{
  "base_url": "http://127.0.0.1:8080/v1",
  "clock_domain_id": null,
  "completed": 6,
  "concurrency": 1,
  "configuration_fingerprint": null,
  "context_distribution": {
    "64000": 6
  },
  "context_seed": 44,
  "context_tokens": 64000,
  "controlled_output_policy": "strict",
  "engine": "ninfer",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-09-19T23:30:34Z",
  "finished_at_unix_ns": 1789860634546952800,
  "gpu": "RTX5090",
  "manifest": null,
  "max_context_tokens": 65536,
  "max_tokens": 512,
  "measurement_protocol": "capacity-v3",
  "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": 33.0,
    "completion_tokens_mean": 33.0,
    "completion_tokens_mean_ci95_high": 33.0,
    "completion_tokens_mean_ci95_low": 33.0,
    "completion_tokens_min": 33.0,
    "completion_tokens_p25": 33.0,
    "completion_tokens_p50": 33.0,
    "completion_tokens_p75": 33.0,
    "completion_tokens_p90": 33.0,
    "completion_tokens_p95": 33.0,
    "completion_tokens_p99": 33.0,
    "completion_tokens_samples": 6,
    "completion_tokens_stddev": 0.0,
    "content_chunks": 57,
    "content_chunks_s": null,
    "controlled_output_exact_adherence": 6,
    "controlled_output_exact_adherence_rate": 1.0,
    "controlled_output_samples": 6,
    "controlled_output_unobservable": 0,
    "decode_tok_s_max": 167.30757675903405,
    "decode_tok_s_mean": 167.0767708213319,
    "decode_tok_s_mean_ci95_high": 167.2690263530955,
    "decode_tok_s_mean_ci95_low": 166.8845152895683,
    "decode_tok_s_min": 166.74795627373288,
    "decode_tok_s_p25": 166.79645520246083,
    "decode_tok_s_p50": 167.1776904587466,
    "decode_tok_s_p75": 167.2184876868321,
    "decode_tok_s_p90": 167.30757675903405,
    "decode_tok_s_p95": 167.30757675903405,
    "decode_tok_s_p99": 167.30757675903405,
    "decode_tok_s_samples": 6,
    "decode_tok_s_stddev": 0.24026936380015076,
    "e2e_max_ms": 12216.776299988851,
    "e2e_mean_ci95_high_ms": 12147.345249887168,
    "e2e_mean_ci95_low_ms": 12039.054583450858,
    "e2e_mean_ms": 12093.199916669013,
    "e2e_min_ms": 12035.063700051978,
    "e2e_p25_ms": 12042.151799891144,
    "e2e_p50_ms": 12063.638800056651,
    "e2e_p75_ms": 12118.794700014405,
    "e2e_p90_ms": 12216.776299988851,
    "e2e_p95_ms": 12216.776299988851,
    "e2e_p99_ms": 12216.776299988851,
    "e2e_samples": 6,
    "e2e_stddev_ms": 67.66757058033146,
    "effective_prefill_tok_s_max": 5130.9176403276515,
    "effective_prefill_tok_s_mean": 5106.324770010286,
    "effective_prefill_tok_s_mean_ci95_high": 5129.449972560762,
    "effective_prefill_tok_s_mean_ci95_low": 5083.199567459809,
    "effective_prefill_tok_s_min": 5053.340016647144,
    "effective_prefill_tok_s_p25": 5095.713020138994,
    "effective_prefill_tok_s_p50": 5111.041164177032,
    "effective_prefill_tok_s_p75": 5128.014553406278,
    "effective_prefill_tok_s_p90": 5130.9176403276515,
    "effective_prefill_tok_s_p95": 5130.9176403276515,
    "effective_prefill_tok_s_p99": 5130.9176403276515,
    "effective_prefill_tok_s_samples": 6,
    "effective_prefill_tok_s_stddev": 28.900482881211794,
    "generation_max_ms": 191.9064000248909,
    "generation_mean_ci95_high_ms": 191.74962087599667,
    "generation_mean_ci95_low_ms": 191.30847913147775,
    "generation_mean_ms": 191.5290500037372,
    "generation_min_ms": 191.2644999101758,
    "generation_p25_ms": 191.36639998760074,
    "generation_p50_ms": 191.37330004014075,
    "generation_p75_ms": 191.85059994924814,
    "generation_p90_ms": 191.9064000248909,
    "generation_p95_ms": 191.9064000248909,
    "generation_p99_ms": 191.9064000248909,
    "generation_samples": 6,
    "generation_stddev_ms": 0.275656167936866,
