{
  "base_url": "[operator identity redacted]",
  "clock_domain_id": null,
  "completed": 6,
  "concurrency": 1,
  "configuration_fingerprint": null,
  "context_distribution": {
    "65536": 6
  },
  "context_seed": 1717,
  "context_tokens": 65536,
  "controlled_output_policy": "observe",
  "engine": "ninfer",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-10-05T11:41:40Z",
  "finished_at_unix_ns": 1791200500799135500,
  "gpu": "rtx5090",
  "manifest": null,
  "max_context_tokens": 262144,
  "max_tokens": 8192,
  "measurement_protocol": "capacity-v4-reasoning",
  "metric_population": {
    "aggregate_throughput_eligible": true,
    "attempted_requests": 6,
    "excluded_requests": 0,
    "failure_policy": "failed or contract-invalid responses retain sanitized timing and metadata but are excluded from performance distributions",
    "p99_interpretation": "descriptive-nearest-rank-only",
    "p99_interpretation_by_metric": {
      "decode_tok_s": "descriptive-nearest-rank-only",
      "e2e": "descriptive-nearest-rank-only",
      "effective_prefill_tok_s": "descriptive-nearest-rank-only",
      "tpot": "descriptive-nearest-rank-only",
      "ttft": "descriptive-nearest-rank-only"
    },
    "p99_minimum_recommended_samples": 100,
    "p99_samples_by_metric": {
      "decode_tok_s": 6,
      "e2e": 6,
      "effective_prefill_tok_s": 6,
      "tpot": 6,
      "ttft": 6
    },
    "performance_eligible_requests": 6,
    "successful_requests": 6
  },
  "metrics": {
    "completion_tokens_max": 413.0,
    "completion_tokens_mean": 306.1666666666667,
    "completion_tokens_mean_ci95_high": 376.8017902557456,
    "completion_tokens_mean_ci95_low": 235.5315430775878,
    "completion_tokens_min": 217.0,
    "completion_tokens_p25": 225.0,
    "completion_tokens_p50": 244.0,
    "completion_tokens_p75": 394.0,
    "completion_tokens_p90": 413.0,
    "completion_tokens_p95": 413.0,
    "completion_tokens_p99": 413.0,
    "completion_tokens_samples": 6,
    "completion_tokens_stddev": 88.27551566921979,
    "content_chunks": 50,
    "content_chunks_s": null,
    "controlled_output_exact_adherence": 0,
    "controlled_output_exact_adherence_rate": null,
    "controlled_output_samples": 0,
    "controlled_output_unobservable": 0,
    "decode_tok_s_max": 245.04366453666688,
    "decode_tok_s_mean": 229.27380787083698,
    "decode_tok_s_mean_ci95_high": 241.36034294285585,
    "decode_tok_s_mean_ci95_low": 217.1872727988181,
    "decode_tok_s_min": 210.77734946731394,
    "decode_tok_s_p25": 215.7331234646854,
    "decode_tok_s_p50": 221.31351052529016,
    "decode_tok_s_p75": 243.600829156665,
    "decode_tok_s_p90": 245.04366453666688,
    "decode_tok_s_p95": 245.04366453666688,
    "decode_tok_s_p99": 245.04366453666688,
    "decode_tok_s_samples": 6,
    "decode_tok_s_stddev": 15.105022288111924,
    "e2e_max_ms": 9662.036300000182,
    "e2e_mean_ci95_high_ms": 9409.804936443781,
    "e2e_mean_ci95_low_ms": 8859.182596888231,
    "e2e_mean_ms": 9134.493766666006,
    "e2e_min_ms": 8684.946299999865,
    "e2e_p25_ms": 8925.67009999766,
    "e2e_p50_ms": 8975.71170000083,
    "e2e_p75_ms": 9298.808199997438,
    "e2e_p90_ms": 9662.036300000182,
    "e2e_p95_ms": 9662.036300000182,
    "e2e_p99_ms": 9662.036300000182,
    "e2e_samples": 6,
    "e2e_stddev_ms": 344.0672890021936,
    "effective_prefill_tok_s_max": 8099.7920159301775,
    "effective_prefill_tok_s_mean": 7998.594150459329,
    "effective_prefill_tok_s_mean_ci95_high": 8110.44955689747,
    "effective_prefill_tok_s_mean_ci95_low": 7886.738744021189,
    "effective_prefill_tok_s_min": 7720.803974934607,
    "effective_prefill_tok_s_p25": 8019.380622270646,
    "effective_prefill_tok_s_p50": 8027.265324278172,
    "effective_prefill_tok_s_p75": 8084.096602204206,
    "effective_prefill_tok_s_p90": 8099.7920159301775,
    "effective_prefill_tok_s_p95": 8099.7920159301775,
    "effective_prefill_tok_s_p99": 8099.7920159301775,
    "effective_prefill_tok_s_samples": 6,
    "effective_prefill_tok_s_stddev": 139.7901381352388,
    "generation_max_ms": 1909.7670000010112,
    "generation_mean_ci95_high_ms": 1667.9730767987296,
    "generation_mean_ci95_low_ms": 1011.6667898666058,
