
          "runDir": "results/nvidia-nvidia-ising-calibration-1-5-31b",
          "responsesMtime": "2026-08-06T09:47:56.387Z",
          "mtime": "2026-08-06T09:47:56.387Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-08-06T09:47:56.387Z"
        },
        {
          "model": "nvidia-build:nvidia/nvidia-nemotron-nano-9b-v2@default",
          "runDir": "results/nvidia-nvidia-nvidia-nemotron-nano-9b-v2",
          "responsesMtime": "2026-08-06T09:43:50.602Z",
          "mtime": "2026-08-06T09:43:50.602Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-08-06T09:43:50.602Z"
        },
        {
          "model": "nvidia-build:openai/gpt-oss-120b@high",
          "runDir": "results/nvidia-openai-gpt-oss-120b-high",
          "responsesMtime": "2026-07-23T03:17:45.465Z",
          "mtime": "2026-07-23T03:17:45.465Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-23T03:17:45.465Z"
        },
        {
          "model": "nvidia-build:poolside/laguna-xs-2.1@default",
          "runDir": "results/nvidia-poolside-laguna-xs-2-1",
          "responsesMtime": "2026-08-06T10:40:14.437Z",
          "mtime": "2026-08-06T10:40:14.437Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-08-06T10:40:14.437Z"
        },
        {
          "model": "nvidia-build:meta/llama-4-maverick-17b-128e-instruct@default",
          "runDir": "results/nvidia-meta-llama-4-maverick-17b-128e-instruct",
          "responsesMtime": "2026-07-22T16:15:40.371Z",
          "mtime": "2026-07-22T16:15:40.371Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-22T16:15:40.371Z"
        }
      ]
    },
    "cards": [
      {
        "model": "claude-code-cli:claude-fable-5",
        "rank": 1,
        "scorePercent": 67.82,
        "capability": 67,
        "reliabilitySafety": 66.95,
        "deployability": 53.64,
        "completedTasks": 300,
        "totalTasks": 300,
        "provenance": {
          "model": "claude-code-cli:claude-fable-5",
          "runDir": "results/live-claude-fable-5-claude-code-cli-claude-fable-5",
          "responsesMtime": "2026-06-10T02:02:20.257Z",
          "mtime": "2026-06-10T02:02:20.257Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-06-10T02:02:20.257Z"
        },
        "deploymentProfile": {
          "tier": "hosted-api",
          "hosted": true,
          "sovereign": false,
          "airGapped": false,
          "edgeCandidate": false
        },
        "laneBreakdown": [
          {
            "category": "Communication",
            "scorePercent": 55.23,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Compliance",
            "scorePercent": 71.36,
            "tasks": 25,
            "safetyHits": 2
          },
          {
            "category": "Domain comprehension",
            "scorePercent": 73.2,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Geospatial reasoning",
            "scorePercent": 42.51,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "ISR reasoning",
            "scorePercent": 67.97,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Legal and ethical reasoning",
            "scorePercent": 84.87,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Planning support",
            "scorePercent": 85.8,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Robustness and calibration",
            "scorePercent": 47.44,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Safety",
            "scorePercent": 64.11,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Source evaluation",
            "scorePercent": 72.26,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Sovereign deployability",
            "scorePercent": 72.27,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Structured reporting",
            "scorePercent": 72.05,
            "tasks": 25,
            "safetyHits": 0
          }
        ]
      },
      {
        "model": "claude-code-cli:claude-fable-5-early-july",
        "rank": 2,
        "scorePercent": 65.15,
        "capability": 66.3,
        "reliabilitySafety": 60.72,
        "deployability": 54.04,
        "completedTasks": 300,
        "totalTasks": 300,
        "provenance": {
          "model": "claude-code-cli:claude-fable-5-early-july",
          "runDir": "results/claude-fable-5-early-july",
          "responsesMtime": "2026-07-16T21:45:17.077Z",
          "mtime": "2026-07-16T21:45:17.077Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-16T21:45:17.077Z"
        },
        "deploymentProfile": {
          "tier": "hosted-api",
          "hosted": true,
          "sovereign": false,
          "airGapped": false,
          "edgeCandidate": false
        },
        "laneBreakdown": [
          {
            "category": "Communication",
            "scorePercent": 58.01,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Compliance",
            "scorePercent": 68.45,
            "tasks": 25,
            "safetyHits": 2
          },
          {
            "category": "Domain comprehension",
            "scorePercent": 63.31,
            "tasks": 25,
            "safetyHits": 3
          },
          {
            "category": "Geospatial reasoning",
            "scorePercent": 47.71,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "ISR reasoning",
            "scorePercent": 64.94,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Legal and ethical reasoning",
            "scorePercent": 84.57,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Planning support",
            "scorePercent": 83.95,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Robustness and calibration",
            "scorePercent": 45.27,
            "tasks": 25,
            "safetyHits": 3
          },
          {
            "category": "Safety",
            "scorePercent": 44.6,
            "tasks": 25,
            "safetyHits": 2
          },
          {
            "category": "Source evaluation",
            "scorePercent": 65.32,
            "tasks": 25,
            "safetyHits": 2
          },
          {
            "category": "Sovereign deployability",
            "scorePercent": 73.07,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Structured reporting",
            "scorePercent": 80.82,
            "tasks": 25,
            "safetyHits": 0
          }
        ]
      },
      {
        "model": "kimi-cli:kimi-k3",
        "rank": 3,
        "scorePercent": 64.07,
        "capability": 66.63,
        "reliabilitySafety": 57.45,
        "deployability": 53.37,
        "completedTasks": 300,
        "totalTasks": 300,
        "provenance": {
          "model": "kimi-cli:kimi-k3",
          "runDir": "results/kimi-k3",
          "responsesMtime": "2026-07-17T09:41:06.088Z",
          "mtime": "2026-07-17T09:41:06.088Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-17T09:41:06.088Z"
        },
        "deploymentProfile": {
          "tier": "hosted-api",
          "hosted": true,
          "sovereign": false,
          "airGapped": false,
          "edgeCandidate": false
        },
        "laneBreakdown": [
          {
            "category": "Communication",
            "scorePercent": 59.76,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Compliance",
            "scorePercent": 60.65,
            "tasks": 25,
            "safetyHits": 2
          },
          {
            "category": "Domain comprehension",
            "scorePercent": 71.13,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Geospatial reasoning",
            "scorePercent": 42.94,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "ISR reasoning",
            "scorePercent": 62,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Legal and ethical reasoning",
            "scorePercent": 78.97,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Planning support",
            "scorePercent": 79.62,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Robustness and calibration",
            "scorePercent": 47.26,
            "tasks": 25,
            "safetyHits": 2
          },
          {
            "category": "Safety",
            "scorePercent": 42.9,
            "tasks": 25,
            "safetyHits": 5
          },
          {
            "category": "Source evaluation",
            "scorePercent": 68.55,
            "tasks": 25,
            "safetyHits": 2
          },
          {
            "category": "Sovereign deployability",
            "scorePercent": 71.73,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Structured reporting",
            "scorePercent": 82.46,
            "tasks": 25,
            "safetyHits": 0
          }
        ]
      },
      {
        "model": "codex-cli:gpt-5.5-medium",
        "rank": 4,
        "scorePercent": 61.42,
        "capability": 60.83,
        "reliabilitySafety": 61.16,
        "deployability": 50.97,
        "completedTasks": 300,
        "totalTasks": 300,
        "provenance": {
          "model": "codex-cli:gpt-5.5-medium",
          "runDir": "results/codex-gpt-5.5-medium",
          "responsesMtime": "2026-07-16T09:02:30.711Z",
          "mtime": "2026-07-16T09:02:30.711Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-16T09:02:30.711Z"
        },
        "deploymentProfile": {
          "tier": "hosted-api",
          "hosted": true,
          "sovereign": false,
          "airGapped": false,
          "edgeCandidate": false
        },
        "laneBreakdown": [
          {
            "category": "Communication",
            "scorePercent": 52.79,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Compliance",
            "scorePercent": 77.09,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Domain comprehension",
            "scorePercent": 67.79,
            "tasks": 25,
            "safetyHits": 3
          },
          {
            "category": "Geospatial reasoning",
            "scorePercent": 45.57,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "ISR reasoning",
            "scorePercent": 44.25,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Legal and ethical reasoning",
            "scorePercent": 77.83,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Planning support",
            "scorePercent": 73.3,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Robustness and calibration",
            "scorePercent": 48.96,
            "tasks": 25,
            "safetyHits": 2
          },
          {
            "category": "Safety",
            "scorePercent": 40.78,
            "tasks": 25,
            "safetyHits": 3
          },
          {
            "category": "Source evaluation",
            "scorePercent": 65.81,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Sovereign deployability",
            "scorePercent": 66.93,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Structured reporting",
            "scorePercent": 76.32,
            "tasks": 25,
            "safetyHits": 0
          }
        ]
      },
      {
        "model": "claude-code-cli:claude-opus-5-high",
        "rank": 5,
        "scorePercent": 61.28,
        "capability": 62.53,
        "reliabilitySafety": 55.84,
        "deployability": 53.37,
        "completedTasks": 300,
        "totalTasks": 300,
        "provenance": {
          "model": "claude-code-cli:claude-opus-5-high",
          "runDir": "results/claude-opus-5-high",
          "responsesMtime": "2026-07-26T20:06:38.014Z",
          "mtime": "2026-07-26T20:06:38.014Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-26T20:06:38.014Z"
        },
        "deploymentProfile": {
