{
  "campaign": "2026-09-10",
  "models": [
    "glm-4.6v-flash",
    "qwen2.5-7b-instruct",
    "qwen3.5-4b",
    "qwen3.5-9b",
    "zai-org/glm-4.7-flash"
  ],
  "tasks": [
    "blink-unix",
    "coop-scheduler",
    "frame-corrupt",
    "mqtt-device",
    "noisy-frames",
    "p-regulator",
    "pid-antiwindup",
    "protocol",
    "protocol-retry",
    "system-id",
    "uart-echo",
    "uart-menu",
    "watchdog"
  ],
  "pass_at_k": 0.52,
  "pass_at_1": 0.42,
  "pass_at_k_reliability": 0.31,
  "groups": [
    {
      "model": "glm-4.6v-flash",
      "task": "blink-unix",
      "attempts": 2,
      "solved": 2,
      "total_iterations": 3,
      "total_duration": 586.48,
      "infra_failures": 0,
      "success_rate": 1.0,
      "avg_iterations": 1.5,
      "avg_duration": 293.2,
      "passed": true,
      "all_solved": true,
      "incomplete": false,
      "tags": [
        "io"
      ],
      "level": 1
    },
    {
      "model": "glm-4.6v-flash",
      "task": "coop-scheduler",
      "attempts": 2,
      "solved": 2,
      "total_iterations": 4,
      "total_duration": 2570.2799999999997,
      "infra_failures": 0,
      "success_rate": 1.0,
      "avg_iterations": 2.0,
      "avg_duration": 1285.1,
      "passed": true,
      "all_solved": true,
      "incomplete": false,
      "tags": [
        "control"
      ],
      "level": 3
    },
    {
      "model": "glm-4.6v-flash",
      "task": "frame-corrupt",
      "attempts": 2,
      "solved": 0,
      "total_iterations": 10,
      "total_duration": 4492.62,
      "infra_failures": 0,
      "success_rate": 0.0,
      "avg_iterations": 5.0,
      "avg_duration": 2246.3,
      "passed": false,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "resilience"
      ],
      "level": 4
    },
    {
      "model": "glm-4.6v-flash",
      "task": "mqtt-device",
      "attempts": 2,
      "solved": 0,
      "total_iterations": 2,
      "total_duration": 1804.12,
      "infra_failures": 0,
      "success_rate": 0.0,
      "avg_iterations": 1.0,
      "avg_duration": 902.1,
      "passed": false,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "protocol"
      ],
      "level": 5
    },
    {
      "model": "glm-4.6v-flash",
      "task": "noisy-frames",
      "attempts": 2,
      "solved": 0,
      "total_iterations": 7,
      "total_duration": 3122.12,
      "infra_failures": 0,
      "success_rate": 0.0,
      "avg_iterations": 3.5,
      "avg_duration": 1561.1,
      "passed": false,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "resilience",
        "protocol"
      ],
      "level": 3
    },
    {
      "model": "glm-4.6v-flash",
      "task": "p-regulator",
      "attempts": 2,
      "solved": 0,
      "total_iterations": 2,
      "total_duration": 1804.1,
      "infra_failures": 0,
      "success_rate": 0.0,
      "avg_iterations": 1.0,
      "avg_duration": 902.0,
      "passed": false,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "control"
      ],
      "level": 2
    },
    {
      "model": "glm-4.6v-flash",
      "task": "pid-antiwindup",
      "attempts": 2,
      "solved": 1,
      "total_iterations": 6,
      "total_duration": 3986.95,
      "infra_failures": 0,
      "success_rate": 0.5,
      "avg_iterations": 3.0,
      "avg_duration": 1993.5,
      "passed": true,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "control"
      ],
      "level": 3
    },
    {
      "model": "glm-4.6v-flash",
      "task": "protocol",
      "attempts": 2,
      "solved": 0,
      "total_iterations": 10,
      "total_duration": 3020.2200000000003,
      "infra_failures": 0,
      "success_rate": 0.0,
      "avg_iterations": 5.0,
      "avg_duration": 1510.1,
      "passed": false,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "protocol"
