{
  "base_url": "redacted://routed-capability",
  "clock_domain_id": null,
  "completed": 20,
  "concurrency": 4,
  "configuration_fingerprint": null,
  "context_distribution": {
    "8192": 20
  },
  "context_seed": 941214,
  "context_tokens": 8192,
  "controlled_output_policy": "observe",
  "engine": "vllm-exl3-v84",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-10-17T00:35:47Z",
  "finished_at_unix_ns": 1792197347326407234,
  "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": 348.0,
    "completion_tokens_mean": 222.3,
    "completion_tokens_mean_ci95_high": 251.79581509656117,
    "completion_tokens_mean_ci95_low": 192.80418490343885,
    "completion_tokens_min": 124.0,
    "completion_tokens_p25": 167.0,
    "completion_tokens_p50": 215.0,
    "completion_tokens_p75": 259.0,
    "completion_tokens_p90": 318.0,
    "completion_tokens_p95": 322.0,
    "completion_tokens_p99": 348.0,
    "completion_tokens_samples": 20,
    "completion_tokens_stddev": 67.30066082415848,
    "content_chunks": 1005,
    "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": 65.35038919443183,
    "decode_tok_s_mean": 34.44629604754468,
    "decode_tok_s_mean_ci95_high": 38.68634422500739,
    "decode_tok_s_mean_ci95_low": 30.20624787008197,
    "decode_tok_s_min": 20.270388913397028,
    "decode_tok_s_p25": 28.207785114282753,
    "decode_tok_s_p50": 32.216135764627595,
    "decode_tok_s_p75": 36.417015712914335,
    "decode_tok_s_p90": 40.488051031777836,
    "decode_tok_s_p95": 49.80376006282441,
    "decode_tok_s_p99": 65.35038919443183,
    "decode_tok_s_samples": 20,
    "decode_tok_s_stddev": 9.674526482327252,
    "e2e_max_ms": 12226.838488015346,
    "e2e_mean_ci95_high_ms": 9775.120449682268,
    "e2e_mean_ci95_low_ms": 7854.450727918614,
    "e2e_mean_ms": 8814.785588800441,
    "e2e_min_ms": 3384.729153010994,
    "e2e_p25_ms": 7232.594506000169,
    "e2e_p50_ms": 9114.818708971143,
    "e2e_p75_ms": 10061.502858996391,
    "e2e_p90_ms": 11353.415417019278,
    "e2e_p95_ms": 11872.091660043225,
    "e2e_p99_ms": 12226.838488015346,
    "e2e_samples": 20,
    "e2e_stddev_ms": 2191.1979898924187,
    "effective_prefill_tok_s_max": 4563.143038700801,
    "effective_prefill_tok_s_mean": 3601.8694766092362,
    "effective_prefill_tok_s_mean_ci95_high": 4135.29451241126,
    "effective_prefill_tok_s_mean_ci95_low": 3068.4444408072122,
    "effective_prefill_tok_s_min": 1234.3684677039062,
    "effective_prefill_tok_s_p25": 2147.225076402358,
    "effective_prefill_tok_s_p50": 4453.242172638876,
    "effective_prefill_tok_s_p75": 4517.651054855912,
    "effective_prefill_tok_s_p90": 4531.926308601494,
    "effective_prefill_tok_s_p95": 4548.422857617778,
    "effective_prefill_tok_s_p99": 4563.143038700801,
    "effective_prefill_tok_s_samples": 20,
    "effective_prefill_tok_s_stddev": 1217.116980564882,
    "generation_max_ms": 10557.271540979855,
    "generation_mean_ci95_high_ms": 7845.423924610325,
    "generation_mean_ci95_low_ms": 5757.878479569378,
    "generation_mean_ms": 6801.651202089852,
    "generation_min_ms": 2019.8808549903333,
    "generation_p25_ms": 4618.125212029554,
    "generation_p50_ms": 6771.180339972489,
    "generation_p75_ms": 8177.08098993171,
    "generation_p90_ms": 10010.778111987747,
    "generation_p95_ms": 10342.157586943358,
    "generation_p99_ms": 10557.271540979855,
    "generation_samples": 20,
    "generation_stddev_ms": 2381.578327158984,
    "mean_inter_token_latency_ms_max": 49.33304458401801,
    "mean_inter_token_latency_ms_mean": 30.83537928699161,
    "mean_inter_token_latency_ms_mean_ci95_high": 34.05577455210734,
    "mean_inter_token_latency_ms_mean_ci95_low": 27.614984021875888,
    "mean_inter_token_latency_ms_min": 15.302127689320708,
    "mean_inter_token_latency_ms_p25": 27.211637869947978,