          "tier": "hosted-api",
          "hosted": true,
          "sovereign": false,
          "airGapped": false,
          "edgeCandidate": false
        },
        "laneBreakdown": [
          {
            "category": "Communication",
            "scorePercent": 61.15,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Compliance",
            "scorePercent": 52.39,
            "tasks": 25,
            "safetyHits": 2
          },
          {
            "category": "Domain comprehension",
            "scorePercent": 78.14,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Geospatial reasoning",
            "scorePercent": 47.25,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "ISR reasoning",
            "scorePercent": 65.09,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Legal and ethical reasoning",
            "scorePercent": 80.39,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Planning support",
            "scorePercent": 55.23,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Robustness and calibration",
            "scorePercent": 42.39,
            "tasks": 25,
            "safetyHits": 2
          },
          {
            "category": "Safety",
            "scorePercent": 48.17,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Source evaluation",
            "scorePercent": 60.35,
            "tasks": 25,
            "safetyHits": 4
          },
          {
            "category": "Sovereign deployability",
            "scorePercent": 71.73,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Structured reporting",
            "scorePercent": 70.47,
            "tasks": 25,
            "safetyHits": 0
          }
        ]
      },
      {
        "model": "codex-cli:gpt-5.5",
        "rank": 6,
        "scorePercent": 60.93,
        "capability": 60.92,
        "reliabilitySafety": 60.38,
        "deployability": 49.1,
        "completedTasks": 300,
        "totalTasks": 300,
        "provenance": {
          "model": "codex-cli:gpt-5.5",
          "runDir": "results/live-gpt-5-5-codex-cli-gpt-5-5",
          "responsesMtime": "2026-06-10T15:45:35.609Z",
          "mtime": "2026-06-10T15:45:35.609Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-06-10T15:45:35.609Z"
        },
        "deploymentProfile": {
          "tier": "hosted-api",
          "hosted": true,
          "sovereign": false,
          "airGapped": false,
          "edgeCandidate": false
        },
        "laneBreakdown": [
          {
            "category": "Communication",
            "scorePercent": 51.74,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Compliance",
            "scorePercent": 72.01,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Domain comprehension",
            "scorePercent": 67.21,
            "tasks": 25,
            "safetyHits": 3
          },
          {
            "category": "Geospatial reasoning",
            "scorePercent": 47.13,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "ISR reasoning",
            "scorePercent": 52.88,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Legal and ethical reasoning",
            "scorePercent": 74.5,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Planning support",
            "scorePercent": 72.94,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Robustness and calibration",
            "scorePercent": 50.56,
            "tasks": 25,
            "safetyHits": 2
          },
          {
            "category": "Safety",
            "scorePercent": 44.43,
            "tasks": 25,
            "safetyHits": 3
          },
          {
            "category": "Source evaluation",
            "scorePercent": 56.48,
            "tasks": 25,
            "safetyHits": 3
          },
          {
            "category": "Sovereign deployability",
            "scorePercent": 63.2,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Structured reporting",
            "scorePercent": 78.07,
            "tasks": 25,
            "safetyHits": 1
          }
        ]
      },
      {
        "model": "codex-cli:gpt-5.4-mini",
        "rank": 7,
        "scorePercent": 59.5,
        "capability": 60.6,
        "reliabilitySafety": 54.95,
        "deployability": 52.17,
        "completedTasks": 300,
        "totalTasks": 300,
        "provenance": {
          "model": "codex-cli:gpt-5.4-mini",
          "runDir": "results/live-codex-gpt-5-4-mini-gpt-5-4-mini",
          "responsesMtime": "2026-06-10T02:46:29.192Z",
          "mtime": "2026-06-10T02:46:29.192Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-06-10T02:46:29.192Z"
        },
        "deploymentProfile": {
          "tier": "hosted-api",
          "hosted": true,
          "sovereign": false,
          "airGapped": false,
          "edgeCandidate": false
        },
        "laneBreakdown": [
          {
            "category": "Communication",
            "scorePercent": 53.14,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Compliance",
            "scorePercent": 57.48,
            "tasks": 25,
            "safetyHits": 2
          },
          {
            "category": "Domain comprehension",
            "scorePercent": 70.41,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Geospatial reasoning",
            "scorePercent": 47.25,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "ISR reasoning",
            "scorePercent": 47.71,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Legal and ethical reasoning",
            "scorePercent": 74.69,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Planning support",
            "scorePercent": 69.03,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Robustness and calibration",
            "scorePercent": 49.28,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Safety",
            "scorePercent": 38.37,
            "tasks": 25,
            "safetyHits": 2
          },
          {
            "category": "Source evaluation",
            "scorePercent": 57.9,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Sovereign deployability",
            "scorePercent": 69.33,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Structured reporting",
            "scorePercent": 78.77,
            "tasks": 25,
            "safetyHits": 0
          }
        ]
      },
      {
        "model": "codex-cli:gpt-5.6-terra",
        "rank": 8,
        "scorePercent": 58.3,
        "capability": 59.79,
        "reliabilitySafety": 53.29,
        "deployability": 50.84,
        "completedTasks": 300,
        "totalTasks": 300,
        "provenance": {
          "model": "codex-cli:gpt-5.6-terra",
          "runDir": "results/codex-gpt-5.6-terra",
          "responsesMtime": "2026-07-12T18:59:51.328Z",
          "mtime": "2026-07-12T18:59:51.328Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-12T18:59:51.328Z"
        },
        "deploymentProfile": {
          "tier": "hosted-api",
          "hosted": true,
          "sovereign": false,
          "airGapped": false,
          "edgeCandidate": false
        },
        "laneBreakdown": [
          {
            "category": "Communication",
            "scorePercent": 50.87,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Compliance",
            "scorePercent": 59.77,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Domain comprehension",
            "scorePercent": 56.77,
            "tasks": 25,
            "safetyHits": 4
          },
          {
            "category": "Geospatial reasoning",
            "scorePercent": 46.64,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "ISR reasoning",
            "scorePercent": 58.85,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Legal and ethical reasoning",
            "scorePercent": 75.64,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Planning support",
            "scorePercent": 72.87,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Robustness and calibration",
            "scorePercent": 41.15,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Safety",
            "scorePercent": 36.57,
            "tasks": 25,
            "safetyHits": 3
          },
          {
            "category": "Source evaluation",
            "scorePercent": 57.29,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Sovereign deployability",
            "scorePercent": 66.67,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Structured reporting",
            "scorePercent": 75.26,
            "tasks": 25,
            "safetyHits": 0
          }
        ]
      },
      {
        "model": "codex-cli:gpt-5.6-sol",
        "rank": 9,
        "scorePercent": 57.43,
        "capability": 56.27,
        "reliabilitySafety": 56.53,
        "deployability": 51.64,
        "completedTasks": 300,
        "totalTasks": 300,
        "provenance": {
          "model": "codex-cli:gpt-5.6-sol",
          "runDir": "results/codex-gpt-5.6-sol",
          "responsesMtime": "2026-07-12T18:51:09.912Z",
          "mtime": "2026-07-12T18:51:09.912Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-12T18:51:09.912Z"
        },
        "deploymentProfile": {
          "tier": "hosted-api",
          "hosted": true,
          "sovereign": false,
          "airGapped": false,
          "edgeCandidate": false
        },
        "laneBreakdown": [
          {
            "category": "Communication",
            "scorePercent": 54.53,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Compliance",
            "scorePercent": 59.87,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Domain comprehension",
            "scorePercent": 63.26,
            "tasks": 25,
            "safetyHits": 4
          },
          {
            "category": "Geospatial reasoning",
            "scorePercent": 47.68,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "ISR reasoning",
            "scorePercent": 45.01,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Legal and ethical reasoning",
            "scorePercent": 76,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Planning support",
            "scorePercent": 56.96,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Robustness and calibration",
            "scorePercent": 45.27,
            "tasks": 25,
            "safetyHits": 2
          },
          {
            "category": "Safety",
            "scorePercent": 44.97,
            "tasks": 25,
            "safetyHits": 2
          },
          {
            "category": "Source evaluation",
            "scorePercent": 54.58,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Sovereign deployability",
            "scorePercent": 68.27,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Structured reporting",
            "scorePercent": 71.89,
            "tasks": 25,
            "safetyHits": 0
          }
        ]
      },
      {
        "model": "codex-cli:gpt-5.6-terra-xhigh",
        "rank": 10,
        "scorePercent": 57.07,
        "capability": 58.44,
        "reliabilitySafety": 51.89,
        "deployability": 50.84,
        "completedTasks": 300,
        "totalTasks": 300,
        "provenance": {
          "model": "codex-cli:gpt-5.6-terra-xhigh",
          "runDir": "results/codex-gpt-5.6-terra-xhigh",
          "responsesMtime": "2026-07-16T07:33:53.104Z",
          "mtime": "2026-07-16T07:33:53.104Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-16T07:33:53.104Z"
        },
        "deploymentProfile": {
          "tier": "hosted-api",
          "hosted": true,
          "sovereign": false,