      ],
      "level": 2
    },
    {
      "model": "glm-4.6v-flash",
      "task": "protocol-retry",
      "attempts": 2,
      "solved": 0,
      "total_iterations": 10,
      "total_duration": 4512.58,
      "infra_failures": 0,
      "success_rate": 0.0,
      "avg_iterations": 5.0,
      "avg_duration": 2256.3,
      "passed": false,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "protocol",
        "resilience"
      ],
      "level": 3
    },
    {
      "model": "glm-4.6v-flash",
      "task": "system-id",
      "attempts": 2,
      "solved": 1,
      "total_iterations": 3,
      "total_duration": 2224.99,
      "infra_failures": 0,
      "success_rate": 0.5,
      "avg_iterations": 1.5,
      "avg_duration": 1112.5,
      "passed": true,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "control"
      ],
      "level": 4
    },
    {
      "model": "glm-4.6v-flash",
      "task": "uart-echo",
      "attempts": 2,
      "solved": 2,
      "total_iterations": 2,
      "total_duration": 163.76,
      "infra_failures": 0,
      "success_rate": 1.0,
      "avg_iterations": 1.0,
      "avg_duration": 81.9,
      "passed": true,
      "all_solved": true,
      "incomplete": false,
      "tags": [
        "io"
      ],
      "level": 1
    },
    {
      "model": "glm-4.6v-flash",
      "task": "uart-menu",
      "attempts": 2,
      "solved": 1,
      "total_iterations": 6,
      "total_duration": 2761.52,
      "infra_failures": 0,
      "success_rate": 0.5,
      "avg_iterations": 3.0,
      "avg_duration": 1380.8,
      "passed": true,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "fsm"
      ],
      "level": 3
    },
    {
      "model": "glm-4.6v-flash",
      "task": "watchdog",
      "attempts": 2,
      "solved": 2,
      "total_iterations": 2,
      "total_duration": 566.78,
      "infra_failures": 0,
      "success_rate": 1.0,
      "avg_iterations": 1.0,
      "avg_duration": 283.4,
      "passed": true,
      "all_solved": true,
      "incomplete": false,
      "tags": [
        "resilience"
      ],
      "level": 3
    },
    {
      "model": "qwen2.5-7b-instruct",
      "task": "blink-unix",
      "attempts": 2,
      "solved": 2,
      "total_iterations": 2,
      "total_duration": 19.42,
      "infra_failures": 0,
      "success_rate": 1.0,
      "avg_iterations": 1.0,
      "avg_duration": 9.7,
      "passed": true,
      "all_solved": true,
      "incomplete": false,
      "tags": [
        "io"
      ],
      "level": 1
    },
    {
      "model": "qwen2.5-7b-instruct",
      "task": "coop-scheduler",
      "attempts": 2,
      "solved": 0,
      "total_iterations": 10,
      "total_duration": 74.42999999999999,
      "infra_failures": 0,
      "success_rate": 0.0,
      "avg_iterations": 5.0,
      "avg_duration": 37.2,
      "passed": false,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "control"
      ],
      "level": 3
    },
    {
      "model": "qwen2.5-7b-instruct",
      "task": "frame-corrupt",
      "attempts": 2,
      "solved": 0,
      "total_iterations": 10,
      "total_duration": 51.15,
      "infra_failures": 0,
      "success_rate": 0.0,
      "avg_iterations": 5.0,
      "avg_duration": 25.6,
      "passed": false,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "resilience"
      ],
      "level": 4
    },
    {
      "model": "qwen2.5-7b-instruct",
      "task": "mqtt-device",
      "attempts": 2,
      "solved": 0,
      "total_iterations": 10,
      "total_duration": 149.34,
      "infra_failures": 0,
      "success_rate": 0.0,
      "avg_iterations": 5.0,
      "avg_duration": 74.7,
      "passed": false,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "protocol"
      ],
      "level": 5
    },
    {
