{
  "levels": [
    12,
    16,
    24
  ],
  "methodology": {
    "warmup_seconds": 120,
    "measurement_seconds": 300
  },
  "arms": [
    {
      "engine": "vLLM",
      "points": [
        {
          "active_sessions": 24,
          "readings": {
            "cache_context": {
              "value": null
            },
            "decode_p10_tps": {
              "value": 21.647357597927805
            },
            "error_rate_pct": {
              "value": 0.1557632398753894
            },
            "gpu_memory_gib": {
              "value": 133.58978271484375
            },
            "gpu_power_w": {
              "value": 591.2128666666667
            },
            "gpu_utilization_pct": {
              "value": 100.0
            },
            "output_tps": {
              "value": 374.7588238735019
            },
            "running_requests_mean": {
              "value": 15.533333333333333
            },
            "tpot_p50_ms": {
              "value": 36.749324468085106
            },
            "ttft_p50_ms": {
              "value": 2327.0245
            },
            "ttft_p95_ms": {
              "value": 14548.081291
            },
            "waiting_requests_mean": {
              "value": 0.03333333333333333
            }
          },
          "users": 12,
          "valid_requests": 641,
          "total_requests": 642,
          "level_valid": true
        },
        {
          "active_sessions": 32,
          "readings": {
            "cache_context": {
              "value": null
            },
            "decode_p10_tps": {
              "value": 15.959221959779452
            },
            "error_rate_pct": {
              "value": 0.15015015015015015
            },
            "gpu_memory_gib": {
              "value": 133.58978271484375
            },
            "gpu_power_w": {
              "value": 600.8785833333333
            },
            "gpu_utilization_pct": {
              "value": 99.91666666666667
            },
            "output_tps": {
              "value": 376.7270594950265
            },
            "running_requests_mean": {
              "value": 23.266666666666666
            },
            "tpot_p50_ms": {
              "value": 49.54259679797979
            },
            "ttft_p50_ms": {
              "value": 2789.812041
            },
            "ttft_p95_ms": {
              "value": 17179.270167
            },
            "waiting_requests_mean": {
              "value": 0.03333333333333333
            }
          },
          "users": 16,
          "valid_requests": 665,
          "total_requests": 666,
          "level_valid": true
        },
        {
          "active_sessions": 48,
          "readings": {
            "cache_context": {
              "value": null
            },
            "decode_p10_tps": {
              "value": 9.581166807869453
            },
            "error_rate_pct": {
              "value": 0.0
            },
            "gpu_memory_gib": {
              "value": 133.58978271484375
            },
            "gpu_power_w": {
              "value": 629.8228833333334
            },
            "gpu_utilization_pct": {
              "value": 100.0
            },
            "output_tps": {
              "value": 379.4803156121838
            },
            "running_requests_mean": {
              "value": 36.61666666666667
            },
            "tpot_p50_ms": {
              "value": 74.48276938613861
            },
            "ttft_p50_ms": {
              "value": 5455.011208
            },
            "ttft_p95_ms": {
              "value": 30787.455542
            },
            "waiting_requests_mean": {
              "value": 0.2
            }
          },
          "users": 24,
          "valid_requests": 662,
          "total_requests": 662,
          "level_valid": true
        }
      ]
    },
    {
      "engine": "SGLang",
      "points": [
        {
          "active_sessions": 24,
          "readings": {
            "cache_context": {
              "value": null
            },
            "decode_p10_tps": {
              "value": 33.005161580540744
            },
            "error_rate_pct": {
              "value": 0.0
            },
            "gpu_memory_gib": {
              "value": 125.672119140625
            },
            "gpu_power_w": {
              "value": 671.2198666666667
            },
            "gpu_utilization_pct": {
              "value": 98.86666666666666
            },
            "output_tps": {
              "value": 497.36994127360805
            },