    "generation_mean_ms": 1339.8199333326677,
    "generation_min_ms": 903.106800000387,
    "generation_p25_ms": 919.5370999987063,
    "generation_p50_ms": 1152.8752999984135,
    "generation_p75_ms": 1603.7957999978971,
    "generation_p90_ms": 1909.7670000010112,
    "generation_p95_ms": 1909.7670000010112,
    "generation_p99_ms": 1909.7670000010112,
    "generation_samples": 6,
    "generation_stddev_ms": 410.10599948070177,
    "mean_inter_token_latency_ms_max": 4.744342798347381,
    "mean_inter_token_latency_ms_mean": 4.377534707530502,
    "mean_inter_token_latency_ms_mean_ci95_high": 4.609917735364123,
    "mean_inter_token_latency_ms_mean_ci95_low": 4.14515167969688,
    "mean_inter_token_latency_ms_min": 4.0809053435061,
    "mean_inter_token_latency_ms_p25": 4.105076339279939,
    "mean_inter_token_latency_ms_p50": 4.181050000001791,
    "mean_inter_token_latency_ms_p75": 4.635356796118959,
    "mean_inter_token_latency_ms_p90": 4.744342798347381,
    "mean_inter_token_latency_ms_p95": 4.744342798347381,
    "mean_inter_token_latency_ms_p99": 4.744342798347381,
    "mean_inter_token_latency_ms_samples": 6,
    "mean_inter_token_latency_ms_stddev": 0.2904182872833438,
    "mean_time_per_output_token_max_ms": 4.744342798347381,
    "mean_time_per_output_token_mean_ci95_high_ms": 4.609917735364123,
    "mean_time_per_output_token_mean_ci95_low_ms": 4.14515167969688,
    "mean_time_per_output_token_mean_ms": 4.377534707530502,
    "mean_time_per_output_token_min_ms": 4.0809053435061,
    "mean_time_per_output_token_p25_ms": 4.105076339279939,
    "mean_time_per_output_token_p50_ms": 4.181050000001791,
    "mean_time_per_output_token_p75_ms": 4.635356796118959,
    "mean_time_per_output_token_p90_ms": 4.744342798347381,
    "mean_time_per_output_token_p95_ms": 4.744342798347381,
    "mean_time_per_output_token_p99_ms": 4.744342798347381,
    "mean_time_per_output_token_samples": 6,
    "mean_time_per_output_token_stddev_ms": 0.2904182872833438,
    "output_token_sources": {
      "usage": 6
    },
    "output_tokens": 1837,
    "prefix_cache_hit_avg": 0.0,
    "prompt_token_samples": 6,
    "prompt_tokens": 373981,
    "prompt_tokens_max": 62335.0,
    "prompt_tokens_mean": 62330.166666666664,
    "prompt_tokens_mean_ci95_high": 62332.718014894126,
    "prompt_tokens_mean_ci95_low": 62327.6153184392,
    "prompt_tokens_min": 62327.0,
    "prompt_tokens_p25": 62328.0,
    "prompt_tokens_p50": 62328.0,
    "prompt_tokens_p75": 62333.0,
    "prompt_tokens_p90": 62335.0,
    "prompt_tokens_p95": 62335.0,
    "prompt_tokens_p99": 62335.0,
    "prompt_tokens_samples": 6,
    "prompt_tokens_stddev": 3.1885210782848317,
    "reasoning_chunks": 406,
    "request_canary_passed": 6,
    "request_canary_samples": 6,
    "throughput_performance_eligible": true,
    "throughput_tok_s": 33.51357879815759,
    "time_to_first_output_max_ms": 8072.604900000442,
    "time_to_first_output_mean_ci95_high_ms": 7906.404753604324,
    "time_to_first_output_mean_ci95_low_ms": 7682.942913062354,
    "time_to_first_output_mean_ms": 7794.673833333339,
    "time_to_first_output_min_ms": 7695.012399999541,
    "time_to_first_output_p25_ms": 7709.952400000475,
    "time_to_first_output_p50_ms": 7752.2692999991705,
    "time_to_first_output_p75_ms": 7772.794799999247,
    "time_to_first_output_p90_ms": 8072.604900000442,
    "time_to_first_output_p95_ms": 8072.604900000442,
    "time_to_first_output_p99_ms": 8072.604900000442,
    "time_to_first_output_samples": 6,
    "time_to_first_output_stddev_ms": 139.63456283443986,
    "tpot_max_ms": 4.744342798347381,
    "tpot_mean_ci95_high_ms": 4.609917735364123,
    "tpot_mean_ci95_low_ms": 4.14515167969688,
    "tpot_mean_ms": 4.377534707530502,
    "tpot_min_ms": 4.0809053435061,
    "tpot_p25_ms": 4.105076339279939,
    "tpot_p50_ms": 4.181050000001791,
    "tpot_p75_ms": 4.635356796118959,
    "tpot_p90_ms": 4.744342798347381,
    "tpot_p95_ms": 4.744342798347381,
    "tpot_p99_ms": 4.744342798347381,
    "tpot_samples": 6,
    "tpot_stddev_ms": 0.2904182872833438,
    "ttft_max_ms": 9589.440100000502,
    "ttft_mean_ci95_high_ms": 9296.28116161328,
    "ttft_mean_ci95_low_ms": 8706.76217171957,