      "model": "qwen2.5-7b-instruct",
      "task": "noisy-frames",
      "attempts": 2,
      "solved": 0,
      "total_iterations": 10,
      "total_duration": 319.61,
      "infra_failures": 0,
      "success_rate": 0.0,
      "avg_iterations": 5.0,
      "avg_duration": 159.8,
      "passed": false,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "resilience",
        "protocol"
      ],
      "level": 3
    },
    {
      "model": "qwen2.5-7b-instruct",
      "task": "p-regulator",
      "attempts": 2,
      "solved": 2,
      "total_iterations": 2,
      "total_duration": 87.54,
      "infra_failures": 0,
      "success_rate": 1.0,
      "avg_iterations": 1.0,
      "avg_duration": 43.8,
      "passed": true,
      "all_solved": true,
      "incomplete": false,
      "tags": [
        "control"
      ],
      "level": 2
    },
    {
      "model": "qwen2.5-7b-instruct",
      "task": "pid-antiwindup",
      "attempts": 2,
      "solved": 1,
      "total_iterations": 8,
      "total_duration": 56.58,
      "infra_failures": 0,
      "success_rate": 0.5,
      "avg_iterations": 4.0,
      "avg_duration": 28.3,
      "passed": true,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "control"
      ],
      "level": 3
    },
    {
      "model": "qwen2.5-7b-instruct",
      "task": "protocol",
      "attempts": 2,
      "solved": 0,
      "total_iterations": 10,
      "total_duration": 51.150000000000006,
      "infra_failures": 0,
      "success_rate": 0.0,
      "avg_iterations": 5.0,
      "avg_duration": 25.6,
      "passed": false,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "protocol"
      ],
      "level": 2
    },
    {
      "model": "qwen2.5-7b-instruct",
      "task": "protocol-retry",
      "attempts": 2,
      "solved": 0,
      "total_iterations": 10,
      "total_duration": 56.019999999999996,
      "infra_failures": 0,
      "success_rate": 0.0,
      "avg_iterations": 5.0,
      "avg_duration": 28.0,
      "passed": false,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "protocol",
        "resilience"
      ],
      "level": 3
    },
    {
      "model": "qwen2.5-7b-instruct",
      "task": "system-id",
      "attempts": 2,
      "solved": 1,
      "total_iterations": 7,
      "total_duration": 131.0,
      "infra_failures": 0,
      "success_rate": 0.5,
      "avg_iterations": 3.5,
      "avg_duration": 65.5,
      "passed": true,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "control"
      ],
      "level": 4
    },
    {
      "model": "qwen2.5-7b-instruct",
      "task": "uart-echo",
      "attempts": 2,
      "solved": 1,
      "total_iterations": 6,
      "total_duration": 29.130000000000003,
      "infra_failures": 0,
      "success_rate": 0.5,
      "avg_iterations": 3.0,
      "avg_duration": 14.6,
      "passed": true,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "io"
      ],
      "level": 1
    },
    {
      "model": "qwen2.5-7b-instruct",
      "task": "uart-menu",
      "attempts": 2,
      "solved": 2,
      "total_iterations": 2,
      "total_duration": 9.239999999999998,
      "infra_failures": 0,
      "success_rate": 1.0,
      "avg_iterations": 1.0,
      "avg_duration": 4.6,
      "passed": true,
      "all_solved": true,
      "incomplete": false,
      "tags": [
        "fsm"
      ],
      "level": 3
    },
    {
      "model": "qwen2.5-7b-instruct",
      "task": "watchdog",
      "attempts": 2,
      "solved": 2,
      "total_iterations": 7,
      "total_duration": 123.33000000000001,
      "infra_failures": 0,
      "success_rate": 1.0,
      "avg_iterations": 3.5,
      "avg_duration": 61.7,
      "passed": true,
      "all_solved": true,
      "incomplete": false,
      "tags": [
        "resilience"
      ],
      "level": 3