    "mean_inter_token_latency_ms_p50": 29.34199461069341,
    "mean_inter_token_latency_ms_p75": 34.738468406608376,
    "mean_inter_token_latency_ms_p90": 36.88312872340355,
    "mean_inter_token_latency_ms_p95": 41.07689564235264,
    "mean_inter_token_latency_ms_p99": 49.33304458401801,
    "mean_inter_token_latency_ms_samples": 20,
    "mean_inter_token_latency_ms_stddev": 7.347982374711444,
    "mean_time_per_output_token_max_ms": 49.33304458401801,
    "mean_time_per_output_token_mean_ci95_high_ms": 34.05577455210734,
    "mean_time_per_output_token_mean_ci95_low_ms": 27.614984021875888,
    "mean_time_per_output_token_mean_ms": 30.83537928699161,
    "mean_time_per_output_token_min_ms": 15.302127689320708,
    "mean_time_per_output_token_p25_ms": 27.211637869947978,
    "mean_time_per_output_token_p50_ms": 29.34199461069341,
    "mean_time_per_output_token_p75_ms": 34.738468406608376,
    "mean_time_per_output_token_p90_ms": 36.88312872340355,
    "mean_time_per_output_token_p95_ms": 41.07689564235264,
    "mean_time_per_output_token_p99_ms": 49.33304458401801,
    "mean_time_per_output_token_samples": 20,
    "mean_time_per_output_token_stddev_ms": 7.347982374711444,
    "output_token_sources": {
      "usage": 20
    },
    "output_tokens": 4446,
    "prefix_cache_hit_avg": null,
    "prompt_token_samples": 20,
    "prompt_tokens": 121453,
    "prompt_tokens_max": 6079.0,
    "prompt_tokens_mean": 6072.65,
    "prompt_tokens_mean_ci95_high": 6074.500016842028,
    "prompt_tokens_mean_ci95_low": 6070.799983157971,
    "prompt_tokens_min": 6066.0,
    "prompt_tokens_p25": 6069.0,
    "prompt_tokens_p50": 6072.0,
    "prompt_tokens_p75": 6076.0,
    "prompt_tokens_p90": 6078.0,
    "prompt_tokens_p95": 6078.0,
    "prompt_tokens_p99": 6079.0,
    "prompt_tokens_samples": 20,
    "prompt_tokens_stddev": 4.221187161525958,
    "reasoning_chunks": 3297,
    "request_canary_passed": 0,
    "request_canary_samples": 0,
    "throughput_performance_eligible": true,
    "throughput_tok_s": 96.70008555568877,
    "time_to_first_output_max_ms": 4915.874116006307,
    "time_to_first_output_mean_ci95_high_ms": 2480.126459579342,
    "time_to_first_output_mean_ci95_low_ms": 1546.1423138418374,
    "time_to_first_output_mean_ms": 2013.1343867105898,
    "time_to_first_output_min_ms": 1330.0043300259858,
    "time_to_first_output_p25_ms": 1344.5274810073897,
    "time_to_first_output_p50_ms": 1356.9266319973394,
    "time_to_first_output_p75_ms": 2383.0119589110836,
    "time_to_first_output_p90_ms": 3580.4555979557335,
    "time_to_first_output_p95_ms": 3982.183259096928,
    "time_to_first_output_p99_ms": 4915.874116006307,
    "time_to_first_output_samples": 20,
    "time_to_first_output_stddev_ms": 1065.5367549878217,
    "tpot_max_ms": 49.33304458401801,
    "tpot_mean_ci95_high_ms": 34.05577455210734,
    "tpot_mean_ci95_low_ms": 27.614984021875888,
    "tpot_mean_ms": 30.83537928699161,
    "tpot_min_ms": 15.302127689320708,
    "tpot_p25_ms": 27.211637869947978,
    "tpot_p50_ms": 29.34199461069341,
    "tpot_p75_ms": 34.738468406608376,
    "tpot_p90_ms": 36.88312872340355,
    "tpot_p95_ms": 41.07689564235264,
    "tpot_p99_ms": 49.33304458401801,
    "tpot_samples": 20,
    "tpot_stddev_ms": 7.347982374711444,
    "ttft_max_ms": 10362.137335003354,
    "ttft_mean_ci95_high_ms": 8152.583534834979,
    "ttft_mean_ci95_low_ms": 6459.015783266092,
    "ttft_mean_ms": 7305.799659050535,
    "ttft_min_ms": 2709.3968259869143,
    "ttft_p25_ms": 5327.715635998175,
    "ttft_p50_ms": 7441.0400099586695,
    "ttft_p75_ms": 8731.24637606088,
    "ttft_p90_ms": 9106.375109986402,
    "ttft_p95_ms": 9769.881407963112,
    "ttft_p99_ms": 10362.137335003354,
    "ttft_samples": 20,
    "ttft_stddev_ms": 1932.108478066075
  },
  "model": "llm.primary",