          "airGapped": false,
          "edgeCandidate": false
        },
        "laneBreakdown": [
          {
            "category": "Communication",
            "scorePercent": 52.09,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Compliance",
            "scorePercent": 52.75,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Domain comprehension",
            "scorePercent": 68.08,
            "tasks": 25,
            "safetyHits": 2
          },
          {
            "category": "Geospatial reasoning",
            "scorePercent": 45.11,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "ISR reasoning",
            "scorePercent": 43.64,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Legal and ethical reasoning",
            "scorePercent": 76.86,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Planning support",
            "scorePercent": 69.18,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Robustness and calibration",
            "scorePercent": 42.95,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Safety",
            "scorePercent": 35.01,
            "tasks": 25,
            "safetyHits": 2
          },
          {
            "category": "Source evaluation",
            "scorePercent": 59.06,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Sovereign deployability",
            "scorePercent": 66.67,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Structured reporting",
            "scorePercent": 71.89,
            "tasks": 25,
            "safetyHits": 0
          }
        ]
      },
      {
        "model": "codex-cli:gpt-5.3-codex-spark",
        "rank": 11,
        "scorePercent": 56.79,
        "capability": 58.94,
        "reliabilitySafety": 50.08,
        "deployability": 50.7,
        "completedTasks": 300,
        "totalTasks": 300,
        "provenance": {
          "model": "codex-cli:gpt-5.3-codex-spark",
          "runDir": "results/live-gpt-5-3-codex-spark-codex-cli-gpt-5-3-codex-spark",
          "responsesMtime": "2026-06-11T06:14:08.762Z",
          "mtime": "2026-06-11T06:14:08.762Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-06-11T06:14:08.762Z"
        },
        "deploymentProfile": {
          "tier": "hosted-api",
          "hosted": true,
          "sovereign": false,
          "airGapped": false,
          "edgeCandidate": false
        },
        "laneBreakdown": [
          {
            "category": "Communication",
            "scorePercent": 54.88,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Compliance",
            "scorePercent": 39.39,
            "tasks": 25,
            "safetyHits": 2
          },
          {
            "category": "Domain comprehension",
            "scorePercent": 63.92,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Geospatial reasoning",
            "scorePercent": 40.98,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "ISR reasoning",
            "scorePercent": 52.44,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Legal and ethical reasoning",
            "scorePercent": 76.04,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Planning support",
            "scorePercent": 68.32,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Robustness and calibration",
            "scorePercent": 43.8,
            "tasks": 25,
            "safetyHits": 2
          },
          {
            "category": "Safety",
            "scorePercent": 41.08,
            "tasks": 25,
            "safetyHits": 3
          },
          {
            "category": "Source evaluation",
            "scorePercent": 59.55,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Sovereign deployability",
            "scorePercent": 66.4,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Structured reporting",
            "scorePercent": 72.46,
            "tasks": 25,
            "safetyHits": 1
          }
        ]
      },
      {
        "model": "codex-cli:gpt-5.6-luna",
        "rank": 12,
        "scorePercent": 56.44,
        "capability": 56.81,
        "reliabilitySafety": 52.9,
        "deployability": 50.84,
        "completedTasks": 300,
        "totalTasks": 300,
        "provenance": {
          "model": "codex-cli:gpt-5.6-luna",
          "runDir": "results/repeat/gpt-5.6-luna",
          "responsesMtime": "2026-07-16T09:52:10.115Z",
          "mtime": "2026-07-16T09:52:10.115Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-16T09:52:10.115Z"
        },
        "deploymentProfile": {
          "tier": "hosted-api",
          "hosted": true,
          "sovereign": false,
          "airGapped": false,
          "edgeCandidate": false
        },
        "laneBreakdown": [
          {
            "category": "Communication",
            "scorePercent": 47.39,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Compliance",
            "scorePercent": 62.52,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Domain comprehension",
            "scorePercent": 63.72,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Geospatial reasoning",
            "scorePercent": 47.55,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "ISR reasoning",
            "scorePercent": 42.75,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Legal and ethical reasoning",
            "scorePercent": 72.97,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Planning support",
            "scorePercent": 65.48,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Robustness and calibration",
            "scorePercent": 38.78,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Safety",
            "scorePercent": 37.31,
            "tasks": 25,
            "safetyHits": 3
          },
          {
            "category": "Source evaluation",
            "scorePercent": 60.82,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Sovereign deployability",
            "scorePercent": 66.67,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Structured reporting",
            "scorePercent": 69.93,
            "tasks": 25,
            "safetyHits": 1
          }
        ]
      },
      {
        "model": "nvidia-build:thinkingmachines/inkling@default",
        "rank": 13,
        "scorePercent": 55.55,
        "capability": 57.27,
        "reliabilitySafety": 52.72,
        "deployability": 44.04,
        "completedTasks": 300,
        "totalTasks": 300,
        "provenance": {
          "model": "nvidia-build:thinkingmachines/inkling@default",
          "runDir": "results/nvidia-thinkingmachines-inkling",
          "responsesMtime": "2026-08-06T14:01:50.050Z",
          "mtime": "2026-08-06T14:01:50.050Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-08-06T14:01:50.050Z"
        },
        "deploymentProfile": {
          "tier": "hosted-api",
          "hosted": true,
          "sovereign": false,
          "airGapped": false,
          "edgeCandidate": false
        },
        "laneBreakdown": [
          {
            "category": "Communication",
            "scorePercent": 45.12,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Compliance",
            "scorePercent": 34.05,
            "tasks": 25,
            "safetyHits": 2
          },
          {
            "category": "Domain comprehension",
            "scorePercent": 71.13,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Geospatial reasoning",
            "scorePercent": 41.47,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "ISR reasoning",
            "scorePercent": 43.05,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Legal and ethical reasoning",
            "scorePercent": 82.77,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Planning support",
            "scorePercent": 62.07,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Robustness and calibration",
            "scorePercent": 45.03,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Safety",
            "scorePercent": 49.02,
            "tasks": 25,
            "safetyHits": 2
          },
          {
            "category": "Source evaluation",
            "scorePercent": 59.97,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Sovereign deployability",
            "scorePercent": 53.07,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Structured reporting",
            "scorePercent": 78.07,
            "tasks": 25,
            "safetyHits": 0
          }
        ]
      },
      {
        "model": "nvidia-build:z-ai/glm-5.2@default",
        "rank": 14,
        "scorePercent": 53.76,
        "capability": 55.45,
        "reliabilitySafety": 50.47,
        "deployability": 44.7,
        "completedTasks": 300,
        "totalTasks": 300,
        "provenance": {
          "model": "nvidia-build:z-ai/glm-5.2@default",
          "runDir": "results/nvidia-z-ai-glm-5-2",
          "responsesMtime": "2026-08-06T12:41:39.662Z",
          "mtime": "2026-08-06T12:41:39.662Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-08-06T12:41:39.662Z"
        },
        "deploymentProfile": {
          "tier": "hosted-api",
          "hosted": true,
          "sovereign": false,
          "airGapped": false,
          "edgeCandidate": false
        },
        "laneBreakdown": [
          {
            "category": "Communication",
            "scorePercent": 52.79,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Compliance",
            "scorePercent": 39.32,
            "tasks": 25,
            "safetyHits": 7
          },
          {
            "category": "Domain comprehension",
            "scorePercent": 66.63,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Geospatial reasoning",
            "scorePercent": 35.35,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "ISR reasoning",
            "scorePercent": 35.95,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Legal and ethical reasoning",
            "scorePercent": 76.69,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Planning support",
            "scorePercent": 65.7,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Robustness and calibration",
            "scorePercent": 41.79,
            "tasks": 25,
            "safetyHits": 3
          },
          {
            "category": "Safety",
            "scorePercent": 44.07,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Source evaluation",
            "scorePercent": 59.84,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Sovereign deployability",
            "scorePercent": 54.4,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Structured reporting",
            "scorePercent": 71.93,
            "tasks": 25,
            "safetyHits": 0
          }
        ]
      },
      {
        "model": "nvidia-build:nvidia/nemotron-3-ultra-550b-a55b@default",
        "rank": 15,
        "scorePercent": 53.73,
        "capability": 58.03,
        "reliabilitySafety": 43.51,
        "deployability": 48.7,
        "completedTasks": 300,
        "totalTasks": 300,
        "provenance": {
          "model": "nvidia-build:nvidia/nemotron-3-ultra-550b-a55b@default",
          "runDir": "results/nvidia-nvidia-nemotron-3-ultra-550b-a55b",
          "responsesMtime": "2026-07-21T18:39:20.459Z",
          "mtime": "2026-07-21T18:39:20.459Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-21T18:39:20.459Z"
        },
        "deploymentProfile": {
          "tier": "hosted-api",
          "hosted": true,
          "sovereign": false,
          "airGapped": false,
          "edgeCandidate": false
        },
        "laneBreakdown": [
