{
  "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": "2026-10-05T11:32:39Z",
  "finished_at_unix_ns": 1791199959462139600,
  "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": 340.0,
    "completion_tokens_mean": 259.5,
    "completion_tokens_mean_ci95_high": 311.5188332561711,
    "completion_tokens_mean_ci95_low": 207.4811667438289,
    "completion_tokens_min": 178.0,
    "completion_tokens_p25": 194.0,
    "completion_tokens_p50": 243.0,
    "completion_tokens_p75": 302.0,
    "completion_tokens_p90": 340.0,
    "completion_tokens_p95": 340.0,
    "completion_tokens_p99": 340.0,
    "completion_tokens_samples": 6,
    "completion_tokens_stddev": 65.00999923088754,
    "content_chunks": 45,
    "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": 268.9881986567831,
    "decode_tok_s_mean": 238.08291327610377,
    "decode_tok_s_mean_ci95_high": 257.4520007711718,
    "decode_tok_s_mean_ci95_low": 218.71382578103572,
    "decode_tok_s_min": 201.9412932236856,
    "decode_tok_s_p25": 223.67286250457076,
    "decode_tok_s_p50": 232.2438170196931,
    "decode_tok_s_p75": 257.9627414354114,
    "decode_tok_s_p90": 268.9881986567831,
    "decode_tok_s_p95": 268.9881986567831,
    "decode_tok_s_p99": 268.9881986567831,
    "decode_tok_s_samples": 6,
    "decode_tok_s_stddev": 24.206316911346477,
    "e2e_max_ms": 9402.902300000278,
    "e2e_mean_ci95_high_ms": 9180.236707829737,
    "e2e_mean_ci95_low_ms": 8607.518058838776,
    "e2e_mean_ms": 8893.877383334257,
    "e2e_min_ms": 8475.547300000471,
    "e2e_p25_ms": 8587.329400001181,
    "e2e_p50_ms": 8711.974600002577,
    "e2e_p75_ms": 9099.811000000045,
    "e2e_p90_ms": 9402.902300000278,
    "e2e_p95_ms": 9402.902300000278,
    "e2e_p99_ms": 9402.902300000278,
    "e2e_samples": 6,
    "e2e_stddev_ms": 357.8746061744903,
    "effective_prefill_tok_s_max": 8074.721698177218,
    "effective_prefill_tok_s_mean": 8005.900313849887,
    "effective_prefill_tok_s_mean_ci95_high": 8080.551499921366,
    "effective_prefill_tok_s_mean_ci95_low": 7931.249127778409,
    "effective_prefill_tok_s_min": 7823.448416445818,
    "effective_prefill_tok_s_p25": 8002.28341029081,
    "effective_prefill_tok_s_p50": 8029.717955883823,
    "effective_prefill_tok_s_p75": 8069.188710188819,
    "effective_prefill_tok_s_p90": 8074.721698177218,
    "effective_prefill_tok_s_p95": 8074.721698177218,
    "effective_prefill_tok_s_p99": 8074.721698177218,
    "effective_prefill_tok_s_samples": 6,
    "effective_prefill_tok_s_stddev": 93.29454824932846,
    "generation_max_ms": 1678.7057000001369,
    "generation_mean_ci95_high_ms": 1395.0802873344228,
    "generation_mean_ci95_low_ms": 819.8287793336193,
    "generation_mean_ms": 1107.454533334021,
    "generation_min_ms": 686.1456000006001,
    "generation_p25_ms": 831.0232000003452,
    "generation_p50_ms": 993.0708000028972,
    "generation_p75_ms": 1336.7737000007764,
    "generation_p90_ms": 1678.7057000001369,
    "generation_p95_ms": 1678.7057000001369,
    "generation_p99_ms": 1678.7057000001369,
    "generation_samples": 6,
    "generation_stddev_ms": 359.45731335931725,
    "mean_inter_token_latency_ms_max": 4.951934218289489,
    "mean_inter_token_latency_ms_mean": 4.237721889034524,
    "mean_inter_token_latency_ms_mean_ci95_high": 4.5934153055126625,
    "mean_inter_token_latency_ms_mean_ci95_low": 3.8820284725563856,
    "mean_inter_token_latency_ms_min": 3.7176352159447537,
    "mean_inter_token_latency_ms_p25": 3.8765288135627123,
    "mean_inter_token_latency_ms_p50": 4.1035983471194095,
    "mean_inter_token_latency_ms_p75": 4.470815050169821,
    "mean_inter_token_latency_ms_p90": 4.951934218289489,
    "mean_inter_token_latency_ms_p95": 4.951934218289489,
    "mean_inter_token_latency_ms_p99": 4.951934218289489,
    "mean_inter_token_latency_ms_samples": 6,
    "mean_inter_token_latency_ms_stddev": 0.4445241710401556,
    "mean_time_per_output_token_max_ms": 4.951934218289489,
