{
  "type": "excalidraw",
  "version": 2,
  "source": "http://127.0.0.1:8766",
  "elements": [
    {
      "id": "06-benchmark-design-1",
      "type": "text",
      "x": 30,
      "y": 10,
      "width": 614.57470703125,
      "height": 31.25,
      "angle": 0,
      "strokeColor": "#292c2b",
      "backgroundColor": "transparent",
      "fillStyle": "solid",
      "strokeWidth": 1.5,
      "strokeStyle": "solid",
      "roughness": 0.5,
      "opacity": 100,
      "groupIds": [],
      "frameId": null,
      "index": "a0",
      "roundness": null,
      "seed": 20260927,
      "version": 2,
      "versionNonce": 838680025,
      "isDeleted": false,
      "boundElements": null,
      "updated": 1790439624242,
      "link": null,
      "locked": false,
      "text": "Two routing jobs. The same two routers on each.",
      "fontSize": 25,
      "fontFamily": 1,
      "textAlign": "left",
      "verticalAlign": "top",
      "containerId": null,
      "originalText": "Two routing jobs. The same two routers on each.",
      "autoResize": true,
      "lineHeight": 1.25
    },
    {
      "id": "06-benchmark-design-2",
      "type": "rectangle",
      "x": 30,
      "y": 80,
      "width": 440,
      "height": 80,
      "angle": 0,
      "strokeColor": "#557f91",
      "backgroundColor": "#557f91",
      "fillStyle": "solid",
      "strokeWidth": 1.5,
      "strokeStyle": "solid",
      "roughness": 0.5,
      "opacity": 100,
      "groupIds": [],
      "frameId": null,
      "index": "a1",
      "roundness": {
        "type": 3
      },
      "seed": 20260928,
      "version": 2,
      "versionNonce": 1709238327,
      "isDeleted": false,
      "boundElements": [
        {
          "type": "text",
          "id": "UvuX8HVQkRADmno3-ROkH"
        }
      ],
      "updated": 1790439624242,
      "link": null,
      "locked": false
    },
    {
      "id": "06-benchmark-design-3",
      "type": "text",
      "x": 40,
      "y": 175,
      "width": 326.93280029296875,
      "height": 71.25,
      "angle": 0,
      "strokeColor": "#292c2b",
      "backgroundColor": "transparent",
      "fillStyle": "solid",
      "strokeWidth": 1.5,
      "strokeStyle": "solid",
      "roughness": 0.5,
      "opacity": 100,
      "groupIds": [],
      "frameId": null,
      "index": "a2",
      "roundness": null,
      "seed": 20260929,
      "version": 2,
      "versionNonce": 321526457,
      "isDeleted": false,
      "boundElements": null,
      "updated": 1790439624242,
      "link": null,
      "locked": false,
      "text": "CLINC150 utterances, 11 routes:\n10 domains + out_of_scope\n110 dev · 330 test (30 per route)",
      "fontSize": 19,
      "fontFamily": 1,
      "textAlign": "left",
      "verticalAlign": "top",
      "containerId": null,
      "originalText": "CLINC150 utterances, 11 routes:\n10 domains + out_of_scope\n110 dev · 330 test (30 per route)",
      "autoResize": true,
      "lineHeight": 1.25
    },
    {
      "id": "06-benchmark-design-4",
      "type": "rectangle",
      "x": 520,
      "y": 80,
      "width": 440,
      "height": 80,
      "angle": 0,
      "strokeColor": "#80658f",
      "backgroundColor": "#80658f",
      "fillStyle": "solid",
      "strokeWidth": 1.5,
      "strokeStyle": "solid",
      "roughness": 0.5,
      "opacity": 100,
      "groupIds": [],
      "frameId": null,
      "index": "a3",
      "roundness": {
        "type": 3
      },
      "seed": 20260930,
      "version": 2,
      "versionNonce": 539281751,
      "isDeleted": false,
      "boundElements": [
        {
          "type": "text",
          "id": "LIVnr40YqDQ4I0pqN7POd"
        }
      ],
      "updated": 1790439624242,
      "link": null,
      "locked": false
    },
    {
      "id": "06-benchmark-design-5",
      "type": "text",
      "x": 530,
      "y": 175,
      "width": 323.2657470703125,
      "height": 71.25,
      "angle": 0,
      "strokeColor": "#292c2b",
      "backgroundColor": "transparent",
      "fillStyle": "solid",
      "strokeWidth": 1.5,
      "strokeStyle": "solid",
      "roughness": 0.5,
      "opacity": 100,
      "groupIds": [],
      "frameId": null,
      "index": "a4",
      "roundness": null,
      "seed": 20260931,
      "version": 2,