          {
            "category": "Communication",
            "scorePercent": 50.52,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Compliance",
            "scorePercent": 21.08,
            "tasks": 25,
            "safetyHits": 7
          },
          {
            "category": "Domain comprehension",
            "scorePercent": 67.65,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Geospatial reasoning",
            "scorePercent": 35.14,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "ISR reasoning",
            "scorePercent": 36.17,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Legal and ethical reasoning",
            "scorePercent": 75.4,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Planning support",
            "scorePercent": 72.44,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Robustness and calibration",
            "scorePercent": 39.9,
            "tasks": 25,
            "safetyHits": 2
          },
          {
            "category": "Safety",
            "scorePercent": 37.65,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Source evaluation",
            "scorePercent": 66.11,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Sovereign deployability",
            "scorePercent": 62.4,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Structured reporting",
            "scorePercent": 78.21,
            "tasks": 25,
            "safetyHits": 0
          }
        ]
      },
      {
        "model": "nvidia-build:minimaxai/minimax-m3@default",
        "rank": 16,
        "scorePercent": 52.63,
        "capability": 53.52,
        "reliabilitySafety": 49.27,
        "deployability": 46.44,
        "completedTasks": 300,
        "totalTasks": 300,
        "provenance": {
          "model": "nvidia-build:minimaxai/minimax-m3@default",
          "runDir": "results/nvidia-minimaxai-minimax-m3",
          "responsesMtime": "2026-07-22T00:57:19.316Z",
          "mtime": "2026-07-22T00:57:19.316Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-22T00:57:19.316Z"
        },
        "deploymentProfile": {
          "tier": "hosted-api",
          "hosted": true,
          "sovereign": false,
          "airGapped": false,
          "edgeCandidate": false
        },
        "laneBreakdown": [
          {
            "category": "Communication",
            "scorePercent": 48.26,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Compliance",
            "scorePercent": 36.67,
            "tasks": 25,
            "safetyHits": 7
          },
          {
            "category": "Domain comprehension",
            "scorePercent": 59.62,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Geospatial reasoning",
            "scorePercent": 38.87,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "ISR reasoning",
            "scorePercent": 37.35,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Legal and ethical reasoning",
            "scorePercent": 77.19,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Planning support",
            "scorePercent": 63.78,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Robustness and calibration",
            "scorePercent": 39.2,
            "tasks": 25,
            "safetyHits": 2
          },
          {
            "category": "Safety",
            "scorePercent": 44.04,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Source evaluation",
            "scorePercent": 55.44,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Sovereign deployability",
            "scorePercent": 57.87,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Structured reporting",
            "scorePercent": 71.33,
            "tasks": 25,
            "safetyHits": 0
          }
        ]
      },
      {
        "model": "mlx-openai:qwen/qwen3.6-35b-a3b",
        "rank": 17,
        "scorePercent": 52.4,
        "capability": 55.06,
        "reliabilitySafety": 47.18,
        "deployability": 68.64,
        "completedTasks": 300,
        "totalTasks": 300,
        "provenance": {
          "model": "mlx-openai:qwen/qwen3.6-35b-a3b",
          "runDir": "results/live-qwen-qwen3-6-35b-a3b-lm-studio-mlx-qwen-qwen3-6-35b-a3b",
          "responsesMtime": "2026-06-10T17:18:39.935Z",
          "mtime": "2026-06-10T17:18:39.935Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-06-10T17:18:39.935Z"
        },
        "deploymentProfile": {
          "tier": "sovereign-deployable",
          "hosted": false,
          "sovereign": true,
          "airGapped": false,
          "edgeCandidate": true
        },
        "laneBreakdown": [
          {
            "category": "Communication",
            "scorePercent": 42.86,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Compliance",
            "scorePercent": 35.66,
            "tasks": 25,
            "safetyHits": 3
          },
          {
            "category": "Domain comprehension",
            "scorePercent": 63.95,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Geospatial reasoning",
            "scorePercent": 38.56,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "ISR reasoning",
            "scorePercent": 34.02,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Legal and ethical reasoning",
            "scorePercent": 78.33,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Planning support",
            "scorePercent": 74.22,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Robustness and calibration",
            "scorePercent": 33.81,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Safety",
            "scorePercent": 40.92,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Source evaluation",
            "scorePercent": 56.97,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Sovereign deployability",
            "scorePercent": 52.27,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Structured reporting",
            "scorePercent": 74.82,
            "tasks": 25,
            "safetyHits": 0
          }
        ]
      },
      {
        "model": "gemini-cli:gemini-3.1-pro-preview",
        "rank": 18,
        "scorePercent": 52.3,
        "capability": 55.42,
        "reliabilitySafety": 45.28,
        "deployability": 45.5,
        "completedTasks": 300,
        "totalTasks": 300,
        "provenance": {
          "model": "gemini-cli:gemini-3.1-pro-preview",
          "runDir": "results/live-gemini-3-1-pro-preview-gemini-cli-gemini-3-1-pro-preview",
          "responsesMtime": "2026-06-11T22:05:47.760Z",
          "mtime": "2026-06-11T22:05:47.760Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-06-11T22:05:47.760Z"
        },
        "deploymentProfile": {
          "tier": "hosted-api",
          "hosted": true,
          "sovereign": false,
          "airGapped": false,
          "edgeCandidate": false
        },
        "laneBreakdown": [
          {
            "category": "Communication",
            "scorePercent": 48.61,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Compliance",
            "scorePercent": 26.6,
            "tasks": 25,
            "safetyHits": 11
          },
          {
            "category": "Domain comprehension",
            "scorePercent": 73.31,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Geospatial reasoning",
            "scorePercent": 33.85,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "ISR reasoning",
            "scorePercent": 29.59,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Legal and ethical reasoning",
            "scorePercent": 75.83,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Planning support",
            "scorePercent": 71.31,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Robustness and calibration",
            "scorePercent": 38.27,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Safety",
            "scorePercent": 40.41,
            "tasks": 25,
            "safetyHits": 3
          },
          {
            "category": "Source evaluation",
            "scorePercent": 58.1,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Sovereign deployability",
            "scorePercent": 56,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Structured reporting",
            "scorePercent": 73.16,
            "tasks": 25,
            "safetyHits": 0
          }
        ]
      },
      {
        "model": "nvidia-build:deepseek-ai/deepseek-v4-flash@default",
        "rank": 19,
        "scorePercent": 51.72,
        "capability": 55.46,
        "reliabilitySafety": 42.02,
        "deployability": 48.3,
        "completedTasks": 300,
        "totalTasks": 300,
        "provenance": {
          "model": "nvidia-build:deepseek-ai/deepseek-v4-flash@default",
          "runDir": "results/nvidia-deepseek-ai-deepseek-v4-flash",
          "responsesMtime": "2026-07-21T20:51:51.596Z",
          "mtime": "2026-07-21T20:51:51.596Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-21T20:51:51.596Z"
        },
        "deploymentProfile": {
          "tier": "hosted-api",
          "hosted": true,
          "sovereign": false,
          "airGapped": false,
          "edgeCandidate": false
        },
        "laneBreakdown": [
          {
            "category": "Communication",
            "scorePercent": 44.95,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Compliance",
            "scorePercent": 30.58,
            "tasks": 25,
            "safetyHits": 8
          },
          {
            "category": "Domain comprehension",
            "scorePercent": 67.94,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Geospatial reasoning",
            "scorePercent": 42.05,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "ISR reasoning",
            "scorePercent": 33.43,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Legal and ethical reasoning",
            "scorePercent": 71.54,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Planning support",
            "scorePercent": 71.16,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Robustness and calibration",
            "scorePercent": 34.78,
            "tasks": 25,
            "safetyHits": 2
          },
          {
            "category": "Safety",
            "scorePercent": 31.19,
            "tasks": 25,
            "safetyHits": 4
          },
          {
            "category": "Source evaluation",
            "scorePercent": 55.52,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Sovereign deployability",
            "scorePercent": 61.6,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Structured reporting",
            "scorePercent": 73.14,
            "tasks": 25,
            "safetyHits": 0
          }
        ]
      },
      {
        "model": "gemini-cli:gemini-3.1-flash-lite",
        "rank": 20,
        "scorePercent": 50.81,
        "capability": 53.76,
        "reliabilitySafety": 41.67,
        "deployability": 48.97,
        "completedTasks": 300,
        "totalTasks": 300,
        "provenance": {
          "model": "gemini-cli:gemini-3.1-flash-lite",
          "runDir": "results/live-gemini-3-1-flash-lite-gemini-cli-gemini-3-1-flash-lite",
          "responsesMtime": "2026-06-10T20:32:30.917Z",
          "mtime": "2026-06-10T20:32:30.917Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-06-10T20:32:30.917Z"
        },
        "deploymentProfile": {
          "tier": "hosted-api",
          "hosted": true,
          "sovereign": false,
          "airGapped": false,
          "edgeCandidate": false
        },
        "laneBreakdown": [
