{
  "schema_version": "2",
  "name": "Decision Models Comparison",
  "references": [
    {
      "id": "prompt",
      "role": "reference",
      "label": "Prompt",
      "asset_type": "text",
      "inputs": [
        {
          "id": "in",
          "label": "Text",
          "type": "text"
        }
      ],
      "outputs": [
        {
          "id": "out",
          "label": "Text",
          "type": "text"
        }
      ],
      "data": {
        "out": "From: GitHub <noreply@github-notifications.com>\nSubject: [GitHub] Action required: unusual OAuth app authorization\n\nHi there,\n\nA third-party OAuth application (\"CI Deploy Helper\") was granted\nfull repo and workflow access to your account from a new device\n(Chrome on Windows, São Paulo, BR).\n\nIf you did not authorize this, revoke access within 12 hours to\nprevent your repositories and secrets from being exposed:\n\nhttps://github.com.security-review.app/settings/applications\n\nIf you made this change, you can ignore this email.\n\nThanks,\nThe GitHub Security Team"
      },
      "width": 514
    },
    {
      "id": "question",
      "role": "reference",
      "label": "Question + options",
      "asset_type": "text",
      "inputs": [
        {
          "id": "in",
          "label": "Text",
          "type": "text"
        }
      ],
      "outputs": [
        {
          "id": "out",
          "label": "Text",
          "type": "text"
        }
      ],
      "data": {
        "out": "Is this email safe to act on?\nphishing\nlegitimate\nneeds human review"
      },
      "width": 200
    }
  ],
  "operators": [
    {
      "id": "fmt",
      "role": "operator",
      "kind": "fn",
      "fn": "format_question",
      "label": "Format question",
      "inputs": [
        {
          "id": "in_question_block",
          "label": "Question + options",
          "type": "text",
          "required": true
        }
      ],
      "outputs": [
        {
          "id": "out_0",
          "label": "Questions JSON",
          "type": "text",
          "output_index": 0
        },
        {
          "id": "out_1",
          "label": "Question",
          "type": "text",
          "output_index": 1
        },
        {
          "id": "out_2",
          "label": "Options",
          "type": "text",
          "output_index": 2
        },
        {
          "id": "out_3",
          "label": "Decider questions",
          "type": "text",
          "output_index": 3
        }
      ],
      "data": {
        "out_0": "{\"decision\": {\"type\": \"choice\", \"instructions\": \"Which team should handle this request?\", \"criteria\": {\"billing\": \"billing\", \"technical\": \"technical\", \"sales\": \"sales\", \"shipping\": \"shipping\"}}}",
        "out_1": "Which team should handle this request?",
        "out_2": "billing\ntechnical\nsales\nshipping",
        "out_3": "choice decision: Which team should handle this request?\n- billing\n- technical\n- sales\n- shipping"
      },
      "width": 200
    },
    {
      "id": "d3",
      "role": "operator",
      "kind": "space",
      "space_id": "vllm-sr/d3-decision-model",
      "endpoint": "/decide",
      "label": "vllm-sr/d3",
      "inputs": [
        {
          "id": "state",
          "label": "State",
          "type": "text",
          "required": true
        },
        {
          "id": "questions",
          "label": "Questions",
          "type": "text",
          "required": true
        }
      ],
      "outputs": [
        {
          "id": "result",
          "label": "Result",
          "type": "json",
          "output_index": 1
        }
      ],
      "data": {
        "result": "{\"model\":\"d3\",\"answers\":{\"decision\":{\"type\":\"choice\",\"choice\":\"billing\",\"probabilities\":{\"billing\":0.9989143687405434,\"technical\":0.0005408171024171331,\"sales\":0.0005116921423538015,\"shipping\":0.00003312201468561358},\"confidence\":0.9932392003570076}},\"usage\":{\"input_tokens\":125,\"output_tokens\":0}}"
      },
      "width": 296
    },
    {
      "id": "clef",
      "role": "operator",
      "kind": "space",
      "space_id": "hugging-apps/clef-decision-model",
      "endpoint": "/decide",
      "label": "Cloudflare/clef",
      "inputs": [
        {
          "id": "state",
          "label": "State",
          "type": "text",
          "required": true
        },
        {
          "id": "questions",
          "label": "Questions",
          "type": "text",
          "required": true
        }
      ],
      "outputs": [
        {
          "id": "result",
          "label": "Result",
          "type": "json",