    "mean_inter_token_latency_ms_max": 5.997075000777841,
    "mean_inter_token_latency_ms_mean": 5.985282812616788,
    "mean_inter_token_latency_ms_mean_ci95_high": 5.992175652374896,
    "mean_inter_token_latency_ms_mean_ci95_low": 5.97838997285868,
    "mean_inter_token_latency_ms_min": 5.977015622192994,
    "mean_inter_token_latency_ms_p25": 5.980199999612523,
    "mean_inter_token_latency_ms_p50": 5.980415626254398,
    "mean_inter_token_latency_ms_p75": 5.995331248414004,
    "mean_inter_token_latency_ms_p90": 5.997075000777841,
    "mean_inter_token_latency_ms_p95": 5.997075000777841,
    "mean_inter_token_latency_ms_p99": 5.997075000777841,
    "mean_inter_token_latency_ms_samples": 6,
    "mean_inter_token_latency_ms_stddev": 0.008614255248027062,
    "mean_time_per_output_token_max_ms": 5.997075000777841,
    "mean_time_per_output_token_mean_ci95_high_ms": 5.992175652374896,
    "mean_time_per_output_token_mean_ci95_low_ms": 5.97838997285868,
    "mean_time_per_output_token_mean_ms": 5.985282812616788,
    "mean_time_per_output_token_min_ms": 5.977015622192994,
    "mean_time_per_output_token_p25_ms": 5.980199999612523,
    "mean_time_per_output_token_p50_ms": 5.980415626254398,
    "mean_time_per_output_token_p75_ms": 5.995331248414004,
    "mean_time_per_output_token_p90_ms": 5.997075000777841,
    "mean_time_per_output_token_p95_ms": 5.997075000777841,
    "mean_time_per_output_token_p99_ms": 5.997075000777841,
    "mean_time_per_output_token_samples": 6,
    "mean_time_per_output_token_stddev_ms": 0.008614255248027062,
    "output_token_sources": {
      "usage": 6
    },
    "output_tokens": 198,
    "prefix_cache_hit_avg": null,
    "prompt_token_samples": 6,
    "prompt_tokens": 364633,
    "prompt_tokens_max": 60776.0,
    "prompt_tokens_mean": 60772.166666666664,
    "prompt_tokens_mean_ci95_high": 60774.56271712059,
    "prompt_tokens_mean_ci95_low": 60769.770616212736,
    "prompt_tokens_min": 60769.0,
    "prompt_tokens_p25": 60769.0,
    "prompt_tokens_p50": 60771.0,
    "prompt_tokens_p75": 60775.0,
    "prompt_tokens_p90": 60776.0,
    "prompt_tokens_p95": 60776.0,
    "prompt_tokens_p99": 60776.0,
    "prompt_tokens_samples": 6,
    "prompt_tokens_stddev": 2.9944392908634274,
    "reasoning_chunks": 0,
    "request_canary_passed": 0,
    "request_canary_samples": 0,
    "throughput_performance_eligible": true,
    "throughput_tok_s": 2.728494780999433,
    "time_to_first_output_max_ms": 12025.511800078675,
    "time_to_first_output_mean_ci95_high_ms": 11955.82989384318,
    "time_to_first_output_mean_ci95_low_ms": 11847.511839487372,
    "time_to_first_output_mean_ms": 11901.670866665276,
    "time_to_first_output_min_ms": 11843.690400011837,
    "time_to_first_output_p25_ms": 11850.785399903543,
    "time_to_first_output_p50_ms": 11872.225699946284,
    "time_to_first_output_p75_ms": 11926.888299989514,
    "time_to_first_output_p90_ms": 12025.511800078675,
    "time_to_first_output_p95_ms": 12025.511800078675,
    "time_to_first_output_p99_ms": 12025.511800078675,
    "time_to_first_output_samples": 6,
    "time_to_first_output_stddev_ms": 67.68468446499509,
    "tpot_max_ms": 5.997075000777841,
    "tpot_mean_ci95_high_ms": 5.992175652374896,
    "tpot_mean_ci95_low_ms": 5.97838997285868,
    "tpot_mean_ms": 5.985282812616788,
    "tpot_min_ms": 5.977015622192994,
    "tpot_p25_ms": 5.980199999612523,
    "tpot_p50_ms": 5.980415626254398,
    "tpot_p75_ms": 5.995331248414004,
    "tpot_p90_ms": 5.997075000777841,
    "tpot_p95_ms": 5.997075000777841,
    "tpot_p99_ms": 5.997075000777841,
    "tpot_samples": 6,
    "tpot_stddev_ms": 0.008614255248027062,