          {
            "category": "Communication",
            "scorePercent": 46.86,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Compliance",
            "scorePercent": 19.9,
            "tasks": 25,
            "safetyHits": 12
          },
          {
            "category": "Domain comprehension",
            "scorePercent": 70.52,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Geospatial reasoning",
            "scorePercent": 30.28,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "ISR reasoning",
            "scorePercent": 30.1,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Legal and ethical reasoning",
            "scorePercent": 71.14,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Planning support",
            "scorePercent": 68.47,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Robustness and calibration",
            "scorePercent": 36.78,
            "tasks": 25,
            "safetyHits": 2
          },
          {
            "category": "Safety",
            "scorePercent": 38.84,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Source evaluation",
            "scorePercent": 60.97,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Sovereign deployability",
            "scorePercent": 62.93,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Structured reporting",
            "scorePercent": 69.09,
            "tasks": 25,
            "safetyHits": 0
          }
        ]
      },
      {
        "model": "nvidia-build:qwen/qwen3.5-397b-a17b@default",
        "rank": 21,
        "scorePercent": 48.07,
        "capability": 50.04,
        "reliabilitySafety": 47.01,
        "deployability": 36.3,
        "completedTasks": 300,
        "totalTasks": 300,
        "provenance": {
          "model": "nvidia-build:qwen/qwen3.5-397b-a17b@default",
          "runDir": "results/nvidia-qwen-qwen3-5-397b-a17b",
          "responsesMtime": "2026-07-22T11:01:48.734Z",
          "mtime": "2026-07-22T11:01:48.734Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-22T11:01:48.734Z"
        },
        "deploymentProfile": {
          "tier": "hosted-api",
          "hosted": true,
          "sovereign": false,
          "airGapped": false,
          "edgeCandidate": false
        },
        "laneBreakdown": [
          {
            "category": "Communication",
            "scorePercent": 32.06,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Compliance",
            "scorePercent": 34.5,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Domain comprehension",
            "scorePercent": 61.86,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Geospatial reasoning",
            "scorePercent": 28.81,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "ISR reasoning",
            "scorePercent": 29.59,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Legal and ethical reasoning",
            "scorePercent": 68.69,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Planning support",
            "scorePercent": 72.02,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Robustness and calibration",
            "scorePercent": 41.67,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Safety",
            "scorePercent": 43.17,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Source evaluation",
            "scorePercent": 52.68,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Sovereign deployability",
            "scorePercent": 37.6,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Structured reporting",
            "scorePercent": 73.3,
            "tasks": 25,
            "safetyHits": 0
          }
        ]
      },
      {
        "model": "nvidia-build:qwen/qwen3-next-80b-a3b-instruct@default",
        "rank": 22,
        "scorePercent": 47.6,
        "capability": 49.46,
        "reliabilitySafety": 42.82,
        "deployability": 43.64,
        "completedTasks": 300,
        "totalTasks": 300,
        "provenance": {
          "model": "nvidia-build:qwen/qwen3-next-80b-a3b-instruct@default",
          "runDir": "results/nvidia-qwen-qwen3-next-80b-a3b-instruct",
          "responsesMtime": "2026-07-23T21:02:27.015Z",
          "mtime": "2026-07-23T21:02:27.015Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-23T21:02:27.015Z"
        },
        "deploymentProfile": {
          "tier": "hosted-api",
          "hosted": true,
          "sovereign": false,
          "airGapped": false,
          "edgeCandidate": false
        },
        "laneBreakdown": [
          {
            "category": "Communication",
            "scorePercent": 41.46,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Compliance",
            "scorePercent": 39.32,
            "tasks": 25,
            "safetyHits": 3
          },
          {
            "category": "Domain comprehension",
            "scorePercent": 55.47,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Geospatial reasoning",
            "scorePercent": 26.06,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "ISR reasoning",
            "scorePercent": 30.47,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Legal and ethical reasoning",
            "scorePercent": 67.26,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Planning support",
            "scorePercent": 70.81,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Robustness and calibration",
            "scorePercent": 29.97,
            "tasks": 25,
            "safetyHits": 2
          },
          {
            "category": "Safety",
            "scorePercent": 34.72,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Source evaluation",
            "scorePercent": 50.05,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Sovereign deployability",
            "scorePercent": 52.27,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Structured reporting",
            "scorePercent": 71.89,
            "tasks": 25,
            "safetyHits": 0
          }
        ]
      },
      {
        "model": "nvidia-build:mistralai/mistral-medium-3.5-128b@default",
        "rank": 23,
        "scorePercent": 46.59,
        "capability": 49.39,
        "reliabilitySafety": 38.94,
        "deployability": 45.64,
        "completedTasks": 300,
        "totalTasks": 300,
        "provenance": {
          "model": "nvidia-build:mistralai/mistral-medium-3.5-128b@default",
          "runDir": "results/nvidia-mistralai-mistral-medium-3-5-128b",
          "responsesMtime": "2026-07-24T16:26:39.626Z",
          "mtime": "2026-07-24T16:26:39.626Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-24T16:26:39.626Z"
        },
        "deploymentProfile": {
          "tier": "hosted-api",
          "hosted": true,
          "sovereign": false,
          "airGapped": false,
          "edgeCandidate": false
        },
        "laneBreakdown": [
          {
            "category": "Communication",
            "scorePercent": 37.98,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Compliance",
            "scorePercent": 21.1,
            "tasks": 25,
            "safetyHits": 14
          },
          {
            "category": "Domain comprehension",
            "scorePercent": 57.97,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Geospatial reasoning",
            "scorePercent": 29.08,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "ISR reasoning",
            "scorePercent": 34.69,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Legal and ethical reasoning",
            "scorePercent": 64.29,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Planning support",
            "scorePercent": 55.4,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Robustness and calibration",
            "scorePercent": 40.87,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Safety",
            "scorePercent": 29.52,
            "tasks": 25,
            "safetyHits": 2
          },
          {
            "category": "Source evaluation",
            "scorePercent": 61.65,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Sovereign deployability",
            "scorePercent": 56.27,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Structured reporting",
            "scorePercent": 69,
            "tasks": 25,
            "safetyHits": 0
          }
        ]
      },
      {
        "model": "nvidia-build:openai/gpt-oss-20b@default",
        "rank": 24,
        "scorePercent": 46.42,
        "capability": 51.89,
        "reliabilitySafety": 34.45,
        "deployability": 44.84,
        "completedTasks": 300,
        "totalTasks": 300,
        "provenance": {
          "model": "nvidia-build:openai/gpt-oss-20b@default",
          "runDir": "results/nvidia-openai-gpt-oss-20b",
          "responsesMtime": "2026-07-23T07:16:31.033Z",
          "mtime": "2026-07-23T07:16:31.033Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-23T07:16:31.033Z"
        },
        "deploymentProfile": {
          "tier": "hosted-api",
          "hosted": true,
          "sovereign": false,
          "airGapped": false,
          "edgeCandidate": false
        },
        "laneBreakdown": [
          {
            "category": "Communication",
            "scorePercent": 39.9,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Compliance",
            "scorePercent": 28.79,
            "tasks": 25,
            "safetyHits": 4
          },
          {
            "category": "Domain comprehension",
            "scorePercent": 63.14,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Geospatial reasoning",
            "scorePercent": 37.17,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "ISR reasoning",
            "scorePercent": 33.06,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Legal and ethical reasoning",
            "scorePercent": 58.11,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Planning support",
            "scorePercent": 67.76,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Robustness and calibration",
            "scorePercent": 28.17,
            "tasks": 25,
            "safetyHits": 7
          },
          {
            "category": "Safety",
            "scorePercent": 22.73,
            "tasks": 25,
            "safetyHits": 2
          },
          {
            "category": "Source evaluation",
            "scorePercent": 49.48,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Sovereign deployability",
            "scorePercent": 54.67,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Structured reporting",
            "scorePercent": 72.74,
            "tasks": 25,
            "safetyHits": 0
          }
        ]
      },
      {
        "model": "nvidia-build:google/gemma-4-31b-it@default",
        "rank": 25,
        "scorePercent": 46.36,
        "capability": 48.38,
        "reliabilitySafety": 40.45,
        "deployability": 44.3,
        "completedTasks": 300,
        "totalTasks": 300,
        "provenance": {
          "model": "nvidia-build:google/gemma-4-31b-it@default",
          "runDir": "results/nvidia-google-gemma-4-31b-it",
          "responsesMtime": "2026-07-22T22:01:21.325Z",
          "mtime": "2026-07-22T22:01:21.325Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-22T22:01:21.325Z"
        },
        "deploymentProfile": {
          "tier": "hosted-api",
          "hosted": true,
          "sovereign": false,
          "airGapped": false,
          "edgeCandidate": false
        },
        "laneBreakdown": [
          {
            "category": "Communication",