      "versionNonce": 2020105113,
      "isDeleted": false,
      "boundElements": null,
      "updated": 1790439624242,
      "link": null,
      "locked": false,
      "text": "MMLU-Pro, 14 subjects\n98 dev · 308 test\nlabel = did Luna answer correctly?",
      "fontSize": 19,
      "fontFamily": 1,
      "textAlign": "left",
      "verticalAlign": "top",
      "containerId": null,
      "originalText": "MMLU-Pro, 14 subjects\n98 dev · 308 test\nlabel = did Luna answer correctly?",
      "autoResize": true,
      "lineHeight": 1.25
    },
    {
      "id": "06-benchmark-design-6",
      "type": "rectangle",
      "x": 30,
      "y": 330,
      "width": 200,
      "height": 75,
      "angle": 0,
      "strokeColor": "#4b7864",
      "backgroundColor": "#4b7864",
      "fillStyle": "solid",
      "strokeWidth": 1.5,
      "strokeStyle": "solid",
      "roughness": 0.5,
      "opacity": 100,
      "groupIds": [],
      "frameId": null,
      "index": "a5",
      "roundness": {
        "type": 3
      },
      "seed": 20260932,
      "version": 2,
      "versionNonce": 1703904887,
      "isDeleted": false,
      "boundElements": [
        {
          "type": "text",
          "id": "FVLfzk55SOlidMnv4ubPI"
        }
      ],
      "updated": 1790439624242,
      "link": null,
      "locked": false
    },
    {
      "id": "06-benchmark-design-7",
      "type": "rectangle",
      "x": 260,
      "y": 330,
      "width": 210,
      "height": 75,
      "angle": 0,
      "strokeColor": "#a38b6e",
      "backgroundColor": "#a38b6e",
      "fillStyle": "solid",
      "strokeWidth": 1.5,
      "strokeStyle": "solid",
      "roughness": 0.5,
      "opacity": 100,
      "groupIds": [],
      "frameId": null,
      "index": "a6",
      "roundness": {
        "type": 3
      },
      "seed": 20260933,
      "version": 2,
      "versionNonce": 569081977,
      "isDeleted": false,
      "boundElements": [
        {
          "type": "text",
          "id": "KTQxN0zBqS4PRTBBSmMww"
        }
      ],
      "updated": 1790439624242,
      "link": null,
      "locked": false
    },
    {
      "id": "06-benchmark-design-8",
      "type": "text",
      "x": 40,
      "y": 420,
      "width": 347.6337890625,
      "height": 67.5,
      "angle": 0,
      "strokeColor": "#292c2b",
      "backgroundColor": "transparent",
      "fillStyle": "solid",
      "strokeWidth": 1.5,
      "strokeStyle": "solid",
      "roughness": 0.5,
      "opacity": 100,
      "groupIds": [],
      "frameId": null,
      "index": "a7",
      "roundness": null,
      "seed": 20260934,
      "version": 2,
      "versionNonce": 1692770199,
      "isDeleted": false,
      "boundElements": null,
      "updated": 1790439624242,
      "link": null,
      "locked": false,
      "text": "Choice over routes · Noul for B\nvs JSON schema + stated confidence\nsame criteria text, frozen before test",
      "fontSize": 18,
      "fontFamily": 1,
      "textAlign": "left",
      "verticalAlign": "top",
      "containerId": null,
      "originalText": "Choice over routes · Noul for B\nvs JSON schema + stated confidence\nsame criteria text, frozen before test",
      "autoResize": true,
      "lineHeight": 1.25
    },
    {
      "id": "06-benchmark-design-9",
      "type": "rectangle",
      "x": 520,
      "y": 330,
      "width": 200,
      "height": 75,
      "angle": 0,
      "strokeColor": "#a38b6e",
      "backgroundColor": "#a38b6e",
      "fillStyle": "solid",
      "strokeWidth": 1.5,
      "strokeStyle": "solid",
      "roughness": 0.5,
      "opacity": 100,
      "groupIds": [],
      "frameId": null,
      "index": "a8",
      "roundness": {
        "type": 3
      },
      "seed": 20260935,
      "version": 2,
      "versionNonce": 2104953177,
      "isDeleted": false,
      "boundElements": [
        {
          "type": "text",
          "id": "UTcm13o2mUI2LctBjF7qP"
        }
      ],
      "updated": 1790439624242,
      "link": null,
      "locked": false
    },
    {
      "id": "06-benchmark-design-10",
      "type": "rectangle",
      "x": 750,
      "y": 330,
      "width": 210,
      "height": 75,
      "angle": 0,
      "strokeColor": "#80658f",
      "backgroundColor": "#80658f",
