{
  "base_url": "redacted://routed-capability",
  "clock_domain_id": null,
  "completed": 20,
  "concurrency": 4,
  "configuration_fingerprint": null,
  "context_distribution": {
    "32768": 20
  },
  "context_seed": 941124,
  "context_tokens": 32768,
  "controlled_output_policy": "observe",
  "engine": "vllm-exl3-v84",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-10-17T00:19:29Z",
  "finished_at_unix_ns": 1792196369555571607,
  "gpu": "dual-rtx-pro-6000-maxq",
  "manifest": null,
  "max_context_tokens": null,
  "max_tokens": 16384,
  "measurement_protocol": "capacity-v4-reasoning",
  "metric_population": {
    "aggregate_throughput_eligible": true,
    "attempted_requests": 20,
    "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": 20,
      "e2e": 20,
      "effective_prefill_tok_s": 20,
      "tpot": 20,
      "ttft": 20
    },
    "performance_eligible_requests": 20,
    "successful_requests": 20
  },
  "metrics": {
    "completion_tokens_max": 469.0,
    "completion_tokens_mean": 238.7,
    "completion_tokens_mean_ci95_high": 277.258026931394,
    "completion_tokens_mean_ci95_low": 200.14197306860598,
    "completion_tokens_min": 110.0,
    "completion_tokens_p25": 143.0,
    "completion_tokens_p50": 238.0,
    "completion_tokens_p75": 278.0,
    "completion_tokens_p90": 322.0,
    "completion_tokens_p95": 327.0,
    "completion_tokens_p99": 469.0,
    "completion_tokens_samples": 20,
    "completion_tokens_stddev": 87.97792785394348,
    "content_chunks": 1134,
    "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": 53.491881995573195,
    "decode_tok_s_mean": 22.146561125127327,
    "decode_tok_s_mean_ci95_high": 27.18890762068267,
    "decode_tok_s_mean_ci95_low": 17.104214629571985,
    "decode_tok_s_min": 12.511338175141708,
    "decode_tok_s_p25": 14.61788647415276,
    "decode_tok_s_p50": 18.613996228685323,
    "decode_tok_s_p75": 24.481811147760194,
    "decode_tok_s_p90": 27.315967281410032,
    "decode_tok_s_p95": 52.86137315470994,
    "decode_tok_s_p99": 53.491881995573195,
    "decode_tok_s_samples": 20,
    "decode_tok_s_stddev": 11.505132173642435,
    "e2e_max_ms": 39367.244750028476,
    "e2e_mean_ci95_high_ms": 28152.878829161564,
    "e2e_mean_ci95_low_ms": 21885.44146138018,
    "e2e_mean_ms": 25019.16014527087,
    "e2e_min_ms": 12225.111917010508,
    "e2e_p25_ms": 19602.86486602854,
    "e2e_p50_ms": 24696.240538032725,
    "e2e_p75_ms": 28665.938585996628,
    "e2e_p90_ms": 35028.085903963074,
    "e2e_p95_ms": 35436.555975000374,
    "e2e_p99_ms": 39367.244750028476,
    "e2e_samples": 20,
    "e2e_stddev_ms": 7150.2122444293545,
    "effective_prefill_tok_s_max": 4177.339532153774,
    "effective_prefill_tok_s_mean": 2184.2311962143804,
    "effective_prefill_tok_s_mean_ci95_high": 2447.4398065789,
    "effective_prefill_tok_s_mean_ci95_low": 1921.0225858498604,
    "effective_prefill_tok_s_min": 1142.7909178010245,
    "effective_prefill_tok_s_p25": 1889.4168935767602,
    "effective_prefill_tok_s_p50": 2184.2433489605382,
    "effective_prefill_tok_s_p75": 2383.6764360197994,
    "effective_prefill_tok_s_p90": 2487.819978663179,
    "effective_prefill_tok_s_p95": 2530.5907808444217,
    "effective_prefill_tok_s_p99": 4177.339532153774,
    "effective_prefill_tok_s_samples": 20,
    "effective_prefill_tok_s_stddev": 600.563617386043,
    "generation_max_ms": 24937.38044903148,
    "generation_mean_ci95_high_ms": 15361.067848627195,
    "generation_mean_ci95_low_ms": 9979.157828390393,
    "generation_mean_ms": 12670.112838508794,