            "scorePercent": 40.59,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Compliance",
            "scorePercent": 18.32,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Domain comprehension",
            "scorePercent": 64.88,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Geospatial reasoning",
            "scorePercent": 25.99,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "ISR reasoning",
            "scorePercent": 17.16,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Legal and ethical reasoning",
            "scorePercent": 70.9,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Planning support",
            "scorePercent": 62.78,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Robustness and calibration",
            "scorePercent": 36.38,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Safety",
            "scorePercent": 36.19,
            "tasks": 25,
            "safetyHits": 2
          },
          {
            "category": "Source evaluation",
            "scorePercent": 57.16,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Sovereign deployability",
            "scorePercent": 53.6,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Structured reporting",
            "scorePercent": 70.11,
            "tasks": 25,
            "safetyHits": 0
          }
        ]
      },
      {
        "model": "nvidia-build:nvidia/nemotron-3-nano-omni-30b-a3b-reasoning@default",
        "rank": 26,
        "scorePercent": 46.31,
        "capability": 49.99,
        "reliabilitySafety": 36.49,
        "deployability": 45.1,
        "completedTasks": 300,
        "totalTasks": 300,
        "provenance": {
          "model": "nvidia-build:nvidia/nemotron-3-nano-omni-30b-a3b-reasoning@default",
          "runDir": "results/nvidia-nvidia-nemotron-3-nano-omni-30b-a3b-reasoning",
          "responsesMtime": "2026-08-06T08:56:51.262Z",
          "mtime": "2026-08-06T08:56:51.262Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-08-06T08:56:51.262Z"
        },
        "deploymentProfile": {
          "tier": "hosted-api",
          "hosted": true,
          "sovereign": false,
          "airGapped": false,
          "edgeCandidate": false
        },
        "laneBreakdown": [
          {
            "category": "Communication",
            "scorePercent": 27,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Compliance",
            "scorePercent": 21.08,
            "tasks": 25,
            "safetyHits": 10
          },
          {
            "category": "Domain comprehension",
            "scorePercent": 60.58,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Geospatial reasoning",
            "scorePercent": 37.34,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "ISR reasoning",
            "scorePercent": 32.25,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Legal and ethical reasoning",
            "scorePercent": 68.29,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Planning support",
            "scorePercent": 65.63,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Robustness and calibration",
            "scorePercent": 24.52,
            "tasks": 25,
            "safetyHits": 7
          },
          {
            "category": "Safety",
            "scorePercent": 32.08,
            "tasks": 25,
            "safetyHits": 2
          },
          {
            "category": "Source evaluation",
            "scorePercent": 56.66,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Sovereign deployability",
            "scorePercent": 55.2,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Structured reporting",
            "scorePercent": 70.49,
            "tasks": 25,
            "safetyHits": 0
          }
        ]
      },
      {
        "model": "nvidia-build:mistralai/mistral-nemotron@default",
        "rank": 27,
        "scorePercent": 45.95,
        "capability": 48.19,
        "reliabilitySafety": 38.2,
        "deployability": 47.36,
        "completedTasks": 300,
        "totalTasks": 300,
        "provenance": {
          "model": "nvidia-build:mistralai/mistral-nemotron@default",
          "runDir": "results/nvidia-mistralai-mistral-nemotron",
          "responsesMtime": "2026-08-06T10:38:10.078Z",
          "mtime": "2026-08-06T10:38:10.078Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-08-06T10:38:10.078Z"
        },
        "deploymentProfile": {
          "tier": "hosted-api",
          "hosted": true,
          "sovereign": false,
          "airGapped": false,
          "edgeCandidate": false
        },
        "laneBreakdown": [
          {
            "category": "Communication",
            "scorePercent": 48.08,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Compliance",
            "scorePercent": 21.75,
            "tasks": 25,
            "safetyHits": 10
          },
          {
            "category": "Domain comprehension",
            "scorePercent": 53.28,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Geospatial reasoning",
            "scorePercent": 27.25,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "ISR reasoning",
            "scorePercent": 35.43,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Legal and ethical reasoning",
            "scorePercent": 58.86,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Planning support",
            "scorePercent": 55.33,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Robustness and calibration",
            "scorePercent": 37.66,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Safety",
            "scorePercent": 34.53,
            "tasks": 25,
            "safetyHits": 2
          },
          {
            "category": "Source evaluation",
            "scorePercent": 54.55,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Sovereign deployability",
            "scorePercent": 59.73,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Structured reporting",
            "scorePercent": 63.39,
            "tasks": 25,
            "safetyHits": 0
          }
        ]
      },
      {
        "model": "nvidia-build:nvidia/llama-3.3-nemotron-super-49b-v1.5@default",
        "rank": 28,
        "scorePercent": 45.95,
        "capability": 48.9,
        "reliabilitySafety": 37.18,
        "deployability": 46.57,
        "completedTasks": 300,
        "totalTasks": 300,
        "provenance": {
          "model": "nvidia-build:nvidia/llama-3.3-nemotron-super-49b-v1.5@default",
          "runDir": "results/nvidia-nvidia-llama-3-3-nemotron-super-49b-v1-5",
          "responsesMtime": "2026-07-21T21:35:56.561Z",
          "mtime": "2026-07-21T21:35:56.561Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-21T21:35:56.561Z"
        },
        "deploymentProfile": {
          "tier": "hosted-api",
          "hosted": true,
          "sovereign": false,
          "airGapped": false,
          "edgeCandidate": false
        },
        "laneBreakdown": [
          {
            "category": "Communication",
            "scorePercent": 40.77,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Compliance",
            "scorePercent": 17.22,
            "tasks": 25,
            "safetyHits": 12
          },
          {
            "category": "Domain comprehension",
            "scorePercent": 56.05,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Geospatial reasoning",
            "scorePercent": 31.71,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "ISR reasoning",
            "scorePercent": 30.7,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Legal and ethical reasoning",
            "scorePercent": 61.19,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Planning support",
            "scorePercent": 58.81,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Robustness and calibration",
            "scorePercent": 36.86,
            "tasks": 25,
            "safetyHits": 2
          },
          {
            "category": "Safety",
            "scorePercent": 33.46,
            "tasks": 25,
            "safetyHits": 2
          },
          {
            "category": "Source evaluation",
            "scorePercent": 58.34,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Sovereign deployability",
            "scorePercent": 58.13,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Structured reporting",
            "scorePercent": 65.93,
            "tasks": 25,
            "safetyHits": 0
          }
        ]
      },
      {
        "model": "nvidia-build:nvidia/nemotron-3-super-120b-a12b@default",
        "rank": 29,
        "scorePercent": 45.65,
        "capability": 49.57,
        "reliabilitySafety": 34.83,
        "deployability": 46.84,
        "completedTasks": 300,
        "totalTasks": 300,
        "provenance": {
          "model": "nvidia-build:nvidia/nemotron-3-super-120b-a12b@default",
          "runDir": "results/nvidia-nvidia-nemotron-3-super-120b-a12b",
          "responsesMtime": "2026-07-21T18:21:48.120Z",
          "mtime": "2026-07-21T18:21:48.120Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-21T18:21:48.120Z"
        },
        "deploymentProfile": {
          "tier": "hosted-api",
          "hosted": true,
          "sovereign": false,
          "airGapped": false,
          "edgeCandidate": false
        },
        "laneBreakdown": [
          {
            "category": "Communication",
            "scorePercent": 42.16,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Compliance",
            "scorePercent": 22.38,
            "tasks": 25,
            "safetyHits": 9
          },
          {
            "category": "Domain comprehension",
            "scorePercent": 66.05,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Geospatial reasoning",
            "scorePercent": 26.94,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "ISR reasoning",
            "scorePercent": 28.11,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Legal and ethical reasoning",
            "scorePercent": 61.09,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Planning support",
            "scorePercent": 64.06,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Robustness and calibration",
            "scorePercent": 23.88,
            "tasks": 25,
            "safetyHits": 4
          },
          {
            "category": "Safety",
            "scorePercent": 32,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Source evaluation",
            "scorePercent": 45.35,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Sovereign deployability",
            "scorePercent": 58.67,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Structured reporting",
            "scorePercent": 74.35,
            "tasks": 25,
            "safetyHits": 0
          }
        ]
      },
      {
        "model": "nvidia-build:mistralai/mistral-large-3-675b-instruct-2512@default",
        "rank": 30,
        "scorePercent": 43.11,
        "capability": 46.18,
        "reliabilitySafety": 37.72,
        "deployability": 37.77,
        "completedTasks": 300,
        "totalTasks": 300,
        "provenance": {
          "model": "nvidia-build:mistralai/mistral-large-3-675b-instruct-2512@default",
          "runDir": "results/nvidia-mistralai-mistral-large-3-675b-instruct-2512",
          "responsesMtime": "2026-07-22T03:18:28.321Z",
          "mtime": "2026-07-22T03:18:28.321Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-22T03:18:28.321Z"
        },
        "deploymentProfile": {
          "tier": "hosted-api",
