{
  "provenance": "Paper-reported values transcribed from the supplied manuscript. No experimental reproduction in this release.",
  "general": {
    "source": "sections/04_evaluation.tex, Table 1",
    "benchmarks": [
      "MMBench v1.1",
      "MMStar",
      "MMMU",
      "AI2D",
      "MathVista-mini",
      "TextVQA",
      "ChartQA",
      "DocVQA",
      "HallusionBench",
      "POPE",
      "Avg."
    ],
    "rows": [
      {
        "model": "LLaVA-OV-7B",
        "scores": [
          80.8,
          61.7,
          48.8,
          82.0,
          63.2,
          77.6,
          80.0,
          87.5,
          47.3,
          87.5,
          71.6
        ]
      },
      {
        "model": "MiniCPM-V2.6",
        "scores": [
          78.0,
          57.5,
          49.8,
          82.1,
          60.6,
          80.1,
          79.3,
          90.5,
          48.1,
          87.0,
          71.3
        ]
      },
      {
        "model": "Cambrian-8B",
        "scores": [
          75.9,
          57.6,
          42.7,
          79.7,
          49.0,
          71.7,
          73.3,
          77.8,
          41.6,
          86.5,
          65.6
        ]
      },
      {
        "model": "InternVL2-8B",
        "scores": [
          79.4,
          61.5,
          51.2,
          83.8,
          58.3,
          77.4,
          83.3,
          91.6,
          45.0,
          86.4,
          71.8
        ]
      },
      {
        "model": "InternVL2.5-8B",
        "scores": [
          82.2,
          62.8,
          56.0,
          84.5,
          64.4,
          78.2,
          84.7,
          92.9,
          49.3,
          88.5,
          74.4
        ]
      },
      {
        "model": "Qwen2-VL-7B",
        "scores": [
          80.5,
          60.7,
          54.1,
          83.0,
          58.2,
          84.3,
          83.0,
          93.9,
          42.4,
          87.3,
          72.7
        ]
      },
      {
        "model": "Qwen3-VL-8B",
        "scores": [
          74.9,
          42.9,
          47.0,
          46.7,
          32.8,
          83.1,
          83.5,
          95.8,
          51.3,
          88.3,
          64.6
        ]
      },
      {
        "model": "SpatialBot-3B",
        "scores": [
          58.2,
          39.1,
          31.4,
          56.7,
          35.0,
          50.2,
          48.1,
          56.3,
          36.2,
          81.5,
          49.3
        ]
      },
      {
        "model": "SpatialRGPT-7B",
        "scores": [
          72.3,
          52.8,
          43.5,
          74.6,
          49.1,
          70.5,
          68.4,
          78.0,
          41.0,
          85.7,
          63.6
        ]
      },
      {
        "model": "ManipVQA-7B",
        "scores": [
          69.5,
          50.2,
          40.3,
          71.5,
          46.8,
          67.0,
          65.3,
          75.2,
          38.8,
          83.7,
          60.8
        ]
      },
      {
        "model": "LEO-7B",
        "scores": [
          70.8,
          51.4,
          41.7,
          72.0,
          47.3,
          68.8,
          66.7,
          76.5,
          39.5,
          84.9,
          62.0
        ]
      },
      {
        "model": "KnowinRouter (8B)",
        "scores": [
          81.9,
          63.2,
          57.8,
          84.3,
          61.5,
          83.9,
          84.2,
          93.8,
          47.9,
          88.3,
          74.7
        ]
      }
    ]
  },
  "embodied": {
    "source": "Figure 3 (Q1), figures/fig03_embodied_results.pdf in the supplied manuscript; transcribed bar labels and legend colors. Native higher-is-better metrics; mixed cross-system protocols.",
    "benchmarks": [
      "UniEQA",
      "OpenEQA",
      "VSI-Bench",