    "generation_min_ms": 2061.997135053389,
    "generation_p25_ms": 7375.455660978332,
    "generation_p50_ms": 13134.593727998435,
    "generation_p75_ms": 15102.337226038799,
    "generation_p90_ms": 18747.096035978757,
    "generation_p95_ms": 24283.52228400763,
    "generation_p99_ms": 24937.38044903148,
    "generation_samples": 20,
    "generation_stddev_ms": 6139.957476549368,
    "mean_inter_token_latency_ms_max": 79.92750143920345,
    "mean_inter_token_latency_ms_mean": 52.93209490303396,
    "mean_inter_token_latency_ms_mean_ci95_high": 60.69015974643824,
    "mean_inter_token_latency_ms_mean_ci95_low": 45.174030059629686,
    "mean_inter_token_latency_ms_min": 18.694425447262383,
    "mean_inter_token_latency_ms_p25": 39.758412999697356,
    "mean_inter_token_latency_ms_p50": 53.259447379212574,
    "mean_inter_token_latency_ms_p75": 65.32692703441148,
    "mean_inter_token_latency_ms_p90": 76.51875933052554,
    "mean_inter_token_latency_ms_p95": 78.15845959826412,
    "mean_inter_token_latency_ms_p99": 79.92750143920345,
    "mean_inter_token_latency_ms_samples": 20,
    "mean_inter_token_latency_ms_stddev": 17.70159220786043,
    "mean_time_per_output_token_max_ms": 79.92750143920345,
    "mean_time_per_output_token_mean_ci95_high_ms": 60.69015974643824,
    "mean_time_per_output_token_mean_ci95_low_ms": 45.174030059629686,
    "mean_time_per_output_token_mean_ms": 52.93209490303396,
    "mean_time_per_output_token_min_ms": 18.694425447262383,
    "mean_time_per_output_token_p25_ms": 39.758412999697356,
    "mean_time_per_output_token_p50_ms": 53.259447379212574,
    "mean_time_per_output_token_p75_ms": 65.32692703441148,
    "mean_time_per_output_token_p90_ms": 76.51875933052554,
    "mean_time_per_output_token_p95_ms": 78.15845959826412,
    "mean_time_per_output_token_p99_ms": 79.92750143920345,
    "mean_time_per_output_token_samples": 20,
    "mean_time_per_output_token_stddev_ms": 17.70159220786043,
    "output_token_sources": {
      "usage": 20
    },
    "output_tokens": 4774,
    "prefix_cache_hit_avg": null,
    "prompt_token_samples": 20,
    "prompt_tokens": 505676,
    "prompt_tokens_max": 25287.0,
    "prompt_tokens_mean": 25283.8,
    "prompt_tokens_mean_ci95_high": 25284.77067139,
    "prompt_tokens_mean_ci95_low": 25282.829328609998,
    "prompt_tokens_min": 25280.0,
    "prompt_tokens_p25": 25281.0,
    "prompt_tokens_p50": 25284.0,
    "prompt_tokens_p75": 25285.0,
    "prompt_tokens_p90": 25286.0,
    "prompt_tokens_p95": 25286.0,
    "prompt_tokens_p99": 25287.0,
    "prompt_tokens_samples": 20,
    "prompt_tokens_stddev": 2.2147828692435603,
    "reasoning_chunks": 3524,
    "request_canary_passed": 0,
    "request_canary_samples": 0,
    "throughput_performance_eligible": true,
    "throughput_tok_s": 36.89307182304518,
    "time_to_first_output_max_ms": 22122.156910947524,
    "time_to_first_output_mean_ci95_high_ms": 13806.147428758615,
    "time_to_first_output_mean_ci95_low_ms": 10891.947184765539,
    "time_to_first_output_mean_ms": 12349.047306762077,
    "time_to_first_output_min_ms": 6051.93803505972,
    "time_to_first_output_p25_ms": 10499.175525968894,
    "time_to_first_output_p50_ms": 11304.872726090252,
    "time_to_first_output_p75_ms": 13223.383108037524,
    "time_to_first_output_p90_ms": 15531.344857998192,
    "time_to_first_output_p95_ms": 17352.744100033306,
    "time_to_first_output_p99_ms": 22122.156910947524,
    "time_to_first_output_samples": 20,
    "time_to_first_output_stddev_ms": 3324.668288579078,
    "tpot_max_ms": 79.92750143920345,
