{
  "schema": "coreai-slot-llm-benchmark/1",
  "status": "PASS",
  "device": "Apple M4 Max GPU",
  "macos": "27.0 (26A428)",
  "build": "Release 0.2.4-zoo f7a75ec + local engine-option patch, Xcode 27.0 (27A266a)",
  "binary": {
    "name": "llm-benchmark",
    "path": "/Users/majimadaisuke/code/codex-conversions/2026-09-23/openthai-systemone-coreai/swift-build-round2/out/Products/Release/llm-benchmark",
    "size_bytes": 17063440,
    "sha256": "1161f9076cd66ca279711feca8287789625d0cfda20c765776f309d9ac6cb4f3"
  },
  "env": {
    "COREAI_CHUNK_THRESHOLD": "1"
  },
  "num_trials_per_engine": 6,
  "num_launches_per_engine": 2,
  "prompt_tokens": 128,
  "generation_tokens": 256,
  "driver_pid": 93377,
  "launches": [
    {
      "launch": 1,
      "engine": "coreai-pipelined",
      "pid": 93383,
      "command": [
        "/Users/majimadaisuke/code/codex-conversions/2026-09-23/openthai-systemone-coreai/swift-build-round2/out/Products/Release/llm-benchmark",
        "--model",
        "/Users/majimadaisuke/code/codex-conversions/2026-09-23/openthai-systemone-coreai/exports/openthai_systemone_decode_int8lin",
        "-p",
        "128",
        "-g",
        "256",
        "-n",
        "3",
        "--inference-engine-variant",
        "coreai-pipelined",
        "--output-json",
        "/Users/majimadaisuke/code/codex-conversions/2026-09-23/openthai-systemone-coreai/results/round2/llm-benchmark-coreai-pipelined-1.json"
      ],
      "stdout_stderr": "logs/llm-benchmark-coreai-pipelined-1.log",
      "tool_json": "results/round2/llm-benchmark-coreai-pipelined-1.json",
      "returncode": 0,
      "wall_seconds": 8.220703084021807,
      "load_seconds": 1.414586083,
      "warmup_seconds": 1.631468334,
      "cache_hit": false,
      "trials": [
        {
          "engine": "coreai-pipelined",
          "launch": 1,
          "trial": 1,
          "prefill_tokens_per_second": 252.52844898699632,
          "decode_tokens_per_second": 253.0164564889887,
          "load_seconds": 1.414586083
        },
        {
          "engine": "coreai-pipelined",
          "launch": 1,
          "trial": 2,
          "prefill_tokens_per_second": 254.7808735202058,
          "decode_tokens_per_second": 252.0935504336627,
          "load_seconds": 1.414586083
        },
        {
          "engine": "coreai-pipelined",
          "launch": 1,
          "trial": 3,
          "prefill_tokens_per_second": 258.56576985378376,
          "decode_tokens_per_second": 253.77845845384445,
          "load_seconds": 1.414586083
        }
      ]
    },
    {
      "launch": 1,
      "engine": "coreai-sequential",
      "pid": 93702,
      "command": [
        "/Users/majimadaisuke/code/codex-conversions/2026-09-23/openthai-systemone-coreai/swift-build-round2/out/Products/Release/llm-benchmark",
        "--model",
        "/Users/majimadaisuke/code/codex-conversions/2026-09-23/openthai-systemone-coreai/exports/openthai_systemone_decode_int8lin",
        "-p",
        "128",
        "-g",
        "256",
        "-n",
        "3",
        "--inference-engine-variant",
        "coreai-sequential",
        "--output-json",
        "/Users/majimadaisuke/code/codex-conversions/2026-09-23/openthai-systemone-coreai/results/round2/llm-benchmark-coreai-sequential-1.json"
      ],
      "stdout_stderr": "logs/llm-benchmark-coreai-sequential-1.log",
      "tool_json": "results/round2/llm-benchmark-coreai-sequential-1.json",
      "returncode": 0,
      "wall_seconds": 8.310072167078033,
      "load_seconds": 0.172054,
      "warmup_seconds": 2.031085,
      "cache_hit": true,
      "trials": [
        {
          "engine": "coreai-sequential",
          "launch": 1,
          "trial": 1,
          "prefill_tokens_per_second": 201.25757145724364,
          "decode_tokens_per_second": 192.64417194062574,
          "load_seconds": 0.172054
        },
        {
          "engine": "coreai-sequential",
          "launch": 1,
          "trial": 2,
          "prefill_tokens_per_second": 196.1141690340596,
          "decode_tokens_per_second": 197.65960041682428,
          "load_seconds": 0.172054