      "fillStyle": "solid",
      "strokeWidth": 1.5,
      "strokeStyle": "solid",
      "roughness": 0.5,
      "opacity": 100,
      "groupIds": [],
      "frameId": null,
      "index": "a9",
      "roundness": {
        "type": 3
      },
      "seed": 20260936,
      "version": 2,
      "versionNonce": 6001847,
      "isDeleted": false,
      "boundElements": [
        {
          "type": "text",
          "id": "qeCXRf2B2k2CF9H5z7dZC"
        }
      ],
      "updated": 1790439624242,
      "link": null,
      "locked": false
    },
    {
      "id": "06-benchmark-design-11",
      "type": "text",
      "x": 530,
      "y": 420,
      "width": 264.725830078125,
      "height": 45,
      "angle": 0,
      "strokeColor": "#292c2b",
      "backgroundColor": "transparent",
      "fillStyle": "solid",
      "strokeWidth": 1.5,
      "strokeStyle": "solid",
      "roughness": 0.5,
      "opacity": 100,
      "groupIds": [],
      "frameId": null,
      "index": "aA",
      "roundness": null,
      "seed": 20260937,
      "version": 2,
      "versionNonce": 1952347705,
      "isDeleted": false,
      "boundElements": null,
      "updated": 1790439624242,
      "link": null,
      "locked": false,
      "text": "Ground truth only.\nRouters never see an answer.",
      "fontSize": 18,
      "fontFamily": 1,
      "textAlign": "left",
      "verticalAlign": "top",
      "containerId": null,
      "originalText": "Ground truth only.\nRouters never see an answer.",
      "autoResize": true,
      "lineHeight": 1.25
    },
    {
      "id": "06-benchmark-design-12",
      "type": "text",
      "x": 30,
      "y": 520,
      "width": 545.0196533203125,
      "height": 21.25,
      "angle": 0,
      "strokeColor": "#666b68",
      "backgroundColor": "transparent",
      "fillStyle": "solid",
      "strokeWidth": 1.5,
      "strokeStyle": "solid",
      "roughness": 0.5,
      "opacity": 100,
      "groupIds": [],
      "frameId": null,
      "index": "aB",
      "roundness": null,
      "seed": 20260938,
      "version": 2,
      "versionNonce": 1603659223,
      "isDeleted": false,
      "boundElements": null,
      "updated": 1790439624242,
      "link": null,
      "locked": false,
      "text": "All calls through OpenRouter, reasoning off for Luna, 8 in flight.",
      "fontSize": 17,
      "fontFamily": 1,
      "textAlign": "left",
      "verticalAlign": "top",
      "containerId": null,
      "originalText": "All calls through OpenRouter, reasoning off for Luna, 8 in flight.",
      "autoResize": true,
      "lineHeight": 1.25
    },
    {
      "id": "UvuX8HVQkRADmno3-ROkH",
      "type": "text",
      "x": 151.57470703125,
      "y": 107.35,
      "width": 196.8505859375,
      "height": 25.299999999999997,
      "angle": 0,
      "strokeColor": "#ffffff",
      "backgroundColor": "transparent",
      "fillStyle": "solid",
      "strokeWidth": 2,
      "strokeStyle": "solid",
      "roughness": 1,
      "opacity": 100,
      "groupIds": [],
      "frameId": null,
      "index": "aC",
      "roundness": null,
      "seed": 2038777625,
      "version": 3,
      "versionNonce": 1648713817,
      "isDeleted": false,
      "boundElements": null,
      "updated": 1790439624242,
      "link": null,
      "locked": false,
      "text": "A  Which specialist?",
      "fontSize": 22,
      "fontFamily": 2,
      "textAlign": "center",
      "verticalAlign": "middle",
      "containerId": "06-benchmark-design-2",
      "originalText": "A  Which specialist?",
      "autoResize": true,
      "lineHeight": 1.15
    },
    {
      "id": "LIVnr40YqDQ4I0pqN7POd",
      "type": "text",
      "x": 596.409423828125,
      "y": 107.925,
      "width": 287.18115234375,
      "height": 24.15,
      "angle": 0,
      "strokeColor": "#ffffff",
      "backgroundColor": "transparent",
      "fillStyle": "solid",
      "strokeWidth": 2,
      "strokeStyle": "solid",
      "roughness": 1,
      "opacity": 100,
      "groupIds": [],
      "frameId": null,
      "index": "aD",
      "roundness": null,
      "seed": 1465842681,
      "version": 3,
      "versionNonce": 1468149175,
      "isDeleted": false,
      "boundElements": null,
      "updated": 1790439624242,
      "link": null,
      "locked": false,