    },
    {
      "model": "qwen3.5-4b",
      "task": "blink-unix",
      "attempts": 2,
      "solved": 2,
      "total_iterations": 3,
      "total_duration": 212.48,
      "infra_failures": 0,
      "success_rate": 1.0,
      "avg_iterations": 1.5,
      "avg_duration": 106.2,
      "passed": true,
      "all_solved": true,
      "incomplete": false,
      "tags": [
        "io"
      ],
      "level": 1
    },
    {
      "model": "qwen3.5-4b",
      "task": "coop-scheduler",
      "attempts": 2,
      "solved": 0,
      "total_iterations": 10,
      "total_duration": 778.16,
      "infra_failures": 0,
      "success_rate": 0.0,
      "avg_iterations": 5.0,
      "avg_duration": 389.1,
      "passed": false,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "control"
      ],
      "level": 3
    },
    {
      "model": "qwen3.5-4b",
      "task": "frame-corrupt",
      "attempts": 2,
      "solved": 0,
      "total_iterations": 10,
      "total_duration": 753.9100000000001,
      "infra_failures": 0,
      "success_rate": 0.0,
      "avg_iterations": 5.0,
      "avg_duration": 377.0,
      "passed": false,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "resilience"
      ],
      "level": 4
    },
    {
      "model": "qwen3.5-4b",
      "task": "mqtt-device",
      "attempts": 2,
      "solved": 0,
      "total_iterations": 10,
      "total_duration": 467.28,
      "infra_failures": 0,
      "success_rate": 0.0,
      "avg_iterations": 5.0,
      "avg_duration": 233.6,
      "passed": false,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "protocol"
      ],
      "level": 5
    },
    {
      "model": "qwen3.5-4b",
      "task": "noisy-frames",
      "attempts": 2,
      "solved": 1,
      "total_iterations": 7,
      "total_duration": 302.03,
      "infra_failures": 0,
      "success_rate": 0.5,
      "avg_iterations": 3.5,
      "avg_duration": 151.0,
      "passed": true,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "resilience",
        "protocol"
      ],
      "level": 3
    },
    {
      "model": "qwen3.5-4b",
      "task": "p-regulator",
      "attempts": 2,
      "solved": 0,
      "total_iterations": 10,
      "total_duration": 735.4200000000001,
      "infra_failures": 0,
      "success_rate": 0.0,
      "avg_iterations": 5.0,
      "avg_duration": 367.7,
      "passed": false,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "control"
      ],
      "level": 2
    },
    {
      "model": "qwen3.5-4b",
      "task": "pid-antiwindup",
      "attempts": 2,
      "solved": 0,
      "total_iterations": 10,
      "total_duration": 622.92,
      "infra_failures": 0,
      "success_rate": 0.0,
      "avg_iterations": 5.0,
      "avg_duration": 311.5,
      "passed": false,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "control"
      ],
      "level": 3
    },
    {
      "model": "qwen3.5-4b",
      "task": "protocol",
      "attempts": 2,
      "solved": 0,
      "total_iterations": 10,
      "total_duration": 324.07,
      "infra_failures": 0,
      "success_rate": 0.0,
      "avg_iterations": 5.0,
      "avg_duration": 162.0,
      "passed": false,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "protocol"
      ],
      "level": 2
    },
    {
      "model": "qwen3.5-4b",
      "task": "protocol-retry",
      "attempts": 2,
      "solved": 0,
      "total_iterations": 10,
      "total_duration": 485.84999999999997,
      "infra_failures": 0,
      "success_rate": 0.0,
      "avg_iterations": 5.0,
      "avg_duration": 242.9,
      "passed": false,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "protocol",
        "resilience"
      ],
      "level": 3
    },
    {
      "model": "qwen3.5-4b",
      "task": "system-id",
      "attempts": 2,
