{
  "experiment": "consolidation",
  "selected_configuration_results": {
    "realtime tenant A": {
      "runs": 3,
      "median_p95_ttft_ms": 508.764744,
      "range_p95_ttft_ms": [
        503.305435,
        558.259726
      ],
      "median_p99_ttft_ms": 753.200054,
      "median_p95_tpot_ms_per_token": 23.284018
    },
    "realtime tenant B": {
      "runs": 3,
      "median_p95_ttft_ms": 555.970907,
      "range_p95_ttft_ms": [
        504.601955,
        598.624945
      ],
      "median_p99_ttft_ms": 693.327188,
      "median_p95_tpot_ms_per_token": 23.431359
    },
    "standard burst": {
      "runs": 3,
      "median_p95_ttft_ms": 25891.952991,
      "range_p95_ttft_ms": [
        22833.296537,
        27306.847334
      ],
      "median_p99_ttft_ms": 26349.029064,
      "median_p95_tpot_ms_per_token": 219.079442
    }
  },
  "matched_detector_comparisons": {
    "queue depth 2": {
      "realtime tenant A": {
        "runs": 3,
        "median_p95_ttft_ms": 4711.060524,
        "range_p95_ttft_ms": [
          4102.216721,
          5359.943867
        ],
        "median_p99_ttft_ms": 5444.03863,
        "median_p95_tpot_ms_per_token": 56.96732
      },
      "realtime tenant B": {
        "runs": 3,
        "median_p95_ttft_ms": 4566.735983,
        "range_p95_ttft_ms": [
          4023.555756,
          5056.685448
        ],
        "median_p99_ttft_ms": 5422.817707,
        "median_p95_tpot_ms_per_token": 56.07767
      },
      "standard burst": {
        "runs": 3,
        "median_p95_ttft_ms": 33176.121712,
        "range_p95_ttft_ms": [
          26404.836893,
          45351.89867
        ],
        "median_p99_ttft_ms": 33886.860847,
        "median_p95_tpot_ms_per_token": 279.401045
      }
    },
    "queue depth 5": {
      "realtime tenant A": {
        "runs": 3,
        "median_p95_ttft_ms": 5117.438555,
        "range_p95_ttft_ms": [
          4728.724957,
          5253.631115
        ],
        "median_p99_ttft_ms": 5889.038086,
        "median_p95_tpot_ms_per_token": 61.769705
      },
      "realtime tenant B": {
        "runs": 3,
        "median_p95_ttft_ms": 4906.474113,
        "range_p95_ttft_ms": [
          4656.904221,
          5498.409033
        ],
        "median_p99_ttft_ms": 5921.866417,
        "median_p95_tpot_ms_per_token": 62.690908
      },
      "standard burst": {
        "runs": 3,
        "median_p95_ttft_ms": 51462.030649,
        "range_p95_ttft_ms": [
          36753.162622,
          60899.498701
        ],
        "median_p99_ttft_ms": 51885.482073,
        "median_p95_tpot_ms_per_token": 424.249319
      }
    },
    "request count 128, 10% headroom": {
      "realtime tenant A": {
        "runs": 3,
        "median_p95_ttft_ms": 508.764744,
        "range_p95_ttft_ms": [
          503.305435,
          558.259726
        ],
        "median_p99_ttft_ms": 753.200054,
        "median_p95_tpot_ms_per_token": 23.284018
      },
      "realtime tenant B": {
        "runs": 3,
        "median_p95_ttft_ms": 555.970907,
        "range_p95_ttft_ms": [
          504.601955,
          598.624945
        ],
        "median_p99_ttft_ms": 693.327188,
        "median_p95_tpot_ms_per_token": 23.431359
      },
      "standard burst": {
        "runs": 3,
        "median_p95_ttft_ms": 25891.952991,
        "range_p95_ttft_ms": [
          22833.296537,
          27306.847334
        ],
        "median_p99_ttft_ms": 26349.029064,
        "median_p95_tpot_ms_per_token": 219.079442
      }
    }
  },
  "calibration_only": {},
  "retained_run_inventory": [
    {
      "scenario": "consolidation",
      "detector": "queue depth 2",
      "evidence_role": "matched three-repeat comparison",
      "repeat": 1
    },
    {
      "scenario": "consolidation",
      "detector": "queue depth 2",
      "evidence_role": "matched three-repeat comparison",
      "repeat": 2
    },
    {
      "scenario": "consolidation",
      "detector": "queue depth 2",
      "evidence_role": "matched three-repeat comparison",
      "repeat": 3
    },
    {
      "scenario": "consolidation",
      "detector": "queue depth 5",
      "evidence_role": "matched three-repeat comparison",
      "repeat": 1
    },
    {
      "scenario": "consolidation",
      "detector": "queue depth 5",
      "evidence_role": "matched three-repeat comparison",
      "repeat": 2
    },
    {
      "scenario": "consolidation",
      "detector": "queue depth 5",
      "evidence_role": "matched three-repeat comparison",
      "repeat": 3
    },
    {
      "scenario": "consolidation",
      "detector": "request count 128, 10% headroom",
      "evidence_role": "matched three-repeat comparison",
      "repeat": 1
    },
    {
      "scenario": "consolidation",
      "detector": "request count 128, 10% headroom",
      "evidence_role": "matched three-repeat comparison",
      "repeat": 2
    },
    {
      "scenario": "consolidation",
      "detector": "request count 128, 10% headroom",
      "evidence_role": "matched three-repeat comparison",
      "repeat": 3
    }
  ],
  "saturation_metric_interpretation": "The pool saturation metric is a detector-normalized score. Request-count admission divides in-flight requests by the configured cap. Utilization admission uses the larger of queue depth divided by its threshold and KV-cache utilization divided by its threshold. Read 1.0 as the configured saturation boundary within each detector; raw magnitudes are not compared across detectors.",
  "statistical_scope": "The consolidation and same-priority comparisons use three matched repeats. Their realtime p95 TTFT ranges do not overlap across the tested detectors. With three repeats, the report presents descriptive medians and ranges rather than a formal significance claim.",
  "claim_boundary": "This package contains production-shaped single-GPU evidence for consolidation. It includes matched repeated detector comparisons."
}