  "performance_eligible": true,
  "prompt_cache_mode": "unique",
  "request_canaries": false,
  "request_timings": [
    {
      "content_chunks": 45,
      "controlled_output": null,
      "decode_tok_s": 28.207785114282753,
      "e2e_ms": 10753.363599069417,
      "effective_prefill_tok_s": 1523.5361120411778,
      "generation_ms": 6771.180339972489,
      "mean_inter_token_latency_ms": 35.45120596844235,
      "mean_time_per_output_token_ms": 35.45120596844235,
      "output_token_source": "usage",
      "output_tokens": 192,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6067,
      "reasoning_chunks": 144,
      "request_canary": null,
      "request_index": 0,
      "time_to_first_output_ms": 3982.183259096928,
      "tpot_ms": 35.45120596844235,
      "ttft_ms": 8731.24637606088,
      "visible_content_truncated": false,
      "visible_generation_ms": 2022.1172230085358
    },
    {
      "content_chunks": 44,
      "controlled_output": null,
      "decode_tok_s": 28.78653106680339,
      "e2e_ms": 7097.873136983253,
      "effective_prefill_tok_s": 2147.225076402358,
      "generation_ms": 4272.83161401283,
      "mean_inter_token_latency_ms": 34.738468406608376,
      "mean_time_per_output_token_ms": 34.738468406608376,
      "output_token_source": "usage",
      "output_tokens": 124,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6066,
      "reasoning_chunks": 75,
      "request_canary": null,
      "request_index": 1,
      "time_to_first_output_ms": 2825.041522970423,
      "tpot_ms": 34.738468406608376,
      "ttft_ms": 6268.555380986072,
      "visible_content_truncated": false,
      "visible_generation_ms": 829.3177559971809
    },
    {
      "content_chunks": 48,
      "controlled_output": null,
      "decode_tok_s": 20.270388913397028,
      "e2e_ms": 12226.838488015346,
      "effective_prefill_tok_s": 3633.876443692574,
      "generation_ms": 10557.271540979855,
      "mean_inter_token_latency_ms": 49.33304458401801,
      "mean_time_per_output_token_ms": 49.33304458401801,
      "output_token_source": "usage",
      "output_tokens": 215,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6067,
      "reasoning_chunks": 158,
      "request_canary": null,
      "request_index": 2,
      "time_to_first_output_ms": 1669.5669470354915,
      "tpot_ms": 49.33304458401801,
      "ttft_ms": 8989.545317017473,
      "visible_content_truncated": false,
      "visible_generation_ms": 3237.293170997873
    },
    {
      "content_chunks": 44,
      "controlled_output": null,
      "decode_tok_s": 38.9399716806677,
      "e2e_ms": 9178.845886024646,
      "effective_prefill_tok_s": 1234.3684677039062,
      "generation_ms": 4262.971770018339,
      "mean_inter_token_latency_ms": 25.68055283143578,
      "mean_time_per_output_token_ms": 25.68055283143578,
      "output_token_source": "usage",
      "output_tokens": 167,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6068,
      "reasoning_chunks": 118,
      "request_canary": null,
      "request_index": 3,
      "time_to_first_output_ms": 4915.874116006307,
      "tpot_ms": 25.68055283143578,
      "ttft_ms": 7171.434137038887,
      "visible_content_truncated": false,
      "visible_generation_ms": 2007.41174898576
    },
    {
      "content_chunks": 51,
      "controlled_output": null,
      "decode_tok_s": 32.216135764627595,
      "e2e_ms": 10061.502858996391,
      "effective_prefill_tok_s": 4531.926308601494,
      "generation_ms": 8722.337218001485,
      "mean_inter_token_latency_ms": 31.040345971535533,
      "mean_time_per_output_token_ms": 31.040345971535533,
      "output_token_source": "usage",
      "output_tokens": 282,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6069,
      "reasoning_chunks": 216,
      "request_canary": null,
      "request_index": 4,
