{
  "source": "Manuscript main results: tables/workbuddy.tex and tables/jobbench_lab.tex. Reported means only; k=5 uses parallel recursion.",
  "configurations": {
    "gemini-opencode": {
      "model": "gemini",
      "harness": "opencode",
      "workbuddy": {
        "columns": ["Code", "Web", "Office", "Average"],
        "primary": 3,
        "rows": [
          {
            "method": "Base harness",
            "values": [73.4, 69.9, 74.0, 72.4]
          },
          {
            "method": "ReasoningBank",
            "values": [72.6, 70.7, 76.8, 73.4]
          },
          {
            "method": "Meta-Harness",
            "values": [75.5, 75.1, 78.6, 76.4]
          },
          {
            "method": "Harness Scaling",
            "values": [74.0, 74.0, 77.9, 75.3]
          },
          {
            "method": "Harness\u00b2 \u00b7 k=1",
            "values": [76.5, 76.4, 79.5, 77.5]
          },
          {
            "method": "Harness\u00b2 \u00b7 k=5",
            "values": [80.8, 78.3, 83.4, 80.8]
          }
        ]
      },
      "jobbench": {
        "columns": ["Score"],
        "primary": 0,
        "rows": [
          {
            "method": "Base harness",
            "values": [31.5]
          },
          {
            "method": "ReasoningBank",
            "values": [30.8]
          },
          {
            "method": "Meta-Harness",
            "values": [35.8]
          },
          {
            "method": "Harness Scaling",
            "values": [32.9]
          },
          {
            "method": "Harness\u00b2 \u00b7 k=1",
            "values": [36.0]
          },
          {
            "method": "Harness\u00b2 \u00b7 k=5",
            "values": [39.0]
          }
        ]
      },
      "lab": {
        "columns": ["Partial (%)"],
        "primary": 0,
        "rows": [
          {
            "method": "Base harness",
            "values": [79.9]
          },
          {
            "method": "ReasoningBank",
            "values": [77.3]
          },
          {
            "method": "Meta-Harness",
            "values": [81.6]
          },
          {
            "method": "Harness Scaling",
            "values": [78.9]
          },
          {
            "method": "Harness\u00b2 \u00b7 k=1",
            "values": [81.9]
          },
          {
            "method": "Harness\u00b2 \u00b7 k=5",
            "values": [84.5]
          }
        ]
      }
    },
    "gemini-codex": {
      "model": "gemini",
      "harness": "codex",
      "workbuddy": {
        "columns": ["Code", "Web", "Office", "Average"],
        "primary": 3,
        "rows": [
          {
            "method": "Base harness",
            "values": [74.0, 72.5, 77.0, 74.5]
          },
          {
            "method": "ReasoningBank",
            "values": [73.5, 65.7, 78.4, 72.5]
          },
          {
            "method": "Meta-Harness",
            "values": [74.6, 67.8, 77.6, 73.3]
          },
          {
            "method": "Harness Scaling",
            "values": [75.5, 69.3, 76.7, 73.8]
          },
          {
            "method": "Harness\u00b2 \u00b7 k=1",
            "values": [79.4, 74.7, 78.5, 77.5]
          },
          {
            "method": "Harness\u00b2 \u00b7 k=5",
            "values": [81.0, 77.3, 82.8, 80.4]
          }
        ]
      },
      "jobbench": {
        "columns": ["Score"],
        "primary": 0,
        "rows": [
          {
            "method": "Base harness",
            "values": [33.9]
          },
          {
            "method": "ReasoningBank",
            "values": [30.2]
          },
          {
            "method": "Meta-Harness",
            "values": [34.8]
          },
          {
            "method": "Harness Scaling",
            "values": [33.6]
          },
          {
            "method": "Harness\u00b2 \u00b7 k=1",
            "values": [35.3]
          },
          {
            "method": "Harness\u00b2 \u00b7 k=5",
            "values": [39.5]
          }
        ]
      },
      "lab": {
        "columns": ["Partial (%)"],
        "primary": 0,
        "rows": [
          {
            "method": "Base harness",
            "values": [71.6]
          },
          {
            "method": "ReasoningBank",
            "values": [74.8]
          },
          {
            "method": "Meta-Harness",
            "values": [76.5]
          },
          {
            "method": "Harness Scaling",
            "values": [67.9]
          },
          {
            "method": "Harness\u00b2 \u00b7 k=1",
            "values": [75.6]
          },
          {
            "method": "Harness\u00b2 \u00b7 k=5",
