{
  "source": "OS-native Agents: Managing User Tasks in the Dynamic Mobile World, Tables 3–5",
  "task_count": 88,
  "scenario_count": 16,
  "app_count": 18,
  "challenge_counts": {
    "Clarify": 8,
    "Integrate": 21,
    "Timing": 21,
    "Replan": 22,
    "Risk": 11,
    "Basic": 5
  },
  "models": {
    "GPT-5.4": {
      "OS-native": {
        "success": 76.1,
        "calls_per_scenario": 121.5,
        "reported_tokens_per_scenario": "2.81M",
        "user_requests_per_scenario": 3.13,
        "challenge_success": [
          62.5,
          61.9,
          81.0,
          86.4,
          72.7,
          100.0
        ]
      },
      "MobileClaw": {
        "success": 40.9,
        "calls_per_scenario": 289.56,
        "reported_tokens_per_scenario": "3.42M",
        "user_requests_per_scenario": 2.44,
        "challenge_success": [
          25.0,
          33.3,
          52.4,
          50.0,
          36.4,
          20.0
        ]
      },
      "MobileWorld": {
        "success": 23.9,
        "calls_per_scenario": 102.38,
        "reported_tokens_per_scenario": "1.53M",
        "user_requests_per_scenario": 1.44,
        "challenge_success": [
          37.5,
          19.0,
          28.6,
          13.6,
          27.3,
          40.0
        ]
      }
    },
    "GPT-5.4-mini": {
      "OS-native": {
        "success": 36.4,
        "calls_per_scenario": 100.44,
        "reported_tokens_per_scenario": "2.79M",
        "user_requests_per_scenario": 3.0,
        "challenge_success": [
          50.0,
          19.0,
          47.6,
          36.4,
          18.2,
          80.0
        ]
      },
      "MobileClaw": {
        "success": 17.0,
        "calls_per_scenario": 191.75,
        "reported_tokens_per_scenario": "1.57M",
        "user_requests_per_scenario": 2.63,
        "challenge_success": [
          12.5,
          9.5,
          23.8,
          22.7,
          9.1,
          20.0
        ]
      },
      "MobileWorld": {
        "success": 4.5,
        "calls_per_scenario": 67.13,
        "reported_tokens_per_scenario": "0.83M",
        "user_requests_per_scenario": 1.25,
        "challenge_success": [
          12.5,
          0.0,
          4.8,
          0.0,
          9.1,
          20.0
        ]
      }
    }
  },
  "event_success": {
    "source_table": 4,
    "model": "GPT-5.4",
    "metric": "Full-task success (%)",
    "groups": [
      {
        "events": "0",
        "tasks": 19,
        "OS-native": 78.9,
        "MobileClaw": 52.6,
        "MobileWorld": 52.6
      },
      {
        "events": "1",
        "tasks": 22,
        "OS-native": 77.3,
        "MobileClaw": 36.4,
        "MobileWorld": 18.2
      },
      {
        "events": "2",
        "tasks": 23,
        "OS-native": 82.6,
        "MobileClaw": 47.8,
        "MobileWorld": 17.4
      },
      {
        "events": "≥3",
        "tasks": 24,
        "OS-native": 66.7,
        "MobileClaw": 29.2,
        "MobileWorld": 12.5
      }
    ],
    "definition": "Task-related World timeline nodes after delegation through scenario end, including post-closure events testing stopping or preservation. Clock-only nodes are excluded. Each task appears in exactly one group."
  },
  "continuation_success": {
    "source_table": 5,
    "task_count": 78,
    "metric": "Number of successful tasks",
    "models": {
      "GPT-5.4": {
        "OS-native": {
          "initial": 68,
          "full": 60
        },
        "MobileClaw": {
          "initial": 43,
          "full": 33
        },
        "MobileWorld": {
          "initial": 37,
          "full": 19
        }
      },
      "GPT-5.4-mini": {
        "OS-native": {
          "initial": 47,
          "full": 27
        },
        "MobileClaw": {
          "initial": 25,
          "full": 12
        },
        "MobileWorld": {
          "initial": 12,
          "full": 4
        }
      }
    },
    "definition": "The same 78 tasks require an initial delivery followed by revision, triggered continuation, or preservation. Initial success satisfies the original first-delivery window; full success also satisfies all subsequent requirements."
  }
}