{
  "testId": "2026-08-03-qwen36-advanced-planning",
  "title": "Qwen3.6 35B deterministic release-planning challenge",
  "scope": "Observed result on one Windows workstation; not hardware purchasing advice.",
  "runtime": "Ollama",
  "model": "qwen3.6:35b",
  "installedModelBytes": 23000000000,
  "contextLength": 16384,
  "outputTokenLimit": 2048,
  "validator": "C:/ComfyUI/LocalAILab/benchmarks/validate_release_schedule.py",
  "validatorCategories": 12,
  "machine": {
    "gpu": "NVIDIA GeForce RTX 5080",
    "installedVramMiB": 16303,
    "installedRamBytes": 68062883840
  },
  "attempts": [
    {
      "label": "cold initial",
      "sessionId": "20260803-061852-ollama-qwen3.6-35b",
      "serviceTotalSeconds": 22.6993448,
      "loadSeconds": 14.1487477,
      "generatedTokens": 595,
      "tokensPerSecond": 76.2161438602132,
      "peakRamUsedBytes": 31947354112,
      "peakVramUsedMiB": 15728,
      "gpuResidentPercent": 63.68417762583768,
      "score": {"passed": 10, "total": 12, "percent": 83.3},
      "failedCategories": ["people_available", "dependencies"],
      "errors": [
        "Task F fell outside Priya's and Leo's availability.",
        "Task H fell outside Mia's availability.",
        "B did not finish before D began.",
        "C did not finish before H began."
      ]
    },
    {
      "label": "warm identical",
      "sessionId": "20260803-061921-ollama-qwen3.6-35b",
      "serviceTotalSeconds": 8.3987515,
      "loadSeconds": 0.3493986,
      "generatedTokens": 595,
      "tokensPerSecond": 75.30938233396306,
      "peakRamUsedBytes": 31977910272,
      "peakVramUsedMiB": 15728,
      "gpuResidentPercent": 63.68417762583768,
      "score": {"passed": 10, "total": 12, "percent": 83.3},
      "note": "Deterministic controls produced the same schedule and failures as the cold attempt."
    },
    {
      "label": "warm repair 1",
      "sessionId": "20260803-062023-ollama-qwen3.6-35b",
      "serviceTotalSeconds": 9.2383612,
      "loadSeconds": 0.3351839,
      "generatedTokens": 595,
      "tokensPerSecond": 76.24337243272868,
      "peakRamUsedBytes": 32180314112,
      "peakVramUsedMiB": 15728,
      "gpuResidentPercent": 63.68417762583768,
      "score": {"passed": 11, "total": 12, "percent": 91.7},
      "failedCategories": ["people_available"],
      "errors": ["Task C crossed Noah's unavailable 12:00-14:00 interval."]
    },
    {
      "label": "warm repair 2",
      "sessionId": "20260803-062114-ollama-qwen3.6-35b",
      "serviceTotalSeconds": 9.2560191,
      "loadSeconds": 0.317397,
      "generatedTokens": 595,
      "tokensPerSecond": 75.94656221184981,
      "peakRamUsedBytes": 32419868672,
      "peakVramUsedMiB": 15728,
      "gpuResidentPercent": 63.68417762583768,
      "score": {"passed": 11, "total": 12, "percent": 91.7},
      "failedCategories": ["people_available"],
      "errors": [
        "Task D fell outside Priya's availability.",
        "Task H fell outside Mia's availability."
      ],
      "note": "The second repair preserved dependencies but moved the availability mistakes to different tasks."
    }
  ],
  "conclusion": "The model reliably produced strict, complete JSON and improved after exact validator feedback, but it did not produce a fully feasible schedule in three distinct prompts. Maximum observed score: 11 of 12 validator categories."
}