      "text": "B  Does it need the big model?",
      "fontSize": 21,
      "fontFamily": 2,
      "textAlign": "center",
      "verticalAlign": "middle",
      "containerId": "06-benchmark-design-4",
      "originalText": "B  Does it need the big model?",
      "autoResize": true,
      "lineHeight": 1.15
    },
    {
      "id": "FVLfzk55SOlidMnv4ubPI",
      "type": "text",
      "x": 88.4169921875,
      "y": 354.85,
      "width": 83.166015625,
      "height": 25.299999999999997,
      "angle": 0,
      "strokeColor": "#ffffff",
      "backgroundColor": "transparent",
      "fillStyle": "solid",
      "strokeWidth": 2,
      "strokeStyle": "solid",
      "roughness": 1,
      "opacity": 100,
      "groupIds": [],
      "frameId": null,
      "index": "aE",
      "roundness": null,
      "seed": 1039078617,
      "version": 3,
      "versionNonce": 2068442425,
      "isDeleted": false,
      "boundElements": null,
      "updated": 1790439624242,
      "link": null,
      "locked": false,
      "text": "Jev 1.13",
      "fontSize": 22,
      "fontFamily": 2,
      "textAlign": "center",
      "verticalAlign": "middle",
      "containerId": "06-benchmark-design-6",
      "originalText": "Jev 1.13",
      "autoResize": true,
      "lineHeight": 1.15
    },
    {
      "id": "KTQxN0zBqS4PRTBBSmMww",
      "type": "text",
      "x": 305.681640625,
      "y": 354.85,
      "width": 118.63671875,
      "height": 25.299999999999997,
      "angle": 0,
      "strokeColor": "#ffffff",
      "backgroundColor": "transparent",
      "fillStyle": "solid",
      "strokeWidth": 2,
      "strokeStyle": "solid",
      "roughness": 1,
      "opacity": 100,
      "groupIds": [],
      "frameId": null,
      "index": "aF",
      "roundness": null,
      "seed": 281612729,
      "version": 3,
      "versionNonce": 679567063,
      "isDeleted": false,
      "boundElements": null,
      "updated": 1790439624242,
      "link": null,
      "locked": false,
      "text": "GPT-6 Luna",
      "fontSize": 22,
      "fontFamily": 2,
      "textAlign": "center",
      "verticalAlign": "middle",
      "containerId": "06-benchmark-design-7",
      "originalText": "GPT-6 Luna",
      "autoResize": true,
      "lineHeight": 1.15
    },
    {
      "id": "UTcm13o2mUI2LctBjF7qP",
      "type": "text",
      "x": 557.7392578125,
      "y": 356,
      "width": 124.521484375,
      "height": 23,
      "angle": 0,
      "strokeColor": "#ffffff",
      "backgroundColor": "transparent",
      "fillStyle": "solid",
      "strokeWidth": 2,
      "strokeStyle": "solid",
      "roughness": 1,
      "opacity": 100,
      "groupIds": [],
      "frameId": null,
      "index": "aG",
      "roundness": null,
      "seed": 1964012185,
      "version": 3,
      "versionNonce": 568974873,
      "isDeleted": false,
      "boundElements": null,
      "updated": 1790439624242,
      "link": null,
      "locked": false,
      "text": "Luna answers",
      "fontSize": 20,
      "fontFamily": 2,
      "textAlign": "center",
      "verticalAlign": "middle",
      "containerId": "06-benchmark-design-9",
      "originalText": "Luna answers",
      "autoResize": true,
      "lineHeight": 1.15
    },
    {
      "id": "qeCXRf2B2k2CF9H5z7dZC",
      "type": "text",
      "x": 800.5322265625,
      "y": 356,
      "width": 108.935546875,
      "height": 23,
      "angle": 0,
      "strokeColor": "#ffffff",
      "backgroundColor": "transparent",
      "fillStyle": "solid",
      "strokeWidth": 2,
      "strokeStyle": "solid",
      "roughness": 1,
      "opacity": 100,
      "groupIds": [],
      "frameId": null,
      "index": "aH",
      "roundness": null,
      "seed": 528004985,
      "version": 3,
      "versionNonce": 817720311,
      "isDeleted": false,
      "boundElements": null,
      "updated": 1790439624242,
      "link": null,
      "locked": false,
      "text": "Sol answers",
      "fontSize": 20,
      "fontFamily": 2,
      "textAlign": "center",
      "verticalAlign": "middle",
      "containerId": "06-benchmark-design-10",
      "originalText": "Sol answers",
      "autoResize": true,
      "lineHeight": 1.15
    }
  ],
  "appState": {
    "viewBackgroundColor": "#ffffff"
  },
  "files": {}
}