      "ERQA",
      "RoboVQA",
      "Where2Place"
    ],
    "scores": [
      71.6,
      58.1,
      64.1,
      46.5,
      63.0,
      65.0
    ],
    "reported_mean": 61.4,
    "coverage": "All 10 models and 24 labeled bars in Figure 3. The figure shows four models per benchmark; a complete model-by-benchmark matrix was not supplied.",
    "missing_value": "null means not shown in Figure 3, not zero. Do not average incomplete rows.",
    "rows": [
      {
        "model": "KnowinRouter (8B)",
        "paper_model": "EKT-Router",
        "scores": [
          71.6,
          58.1,
          64.1,
          46.5,
          63.0,
          65.0
        ]
      },
      {
        "model": "Qwen3-VL-8B-Instruct",
        "paper_model": "Qwen3-VL-8B-Instruct",
        "scores": [
          null,
          null,
          61.2,
          42.8,
          55.3,
          63.0
        ]
      },
      {
        "model": "InternVL3.5-14B",
        "paper_model": "InternVL3.5-14B",
        "scores": [
          70.6,
          null,
          60.5,
          null,
          null,
          null
        ]
      },
      {
        "model": "Pelican1.0-VL-7B",
        "paper_model": "Pelican1.0-VL-7B",
        "scores": [
          70.3,
          57.2,
          null,
          null,
          62.6,
          null
        ]
      },
      {
        "model": "Gemini-2.5-Pro",
        "paper_model": "Gemini-2.5-Pro",
        "scores": [
          60.6,
          null,
          null,
          null,
          null,
          49.6
        ]
      },
      {
        "model": "MiMo-Embodied-7B",
        "paper_model": "MiMo-Embodied-7B",
        "scores": [
          null,
          null,
          null,
          null,
          null,
          63.6
        ]
      },
      {
        "model": "OpenAI o3",
        "paper_model": "OpenAI o3",
        "scores": [
          null,
          null,
          57.8,
          41.5,
          52.2,
          null
        ]
      },
      {
        "model": "Qwen-VL-Max",
        "paper_model": "Qwen-VL-Max",
        "scores": [
          null,
          52.2,
          null,
          null,
          null,
          null
        ]
      },
      {
        "model": "InternVL3-38B",
        "paper_model": "InternVL3-38B",
        "scores": [
          null,
          null,
          null,
          41.7,
          null,
          null
        ]
      },
      {
        "model": "GPT-4o",
        "paper_model": "GPT-4o",
        "scores": [
          null,
          51.1,
          null,
          null,
          null,
          null
        ]
      }
    ],
    "source_figure_sha256": "d657c85479dc31f050b6853c73ddff61d36a91a37f4a5f055712b506f790f780"
  },
  "ablation": {
    "source": "Section 4.4/Figure 5, separate frozen protocol; not the Q1 61.4 protocol",
    "full": 60.2,
    "variants": [
      "Without router",
      "Random routing",
      "Without knowledge",
      "Without perception",
      "Without both",
      "Heuristic labels",
      "Joint end-to-end",
      "Joint expert update"
    ],
    "scores": [
      54.1,
      52.4,
      57.9,
      57.4,
      55.8,
      58.3,
      57.1,
      56.2
    ]
  },
  "matched_general_delta": {
    "source": "Section 4.2, matched comparison against the General path",
    "benchmarks": [
      "UniEQA",
      "OpenEQA",
      "VSI-Bench",
      "ERQA",
      "RoboVQA",
      "Where2Place"
    ],
    "deltas": [
      -3.37,
      -2.43,
      2.94,
      3.75,
      7.78,
      2.0
    ],
    "reported_mean": 1.78
  },
  "evidence_null": {
    "source": "Section 4.3, nine benchmarks, Knowledge and Tools disabled",
    "best_fixed": 58.49,
    "routed": 60.24,
    "gain": 1.75,
    "ci95": [
      1.25,
      2.26
    ]
  }
}