      "solved": 1,
      "total_iterations": 9,
      "total_duration": 192.45,
      "infra_failures": 0,
      "success_rate": 0.5,
      "avg_iterations": 4.5,
      "avg_duration": 96.2,
      "passed": true,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "control"
      ],
      "level": 4
    },
    {
      "model": "qwen3.5-4b",
      "task": "uart-echo",
      "attempts": 2,
      "solved": 2,
      "total_iterations": 2,
      "total_duration": 140.62,
      "infra_failures": 0,
      "success_rate": 1.0,
      "avg_iterations": 1.0,
      "avg_duration": 70.3,
      "passed": true,
      "all_solved": true,
      "incomplete": false,
      "tags": [
        "io"
      ],
      "level": 1
    },
    {
      "model": "qwen3.5-4b",
      "task": "uart-menu",
      "attempts": 2,
      "solved": 1,
      "total_iterations": 6,
      "total_duration": 251.21999999999997,
      "infra_failures": 0,
      "success_rate": 0.5,
      "avg_iterations": 3.0,
      "avg_duration": 125.6,
      "passed": true,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "fsm"
      ],
      "level": 3
    },
    {
      "model": "qwen3.5-4b",
      "task": "watchdog",
      "attempts": 2,
      "solved": 2,
      "total_iterations": 4,
      "total_duration": 267.45000000000005,
      "infra_failures": 0,
      "success_rate": 1.0,
      "avg_iterations": 2.0,
      "avg_duration": 133.7,
      "passed": true,
      "all_solved": true,
      "incomplete": false,
      "tags": [
        "resilience"
      ],
      "level": 3
    },
    {
      "model": "qwen3.5-9b",
      "task": "blink-unix",
      "attempts": 2,
      "solved": 2,
      "total_iterations": 2,
      "total_duration": 735.4399999999999,
      "infra_failures": 0,
      "success_rate": 1.0,
      "avg_iterations": 1.0,
      "avg_duration": 367.7,
      "passed": true,
      "all_solved": true,
      "incomplete": false,
      "tags": [
        "io"
      ],
      "level": 1
    },
    {
      "model": "qwen3.5-9b",
      "task": "coop-scheduler",
      "attempts": 2,
      "solved": 2,
      "total_iterations": 3,
      "total_duration": 643.4300000000001,
      "infra_failures": 0,
      "success_rate": 1.0,
      "avg_iterations": 1.5,
      "avg_duration": 321.7,
      "passed": true,
      "all_solved": true,
      "incomplete": false,
      "tags": [
        "control"
      ],
      "level": 3
    },
    {
      "model": "qwen3.5-9b",
      "task": "frame-corrupt",
      "attempts": 2,
      "solved": 0,
      "total_iterations": 2,
      "total_duration": 1804.09,
      "infra_failures": 0,
      "success_rate": 0.0,
      "avg_iterations": 1.0,
      "avg_duration": 902.0,
      "passed": false,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "resilience"
      ],
      "level": 4
    },
    {
      "model": "qwen3.5-9b",
      "task": "mqtt-device",
      "attempts": 2,
      "solved": 0,
      "total_iterations": 2,
      "total_duration": 1804.09,
      "infra_failures": 0,
      "success_rate": 0.0,
      "avg_iterations": 1.0,
      "avg_duration": 902.0,
      "passed": false,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "protocol"
      ],
      "level": 5
    },
    {
      "model": "qwen3.5-9b",
      "task": "noisy-frames",
      "attempts": 2,
      "solved": 0,
      "total_iterations": 2,
      "total_duration": 1804.09,
      "infra_failures": 0,
      "success_rate": 0.0,
      "avg_iterations": 1.0,
      "avg_duration": 902.0,
      "passed": false,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "resilience",
        "protocol"
      ],
      "level": 3
    },
    {
      "model": "qwen3.5-9b",
      "task": "p-regulator",
      "attempts": 2,
      "solved": 1,
      "total_iterations": 2,