      "time_to_first_output_ms": 1339.1656409949064,
      "tpot_ms": 31.040345971535533,
      "ttft_ms": 9106.375109986402,
      "visible_content_truncated": false,
      "visible_generation_ms": 955.1277490099892
    },
    {
      "content_chunks": 42,
      "controlled_output": null,
      "decode_tok_s": 30.695550198053766,
      "e2e_ms": 9507.085319957696,
      "effective_prefill_tok_s": 4563.143038700801,
      "generation_ms": 8177.08098993171,
      "mean_inter_token_latency_ms": 32.57801191207852,
      "mean_time_per_output_token_ms": 32.57801191207852,
      "output_token_source": "usage",
      "output_tokens": 252,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6069,
      "reasoning_chunks": 196,
      "request_canary": null,
      "request_index": 5,
      "time_to_first_output_ms": 1330.0043300259858,
      "tpot_ms": 32.57801191207852,
      "ttft_ms": 7542.383105959743,
      "visible_content_truncated": false,
      "visible_generation_ms": 1964.7022139979526
    },
    {
      "content_chunks": 46,
      "controlled_output": null,
      "decode_tok_s": 34.97232446690209,
      "e2e_ms": 8043.4263839852065,
      "effective_prefill_tok_s": 4488.985715056204,
      "generation_ms": 6691.005060914904,
      "mean_inter_token_latency_ms": 28.594038721858563,
      "mean_time_per_output_token_ms": 28.594038721858563,
      "output_token_source": "usage",
      "output_tokens": 235,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6071,
      "reasoning_chunks": 182,
      "request_canary": null,
      "request_index": 6,
      "time_to_first_output_ms": 1352.4213230703026,
      "tpot_ms": 28.594038721858563,
      "ttft_ms": 6009.3198310351,
      "visible_content_truncated": false,
      "visible_generation_ms": 2034.1065529501066
    },
    {
      "content_chunks": 53,
      "controlled_output": null,
      "decode_tok_s": 34.080846011591845,
      "e2e_ms": 10773.584029986523,
      "effective_prefill_tok_s": 4481.091785725784,
      "generation_ms": 9418.780270032585,
      "mean_inter_token_latency_ms": 29.34199461069341,
      "mean_time_per_output_token_ms": 29.34199461069341,
      "output_token_source": "usage",
      "output_tokens": 322,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6071,
      "reasoning_chunks": 262,
      "request_canary": null,
      "request_index": 7,
      "time_to_first_output_ms": 1354.8037599539384,
      "tpot_ms": 29.34199461069341,
      "ttft_ms": 9769.881407963112,
      "visible_content_truncated": false,
      "visible_generation_ms": 1003.7026220234111
    },
    {
      "content_chunks": 48,
      "controlled_output": null,
      "decode_tok_s": 27.377416132123155,
      "e2e_ms": 8347.828023019247,
      "effective_prefill_tok_s": 4548.422857617778,
      "generation_ms": 7013.079651980661,
      "mean_inter_token_latency_ms": 36.52645652073261,
      "mean_time_per_output_token_ms": 36.52645652073261,
      "output_token_source": "usage",
      "output_tokens": 193,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6071,
      "reasoning_chunks": 141,
      "request_canary": null,
      "request_index": 8,
      "time_to_first_output_ms": 1334.748371038586,
      "tpot_ms": 36.52645652073261,
      "ttft_ms": 7441.0400099586695,
      "visible_content_truncated": false,
      "visible_generation_ms": 906.7880130605772
    },
    {
      "content_chunks": 85,
      "controlled_output": null,
      "decode_tok_s": 30.651244417335345,
      "e2e_ms": 11872.091660043225,
      "effective_prefill_tok_s": 3968.798464431382,
      "generation_ms": 10342.157586943358,
      "mean_inter_token_latency_ms": 32.62510279792858,
      "mean_time_per_output_token_ms": 32.62510279792858,
      "output_token_source": "usage",
      "output_tokens": 318,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6072,