        },
        {
          "engine": "coreai-sequential",
          "launch": 1,
          "trial": 3,
          "prefill_tokens_per_second": 197.26953777036385,
          "decode_tokens_per_second": 193.31968059713273,
          "load_seconds": 0.172054
        }
      ]
    },
    {
      "launch": 2,
      "engine": "coreai-pipelined",
      "pid": 94066,
      "command": [
        "/Users/majimadaisuke/code/codex-conversions/2026-09-23/openthai-systemone-coreai/swift-build-round2/out/Products/Release/llm-benchmark",
        "--model",
        "/Users/majimadaisuke/code/codex-conversions/2026-09-23/openthai-systemone-coreai/exports/openthai_systemone_decode_int8lin",
        "-p",
        "128",
        "-g",
        "256",
        "-n",
        "3",
        "--inference-engine-variant",
        "coreai-pipelined",
        "--output-json",
        "/Users/majimadaisuke/code/codex-conversions/2026-09-23/openthai-systemone-coreai/results/round2/llm-benchmark-coreai-pipelined-2.json"
      ],
      "stdout_stderr": "logs/llm-benchmark-coreai-pipelined-2.log",
      "tool_json": "results/round2/llm-benchmark-coreai-pipelined-2.json",
      "returncode": 0,
      "wall_seconds": 6.742144333082251,
      "load_seconds": 0.16701425,
      "warmup_seconds": 1.66963975,
      "cache_hit": true,
      "trials": [
        {
          "engine": "coreai-pipelined",
          "launch": 2,
          "trial": 1,
          "prefill_tokens_per_second": 252.8524716983957,
          "decode_tokens_per_second": 249.45770462960482,
          "load_seconds": 0.16701425
        },
        {
          "engine": "coreai-pipelined",
          "launch": 2,
          "trial": 2,
          "prefill_tokens_per_second": 244.70684048818057,
          "decode_tokens_per_second": 244.79629883663122,
          "load_seconds": 0.16701425
        },
        {
          "engine": "coreai-pipelined",
          "launch": 2,
          "trial": 3,
          "prefill_tokens_per_second": 243.79336750696703,
          "decode_tokens_per_second": 244.51588224721894,
          "load_seconds": 0.16701425
        }
      ]
    },
    {
      "launch": 2,
      "engine": "coreai-sequential",
      "pid": 94304,
      "command": [
        "/Users/majimadaisuke/code/codex-conversions/2026-09-23/openthai-systemone-coreai/swift-build-round2/out/Products/Release/llm-benchmark",
        "--model",
        "/Users/majimadaisuke/code/codex-conversions/2026-09-23/openthai-systemone-coreai/exports/openthai_systemone_decode_int8lin",
        "-p",
        "128",
        "-g",
        "256",
        "-n",
        "3",
        "--inference-engine-variant",
        "coreai-sequential",
        "--output-json",
        "/Users/majimadaisuke/code/codex-conversions/2026-09-23/openthai-systemone-coreai/results/round2/llm-benchmark-coreai-sequential-2.json"
      ],
      "stdout_stderr": "logs/llm-benchmark-coreai-sequential-2.log",
      "tool_json": "results/round2/llm-benchmark-coreai-sequential-2.json",
      "returncode": 0,
      "wall_seconds": 8.31699312501587,
      "load_seconds": 0.165566875,
      "warmup_seconds": 2.042509167,
      "cache_hit": true,
      "trials": [
        {
          "engine": "coreai-sequential",
          "launch": 2,
          "trial": 1,
          "prefill_tokens_per_second": 197.01550993211984,
          "decode_tokens_per_second": 194.46614805285301,
          "load_seconds": 0.165566875
        },
        {
          "engine": "coreai-sequential",
          "launch": 2,
          "trial": 2,
          "prefill_tokens_per_second": 196.78542824074384,
          "decode_tokens_per_second": 197.51888519114257,
          "load_seconds": 0.165566875
        },
        {
          "engine": "coreai-sequential",
          "launch": 2,
          "trial": 3,
          "prefill_tokens_per_second": 197.7727390365532,
          "decode_tokens_per_second": 195.1917448423383,
          "load_seconds": 0.165566875
        }
      ]
    }
  ],
  "trials": [
    {
      "engine": "coreai-pipelined",
      "launch": 1,
      "trial": 1,
      "prefill_tokens_per_second": 252.52844898699632,