      "total_duration": 1309.07,
      "infra_failures": 0,
      "success_rate": 0.5,
      "avg_iterations": 1.0,
      "avg_duration": 654.5,
      "passed": true,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "control"
      ],
      "level": 2
    },
    {
      "model": "qwen3.5-9b",
      "task": "pid-antiwindup",
      "attempts": 2,
      "solved": 0,
      "total_iterations": 2,
      "total_duration": 1804.15,
      "infra_failures": 0,
      "success_rate": 0.0,
      "avg_iterations": 1.0,
      "avg_duration": 902.1,
      "passed": false,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "control"
      ],
      "level": 3
    },
    {
      "model": "qwen3.5-9b",
      "task": "protocol",
      "attempts": 2,
      "solved": 0,
      "total_iterations": 10,
      "total_duration": 2649.44,
      "infra_failures": 0,
      "success_rate": 0.0,
      "avg_iterations": 5.0,
      "avg_duration": 1324.7,
      "passed": false,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "protocol"
      ],
      "level": 2
    },
    {
      "model": "qwen3.5-9b",
      "task": "protocol-retry",
      "attempts": 2,
      "solved": 0,
      "total_iterations": 2,
      "total_duration": 1804.08,
      "infra_failures": 0,
      "success_rate": 0.0,
      "avg_iterations": 1.0,
      "avg_duration": 902.0,
      "passed": false,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "protocol",
        "resilience"
      ],
      "level": 3
    },
    {
      "model": "qwen3.5-9b",
      "task": "system-id",
      "attempts": 2,
      "solved": 0,
      "total_iterations": 2,
      "total_duration": 1804.1,
      "infra_failures": 0,
      "success_rate": 0.0,
      "avg_iterations": 1.0,
      "avg_duration": 902.0,
      "passed": false,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "control"
      ],
      "level": 4
    },
    {
      "model": "qwen3.5-9b",
      "task": "uart-echo",
      "attempts": 2,
      "solved": 2,
      "total_iterations": 2,
      "total_duration": 576.33,
      "infra_failures": 0,
      "success_rate": 1.0,
      "avg_iterations": 1.0,
      "avg_duration": 288.2,
      "passed": true,
      "all_solved": true,
      "incomplete": false,
      "tags": [
        "io"
      ],
      "level": 1
    },
    {
      "model": "qwen3.5-9b",
      "task": "uart-menu",
      "attempts": 2,
      "solved": 0,
      "total_iterations": 6,
      "total_duration": 2111.5299999999997,
      "infra_failures": 0,
      "success_rate": 0.0,
      "avg_iterations": 3.0,
      "avg_duration": 1055.8,
      "passed": false,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "fsm"
      ],
      "level": 3
    },
    {
      "model": "qwen3.5-9b",
      "task": "watchdog",
      "attempts": 2,
      "solved": 1,
      "total_iterations": 9,
      "total_duration": 1100.98,
      "infra_failures": 0,
      "success_rate": 0.5,
      "avg_iterations": 4.5,
      "avg_duration": 550.5,
      "passed": true,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "resilience"
      ],
      "level": 3
    },
    {
      "model": "zai-org/glm-4.7-flash",
      "task": "blink-unix",
      "attempts": 2,
      "solved": 2,
      "total_iterations": 2,
      "total_duration": 199.70999999999998,
      "infra_failures": 0,
      "success_rate": 1.0,
      "avg_iterations": 1.0,
      "avg_duration": 99.9,
      "passed": true,
      "all_solved": true,
      "incomplete": false,
      "tags": [
        "io"
      ],
      "level": 1
    },
    {
      "model": "zai-org/glm-4.7-flash",
      "task": "coop-scheduler",
      "attempts": 2,
      "solved": 0,
      "total_iterations": 10,
      "total_duration": 3744.71,
      "infra_failures": 0,
      "success_rate": 0.0,
      "avg_iterations": 5.0,