    "tpot_mean_ci95_high_ms": 60.69015974643824,
    "tpot_mean_ci95_low_ms": 45.174030059629686,
    "tpot_mean_ms": 52.93209490303396,
    "tpot_min_ms": 18.694425447262383,
    "tpot_p25_ms": 39.758412999697356,
    "tpot_p50_ms": 53.259447379212574,
    "tpot_p75_ms": 65.32692703441148,
    "tpot_p90_ms": 76.51875933052554,
    "tpot_p95_ms": 78.15845959826412,
    "tpot_p99_ms": 79.92750143920345,
    "tpot_samples": 20,
    "tpot_stddev_ms": 17.70159220786043,
    "ttft_max_ms": 36313.15924599767,
    "ttft_mean_ci95_high_ms": 24869.797760061847,
    "ttft_mean_ci95_low_ms": 19238.919391960444,
    "ttft_mean_ms": 22054.358576011145,
    "ttft_min_ms": 11317.453522933647,
    "ttft_p25_ms": 17102.394278044812,
    "ttft_p50_ms": 19949.374695075676,
    "ttft_p75_ms": 24994.390050997026,
    "ttft_p90_ms": 27779.985839035362,
    "ttft_p95_ms": 34061.8464150466,
    "ttft_p99_ms": 36313.15924599767,
    "ttft_samples": 20,
    "ttft_stddev_ms": 6423.993267401951
  },
  "model": "llm.primary",
  "performance_eligible": true,
  "prompt_cache_mode": "unique",
  "request_canaries": false,
  "request_timings": [
    {
      "content_chunks": 44,
      "controlled_output": null,
      "decode_tok_s": 19.000926469383998,
      "e2e_ms": 16942.471851943992,
      "effective_prefill_tok_s": 2277.3438230441793,
      "generation_ms": 5841.820406960323,
      "mean_inter_token_latency_ms": 52.62901267531823,
      "mean_time_per_output_token_ms": 52.62901267531823,
      "output_token_source": "usage",
      "output_tokens": 112,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25280,
      "reasoning_chunks": 66,
      "request_canary": null,
      "request_index": 0,
      "time_to_first_output_ms": 11100.651444983669,
      "tpot_ms": 52.62901267531823,
      "ttft_ms": 16099.474417977035,
      "visible_content_truncated": false,
      "visible_generation_ms": 842.997433966957
    },
    {
      "content_chunks": 75,
      "controlled_output": null,
      "decode_tok_s": 18.613996228685323,
      "e2e_ms": 39367.244750028476,
      "effective_prefill_tok_s": 1142.7909178010245,
      "generation_ms": 17245.087839080952,
      "mean_inter_token_latency_ms": 53.723015075018544,
      "mean_time_per_output_token_ms": 53.723015075018544,
      "output_token_source": "usage",
      "output_tokens": 322,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25281,
      "reasoning_chunks": 244,
      "request_canary": null,
      "request_index": 1,
      "time_to_first_output_ms": 22122.156910947524,
      "tpot_ms": 53.723015075018544,
      "ttft_ms": 36313.15924599767,
      "visible_content_truncated": false,
      "visible_generation_ms": 3054.085504030809
    },
    {
      "content_chunks": 60,
      "controlled_output": null,
      "decode_tok_s": 13.424741114047258,
      "e2e_ms": 30335.46031906735,
      "effective_prefill_tok_s": 4177.339532153774,
      "generation_ms": 24283.52228400763,
      "mean_inter_token_latency_ms": 74.48933215953261,
      "mean_time_per_output_token_ms": 74.48933215953261,
      "output_token_source": "usage",
      "output_tokens": 327,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25281,
      "reasoning_chunks": 255,
      "request_canary": null,
      "request_index": 2,
      "time_to_first_output_ms": 6051.93803505972,
      "tpot_ms": 74.48933215953261,
      "ttft_ms": 24404.102302039973,
      "visible_content_truncated": false,
      "visible_generation_ms": 5931.358017027378
    },
    {
      "content_chunks": 50,
      "controlled_output": null,