      "decode_tokens_per_second": 253.0164564889887,
      "load_seconds": 1.414586083
    },
    {
      "engine": "coreai-pipelined",
      "launch": 1,
      "trial": 2,
      "prefill_tokens_per_second": 254.7808735202058,
      "decode_tokens_per_second": 252.0935504336627,
      "load_seconds": 1.414586083
    },
    {
      "engine": "coreai-pipelined",
      "launch": 1,
      "trial": 3,
      "prefill_tokens_per_second": 258.56576985378376,
      "decode_tokens_per_second": 253.77845845384445,
      "load_seconds": 1.414586083
    },
    {
      "engine": "coreai-sequential",
      "launch": 1,
      "trial": 1,
      "prefill_tokens_per_second": 201.25757145724364,
      "decode_tokens_per_second": 192.64417194062574,
      "load_seconds": 0.172054
    },
    {
      "engine": "coreai-sequential",
      "launch": 1,
      "trial": 2,
      "prefill_tokens_per_second": 196.1141690340596,
      "decode_tokens_per_second": 197.65960041682428,
      "load_seconds": 0.172054
    },
    {
      "engine": "coreai-sequential",
      "launch": 1,
      "trial": 3,
      "prefill_tokens_per_second": 197.26953777036385,
      "decode_tokens_per_second": 193.31968059713273,
      "load_seconds": 0.172054
    },
    {
      "engine": "coreai-pipelined",
      "launch": 2,
      "trial": 1,
      "prefill_tokens_per_second": 252.8524716983957,
      "decode_tokens_per_second": 249.45770462960482,
      "load_seconds": 0.16701425
    },
    {
      "engine": "coreai-pipelined",
      "launch": 2,
      "trial": 2,
      "prefill_tokens_per_second": 244.70684048818057,
      "decode_tokens_per_second": 244.79629883663122,
      "load_seconds": 0.16701425
    },
    {
      "engine": "coreai-pipelined",
      "launch": 2,
      "trial": 3,
      "prefill_tokens_per_second": 243.79336750696703,
      "decode_tokens_per_second": 244.51588224721894,
      "load_seconds": 0.16701425
    },
    {
      "engine": "coreai-sequential",
      "launch": 2,
      "trial": 1,
      "prefill_tokens_per_second": 197.01550993211984,
      "decode_tokens_per_second": 194.46614805285301,
      "load_seconds": 0.165566875
    },
    {
      "engine": "coreai-sequential",
      "launch": 2,
      "trial": 2,
      "prefill_tokens_per_second": 196.78542824074384,
      "decode_tokens_per_second": 197.51888519114257,
      "load_seconds": 0.165566875
    },
    {
      "engine": "coreai-sequential",
      "launch": 2,
      "trial": 3,
      "prefill_tokens_per_second": 197.7727390365532,
      "decode_tokens_per_second": 195.1917448423383,
      "load_seconds": 0.165566875
    }
  ],
  "engines": {
    "coreai-pipelined": {
      "engine": "coreai-pipelined",
      "num_trials": 6,
      "num_launches": 2,
      "trials": [
        {
          "engine": "coreai-pipelined",
          "launch": 1,
          "trial": 1,
          "prefill_tokens_per_second": 252.52844898699632,
          "decode_tokens_per_second": 253.0164564889887,
          "load_seconds": 1.414586083
        },
        {
          "engine": "coreai-pipelined",
          "launch": 1,
          "trial": 2,
          "prefill_tokens_per_second": 254.7808735202058,
          "decode_tokens_per_second": 252.0935504336627,
          "load_seconds": 1.414586083
        },
        {
          "engine": "coreai-pipelined",
          "launch": 1,
          "trial": 3,
          "prefill_tokens_per_second": 258.56576985378376,
          "decode_tokens_per_second": 253.77845845384445,
          "load_seconds": 1.414586083
        },
        {
          "engine": "coreai-pipelined",
          "launch": 2,
          "trial": 1,
          "prefill_tokens_per_second": 252.8524716983957,
          "decode_tokens_per_second": 249.45770462960482,
          "load_seconds": 0.16701425
        },
        {
          "engine": "coreai-pipelined",
          "launch": 2,
          "trial": 2,
          "prefill_tokens_per_second": 244.70684048818057,
          "decode_tokens_per_second": 244.79629883663122,
          "load_seconds": 0.16701425
        },
        {
          "engine": "coreai-pipelined",
          "launch": 2,