      "avg_duration": 1872.4,
      "passed": false,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "control"
      ],
      "level": 3
    },
    {
      "model": "zai-org/glm-4.7-flash",
      "task": "frame-corrupt",
      "attempts": 2,
      "solved": 0,
      "total_iterations": 10,
      "total_duration": 3137.9300000000003,
      "infra_failures": 0,
      "success_rate": 0.0,
      "avg_iterations": 5.0,
      "avg_duration": 1569.0,
      "passed": false,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "resilience"
      ],
      "level": 4
    },
    {
      "model": "zai-org/glm-4.7-flash",
      "task": "mqtt-device",
      "attempts": 2,
      "solved": 0,
      "total_iterations": 10,
      "total_duration": 3706.15,
      "infra_failures": 0,
      "success_rate": 0.0,
      "avg_iterations": 5.0,
      "avg_duration": 1853.1,
      "passed": false,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "protocol"
      ],
      "level": 5
    },
    {
      "model": "zai-org/glm-4.7-flash",
      "task": "noisy-frames",
      "attempts": 2,
      "solved": 1,
      "total_iterations": 8,
      "total_duration": 1646.04,
      "infra_failures": 0,
      "success_rate": 0.5,
      "avg_iterations": 4.0,
      "avg_duration": 823.0,
      "passed": true,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "resilience",
        "protocol"
      ],
      "level": 3
    },
    {
      "model": "zai-org/glm-4.7-flash",
      "task": "p-regulator",
      "attempts": 2,
      "solved": 2,
      "total_iterations": 2,
      "total_duration": 548.62,
      "infra_failures": 0,
      "success_rate": 1.0,
      "avg_iterations": 1.0,
      "avg_duration": 274.3,
      "passed": true,
      "all_solved": true,
      "incomplete": false,
      "tags": [
        "control"
      ],
      "level": 2
    },
    {
      "model": "zai-org/glm-4.7-flash",
      "task": "pid-antiwindup",
      "attempts": 2,
      "solved": 2,
      "total_iterations": 4,
      "total_duration": 1414.05,
      "infra_failures": 0,
      "success_rate": 1.0,
      "avg_iterations": 2.0,
      "avg_duration": 707.0,
      "passed": true,
      "all_solved": true,
      "incomplete": false,
      "tags": [
        "control"
      ],
      "level": 3
    },
    {
      "model": "zai-org/glm-4.7-flash",
      "task": "protocol",
      "attempts": 2,
      "solved": 0,
      "total_iterations": 10,
      "total_duration": 2285.92,
      "infra_failures": 0,
      "success_rate": 0.0,
      "avg_iterations": 5.0,
      "avg_duration": 1143.0,
      "passed": false,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "protocol"
      ],
      "level": 2
    },
    {
      "model": "zai-org/glm-4.7-flash",
      "task": "protocol-retry",
      "attempts": 2,
      "solved": 1,
      "total_iterations": 7,
      "total_duration": 1698.78,
      "infra_failures": 0,
      "success_rate": 0.5,
      "avg_iterations": 3.5,
      "avg_duration": 849.4,
      "passed": true,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "protocol",
        "resilience"
      ],
      "level": 3
    },
    {
      "model": "zai-org/glm-4.7-flash",
      "task": "system-id",
      "attempts": 2,
      "solved": 2,
      "total_iterations": 2,
      "total_duration": 649.62,
      "infra_failures": 0,
      "success_rate": 1.0,
      "avg_iterations": 1.0,
      "avg_duration": 324.8,
      "passed": true,
      "all_solved": true,
      "incomplete": false,
      "tags": [
        "control"
      ],
      "level": 4
    },
    {
      "model": "zai-org/glm-4.7-flash",
      "task": "uart-echo",
      "attempts": 2,
      "solved": 2,
      "total_iterations": 2,
      "total_duration": 329.75,
      "infra_failures": 0,