    "ttft_mean_ms": 9001.521666666426,
    "ttft_min_ms": 8540.200900002674,
    "ttft_p25_ms": 8781.176599997707,
    "ttft_p50_ms": 8828.89989999967,
    "ttft_p75_ms": 9190.386399997806,
    "ttft_p90_ms": 9589.440100000502,
    "ttft_p95_ms": 9589.440100000502,
    "ttft_p99_ms": 9589.440100000502,
    "ttft_samples": 6,
    "ttft_stddev_ms": 368.37263237768957
  },
  "model": "qwen38-swift15-ninfer-dflash2-c4-262k",
  "performance_eligible": true,
  "prompt_cache_mode": "unique",
  "request_canaries": true,
  "request_timings": [
    {
      "content_chunks": 9,
      "controlled_output": null,
      "decode_tok_s": 239.17437007440031,
      "e2e_ms": 8975.71170000083,
      "effective_prefill_tok_s": 7720.803974934607,
      "generation_ms": 903.106800000387,
      "mean_inter_token_latency_ms": 4.181050000001791,
      "mean_time_per_output_token_ms": 4.181050000001791,
      "output_token_source": "usage",
      "output_tokens": 217,
      "planned_context_tokens": 65536,
      "prompt_tokens": 62327,
      "reasoning_chunks": 42,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_1717_00000",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 0,
      "time_to_first_output_ms": 8072.604900000442,
      "tpot_ms": 4.181050000001791,
      "ttft_ms": 8828.89989999967,
      "visible_content_truncated": false,
      "visible_generation_ms": 146.81180000116
    },
    {
      "content_chunks": 7,
      "controlled_output": null,
      "decode_tok_s": 245.04366453666688,
      "e2e_ms": 9298.808199997438,
      "effective_prefill_tok_s": 8099.7920159301775,
      "generation_ms": 1603.7957999978971,
      "mean_inter_token_latency_ms": 4.0809053435061,
      "mean_time_per_output_token_ms": 4.0809053435061,
      "output_token_source": "usage",
      "output_tokens": 394,
      "planned_context_tokens": 65536,
      "prompt_tokens": 62328,
      "reasoning_chunks": 84,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_1717_00001",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 1,
      "time_to_first_output_ms": 7695.012399999541,
      "tpot_ms": 4.0809053435061,
      "ttft_ms": 9190.386399997806,
      "visible_content_truncated": false,
      "visible_generation_ms": 108.42179999963264
    },
    {
      "content_chunks": 11,
      "controlled_output": null,
      "decode_tok_s": 221.31351052529016,
      "e2e_ms": 9259.790000000066,
      "effective_prefill_tok_s": 8084.096602204206,
      "generation_ms": 1549.8375999995915,
      "mean_inter_token_latency_ms": 4.518476967928838,
      "mean_time_per_output_token_ms": 4.518476967928838,
      "output_token_source": "usage",
      "output_tokens": 344,
      "planned_context_tokens": 65536,
      "prompt_tokens": 62328,
      "reasoning_chunks": 77,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_1717_00002",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 2,
      "time_to_first_output_ms": 7709.952400000475,
      "tpot_ms": 4.518476967928838,
      "ttft_ms": 9079.026100000192,
      "visible_content_truncated": false,
      "visible_generation_ms": 180.7638999998744
    },
    {
      "content_chunks": 5,
      "controlled_output": null,
      "decode_tok_s": 215.7331234646854,
      "e2e_ms": 9662.036300000182,
      "effective_prefill_tok_s": 8040.2263631381675,
      "generation_ms": 1909.7670000010112,
      "mean_inter_token_latency_ms": 4.635356796118959,
      "mean_time_per_output_token_ms": 4.635356796118959,
      "output_token_source": "usage",
      "output_tokens": 413,
      "planned_context_tokens": 65536,
      "prompt_tokens": 62330,
      "reasoning_chunks": 102,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_1717_00003",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 3,
      "time_to_first_output_ms": 7752.2692999991705,
      "tpot_ms": 4.635356796118959,
      "ttft_ms": 9589.440100000502,
      "visible_content_truncated": false,
      "visible_generation_ms": 72.59619999967981
    },
    {
      "content_chunks": 9,
      "controlled_output": null,
      "decode_tok_s": 210.77734946731394,
      "e2e_ms": 8925.67009999766,
      "effective_prefill_tok_s": 8019.380622270646,