    "ttft_max_ms": 12025.513299973682,
    "ttft_mean_ci95_high_ms": 11955.83093645508,
    "ttft_mean_ci95_low_ms": 11847.512530183198,
    "ttft_mean_ms": 11901.67173331914,
    "ttft_min_ms": 11843.691500020213,
    "ttft_p25_ms": 11850.786099908873,
    "ttft_p50_ms": 11872.226399951614,
    "ttft_p75_ms": 11926.888899994083,
    "ttft_p90_ms": 12025.513299973682,
    "ttft_p95_ms": 12025.513299973682,
    "ttft_p99_ms": 12025.513299973682,
    "ttft_samples": 6,
    "ttft_stddev_ms": 67.68490436673352
  },
  "model": "qwen38-huihui-ninfer-schemafix-mtp3-64k",
  "performance_eligible": true,
  "prompt_cache_mode": "unique",
  "request_canaries": false,
  "request_timings": [
    {
      "content_chunks": 9,
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": true,
        "observed_code_words": 32,
        "observed_extra_words": 0,
        "passed": true,
        "policy": "strict",
        "requested_words": 32
      },
      "decode_tok_s": 167.30757675903405,
      "e2e_ms": 12216.776299988851,
      "effective_prefill_tok_s": 5053.340016647144,
      "generation_ms": 191.2644999101758,
      "mean_inter_token_latency_ms": 5.977015622192994,
      "mean_time_per_output_token_ms": 5.977015622192994,
      "output_token_source": "usage",
      "output_tokens": 33,
      "planned_context_tokens": 64000,
      "prompt_tokens": 60769,
      "reasoning_chunks": 0,
      "request_canary": null,
      "request_index": 0,
      "time_to_first_output_ms": 12025.511800078675,
      "tpot_ms": 5.977015622192994,
      "ttft_ms": 12025.513299973682,
      "visible_content_truncated": false,
      "visible_generation_ms": 191.26300001516938
    },
    {
      "content_chunks": 10,
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": true,
        "observed_code_words": 32,
        "observed_extra_words": 0,
        "passed": true,
        "policy": "strict",
        "requested_words": 32
      },
      "decode_tok_s": 167.21245854718484,
      "e2e_ms": 12035.063700051978,
      "effective_prefill_tok_s": 5130.9176403276515,
      "generation_ms": 191.37330004014075,
      "mean_inter_token_latency_ms": 5.980415626254398,
      "mean_time_per_output_token_ms": 5.980415626254398,
      "output_token_source": "usage",
      "output_tokens": 33,
      "planned_context_tokens": 64000,
      "prompt_tokens": 60769,
      "reasoning_chunks": 0,
      "request_canary": null,
      "request_index": 1,
      "time_to_first_output_ms": 11843.690400011837,
      "tpot_ms": 5.980415626254398,
      "ttft_ms": 11843.691500020213,
      "visible_content_truncated": false,
      "visible_generation_ms": 191.3722000317648
    },
    {
      "content_chunks": 10,
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": true,
        "observed_code_words": 32,
        "observed_extra_words": 0,
        "passed": true,
        "policy": "strict",
        "requested_words": 32
      },
      "decode_tok_s": 167.2184876868321,
      "e2e_ms": 12042.151799891144,
      "effective_prefill_tok_s": 5128.014553406278,
      "generation_ms": 191.36639998760074,
      "mean_inter_token_latency_ms": 5.980199999612523,
      "mean_time_per_output_token_ms": 5.980199999612523,
      "output_token_source": "usage",
      "output_tokens": 33,
      "planned_context_tokens": 64000,
      "prompt_tokens": 60771,
      "reasoning_chunks": 0,
      "request_canary": null,
      "request_index": 2,
      "time_to_first_output_ms": 11850.785399903543,
      "tpot_ms": 5.980199999612523,
      "ttft_ms": 11850.786099908873,
      "visible_content_truncated": false,
      "visible_generation_ms": 191.3656999822706
    },
    {
      "content_chunks": 10,