      "decode_tok_s": 27.315967281410032,
      "e2e_ms": 24989.118539029732,
      "effective_prefill_tok_s": 1677.7074604808877,
      "generation_ms": 9920.937348040752,
      "mean_inter_token_latency_ms": 36.60862490051938,
      "mean_time_per_output_token_ms": 36.60862490051938,
      "output_token_source": "usage",
      "output_tokens": 272,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25280,
      "reasoning_chunks": 212,
      "request_canary": null,
      "request_index": 3,
      "time_to_first_output_ms": 15068.18119098898,
      "tpot_ms": 36.60862490051938,
      "ttft_ms": 24025.39580897428,
      "visible_content_truncated": false,
      "visible_generation_ms": 963.7227300554514
    },
    {
      "content_chunks": 43,
      "controlled_output": null,
      "decode_tok_s": 18.57512353098837,
      "e2e_ms": 20598.838769015856,
      "effective_prefill_tok_s": 1911.8405474188776,
      "generation_ms": 7375.455660978332,
      "mean_inter_token_latency_ms": 53.83544278086374,
      "mean_time_per_output_token_ms": 53.83544278086374,
      "output_token_source": "usage",
      "output_tokens": 138,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25281,
      "reasoning_chunks": 91,
      "request_canary": null,
      "request_index": 4,
      "time_to_first_output_ms": 13223.383108037524,
      "tpot_ms": 53.83544278086374,
      "ttft_ms": 19769.665942061692,
      "visible_content_truncated": false,
      "visible_generation_ms": 829.1728269541636
    },
    {
      "content_chunks": 77,
      "controlled_output": null,
      "decode_tok_s": 12.511338175141708,
      "e2e_ms": 35436.555975000374,
      "effective_prefill_tok_s": 2407.998603077636,
      "generation_ms": 24937.38044903148,
      "mean_inter_token_latency_ms": 79.92750143920345,
      "mean_time_per_output_token_ms": 79.92750143920345,
      "output_token_source": "usage",
      "output_tokens": 313,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25282,
      "reasoning_chunks": 229,
      "request_canary": null,
      "request_index": 5,
      "time_to_first_output_ms": 10499.175525968894,
      "tpot_ms": 79.92750143920345,
      "ttft_ms": 24426.710749976337,
      "visible_content_truncated": false,
      "visible_generation_ms": 11009.845225024037
    },
    {
      "content_chunks": 62,
      "controlled_output": null,
      "decode_tok_s": 12.794520326270732,
      "e2e_ms": 23913.01316092722,
      "effective_prefill_tok_s": 1972.9975048053022,
      "generation_ms": 11098.501262953505,
      "mean_inter_token_latency_ms": 78.15845959826412,
      "mean_time_per_output_token_ms": 78.15845959826412,
      "output_token_source": "usage",
      "output_tokens": 143,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25283,
      "reasoning_chunks": 78,
      "request_canary": null,
      "request_index": 6,
      "time_to_first_output_ms": 12814.511897973716,
      "tpot_ms": 78.15845959826412,
      "ttft_ms": 18593.702723970637,
      "visible_content_truncated": false,
      "visible_generation_ms": 5319.310436956584
    },
    {
      "content_chunks": 47,
      "controlled_output": null,
      "decode_tok_s": 52.86137315470994,
      "e2e_ms": 12225.111917010508,
      "effective_prefill_tok_s": 2487.819978663179,
      "generation_ms": 2061.997135053389,
      "mean_inter_token_latency_ms": 18.917404908746686,
      "mean_time_per_output_token_ms": 18.917404908746686,
      "output_token_source": "usage",
      "output_tokens": 110,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25284,