      "generation_ms": 1152.8752999984135,
      "mean_inter_token_latency_ms": 4.744342798347381,
      "mean_time_per_output_token_ms": 4.744342798347381,
      "output_token_source": "usage",
      "output_tokens": 244,
      "planned_context_tokens": 65536,
      "prompt_tokens": 62333,
      "reasoning_chunks": 57,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_1717_00004",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 4,
      "time_to_first_output_ms": 7772.794799999247,
      "tpot_ms": 4.744342798347381,
      "ttft_ms": 8781.176599997707,
      "visible_content_truncated": false,
      "visible_generation_ms": 144.4934999999532
    },
    {
      "content_chunks": 9,
      "controlled_output": null,
      "decode_tok_s": 243.600829156665,
      "e2e_ms": 8684.946299999865,
      "effective_prefill_tok_s": 8027.265324278172,
      "generation_ms": 919.5370999987063,
      "mean_inter_token_latency_ms": 4.105076339279939,
      "mean_time_per_output_token_ms": 4.105076339279939,
      "output_token_source": "usage",
      "output_tokens": 225,
      "planned_context_tokens": 65536,
      "prompt_tokens": 62335,
      "reasoning_chunks": 44,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_1717_00005",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 5,
      "time_to_first_output_ms": 7765.409200001159,
      "tpot_ms": 4.105076339279939,
      "ttft_ms": 8540.200900002674,
      "visible_content_truncated": false,
      "visible_generation_ms": 144.74539999719127
    }
  ],
  "requests": 6,
  "response_words": 0,
  "responses_received": 6,
  "run_id": "benchmark-20261005T114045Z",
  "sampling": {
    "temperature": {
      "effective_request": 0.9,
      "requested": 0.9,
      "sent": true
    },
    "top_p": {
      "effective_request": 0.95,
      "requested": 0.95,
      "sent": true
    }
  },
  "schema": "anvil-serving.benchmark/v1",
  "serve_flags": {
    "no_thinking": false,
    "reasoning_effort": null,
    "shared_prefix_burst": false,
    "thinking_mode": "default"
  },
  "shared_prefix_tokens": 0,
  "source_recipe": null,
  "started_at": "2026-10-05T11:40:45Z",
  "started_at_unix_ns": 1791200445985529300,
  "tier": null,
  "timestamp_precision": "nanosecond-unix-fields; ISO labels are second precision",
  "timing_methodology": {
    "artifact_timestamps": "time.time_ns Unix timestamps; independent artifacts may align them only when the operator supplied the same explicit clock_domain_id",
    "clock": "client time.perf_counter",
    "controlled_output": "--response-words adherence is recorded; observe preserves historical behavior while strict rejects non-exact or truncated completions; visible response capture is capped at 8192 characters",
    "decode": "usage completion tokens after the first token divided by client-observed generation time; completion tokens may include reasoning tokens",
    "distributions": "request-level mean, sample standard deviation, normal-approximation 95% confidence interval for the mean, min, p25, p50, p75, p90, p95, p99, and max using the nearest-rank percentile method",
    "effective_prefill": "usage.prompt_tokens divided by client-observed time to first output; includes queueing, scheduling, prefill, and first-token work",
    "generation": "client-observed E2E minus time to first output",
    "mean_inter_token_latency": "legacy key for a TPOT proxy: client-observed generation time divided by completion tokens after the first token; not raw per-token ITL",
    "p99": "numeric nearest-rank p99 is always retained for compatibility; it is descriptive only below 100 eligible samples",
    "request_canaries": "when enabled, each response must begin with its own synthetic marker and contain no other ANVIL_REQ marker; bounded/truncated captures fail closed and are excluded from performance metrics",
    "time_to_first_output": "request start through the first non-empty streamed reasoning or content delta",
    "tpot": "client-observed mean time per output token after the first output; identical to the legacy mean_inter_token_latency key",
    "ttft": "request start through the first non-empty streamed content delta"
  },
  "wall_clock_ms": 54813.60289999793
}