          "trial": 3,
          "prefill_tokens_per_second": 243.79336750696703,
          "decode_tokens_per_second": 244.51588224721894,
          "load_seconds": 0.16701425
        }
      ],
      "launches": [
        {
          "launch": 1,
          "engine": "coreai-pipelined",
          "pid": 93383,
          "command": [
            "/Users/majimadaisuke/code/codex-conversions/2026-09-23/openthai-systemone-coreai/swift-build-round2/out/Products/Release/llm-benchmark",
            "--model",
            "/Users/majimadaisuke/code/codex-conversions/2026-09-23/openthai-systemone-coreai/exports/openthai_systemone_decode_int8lin",
            "-p",
            "128",
            "-g",
            "256",
            "-n",
            "3",
            "--inference-engine-variant",
            "coreai-pipelined",
            "--output-json",
            "/Users/majimadaisuke/code/codex-conversions/2026-09-23/openthai-systemone-coreai/results/round2/llm-benchmark-coreai-pipelined-1.json"
          ],
          "stdout_stderr": "logs/llm-benchmark-coreai-pipelined-1.log",
          "tool_json": "results/round2/llm-benchmark-coreai-pipelined-1.json",
          "returncode": 0,
          "wall_seconds": 8.220703084021807,
          "load_seconds": 1.414586083,
          "warmup_seconds": 1.631468334,
          "cache_hit": false,
          "trials": [
            {
              "engine": "coreai-pipelined",
              "launch": 1,
              "trial": 1,
              "prefill_tokens_per_second": 252.52844898699632,
              "decode_tokens_per_second": 253.0164564889887,
              "load_seconds": 1.414586083
            },
            {
              "engine": "coreai-pipelined",
              "launch": 1,
              "trial": 2,
              "prefill_tokens_per_second": 254.7808735202058,
              "decode_tokens_per_second": 252.0935504336627,
              "load_seconds": 1.414586083
            },
            {
              "engine": "coreai-pipelined",
              "launch": 1,
              "trial": 3,
              "prefill_tokens_per_second": 258.56576985378376,
              "decode_tokens_per_second": 253.77845845384445,
              "load_seconds": 1.414586083
            }
          ]
        },
        {
          "launch": 2,
          "engine": "coreai-pipelined",
          "pid": 94066,
          "command": [
            "/Users/majimadaisuke/code/codex-conversions/2026-09-23/openthai-systemone-coreai/swift-build-round2/out/Products/Release/llm-benchmark",
            "--model",
            "/Users/majimadaisuke/code/codex-conversions/2026-09-23/openthai-systemone-coreai/exports/openthai_systemone_decode_int8lin",
            "-p",
            "128",
            "-g",
            "256",
            "-n",
            "3",
            "--inference-engine-variant",
            "coreai-pipelined",
            "--output-json",
            "/Users/majimadaisuke/code/codex-conversions/2026-09-23/openthai-systemone-coreai/results/round2/llm-benchmark-coreai-pipelined-2.json"
          ],
          "stdout_stderr": "logs/llm-benchmark-coreai-pipelined-2.log",
          "tool_json": "results/round2/llm-benchmark-coreai-pipelined-2.json",
          "returncode": 0,
          "wall_seconds": 6.742144333082251,
          "load_seconds": 0.16701425,
          "warmup_seconds": 1.66963975,
          "cache_hit": true,
          "trials": [
            {
              "engine": "coreai-pipelined",
              "launch": 2,
              "trial": 1,
              "prefill_tokens_per_second": 252.8524716983957,
              "decode_tokens_per_second": 249.45770462960482,
              "load_seconds": 0.16701425
            },
            {
              "engine": "coreai-pipelined",
              "launch": 2,
              "trial": 2,
              "prefill_tokens_per_second": 244.70684048818057,
              "decode_tokens_per_second": 244.79629883663122,
              "load_seconds": 0.16701425
            },
            {
              "engine": "coreai-pipelined",
              "launch": 2,
              "trial": 3,