            "values": [78.5]
          }
        ]
      }
    },
    "glm-opencode": {
      "model": "glm",
      "harness": "opencode",
      "workbuddy": {
        "columns": ["Code", "Web", "Office", "Average"],
        "primary": 3,
        "rows": [
          {
            "method": "Base harness",
            "values": [58.4, 59.2, 75.1, 64.2]
          },
          {
            "method": "ReasoningBank",
            "values": [63.3, 61.4, 72.0, 65.6]
          },
          {
            "method": "Meta-Harness",
            "values": [69.0, 61.9, 78.1, 69.7]
          },
          {
            "method": "Harness Scaling",
            "values": [60.3, 63.4, 79.8, 67.8]
          },
          {
            "method": "Harness\u00b2 \u00b7 k=1",
            "values": [77.1, 69.1, 80.8, 75.7]
          },
          {
            "method": "Harness\u00b2 \u00b7 k=5",
            "values": [81.9, 74.4, 82.5, 79.6]
          }
        ]
      },
      "jobbench": {
        "columns": ["Score"],
        "primary": 0,
        "rows": [
          {
            "method": "Base harness",
            "values": [20.6]
          },
          {
            "method": "ReasoningBank",
            "values": [20.7]
          },
          {
            "method": "Meta-Harness",
            "values": [20.9]
          },
          {
            "method": "Harness Scaling",
            "values": [19.3]
          },
          {
            "method": "Harness\u00b2 \u00b7 k=1",
            "values": [24.2]
          },
          {
            "method": "Harness\u00b2 \u00b7 k=5",
            "values": [28.4]
          }
        ]
      },
      "lab": {
        "columns": ["Partial (%)"],
        "primary": 0,
        "rows": [
          {
            "method": "Base harness",
            "values": [59.6]
          },
          {
            "method": "ReasoningBank",
            "values": [71.3]
          },
          {
            "method": "Meta-Harness",
            "values": [60.0]
          },
          {
            "method": "Harness Scaling",
            "values": [67.8]
          },
          {
            "method": "Harness\u00b2 \u00b7 k=1",
            "values": [71.4]
          },
          {
            "method": "Harness\u00b2 \u00b7 k=5",
            "values": [78.1]
          }
        ]
      }
    },
    "glm-codex": {
      "model": "glm",
      "harness": "codex",
      "workbuddy": {
        "columns": ["Code", "Web", "Office", "Average"],
        "primary": 3,
        "rows": [
          {
            "method": "Base harness",
            "values": [56.8, 59.6, 76.9, 64.4]
          },
          {
            "method": "ReasoningBank",
            "values": [50.2, 66.3, 75.8, 64.1]
          },
          {
            "method": "Meta-Harness",
            "values": [76.8, 67.8, 77.9, 74.2]
          },
          {
            "method": "Harness Scaling",
            "values": [49.5, 60.9, 78.2, 62.9]
          },
          {
            "method": "Harness\u00b2 \u00b7 k=1",
            "values": [74.9, 70.0, 81.1, 75.3]
          },
          {
            "method": "Harness\u00b2 \u00b7 k=5",
            "values": [77.2, 73.3, 84.3, 78.3]
          }
        ]
      },
      "jobbench": {
        "columns": ["Score"],
        "primary": 0,
        "rows": [
          {
            "method": "Base harness",
            "values": [23.6]
          },
          {
            "method": "ReasoningBank",
            "values": [22.8]
          },
          {
            "method": "Meta-Harness",
            "values": [23.9]
          },
          {
            "method": "Harness Scaling",
            "values": [17.5]
          },
          {
            "method": "Harness\u00b2 \u00b7 k=1",
            "values": [24.8]
          },
          {
            "method": "Harness\u00b2 \u00b7 k=5",
            "values": [27.9]
          }
        ]
      },
      "lab": {
        "columns": ["Partial (%)"],
        "primary": 0,
        "rows": [
          {
            "method": "Base harness",
            "values": [58.1]
          },
          {
            "method": "ReasoningBank",
            "values": [63.8]
          },
          {
            "method": "Meta-Harness",
            "values": [59.0]
          },
          {
            "method": "Harness Scaling",
            "values": [63.7]
          },
          {
            "method": "Harness\u00b2 \u00b7 k=1",
            "values": [65.9]
          },
          {
            "method": "Harness\u00b2 \u00b7 k=5",
            "values": [70.7]
          }
        ]
      }
    }
  }
}