      "reasoning_chunks": 222,
      "request_canary": null,
      "request_index": 9,
      "time_to_first_output_ms": 1529.9340730998665,
      "tpot_ms": 32.62510279792858,
      "ttft_ms": 9092.046683072113,
      "visible_content_truncated": false,
      "visible_generation_ms": 2780.044976971112
    },
    {
      "content_chunks": 59,
      "controlled_output": null,
      "decode_tok_s": 36.060095046962005,
      "e2e_ms": 9537.735196994618,
      "effective_prefill_tok_s": 2548.035891005179,
      "generation_ms": 7154.723238083534,
      "mean_inter_token_latency_ms": 27.731485418928425,
      "mean_time_per_output_token_ms": 27.731485418928425,
      "output_token_source": "usage",
      "output_tokens": 259,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6072,
      "reasoning_chunks": 190,
      "request_canary": null,
      "request_index": 10,
      "time_to_first_output_ms": 2383.0119589110836,
      "tpot_ms": 27.731485418928425,
      "ttft_ms": 8427.884906996042,
      "visible_content_truncated": false,
      "visible_generation_ms": 1109.850289998576
    },
    {
      "content_chunks": 52,
      "controlled_output": null,
      "decode_tok_s": 34.66264021819373,
      "e2e_ms": 11353.415417019278,
      "effective_prefill_tok_s": 4523.187295065797,
      "generation_ms": 10010.778111987747,
      "mean_inter_token_latency_ms": 28.84950464549783,
      "mean_time_per_output_token_ms": 28.84950464549783,
      "output_token_source": "usage",
      "output_tokens": 348,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6073,
      "reasoning_chunks": 287,
      "request_canary": null,
      "request_index": 11,
      "time_to_first_output_ms": 1342.6373050315306,
      "tpot_ms": 28.84950464549783,
      "ttft_ms": 10362.137335003354,
      "visible_content_truncated": false,
      "visible_generation_ms": 991.2780820159242
    },
    {
      "content_chunks": 45,
      "controlled_output": null,
      "decode_tok_s": 36.74898235744866,
      "e2e_ms": 8882.15117098298,
      "effective_prefill_tok_s": 4518.315977780012,
      "generation_ms": 7537.6236899755895,
      "mean_inter_token_latency_ms": 27.211637869947978,
      "mean_time_per_output_token_ms": 27.211637869947978,
      "output_token_source": "usage",
      "output_tokens": 278,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6075,
      "reasoning_chunks": 221,
      "request_canary": null,
      "request_index": 12,
      "time_to_first_output_ms": 1344.5274810073897,
      "tpot_ms": 27.211637869947978,
      "ttft_ms": 8022.920230985619,
      "visible_content_truncated": false,
      "visible_generation_ms": 859.2309399973601
    },
    {
      "content_chunks": 45,
      "controlled_output": null,
      "decode_tok_s": 40.488051031777836,
      "e2e_ms": 6173.162410967052,
      "effective_prefill_tok_s": 4477.766046242589,
      "generation_ms": 4816.235778969713,
      "mean_inter_token_latency_ms": 24.6986450203575,
      "mean_time_per_output_token_ms": 24.6986450203575,
      "output_token_source": "usage",
      "output_tokens": 196,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6076,
      "reasoning_chunks": 147,
      "request_canary": null,
      "request_index": 13,
      "time_to_first_output_ms": 1356.9266319973394,
      "tpot_ms": 24.6986450203575,
      "ttft_ms": 5327.715635998175,
      "visible_content_truncated": false,
      "visible_generation_ms": 845.4467749688774
    },
    {
      "content_chunks": 54,
      "controlled_output": null,
      "decode_tok_s": 31.739540660257404,
      "e2e_ms": 9445.427227998152,
      "effective_prefill_tok_s": 4506.5103790741305,
      "generation_ms": 8097.155618946999,
      "mean_inter_token_latency_ms": 31.506442097070035,
      "mean_time_per_output_token_ms": 31.506442097070035,
