{
  "title": "Qwen3.8-27B with Qwen Code: sanitized metrics for the 150k Go repair run",
  "measurement_date": "2026-08-20",
  "task": {
    "language": "Go",
    "request": "Repair two code paths that ignored json.Marshal errors, preserve successful response formats, add deterministic failure behavior and regression tests, avoid unrelated cleanup, run focused tests and build, and update relevant documentation.",
    "prompt_file": "everyday-json-marshal-fix.md"
  },
  "system": {
    "model": "Qwen3.8-27B",
    "weights": "fast W4A16 AutoRound variant",
    "server": "frozen third-party patched vLLM stack",
    "harness": "Qwen Code 0.21.13",
    "reasoning_effort": "medium",
    "advertised_context_tokens": 150000,
    "gpu": "NVIDIA GeForce RTX 3090 24 GB",
    "gpu_power_limit_w": 250,
    "inference_slots": 1
  },
  "run": {
    "completed": true,
    "timed_out": false,
    "interrupted": false,
    "wall_seconds": 502.140624952,
    "event_count": 137,
    "tool_calls": 52,
    "tool_calls_by_name": {
      "read_file": 16,
      "grep_search": 11,
      "run_shell_command": 9,
      "glob": 2,
      "todo_write": 4,
      "edit": 8,
      "write_file": 2
    },
    "token_usage": {
      "cumulative_input_tokens_across_turns": 1891966,
      "output_tokens": 28916,
      "total_tokens": 1920882,
      "note": "Cumulative input repeats conversation and repository context across turns; it is not the size of one prompt."
    },
    "context_compactions": 0
  },
  "gpu_measurement": {
    "samples": 503,
    "average_power_w": 242.88491053677976,
    "maximum_sampled_power_w": 250.32,
    "estimated_energy_wh": 33.87843910231922,
    "average_temperature_c": 63.55268389662028,
    "maximum_temperature_c": 71,
    "maximum_fan_percent": 58,
    "average_gpu_utilization_percent": 89.20477137176938,
    "maximum_gpu_utilization_percent": 100,
    "maximum_vram_used_mib": 22539,
    "average_graphics_clock_mhz": 1542.4950298210736
  },
  "patch": {
    "tracked_files_changed": 5,
    "new_regression_test_files": 2,
    "tracked_lines_added": 28,
    "tracked_lines_removed": 7,
    "focused_test_suites_passed": 2,
    "go_build_all_passed": true,
    "diff_check_passed": true,
    "broad_preexisting_failures_reproduced": true
  },
  "evaluation": {
    "evaluator": "OpenAI Codex",
    "independent_external_auditor": false,
    "rubric_owner": "Syntalith",
    "score": 100,
    "score_max": 100,
    "breakdown": {
      "correctness": "40/40",
      "regression_tests": "20/20",
      "compatibility": "15/15",
      "scope_discipline": "10/10",
      "verification": "10/10",
      "documentation": "5/5"
    },
    "meaning": "Codex found no deduction under this rubric in this single run. The score does not establish repeatability or general model reliability."
  },
  "source_artifact": {
    "type": "sanitized extract of the private run summary",
    "private_summary_sha256": "5b5e5c4b4d866031cc850540dee012dd0cfd53a62cb75471e1b90b6084ec7129"
  }
}