      "reasoning_chunks": 60,
      "request_canary": null,
      "request_index": 7,
      "time_to_first_output_ms": 10163.11478195712,
      "tpot_ms": 18.917404908746686,
      "ttft_ms": 11317.453522933647,
      "visible_content_truncated": false,
      "visible_generation_ms": 907.6583940768614
    },
    {
      "content_chunks": 47,
      "controlled_output": null,
      "decode_tok_s": 14.61788647415276,
      "e2e_ms": 28665.938585996628,
      "effective_prefill_tok_s": 1627.9337192734777,
      "generation_ms": 13134.593727998435,
      "mean_inter_token_latency_ms": 68.40934233332518,
      "mean_time_per_output_token_ms": 68.40934233332518,
      "output_token_source": "usage",
      "output_tokens": 193,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25284,
      "reasoning_chunks": 140,
      "request_canary": null,
      "request_index": 8,
      "time_to_first_output_ms": 15531.344857998192,
      "tpot_ms": 68.40934233332518,
      "ttft_ms": 27779.985839035362,
      "visible_content_truncated": false,
      "visible_generation_ms": 885.9527469612658
    },
    {
      "content_chunks": 65,
      "controlled_output": null,
      "decode_tok_s": 18.292247883599753,
      "e2e_ms": 24696.240538032725,
      "effective_prefill_tok_s": 2476.582626157068,
      "generation_ms": 14487.011202028953,
      "mean_inter_token_latency_ms": 54.667966800109255,
      "mean_time_per_output_token_ms": 54.667966800109255,
      "output_token_source": "usage",
      "output_tokens": 266,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25284,
      "reasoning_chunks": 198,
      "request_canary": null,
      "request_index": 9,
      "time_to_first_output_ms": 10209.229336003773,
      "tpot_ms": 54.667966800109255,
      "ttft_ms": 23456.88875997439,
      "visible_content_truncated": false,
      "visible_generation_ms": 1239.3517780583352
    },
    {
      "content_chunks": 56,
      "controlled_output": null,
      "decode_tok_s": 18.158502885057352,
      "e2e_ms": 25788.445610087365,
      "effective_prefill_tok_s": 2236.646144776609,
      "generation_ms": 14483.572883997113,
      "mean_inter_token_latency_ms": 55.0706193307875,
      "mean_time_per_output_token_ms": 55.0706193307875,
      "output_token_source": "usage",
      "output_tokens": 264,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25285,
      "reasoning_chunks": 201,
      "request_canary": null,
      "request_index": 10,
      "time_to_first_output_ms": 11304.872726090252,
      "tpot_ms": 55.0706193307875,
      "ttft_ms": 19949.374695075676,
      "visible_content_truncated": false,
      "visible_generation_ms": 5839.070915011689
    },
    {
      "content_chunks": 51,
      "controlled_output": null,
      "decode_tok_s": 19.533400399202694,
      "e2e_ms": 27853.62353909295,
      "effective_prefill_tok_s": 1983.0156251854696,
      "generation_ms": 15102.337226038799,
      "mean_inter_token_latency_ms": 51.19436347809762,
      "mean_time_per_output_token_ms": 51.19436347809762,
      "output_token_source": "usage",
      "output_tokens": 296,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25286,
      "reasoning_chunks": 240,
      "request_canary": null,
      "request_index": 11,
      "time_to_first_output_ms": 12751.286313054152,
      "tpot_ms": 51.19436347809762,
      "ttft_ms": 26878.377125016414,
      "visible_content_truncated": false,
      "visible_generation_ms": 975.2464140765369
    },
    {
      "content_chunks": 51,
      "controlled_output": null,
      "decode_tok_s": 18.776011566171544,