            "running_requests_mean": {
              "value": 13.716666666666667
            },
            "tpot_p50_ms": {
              "value": 20.63596386725664
            },
            "ttft_p50_ms": {
              "value": 1667.396959
            },
            "ttft_p95_ms": {
              "value": 8229.3205
            },
            "waiting_requests_mean": {
              "value": 0.05
            }
          },
          "users": 12,
          "valid_requests": 747,
          "total_requests": 747,
          "level_valid": true
        },
        {
          "active_sessions": 32,
          "readings": {
            "cache_context": {
              "value": null
            },
            "decode_p10_tps": {
              "value": 17.482769989167465
            },
            "error_rate_pct": {
              "value": 0.0
            },
            "gpu_memory_gib": {
              "value": 125.672119140625
            },
            "gpu_power_w": {
              "value": 682.3903999999999
            },
            "gpu_utilization_pct": {
              "value": 97.73333333333333
            },
            "output_tps": {
              "value": 480.43621543584874
            },
            "running_requests_mean": {
              "value": 19.05
            },
            "tpot_p50_ms": {
              "value": 27.706553358574613
            },
            "ttft_p50_ms": {
              "value": 2041.941292
            },
            "ttft_p95_ms": {
              "value": 9764.01125
            },
            "waiting_requests_mean": {
              "value": 0.5833333333333334
            }
          },
          "users": 16,
          "valid_requests": 889,
          "total_requests": 889,
          "level_valid": true
        },
        {
          "active_sessions": 48,
          "readings": {
            "cache_context": {
              "value": null
            },
            "decode_p10_tps": {
              "value": 8.162312232411331
            },
            "error_rate_pct": {
              "value": 0.0
            },
            "gpu_memory_gib": {
              "value": 125.676025390625
            },
            "gpu_power_w": {
              "value": 692.1702000000001
            },
            "gpu_utilization_pct": {
              "value": 99.26666666666667
            },
            "output_tps": {
              "value": 365.3459253216277
            },
            "running_requests_mean": {
              "value": 34.516666666666666
            },
            "tpot_p50_ms": {
              "value": 68.29774684126984
            },
            "ttft_p50_ms": {
              "value": 5451.441167
            },
            "ttft_p95_ms": {
              "value": 30014.645167
            },
            "waiting_requests_mean": {
              "value": 1.2
            }
          },
          "users": 24,
          "valid_requests": 601,
          "total_requests": 601,
          "level_valid": true
        }
      ]
    }
  ],
  "metric_definitions": [
    {
      "id": "output_tps",
      "explanation": "Successful output tokens per measured second."
    },
    {
      "id": "ttft_p50_ms",
      "explanation": "Median client-visible time to first token."
    },
    {
      "id": "ttft_p95_ms",
      "explanation": "95th-percentile client-visible time to first token."
    },
    {
      "id": "tpot_p50_ms",
      "explanation": "Median time per output token after the first token."
    },
    {
      "id": "decode_p10_tps",
      "explanation": "Ten-percentile per-request decode speed; 90% of valid requests are at least this fast."
    },
    {
      "id": "error_rate_pct",
      "explanation": "Share of measured requests that did not satisfy the validity contract."
    },
    {
      "id": "running_requests_mean",
      "explanation": "Mean runtime-native running-request gauge in the measured window."
    },
    {
      "id": "waiting_requests_mean",
      "explanation": "Mean runtime-native waiting-request gauge in the measured window."
    },
    {
      "id": "cache_context",
      "explanation": "Available cache context retained without cross-engine normalization."
    },
    {
      "id": "gpu_utilization_pct",
      "explanation": "Mean sampled device utilization in the measured window."
    },
    {
      "id": "gpu_memory_gib",
      "explanation": "Maximum sampled device memory use in the measured window."
    },
    {
      "id": "gpu_power_w",
      "explanation": "Mean sampled board power in the measured window."
    }
  ],
  "schema_version": "token-by-token.public-site-projection.v1"
}