      "success_rate": 1.0,
      "avg_iterations": 1.0,
      "avg_duration": 164.9,
      "passed": true,
      "all_solved": true,
      "incomplete": false,
      "tags": [
        "io"
      ],
      "level": 1
    },
    {
      "model": "zai-org/glm-4.7-flash",
      "task": "uart-menu",
      "attempts": 2,
      "solved": 1,
      "total_iterations": 7,
      "total_duration": 1602.53,
      "infra_failures": 0,
      "success_rate": 0.5,
      "avg_iterations": 3.5,
      "avg_duration": 801.3,
      "passed": true,
      "all_solved": false,
      "incomplete": false,
      "tags": [
        "fsm"
      ],
      "level": 3
    },
    {
      "model": "zai-org/glm-4.7-flash",
      "task": "watchdog",
      "attempts": 2,
      "solved": 2,
      "total_iterations": 2,
      "total_duration": 261.96000000000004,
      "infra_failures": 0,
      "success_rate": 1.0,
      "avg_iterations": 1.0,
      "avg_duration": 131.0,
      "passed": true,
      "all_solved": true,
      "incomplete": false,
      "tags": [
        "resilience"
      ],
      "level": 3
    }
  ],
  "class_profile": {
    "glm-4.6v-flash": {
      "control": {
        "attempts": 8,
        "solved": 4,
        "success_rate": 0.5
      },
      "fsm": {
        "attempts": 2,
        "solved": 1,
        "success_rate": 0.5
      },
      "io": {
        "attempts": 4,
        "solved": 4,
        "success_rate": 1.0
      },
      "protocol": {
        "attempts": 8,
        "solved": 0,
        "success_rate": 0.0
      },
      "resilience": {
        "attempts": 8,
        "solved": 2,
        "success_rate": 0.25
      }
    },
    "qwen2.5-7b-instruct": {
      "control": {
        "attempts": 8,
        "solved": 4,
        "success_rate": 0.5
      },
      "fsm": {
        "attempts": 2,
        "solved": 2,
        "success_rate": 1.0
      },
      "io": {
        "attempts": 4,
        "solved": 3,
        "success_rate": 0.75
      },
      "protocol": {
        "attempts": 8,
        "solved": 0,
        "success_rate": 0.0
      },
      "resilience": {
        "attempts": 8,
        "solved": 2,
        "success_rate": 0.25
      }
    },
    "qwen3.5-4b": {
      "control": {
        "attempts": 8,
        "solved": 1,
        "success_rate": 0.12
      },
      "fsm": {
        "attempts": 2,
        "solved": 1,
        "success_rate": 0.5
      },
      "io": {
        "attempts": 4,
        "solved": 4,
        "success_rate": 1.0
      },
      "protocol": {
        "attempts": 8,
        "solved": 1,
        "success_rate": 0.12
      },
      "resilience": {
        "attempts": 8,
        "solved": 3,
        "success_rate": 0.38
      }
    },
    "qwen3.5-9b": {
      "control": {
        "attempts": 8,
        "solved": 3,
        "success_rate": 0.38
      },
      "fsm": {
        "attempts": 2,
        "solved": 0,
        "success_rate": 0.0
      },
      "io": {
        "attempts": 4,
        "solved": 4,
        "success_rate": 1.0
      },
      "protocol": {
        "attempts": 8,
        "solved": 0,
        "success_rate": 0.0
      },
      "resilience": {
        "attempts": 8,
        "solved": 1,
        "success_rate": 0.12
      }
    },
    "zai-org/glm-4.7-flash": {
      "control": {
        "attempts": 8,
        "solved": 6,
        "success_rate": 0.75
      },
      "fsm": {
        "attempts": 2,
        "solved": 1,
        "success_rate": 0.5
      },
      "io": {
        "attempts": 4,
        "solved": 4,
        "success_rate": 1.0
      },
      "protocol": {
        "attempts": 8,
        "solved": 2,
        "success_rate": 0.25
      },
      "resilience": {
        "attempts": 8,
        "solved": 4,
        "success_rate": 0.5
      }
    }
  },
  "unparseable": []
}