      "e2e_ms": 26329.41585208755,
      "effective_prefill_tok_s": 2184.2433489605382,
      "generation_ms": 14752.866924041882,
      "mean_inter_token_latency_ms": 53.259447379212574,
      "mean_time_per_output_token_ms": 53.259447379212574,
      "output_token_source": "usage",
      "output_tokens": 278,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25286,
      "reasoning_chunks": 221,
      "request_canary": null,
      "request_index": 12,
      "time_to_first_output_ms": 11576.548928045668,
      "tpot_ms": 53.259447379212574,
      "ttft_ms": 24994.390050997026,
      "visible_content_truncated": false,
      "visible_generation_ms": 1335.0258010905236
    },
    {
      "content_chunks": 54,
      "controlled_output": null,
      "decode_tok_s": 25.15190935834416,
      "e2e_ms": 19800.40037201252,
      "effective_prefill_tok_s": 2355.273802323086,
      "generation_ms": 9064.918163930997,
      "mean_inter_token_latency_ms": 39.758412999697356,
      "mean_time_per_output_token_ms": 39.758412999697356,
      "output_token_source": "usage",
      "output_tokens": 229,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25285,
      "reasoning_chunks": 166,
      "request_canary": null,
      "request_index": 13,
      "time_to_first_output_ms": 10735.482208081521,
      "tpot_ms": 39.758412999697356,
      "ttft_ms": 15618.51087899413,
      "visible_content_truncated": false,
      "visible_generation_ms": 4181.889493018389
    },
    {
      "content_chunks": 61,
      "controlled_output": null,
      "decode_tok_s": 13.068690720408364,
      "e2e_ms": 32129.53168898821,
      "effective_prefill_tok_s": 1889.4168935767602,
      "generation_ms": 18747.096035978757,
      "mean_inter_token_latency_ms": 76.51875933052554,
      "mean_time_per_output_token_ms": 76.51875933052554,
      "output_token_source": "usage",
      "output_tokens": 246,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25285,
      "reasoning_chunks": 174,
      "request_canary": null,
      "request_index": 14,
      "time_to_first_output_ms": 13382.435653009452,
      "tpot_ms": 76.51875933052554,
      "ttft_ms": 26192.518273950554,
      "visible_content_truncated": false,
      "visible_generation_ms": 5937.013415037654
    },
    {
      "content_chunks": 60,
      "controlled_output": null,
      "decode_tok_s": 20.475705995756126,
      "e2e_ms": 18257.70576298237,
      "effective_prefill_tok_s": 2158.645297420804,
      "generation_ms": 6544.340890017338,
      "mean_inter_token_latency_ms": 48.83836485087566,
      "mean_time_per_output_token_ms": 48.83836485087566,
      "output_token_source": "usage",
      "output_tokens": 135,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25285,
      "reasoning_chunks": 73,
      "request_canary": null,
      "request_index": 15,
      "time_to_first_output_ms": 11713.36487296503,
      "tpot_ms": 48.83836485087566,
      "ttft_ms": 17102.394278044812,
      "visible_content_truncated": false,
      "visible_generation_ms": 1155.311484937556
    },
    {
      "content_chunks": 51,
      "controlled_output": null,
      "decode_tok_s": 15.307623447116102,
      "e2e_ms": 23384.153180057183,
      "effective_prefill_tok_s": 2530.5907808444217,
      "generation_ms": 13392.020042054355,
      "mean_inter_token_latency_ms": 65.32692703441148,
      "mean_time_per_output_token_ms": 65.32692703441148,
      "output_token_source": "usage",
      "output_tokens": 206,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25286,
      "reasoning_chunks": 150,