              "prefill_tokens_per_second": 243.79336750696703,
              "decode_tokens_per_second": 244.51588224721894,
              "load_seconds": 0.16701425
            }
          ]
        }
      ],
      "load_seconds_per_launch": [
        1.414586083,
        0.16701425
      ],
      "median_prefill_tokens_per_second": 252.690460342696,
      "min_prefill_tokens_per_second": 243.79336750696703,
      "max_prefill_tokens_per_second": 258.56576985378376,
      "median_decode_tokens_per_second": 250.77562753163375,
      "min_decode_tokens_per_second": 244.51588224721894,
      "max_decode_tokens_per_second": 253.77845845384445,
      "median_load_seconds": 0.7908001665000001,
      "min_load_seconds": 0.16701425,
      "max_load_seconds": 1.414586083
    },
    "coreai-sequential": {
      "engine": "coreai-sequential",
      "num_trials": 6,
      "num_launches": 2,
      "trials": [
        {
          "engine": "coreai-sequential",
          "launch": 1,
          "trial": 1,
          "prefill_tokens_per_second": 201.25757145724364,
          "decode_tokens_per_second": 192.64417194062574,
          "load_seconds": 0.172054
        },
        {
          "engine": "coreai-sequential",
          "launch": 1,
          "trial": 2,
          "prefill_tokens_per_second": 196.1141690340596,
          "decode_tokens_per_second": 197.65960041682428,
          "load_seconds": 0.172054
        },
        {
          "engine": "coreai-sequential",
          "launch": 1,
          "trial": 3,
          "prefill_tokens_per_second": 197.26953777036385,
          "decode_tokens_per_second": 193.31968059713273,
          "load_seconds": 0.172054
        },
        {
          "engine": "coreai-sequential",
          "launch": 2,
          "trial": 1,
          "prefill_tokens_per_second": 197.01550993211984,
          "decode_tokens_per_second": 194.46614805285301,
          "load_seconds": 0.165566875
        },
        {
          "engine": "coreai-sequential",
          "launch": 2,
          "trial": 2,
          "prefill_tokens_per_second": 196.78542824074384,
          "decode_tokens_per_second": 197.51888519114257,
          "load_seconds": 0.165566875
        },
        {
          "engine": "coreai-sequential",
          "launch": 2,
          "trial": 3,
          "prefill_tokens_per_second": 197.7727390365532,
          "decode_tokens_per_second": 195.1917448423383,
          "load_seconds": 0.165566875
        }
      ],
      "launches": [
        {
          "launch": 1,
          "engine": "coreai-sequential",
          "pid": 93702,
          "command": [
            "/Users/majimadaisuke/code/codex-conversions/2026-09-23/openthai-systemone-coreai/swift-build-round2/out/Products/Release/llm-benchmark",
            "--model",
            "/Users/majimadaisuke/code/codex-conversions/2026-09-23/openthai-systemone-coreai/exports/openthai_systemone_decode_int8lin",
            "-p",
            "128",
            "-g",
            "256",
            "-n",
            "3",
            "--inference-engine-variant",
            "coreai-sequential",
            "--output-json",
            "/Users/majimadaisuke/code/codex-conversions/2026-09-23/openthai-systemone-coreai/results/round2/llm-benchmark-coreai-sequential-1.json"
          ],
          "stdout_stderr": "logs/llm-benchmark-coreai-sequential-1.log",
          "tool_json": "results/round2/llm-benchmark-coreai-sequential-1.json",
          "returncode": 0,
          "wall_seconds": 8.310072167078033,
          "load_seconds": 0.172054,
          "warmup_seconds": 2.031085,
          "cache_hit": true,
          "trials": [
            {
              "engine": "coreai-sequential",
              "launch": 1,
              "trial": 1,
              "prefill_tokens_per_second": 201.25757145724364,
              "decode_tokens_per_second": 192.64417194062574,
              "load_seconds": 0.172054
            },
            {
              "engine": "coreai-sequential",
              "launch": 1,
              "trial": 2,
              "prefill_tokens_per_second": 196.1141690340596,