          "output_index": 1
        }
      ],
      "data": {
        "result": "{\"model\":\"clef\",\"answers\":{\"decision\":{\"type\":\"choice\",\"choice\":\"billing\",\"confidence\":0.9871,\"probabilities\":{\"billing\":0.9871,\"technical\":0.0073,\"sales\":0.0029,\"shipping\":0.0028}}},\"usage\":{\"input_tokens\":180,\"output_tokens\":0}}"
      },
      "width": 295
    },
    {
      "id": "jev",
      "role": "operator",
      "kind": "space",
      "space_id": "hugging-apps/jev-27b-vl",
      "endpoint": "/decide",
      "label": "autotrust/JEV-27B-VL",
      "inputs": [
        {
          "id": "state_text",
          "label": "State",
          "type": "text",
          "required": true
        },
        {
          "id": "images",
          "label": "Images",
          "type": "json"
        },
        {
          "id": "kind",
          "label": "Kind",
          "type": "text"
        },
        {
          "id": "question",
          "label": "Question",
          "type": "text",
          "required": true
        },
        {
          "id": "options_text",
          "label": "Options",
          "type": "text",
          "required": true
        }
      ],
      "outputs": [
        {
          "id": "result",
          "label": "Result",
          "type": "text",
          "output_index": 2
        }
      ],
      "data": {
        "images": [],
        "kind": "choice",
        "result": "{\n  \"kind\": \"choice\",\n  \"question\": \"Which team should handle this request?\",\n  \"options\": [\n    \"billing\",\n    \"technical\",\n    \"sales\",\n    \"shipping\"\n  ],\n  \"probabilities\": {\n    \"billing\": 0.9998358433869672,\n    \"technical\": 4.562085097251872e-05,\n    \"sales\": 0.0001183599322252175,\n    \"shipping\": 1.758298351067686e-07\n  }\n}"
      },
      "width": 296
    },
    {
      "id": "decider",
      "role": "operator",
      "kind": "space",
      "space_id": "hugging-apps/decider-2b-demo",
      "endpoint": "/decide",
      "label": "Mapika/decider-2b",
      "inputs": [
        {
          "id": "state_text",
          "label": "State",
          "type": "text",
          "required": true
        },
        {
          "id": "questions_text",
          "label": "Questions",
          "type": "text",
          "required": true
        },
        {
          "id": "temperature",
          "label": "Temperature",
          "type": "number"
        },
        {
          "id": "independent",
          "label": "Independent",
          "type": "boolean"
        },
        {
          "id": "isolated",
          "label": "Isolated",
          "type": "boolean"
        }
      ],
      "outputs": [
        {
          "id": "result",
          "label": "Result",
          "type": "text",
          "output_index": 1
        }
      ],
      "data": {
        "temperature": 1.3,
        "independent": true,
        "isolated": true,
        "result": "{\n  \"model\": \"decider-v10\",\n  \"answers\": {\n    \"decision\": {\n      \"type\": \"choice\",\n      \"choice\": \"billing\",\n      \"confidence\": 0.9899,\n      \"certainty\": 0.9548,\n      \"probabilities\": {\n        \"billing\": 0.9899,\n        \"technical\": 0.0081,\n        \"sales\": 0.0011,\n        \"shipping\": 0.0009\n      }\n    }\n  },\n  \"usage\": {\n    \"input_tokens\": 58,\n    \"output_tokens\": 0\n  }\n}"
      },
      "width": 316
    },
    {
      "id": "cmp",
      "role": "operator",
      "kind": "fn",
      "fn": "compare_decisions",
      "label": "Compare decisions",
      "inputs": [
        {
          "id": "in_d3",
          "label": "d3",
          "type": "json"
        },
        {
          "id": "in_clef",
          "label": "clef",
          "type": "json"
        },
        {
          "id": "in_jev",
          "label": "jev",
          "type": "json"
        },
        {
          "id": "in_decider",
          "label": "decider",
          "type": "json"
        }
      ],
      "outputs": [
        {
          "id": "out_0",
          "label": "Comparison",
          "type": "markdown",
          "output_index": 0
        }
      ],
      "data": {