      "request_canary": null,
      "request_index": 16,
      "time_to_first_output_ms": 9992.133138002828,
      "tpot_ms": 65.32692703441148,
      "ttft_ms": 17642.867294023745,
      "visible_content_truncated": false,
      "visible_generation_ms": 5741.285886033438
    },
    {
      "content_chunks": 65,
      "controlled_output": null,
      "decode_tok_s": 26.477564348766897,
      "e2e_ms": 35028.085903963074,
      "effective_prefill_tok_s": 1457.1182433302563,
      "generation_ms": 17675.34180392977,
      "mean_inter_token_latency_ms": 37.7678243673713,
      "mean_time_per_output_token_ms": 37.7678243673713,
      "output_token_source": "usage",
      "output_tokens": 469,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25285,
      "reasoning_chunks": 396,
      "request_canary": null,
      "request_index": 17,
      "time_to_first_output_ms": 17352.744100033306,
      "tpot_ms": 37.7678243673713,
      "ttft_ms": 34061.8464150466,
      "visible_content_truncated": false,
      "visible_generation_ms": 966.2394889164716
    },
    {
      "content_chunks": 47,
      "controlled_output": null,
      "decode_tok_s": 24.481811147760194,
      "e2e_ms": 19602.86486602854,
      "effective_prefill_tok_s": 2345.6426389744584,
      "generation_ms": 8822.876653051935,
      "mean_inter_token_latency_ms": 40.846651171536735,
      "mean_time_per_output_token_ms": 40.846651171536735,
      "output_token_source": "usage",
      "output_tokens": 217,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25286,
      "reasoning_chunks": 164,
      "request_canary": null,
      "request_index": 18,
      "time_to_first_output_ms": 10779.988212976605,
      "tpot_ms": 40.846651171536735,
      "ttft_ms": 18695.45271503739,
      "visible_content_truncated": false,
      "visible_generation_ms": 907.4121509911492
    },
    {
      "content_chunks": 68,
      "controlled_output": null,
      "decode_tok_s": 53.491881995573195,
      "e2e_ms": 15038.981724064797,
      "effective_prefill_tok_s": 2383.6764360197994,
      "generation_ms": 4430.578831001185,
      "mean_inter_token_latency_ms": 18.694425447262383,
      "mean_time_per_output_token_ms": 18.694425447262383,
      "output_token_source": "usage",
      "output_tokens": 238,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25287,
      "reasoning_chunks": 166,
      "request_canary": null,
      "request_index": 19,
      "time_to_first_output_ms": 10608.402893063612,
      "tpot_ms": 18.694425447262383,
      "ttft_ms": 13764.900481095538,
      "visible_content_truncated": false,
      "visible_generation_ms": 1274.0812429692596
    }
  ],
  "requests": 20,
  "response_words": 0,
  "responses_received": 20,
  "run_id": "benchmark-20261017T001720Z",
  "sampling": {
    "temperature": {
      "effective_request": 0.6,
      "requested": 0.6,
      "sent": true
    },
    "top_p": {
      "effective_request": 0.95,
      "requested": 0.95,
      "sent": true
    }
  },
  "schema": "anvil-serving.benchmark/v1",
  "serve_flags": {
    "chat_template_kwargs": null,
    "force_nonempty_content": false,
    "no_thinking": false,
    "reasoning_effort": "max",
    "shared_prefix_burst": false,
    "thinking_mode": "default"
  },
  "shared_prefix_tokens": 0,
  "source_recipe": null,
  "started_at": "2026-10-17T00:17:20Z",
  "started_at_unix_ns": 1792196240154577460,
  "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"
  },
  "unique_prefix_namespace": "capacity-v1:seed-request-index",
  "wall_clock_ms": 129400.98951093387
}