              "decode_tokens_per_second": 197.65960041682428,
              "load_seconds": 0.172054
            },
            {
              "engine": "coreai-sequential",
              "launch": 1,
              "trial": 3,
              "prefill_tokens_per_second": 197.26953777036385,
              "decode_tokens_per_second": 193.31968059713273,
              "load_seconds": 0.172054
            }
          ]
        },
        {
          "launch": 2,
          "engine": "coreai-sequential",
          "pid": 94304,
          "command": [
            "/Users/majimadaisuke/code/codex-conversions/2026-09-23/openthai-systemone-coreai/swift-build-round2/out/Products/Release/llm-benchmark",
            "--model",
            "/Users/majimadaisuke/code/codex-conversions/2026-09-23/openthai-systemone-coreai/exports/openthai_systemone_decode_int8lin",
            "-p",
            "128",
            "-g",
            "256",
            "-n",
            "3",
            "--inference-engine-variant",
            "coreai-sequential",
            "--output-json",
            "/Users/majimadaisuke/code/codex-conversions/2026-09-23/openthai-systemone-coreai/results/round2/llm-benchmark-coreai-sequential-2.json"
          ],
          "stdout_stderr": "logs/llm-benchmark-coreai-sequential-2.log",
          "tool_json": "results/round2/llm-benchmark-coreai-sequential-2.json",
          "returncode": 0,
          "wall_seconds": 8.31699312501587,
          "load_seconds": 0.165566875,
          "warmup_seconds": 2.042509167,
          "cache_hit": true,
          "trials": [
            {
              "engine": "coreai-sequential",
              "launch": 2,
              "trial": 1,
              "prefill_tokens_per_second": 197.01550993211984,
              "decode_tokens_per_second": 194.46614805285301,
              "load_seconds": 0.165566875
            },
            {
              "engine": "coreai-sequential",
              "launch": 2,
              "trial": 2,
              "prefill_tokens_per_second": 196.78542824074384,
              "decode_tokens_per_second": 197.51888519114257,
              "load_seconds": 0.165566875
            },
            {
              "engine": "coreai-sequential",
              "launch": 2,
              "trial": 3,
              "prefill_tokens_per_second": 197.7727390365532,
              "decode_tokens_per_second": 195.1917448423383,
              "load_seconds": 0.165566875
            }
          ]
        }
      ],
      "load_seconds_per_launch": [
        0.172054,
        0.165566875
      ],
      "median_prefill_tokens_per_second": 197.14252385124183,
      "min_prefill_tokens_per_second": 196.1141690340596,
      "max_prefill_tokens_per_second": 201.25757145724364,
      "median_decode_tokens_per_second": 194.82894644759565,
      "min_decode_tokens_per_second": 192.64417194062574,
      "max_decode_tokens_per_second": 197.65960041682428,
      "median_load_seconds": 0.16881043750000002,
      "min_load_seconds": 0.165566875,
      "max_load_seconds": 0.172054
    }
  },
  "interpretation": "Prefill-rate proxy: this decision graph is decode-only S=1; synthetic generation throughput is not decision latency. Input IDs are sampled from metadata logits width 256 by llm-benchmark.",
  "contention_rule": "Mandated top-eight ps snapshot; non-WindowServer/Chrome Core AI, Python, Swift engine jobs; exclude this foreground driver only.",
  "snapshots": [
    {
      "path": "results/ps-before.txt",
      "at": "2026-09-22T19:19:24.177433+00:00",
      "engine_jobs": []
    },
    {
      "path": "results/ps-after.txt",
      "at": "2026-09-22T19:19:55.799761+00:00",
      "engine_jobs": []
    }
  ],
  "contended": false,
  "wait_seconds": 3.3294782042503357e-07,
  "benchmark_start_snapshot": {
    "path": "results/ps-before.txt",
    "at": "2026-09-22T19:19:24.177433+00:00",
    "engine_jobs": []
  },
  "benchmark_process_wall_seconds": 31.62205179198645,
  "load_measurement": "llm-benchmark prepare_seconds: EngineFactory creation; per launch, not per trial; excluded warmup.",
  "completed_utc": "2026-09-22T19:19:55.799819+00:00"
}