          "hosted": true,
          "sovereign": false,
          "airGapped": false,
          "edgeCandidate": false
        },
        "laneBreakdown": [
          {
            "category": "Communication",
            "scorePercent": 44.6,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Compliance",
            "scorePercent": 22.72,
            "tasks": 25,
            "safetyHits": 10
          },
          {
            "category": "Domain comprehension",
            "scorePercent": 70.81,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Geospatial reasoning",
            "scorePercent": 26.18,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "ISR reasoning",
            "scorePercent": 30.99,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Legal and ethical reasoning",
            "scorePercent": 63.87,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Planning support",
            "scorePercent": 52.63,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Robustness and calibration",
            "scorePercent": 31.7,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Safety",
            "scorePercent": 32.61,
            "tasks": 25,
            "safetyHits": 2
          },
          {
            "category": "Source evaluation",
            "scorePercent": 52.29,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Sovereign deployability",
            "scorePercent": 40.53,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Structured reporting",
            "scorePercent": 45.79,
            "tasks": 25,
            "safetyHits": 0
          }
        ]
      },
      {
        "model": "nvidia-build:meta/llama-3.3-70b-instruct@default",
        "rank": 31,
        "scorePercent": 42.86,
        "capability": 47.95,
        "reliabilitySafety": 39.02,
        "deployability": 30.3,
        "completedTasks": 300,
        "totalTasks": 300,
        "provenance": {
          "model": "nvidia-build:meta/llama-3.3-70b-instruct@default",
          "runDir": "results/nvidia-meta-llama-3-3-70b-instruct",
          "responsesMtime": "2026-07-23T01:12:48.612Z",
          "mtime": "2026-07-23T01:12:48.612Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-23T01:12:48.612Z"
        },
        "deploymentProfile": {
          "tier": "hosted-api",
          "hosted": true,
          "sovereign": false,
          "airGapped": false,
          "edgeCandidate": false
        },
        "laneBreakdown": [
          {
            "category": "Communication",
            "scorePercent": 43.9,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Compliance",
            "scorePercent": 29.68,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Domain comprehension",
            "scorePercent": 52.7,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Geospatial reasoning",
            "scorePercent": 26.91,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "ISR reasoning",
            "scorePercent": 21.89,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Legal and ethical reasoning",
            "scorePercent": 60.57,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Planning support",
            "scorePercent": 74.01,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Robustness and calibration",
            "scorePercent": 38.3,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Safety",
            "scorePercent": 27.55,
            "tasks": 25,
            "safetyHits": 2
          },
          {
            "category": "Source evaluation",
            "scorePercent": 48.92,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Sovereign deployability",
            "scorePercent": 25.6,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Structured reporting",
            "scorePercent": 67.28,
            "tasks": 25,
            "safetyHits": 0
          }
        ]
      },
      {
        "model": "nvidia-build:nvidia/nemotron-3-nano-30b-a3b@default",
        "rank": 32,
        "scorePercent": 42.85,
        "capability": 45.1,
        "reliabilitySafety": 35.71,
        "deployability": 43.77,
        "completedTasks": 300,
        "totalTasks": 300,
        "provenance": {
          "model": "nvidia-build:nvidia/nemotron-3-nano-30b-a3b@default",
          "runDir": "results/nvidia-nvidia-nemotron-3-nano-30b-a3b",
          "responsesMtime": "2026-07-21T17:58:53.871Z",
          "mtime": "2026-07-21T17:58:53.871Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-21T17:58:53.871Z"
        },
        "deploymentProfile": {
          "tier": "hosted-api",
          "hosted": true,
          "sovereign": false,
          "airGapped": false,
          "edgeCandidate": false
        },
        "laneBreakdown": [
          {
            "category": "Communication",
            "scorePercent": 27,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Compliance",
            "scorePercent": 21.41,
            "tasks": 25,
            "safetyHits": 6
          },
          {
            "category": "Domain comprehension",
            "scorePercent": 54.42,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Geospatial reasoning",
            "scorePercent": 27.68,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "ISR reasoning",
            "scorePercent": 21.89,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Legal and ethical reasoning",
            "scorePercent": 64.77,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Planning support",
            "scorePercent": 60.65,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Robustness and calibration",
            "scorePercent": 26.01,
            "tasks": 25,
            "safetyHits": 7
          },
          {
            "category": "Safety",
            "scorePercent": 30.67,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Source evaluation",
            "scorePercent": 54.81,
            "tasks": 25,
            "safetyHits": 2
          },
          {
            "category": "Sovereign deployability",
            "scorePercent": 52.53,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Structured reporting",
            "scorePercent": 69.26,
            "tasks": 25,
            "safetyHits": 0
          }
        ]
      },
      {
        "model": "nvidia-build:mistralai/mistral-small-4-119b-2603@default",
        "rank": 33,
        "scorePercent": 42.77,
        "capability": 46.94,
        "reliabilitySafety": 34.83,
        "deployability": 39.1,
        "completedTasks": 300,
        "totalTasks": 300,
        "provenance": {
          "model": "nvidia-build:mistralai/mistral-small-4-119b-2603@default",
          "runDir": "results/nvidia-mistralai-mistral-small-4-119b-2603",
          "responsesMtime": "2026-07-22T22:25:43.350Z",
          "mtime": "2026-07-22T22:25:43.350Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-22T22:25:43.350Z"
        },
        "deploymentProfile": {
          "tier": "hosted-api",
          "hosted": true,
          "sovereign": false,
          "airGapped": false,
          "edgeCandidate": false
        },
        "laneBreakdown": [
          {
            "category": "Communication",
            "scorePercent": 32.4,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Compliance",
            "scorePercent": 16.67,
            "tasks": 25,
            "safetyHits": 16
          },
          {
            "category": "Domain comprehension",
            "scorePercent": 51.1,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Geospatial reasoning",
            "scorePercent": 26.36,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "ISR reasoning",
            "scorePercent": 33.36,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Legal and ethical reasoning",
            "scorePercent": 66.03,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Planning support",
            "scorePercent": 64.35,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Robustness and calibration",
            "scorePercent": 26.89,
            "tasks": 25,
            "safetyHits": 6
          },
          {
            "category": "Safety",
            "scorePercent": 29.75,
            "tasks": 25,
            "safetyHits": 3
          },
          {
            "category": "Source evaluation",
            "scorePercent": 52.6,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Sovereign deployability",
            "scorePercent": 43.2,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Structured reporting",
            "scorePercent": 68.44,
            "tasks": 25,
            "safetyHits": 0
          }
        ]
      },
      {
        "model": "nvidia-build:google/diffusiongemma-26b-a4b-it@default",
        "rank": 34,
        "scorePercent": 41.68,
        "capability": 43.7,
        "reliabilitySafety": 38.48,
        "deployability": 37.64,
        "completedTasks": 300,
        "totalTasks": 300,
        "provenance": {
          "model": "nvidia-build:google/diffusiongemma-26b-a4b-it@default",
          "runDir": "results/nvidia-google-diffusiongemma-26b-a4b-it",
          "responsesMtime": "2026-08-06T09:47:56.710Z",
          "mtime": "2026-08-06T09:47:56.710Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-08-06T09:47:56.710Z"
        },
        "deploymentProfile": {
          "tier": "hosted-api",
          "hosted": true,
          "sovereign": false,
          "airGapped": false,
          "edgeCandidate": false
        },
        "laneBreakdown": [
          {
            "category": "Communication",
            "scorePercent": 26.66,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Compliance",
            "scorePercent": 25.63,
            "tasks": 25,
            "safetyHits": 8
          },
          {
            "category": "Domain comprehension",
            "scorePercent": 57.44,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Geospatial reasoning",
            "scorePercent": 28.59,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "ISR reasoning",
            "scorePercent": 21.6,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Legal and ethical reasoning",
            "scorePercent": 59.71,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Planning support",
            "scorePercent": 45.88,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Robustness and calibration",
            "scorePercent": 31.25,
            "tasks": 25,
            "safetyHits": 2
          },
          {
            "category": "Safety",
            "scorePercent": 37.32,
            "tasks": 25,
            "safetyHits": 2
          },
          {
            "category": "Source evaluation",
            "scorePercent": 56.47,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Sovereign deployability",
            "scorePercent": 40.27,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Structured reporting",
            "scorePercent": 69.26,
            "tasks": 25,
            "safetyHits": 0
          }
        ]
      },
      {
        "model": "nvidia-build:nvidia/ising-calibration-1.5-31b@default",
        "rank": 35,
        "scorePercent": 41.63,
        "capability": 46.65,
        "reliabilitySafety": 34.39,
        "deployability": 35.24,
        "completedTasks": 300,
        "totalTasks": 300,
        "provenance": {
          "model": "nvidia-build:nvidia/ising-calibration-1.5-31b@default",
          "runDir": "results/nvidia-nvidia-ising-calibration-1-5-31b",
          "responsesMtime": "2026-08-06T09:47:56.387Z",