    "mean_time_per_output_token_mean_ci95_high_ms": 4.5934153055126625,
    "mean_time_per_output_token_mean_ci95_low_ms": 3.8820284725563856,
    "mean_time_per_output_token_mean_ms": 4.237721889034524,
    "mean_time_per_output_token_min_ms": 3.7176352159447537,
    "mean_time_per_output_token_p25_ms": 3.8765288135627123,
    "mean_time_per_output_token_p50_ms": 4.1035983471194095,
    "mean_time_per_output_token_p75_ms": 4.470815050169821,
    "mean_time_per_output_token_p90_ms": 4.951934218289489,
    "mean_time_per_output_token_p95_ms": 4.951934218289489,
    "mean_time_per_output_token_p99_ms": 4.951934218289489,
    "mean_time_per_output_token_samples": 6,
    "mean_time_per_output_token_stddev_ms": 0.4445241710401556,
    "output_token_sources": {
      "usage": 6
    },
    "output_tokens": 1557,
    "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": 334,
    "request_canary_passed": 6,
    "request_canary_samples": 6,
    "throughput_performance_eligible": true,
    "throughput_tok_s": 29.17369079015459,
    "time_to_first_output_max_ms": 7966.691500001616,
    "time_to_first_output_mean_ci95_high_ms": 7860.099459937997,
    "time_to_first_output_mean_ci95_low_ms": 7712.746240062474,
    "time_to_first_output_mean_ms": 7786.4228500002355,
    "time_to_first_output_min_ms": 7718.90379999968,
    "time_to_first_output_p25_ms": 7724.196600000141,
    "time_to_first_output_p50_ms": 7756.306200000836,
    "time_to_first_output_p75_ms": 7789.401699999871,
    "time_to_first_output_p90_ms": 7966.691500001616,
    "time_to_first_output_p95_ms": 7966.691500001616,
    "time_to_first_output_p99_ms": 7966.691500001616,
    "time_to_first_output_samples": 6,
    "time_to_first_output_stddev_ms": 92.07658179876836,
    "tpot_max_ms": 4.951934218289489,
    "tpot_mean_ci95_high_ms": 4.5934153055126625,
    "tpot_mean_ci95_low_ms": 3.8820284725563856,
    "tpot_mean_ms": 4.237721889034524,
    "tpot_min_ms": 3.7176352159447537,
    "tpot_p25_ms": 3.8765288135627123,
    "tpot_p50_ms": 4.1035983471194095,
    "tpot_p75_ms": 4.470815050169821,
    "tpot_p90_ms": 4.951934218289489,
    "tpot_p95_ms": 4.951934218289489,
    "tpot_p99_ms": 4.951934218289489,
    "tpot_samples": 6,
    "tpot_stddev_ms": 0.4445241710401556,
    "ttft_max_ms": 9222.088800001075,
    "ttft_mean_ci95_high_ms": 9052.495580926725,
    "ttft_mean_ci95_low_ms": 8499.859719076578,
    "ttft_mean_ms": 8776.177650001651,
    "ttft_min_ms": 8384.973300002457,
    "ttft_p25_ms": 8442.63940000019,
    "ttft_p50_ms": 8603.36290000123,
    "ttft_p75_ms": 9027.16010000222,
    "ttft_p90_ms": 9222.088800001075,
    "ttft_p95_ms": 9222.088800001075,
    "ttft_p99_ms": 9222.088800001075,
    "ttft_samples": 6,
    "ttft_stddev_ms": 345.3254783408354
  },
  "model": "qwen38-swift15-ninfer-dflash2-c4-262k",
  "performance_eligible": true,
  "prompt_cache_mode": "unique",
  "request_canaries": true,
  "request_timings": [
    {
      "content_chunks": 7,
      "controlled_output": null,
      "decode_tok_s": 268.9881986567831,
      "e2e_ms": 9085.699700000987,
      "effective_prefill_tok_s": 7823.448416445818,
      "generation_ms": 1119.0081999993708,
      "mean_inter_token_latency_ms": 3.7176352159447537,
      "mean_time_per_output_token_ms": 3.7176352159447537,
      "output_token_source": "usage",
      "output_tokens": 302,
      "planned_context_tokens": 65536,
      "prompt_tokens": 62327,
      "reasoning_chunks": 56,
      "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": 7966.691500001616,
      "tpot_ms": 3.7176352159447537,
      "ttft_ms": 8976.84140000274,
      "visible_content_truncated": false,
      "visible_generation_ms": 108.85829999824637
    },
    {
      "content_chunks": 7,
      "controlled_output": null,
      "decode_tok_s": 243.68856681647873,
      "e2e_ms": 8711.974600002577,
      "effective_prefill_tok_s": 8074.721698177218,
      "generation_ms": 993.0708000028972,