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": true,
        "observed_code_words": 32,
        "observed_extra_words": 0,
        "passed": true,
        "policy": "strict",
        "requested_words": 32
      },
      "decode_tok_s": 167.1776904587466,
      "e2e_ms": 12063.638800056651,
      "effective_prefill_tok_s": 5118.922225364614,
      "generation_ms": 191.41310011036694,
      "mean_inter_token_latency_ms": 5.981659378448967,
      "mean_time_per_output_token_ms": 5.981659378448967,
      "output_token_source": "usage",
      "output_tokens": 33,
      "planned_context_tokens": 64000,
      "prompt_tokens": 60773,
      "reasoning_chunks": 0,
      "request_canary": null,
      "request_index": 3,
      "time_to_first_output_ms": 11872.225699946284,
      "tpot_ms": 5.981659378448967,
      "ttft_ms": 11872.226399951614,
      "visible_content_truncated": false,
      "visible_generation_ms": 191.4124001050368
    },
    {
      "content_chunks": 9,
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": true,
        "observed_code_words": 32,
        "observed_extra_words": 0,
        "passed": true,
        "policy": "strict",
        "requested_words": 32
      },
      "decode_tok_s": 166.79645520246083,
      "e2e_ms": 12082.774200011045,
      "effective_prefill_tok_s": 5111.041164177032,
      "generation_ms": 191.85059994924814,
      "mean_inter_token_latency_ms": 5.995331248414004,
      "mean_time_per_output_token_ms": 5.995331248414004,
      "output_token_source": "usage",
      "output_tokens": 33,
      "planned_context_tokens": 64000,
      "prompt_tokens": 60775,
      "reasoning_chunks": 0,
      "request_canary": null,
      "request_index": 4,
      "time_to_first_output_ms": 11890.923600061797,
      "tpot_ms": 5.995331248414004,
      "ttft_ms": 11890.924200066365,
      "visible_content_truncated": false,
      "visible_generation_ms": 191.84999994467944
    },
    {
      "content_chunks": 9,
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": true,
        "observed_code_words": 32,
        "observed_extra_words": 0,
        "passed": true,
        "policy": "strict",
        "requested_words": 32
      },
      "decode_tok_s": 166.74795627373288,
      "e2e_ms": 12118.794700014405,
      "effective_prefill_tok_s": 5095.713020138994,
      "generation_ms": 191.9064000248909,
      "mean_inter_token_latency_ms": 5.997075000777841,
      "mean_time_per_output_token_ms": 5.997075000777841,
      "output_token_source": "usage",
      "output_tokens": 33,
      "planned_context_tokens": 64000,
      "prompt_tokens": 60776,
      "reasoning_chunks": 0,
      "request_canary": null,
      "request_index": 5,
      "time_to_first_output_ms": 11926.888299989514,
      "tpot_ms": 5.997075000777841,
      "ttft_ms": 11926.888899994083,
      "visible_content_truncated": false,
      "visible_generation_ms": 191.9058000203222
    }
  ],
  "requests": 6,
  "response_words": 32,
  "responses_received": 6,
  "run_id": "benchmark-20260919T232921Z",
  "sampling": {
    "temperature": {
      "effective_request": 0.0,
      "requested": null,
      "sent": true
    },
    "top_p": {
      "effective_request": null,
      "requested": null,
      "sent": false
    }
  },
  "schema": "anvil-serving.benchmark/v1",
  "serve_flags": {
    "no_thinking": false,
    "reasoning_effort": null,
    "shared_prefix_burst": false,
    "thinking_mode": "unsupported"
  },
  "shared_prefix_tokens": 0,
  "source_recipe": null,
  "started_at": "2026-09-19T23:29:21Z",
  "started_at_unix_ns": 1789860561980641600,
  "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": 72567.48349999543
}