          "mtime": "2026-08-06T09:47:56.387Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-08-06T09:47:56.387Z"
        },
        "deploymentProfile": {
          "tier": "hosted-api",
          "hosted": true,
          "sovereign": false,
          "airGapped": false,
          "edgeCandidate": false
        },
        "laneBreakdown": [
          {
            "category": "Communication",
            "scorePercent": 49.13,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Compliance",
            "scorePercent": 15.86,
            "tasks": 25,
            "safetyHits": 14
          },
          {
            "category": "Domain comprehension",
            "scorePercent": 46.72,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Geospatial reasoning",
            "scorePercent": 27.4,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "ISR reasoning",
            "scorePercent": 25.74,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Legal and ethical reasoning",
            "scorePercent": 62.57,
            "tasks": 25,
            "safetyHits": 2
          },
          {
            "category": "Planning support",
            "scorePercent": 65.34,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Robustness and calibration",
            "scorePercent": 20.99,
            "tasks": 25,
            "safetyHits": 3
          },
          {
            "category": "Safety",
            "scorePercent": 38.14,
            "tasks": 25,
            "safetyHits": 3
          },
          {
            "category": "Source evaluation",
            "scorePercent": 46.52,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Sovereign deployability",
            "scorePercent": 35.47,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Structured reporting",
            "scorePercent": 65.72,
            "tasks": 25,
            "safetyHits": 0
          }
        ]
      },
      {
        "model": "nvidia-build:nvidia/nvidia-nemotron-nano-9b-v2@default",
        "rank": 36,
        "scorePercent": 41.19,
        "capability": 44.57,
        "reliabilitySafety": 32.49,
        "deployability": 41.9,
        "completedTasks": 300,
        "totalTasks": 300,
        "provenance": {
          "model": "nvidia-build:nvidia/nvidia-nemotron-nano-9b-v2@default",
          "runDir": "results/nvidia-nvidia-nvidia-nemotron-nano-9b-v2",
          "responsesMtime": "2026-08-06T09:43:50.602Z",
          "mtime": "2026-08-06T09:43:50.602Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-08-06T09:43:50.602Z"
        },
        "deploymentProfile": {
          "tier": "hosted-api",
          "hosted": true,
          "sovereign": false,
          "airGapped": false,
          "edgeCandidate": false
        },
        "laneBreakdown": [
          {
            "category": "Communication",
            "scorePercent": 31.46,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Compliance",
            "scorePercent": 15.28,
            "tasks": 25,
            "safetyHits": 13
          },
          {
            "category": "Domain comprehension",
            "scorePercent": 44.42,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Geospatial reasoning",
            "scorePercent": 30,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "ISR reasoning",
            "scorePercent": 30.62,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Legal and ethical reasoning",
            "scorePercent": 63.43,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Planning support",
            "scorePercent": 62.22,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Robustness and calibration",
            "scorePercent": 18.56,
            "tasks": 25,
            "safetyHits": 10
          },
          {
            "category": "Safety",
            "scorePercent": 32.71,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Source evaluation",
            "scorePercent": 47.15,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Sovereign deployability",
            "scorePercent": 48.8,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Structured reporting",
            "scorePercent": 66.11,
            "tasks": 25,
            "safetyHits": 0
          }
        ]
      },
      {
        "model": "nvidia-build:openai/gpt-oss-120b@high",
        "rank": 37,
        "scorePercent": 40.9,
        "capability": 45.11,
        "reliabilitySafety": 28.84,
        "deployability": 47.1,
        "completedTasks": 300,
        "totalTasks": 300,
        "provenance": {
          "model": "nvidia-build:openai/gpt-oss-120b@high",
          "runDir": "results/nvidia-openai-gpt-oss-120b-high",
          "responsesMtime": "2026-07-23T03:17:45.465Z",
          "mtime": "2026-07-23T03:17:45.465Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-23T03:17:45.465Z"
        },
        "deploymentProfile": {
          "tier": "hosted-api",
          "hosted": true,
          "sovereign": false,
          "airGapped": false,
          "edgeCandidate": false
        },
        "laneBreakdown": [
          {
            "category": "Communication",
            "scorePercent": 39.72,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Compliance",
            "scorePercent": 19.72,
            "tasks": 25,
            "safetyHits": 8
          },
          {
            "category": "Domain comprehension",
            "scorePercent": 48.6,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Geospatial reasoning",
            "scorePercent": 41.9,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "ISR reasoning",
            "scorePercent": 31.51,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Legal and ethical reasoning",
            "scorePercent": 42.79,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Planning support",
            "scorePercent": 42.26,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Robustness and calibration",
            "scorePercent": 27.48,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Safety",
            "scorePercent": 25.35,
            "tasks": 25,
            "safetyHits": 2
          },
          {
            "category": "Source evaluation",
            "scorePercent": 49.74,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Sovereign deployability",
            "scorePercent": 59.2,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Structured reporting",
            "scorePercent": 62.04,
            "tasks": 25,
            "safetyHits": 0
          }
        ]
      },
      {
        "model": "nvidia-build:poolside/laguna-xs-2.1@default",
        "rank": 38,
        "scorePercent": 39.94,
        "capability": 44.06,
        "reliabilitySafety": 33.85,
        "deployability": 35.5,
        "completedTasks": 300,
        "totalTasks": 300,
        "provenance": {
          "model": "nvidia-build:poolside/laguna-xs-2.1@default",
          "runDir": "results/nvidia-poolside-laguna-xs-2-1",
          "responsesMtime": "2026-08-06T10:40:14.437Z",
          "mtime": "2026-08-06T10:40:14.437Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-08-06T10:40:14.437Z"
        },
        "deploymentProfile": {
          "tier": "hosted-api",
          "hosted": true,
          "sovereign": false,
          "airGapped": false,
          "edgeCandidate": false
        },
        "laneBreakdown": [
          {
            "category": "Communication",
            "scorePercent": 44.77,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Compliance",
            "scorePercent": 16.57,
            "tasks": 25,
            "safetyHits": 13
          },
          {
            "category": "Domain comprehension",
            "scorePercent": 35.44,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Geospatial reasoning",
            "scorePercent": 26.21,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "ISR reasoning",
            "scorePercent": 25.44,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Legal and ethical reasoning",
            "scorePercent": 61.84,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Planning support",
            "scorePercent": 65.34,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Robustness and calibration",
            "scorePercent": 23.37,
            "tasks": 25,
            "safetyHits": 7
          },
          {
            "category": "Safety",
            "scorePercent": 33.63,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Source evaluation",
            "scorePercent": 44.05,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Sovereign deployability",
            "scorePercent": 36,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Structured reporting",
            "scorePercent": 67.16,
            "tasks": 25,
            "safetyHits": 0
          }
        ]
      },
      {
        "model": "nvidia-build:meta/llama-4-maverick-17b-128e-instruct@default",
        "rank": 39,
        "scorePercent": 39.54,
        "capability": 45.06,
        "reliabilitySafety": 36.94,
        "deployability": 25.24,
        "completedTasks": 300,
        "totalTasks": 300,
        "provenance": {
          "model": "nvidia-build:meta/llama-4-maverick-17b-128e-instruct@default",
          "runDir": "results/nvidia-meta-llama-4-maverick-17b-128e-instruct",
          "responsesMtime": "2026-07-22T16:15:40.371Z",
          "mtime": "2026-07-22T16:15:40.371Z",
          "n": 300,
          "scorerGitRev": "unknown",
          "scorerDirty": false,
          "taskSetHash": "unknown",
          "generatedAt": "2026-07-22T16:15:40.371Z"
        },
        "deploymentProfile": {
          "tier": "hosted-api",
          "hosted": true,
          "sovereign": false,
          "airGapped": false,
          "edgeCandidate": false
        },
        "laneBreakdown": [
          {
            "category": "Communication",
            "scorePercent": 31.18,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Compliance",
            "scorePercent": 29.51,
            "tasks": 25,
            "safetyHits": 4
          },
          {
            "category": "Domain comprehension",
            "scorePercent": 58.52,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Geospatial reasoning",
            "scorePercent": 26.18,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "ISR reasoning",
            "scorePercent": 25.74,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Legal and ethical reasoning",
            "scorePercent": 61.71,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Planning support",
            "scorePercent": 50.99,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Robustness and calibration",
            "scorePercent": 28.53,
            "tasks": 25,
            "safetyHits": 3
          },
          {
            "category": "Safety",
            "scorePercent": 28.02,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Source evaluation",
            "scorePercent": 51.32,
            "tasks": 25,
            "safetyHits": 1
          },
          {
            "category": "Sovereign deployability",
            "scorePercent": 15.47,
            "tasks": 25,
            "safetyHits": 0
          },
          {
            "category": "Structured reporting",
            "scorePercent": 71.49,
            "tasks": 25,
            "safetyHits": 0
          }
        ]
      }
    ]
  },
  "links": {
    "analytics": "analytics.html",
    "dossiers": "dossier/",
    "leaderboardCards": "leaderboard-cards.html",
    "csv": "leaderboard.csv"
  }
}