      "mean_inter_token_latency_ms": 4.1035983471194095,
      "mean_time_per_output_token_ms": 4.1035983471194095,
      "output_token_source": "usage",
      "output_tokens": 243,
      "planned_context_tokens": 65536,
      "prompt_tokens": 62328,
      "reasoning_chunks": 50,
      "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": 7718.90379999968,
      "tpot_ms": 4.1035983471194095,
      "ttft_ms": 8603.36290000123,
      "visible_content_truncated": false,
      "visible_generation_ms": 108.61170000134734
    },
    {
      "content_chunks": 11,
      "controlled_output": null,
      "decode_tok_s": 201.9412932236856,
      "e2e_ms": 9402.902300000278,
      "effective_prefill_tok_s": 8069.188710188819,
      "generation_ms": 1678.7057000001369,
      "mean_inter_token_latency_ms": 4.951934218289489,
      "mean_time_per_output_token_ms": 4.951934218289489,
      "output_token_source": "usage",
      "output_tokens": 340,
      "planned_context_tokens": 65536,
      "prompt_tokens": 62328,
      "reasoning_chunks": 84,
      "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": 7724.196600000141,
      "tpot_ms": 4.951934218289489,
      "ttft_ms": 9222.088800001075,
      "visible_content_truncated": false,
      "visible_generation_ms": 180.8134999992035
    },
    {
      "content_chunks": 9,
      "controlled_output": null,
      "decode_tok_s": 232.2438170196931,
      "e2e_ms": 8587.329400001181,
      "effective_prefill_tok_s": 8036.041692112836,
      "generation_ms": 831.0232000003452,
      "mean_inter_token_latency_ms": 4.30581968912096,
      "mean_time_per_output_token_ms": 4.30581968912096,
      "output_token_source": "usage",
      "output_tokens": 194,
      "planned_context_tokens": 65536,
      "prompt_tokens": 62330,
      "reasoning_chunks": 39,
      "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": 7756.306200000836,
      "tpot_ms": 4.30581968912096,
      "ttft_ms": 8442.63940000019,
      "visible_content_truncated": false,
      "visible_generation_ms": 144.69000000099186
    },
    {
      "content_chunks": 6,
      "controlled_output": null,
      "decode_tok_s": 257.9627414354114,
      "e2e_ms": 8475.547300000471,
      "effective_prefill_tok_s": 8002.28341029081,
      "generation_ms": 686.1456000006001,
      "mean_inter_token_latency_ms": 3.8765288135627123,
      "mean_time_per_output_token_ms": 3.8765288135627123,
      "output_token_source": "usage",
      "output_tokens": 178,
      "planned_context_tokens": 65536,
      "prompt_tokens": 62333,
      "reasoning_chunks": 34,
      "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": 7789.401699999871,
      "tpot_ms": 3.8765288135627123,
      "ttft_ms": 8384.973300002457,
      "visible_content_truncated": false,
      "visible_generation_ms": 90.57399999801419
    },
    {
      "content_chunks": 5,
      "controlled_output": null,
      "decode_tok_s": 223.67286250457076,
      "e2e_ms": 9099.811000000045,
      "effective_prefill_tok_s": 8029.717955883823,
      "generation_ms": 1336.7737000007764,
      "mean_inter_token_latency_ms": 4.470815050169821,
      "mean_time_per_output_token_ms": 4.470815050169821,
      "output_token_source": "usage",
      "output_tokens": 300,
      "planned_context_tokens": 65536,
      "prompt_tokens": 62335,
      "reasoning_chunks": 71,
      "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": 7763.037299999269,
      "tpot_ms": 4.470815050169821,
      "ttft_ms": 9027.16010000222,
      "visible_content_truncated": false,
      "visible_generation_ms": 72.65089999782504
    }
  ],
  "requests": 6,
  "response_words": 0,
  "responses_received": 6,
  "run_id": "benchmark-20261005T113146Z",
  "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": "2026-10-05T11:31:46Z",
  "started_at_unix_ns": 1791199906092132400,
  "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": 53370.00419999822
}