        "out_0": "### Decision comparison\n\n| Model | Decision | Confidence | Margin vs 2nd | billing | sales | shipping | technical |\n|---|---|---|---|---|---|---|---|\n| vllm-sr/d3 | **billing** | 99.9% | 99.8% | **99.9%** | 0.1% | 0.0% | 0.1% |\n| Cloudflare/clef | **billing** | 98.7% | 98.0% | **98.7%** | 0.3% | 0.3% | 0.7% |\n| autotrust/JEV-27B-VL | **billing** | 100.0% | 100.0% | **100.0%** | 0.0% | 0.0% | 0.0% |\n| Mapika/decider-2b | **billing** | 99.0% | 98.2% | **99.0%** | 0.1% | 0.1% | 0.8% |\n\n**Votes:** billing: 4/4 (✅ all models agree)\n\n#### Probabilities\n```\nvllm-sr/d3             billing:██████████  sales:░░░░░░░░░░  shipping:░░░░░░░░░░  technical:░░░░░░░░░░\nCloudflare/clef        billing:██████████  sales:░░░░░░░░░░  shipping:░░░░░░░░░░  technical:░░░░░░░░░░\nautotrust/JEV-27B-VL   billing:██████████  sales:░░░░░░░░░░  shipping:░░░░░░░░░░  technical:░░░░░░░░░░\nMapika/decider-2b      billing:██████████  sales:░░░░░░░░░░  shipping:░░░░░░░░░░  technical:░░░░░░░░░░\n```"
      },
      "width": 200
    }
  ],
  "subjects": [
    {
      "id": "out",
      "role": "subject",
      "label": "Comparison",
      "asset_type": "markdown",
      "inputs": [
        {
          "id": "in",
          "label": "Comparison",
          "type": "markdown"
        }
      ],
      "outputs": [
        {
          "id": "out",
          "label": "Comparison",
          "type": "markdown"
        }
      ],
      "data": {
        "in": "### Decision comparison\n\n| Model | Decision | Confidence | Margin vs 2nd | billing | sales | shipping | technical |\n|---|---|---|---|---|---|---|---|\n| vllm-sr/d3 | **billing** | 99.9% | 99.8% | **99.9%** | 0.1% | 0.0% | 0.1% |\n| Cloudflare/clef | **billing** | 98.7% | 98.0% | **98.7%** | 0.3% | 0.3% | 0.7% |\n| autotrust/JEV-27B-VL | **billing** | 100.0% | 100.0% | **100.0%** | 0.0% | 0.0% | 0.0% |\n| Mapika/decider-2b | **billing** | 99.0% | 98.2% | **99.0%** | 0.1% | 0.1% | 0.8% |\n\n**Votes:** billing: 4/4 (✅ all models agree)\n\n#### Probabilities\n```\nvllm-sr/d3             billing:██████████  sales:░░░░░░░░░░  shipping:░░░░░░░░░░  technical:░░░░░░░░░░\nCloudflare/clef        billing:██████████  sales:░░░░░░░░░░  shipping:░░░░░░░░░░  technical:░░░░░░░░░░\nautotrust/JEV-27B-VL   billing:██████████  sales:░░░░░░░░░░  shipping:░░░░░░░░░░  technical:░░░░░░░░░░\nMapika/decider-2b      billing:██████████  sales:░░░░░░░░░░  shipping:░░░░░░░░░░  technical:░░░░░░░░░░\n```"
      },
      "width": 839
    }
  ],
  "edges": [
    {
      "id": "e0",
      "from_node_id": "question",
      "from_port_id": "out",
      "to_node_id": "fmt",
      "to_port_id": "in_question_block",
      "type": "text"
    },
    {
      "id": "e1",
      "from_node_id": "prompt",
      "from_port_id": "out",
      "to_node_id": "d3",
      "to_port_id": "state",
      "type": "text"
    },
    {
      "id": "e2",
      "from_node_id": "prompt",
      "from_port_id": "out",
      "to_node_id": "clef",
      "to_port_id": "state",
      "type": "text"
    },
    {
      "id": "e3",
      "from_node_id": "prompt",
      "from_port_id": "out",
      "to_node_id": "jev",
      "to_port_id": "state_text",
      "type": "text"
    },
    {
      "id": "e4",
      "from_node_id": "prompt",
      "from_port_id": "out",
      "to_node_id": "decider",
      "to_port_id": "state_text",
      "type": "text"
    },
    {
      "id": "e5",
      "from_node_id": "fmt",
      "from_port_id": "out_0",
      "to_node_id": "d3",
      "to_port_id": "questions",
      "type": "text"
    },
    {
      "id": "e6",
      "from_node_id": "fmt",
      "from_port_id": "out_0",
      "to_node_id": "clef",
      "to_port_id": "questions",
      "type": "text"
    },
    {
      "id": "e7",
      "from_node_id": "fmt",
      "from_port_id": "out_1",
      "to_node_id": "jev",
      "to_port_id": "question",
      "type": "text"
    },
    {
      "id": "e8",
      "from_node_id": "fmt",
      "from_port_id": "out_2",
      "to_node_id": "jev",
      "to_port_id": "options_text",
      "type": "text"
    },
    {
      "id": "e9",
      "from_node_id": "fmt",
      "from_port_id": "out_3",
      "to_node_id": "decider",
      "to_port_id": "questions_text",
      "type": "text"
    },
    {
      "id": "e10",
      "from_node_id": "d3",
      "from_port_id": "result",
      "to_node_id": "cmp",
      "to_port_id": "in_d3",
      "type": "text"
    },
    {
      "id": "e11",
      "from_node_id": "clef",
      "from_port_id": "result",
      "to_node_id": "cmp",
      "to_port_id": "in_clef",
      "type": "text"
    },
    {
      "id": "e12",
      "from_node_id": "jev",
      "from_port_id": "result",
      "to_node_id": "cmp",
      "to_port_id": "in_jev",
      "type": "text"
    },
    {
      "id": "e13",
      "from_node_id": "decider",
      "from_port_id": "result",
      "to_node_id": "cmp",
      "to_port_id": "in_decider",
      "type": "text"
    },
    {
      "id": "e14",
      "from_node_id": "cmp",
      "from_port_id": "out_0",
      "to_node_id": "out",
      "to_port_id": "in",
      "type": "markdown"
    }
  ]
}