      "output_token_source": "usage",
      "output_tokens": 258,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6076,
      "reasoning_chunks": 200,
      "request_canary": null,
      "request_index": 14,
      "time_to_first_output_ms": 1348.271609051153,
      "tpot_ms": 31.506442097070035,
      "ttft_ms": 8424.0511150565,
      "visible_content_truncated": false,
      "visible_generation_ms": 1021.3761129416525
    },
    {
      "content_chunks": 51,
      "controlled_output": null,
      "decode_tok_s": 24.344585547719497,
      "e2e_ms": 7345.723951002583,
      "effective_prefill_tok_s": 3536.8630126586845,
      "generation_ms": 5627.534703002311,
      "mean_inter_token_latency_ms": 41.07689564235264,
      "mean_time_per_output_token_ms": 41.07689564235264,
      "output_token_source": "usage",
      "output_tokens": 138,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6077,
      "reasoning_chunks": 84,
      "request_canary": null,
      "request_index": 15,
      "time_to_first_output_ms": 1718.1892480002716,
      "tpot_ms": 41.07689564235264,
      "ttft_ms": 5207.0938990218565,
      "visible_content_truncated": false,
      "visible_generation_ms": 2138.6300519807264
    },
    {
      "content_chunks": 46,
      "controlled_output": null,
      "decode_tok_s": 27.112667352579212,
      "e2e_ms": 9114.818708971143,
      "effective_prefill_tok_s": 2136.614809216943,
      "generation_ms": 6270.131882978603,
      "mean_inter_token_latency_ms": 36.88312872340355,
      "mean_time_per_output_token_ms": 36.88312872340355,
      "output_token_source": "usage",
      "output_tokens": 171,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6078,
      "reasoning_chunks": 120,
      "request_canary": null,
      "request_index": 16,
      "time_to_first_output_ms": 2844.6868259925395,
      "tpot_ms": 36.88312872340355,
      "ttft_ms": 7110.395187977701,
      "visible_content_truncated": false,
      "visible_generation_ms": 2004.4235209934413
    },
    {
      "content_chunks": 47,
      "controlled_output": null,
      "decode_tok_s": 36.417015712914335,
      "e2e_ms": 7232.594506000169,
      "effective_prefill_tok_s": 1697.8286236731476,
      "generation_ms": 3652.138908044435,
      "mean_inter_token_latency_ms": 27.459691037928085,
      "mean_time_per_output_token_ms": 27.459691037928085,
      "output_token_source": "usage",
      "output_tokens": 134,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6079,
      "reasoning_chunks": 81,
      "request_canary": null,
      "request_index": 17,
      "time_to_first_output_ms": 3580.4555979557335,
      "tpot_ms": 27.459691037928085,
      "ttft_ms": 5170.035702991299,
      "visible_content_truncated": false,
      "visible_generation_ms": 2062.5588030088693
    },
    {
      "content_chunks": 55,
      "controlled_output": null,
      "decode_tok_s": 49.80376006282441,
      "e2e_ms": 5963.5146469809115,
      "effective_prefill_tok_s": 4517.651054855912,
      "generation_ms": 4618.125212029554,
      "mean_inter_token_latency_ms": 20.078805269693714,
      "mean_time_per_output_token_ms": 20.078805269693714,
      "output_token_source": "usage",
      "output_tokens": 231,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6078,
      "reasoning_chunks": 168,
      "request_canary": null,
      "request_index": 18,
      "time_to_first_output_ms": 1345.3894349513575,
      "tpot_ms": 20.078805269693714,
      "ttft_ms": 5232.534981914796,
      "visible_content_truncated": false,
      "visible_generation_ms": 730.9796650661156
    },
    {
      "content_chunks": 45,
      "controlled_output": null,
      "decode_tok_s": 65.35038919443183,
      "e2e_ms": 3384.729153010994,
      "effective_prefill_tok_s": 4453.242172638876,
      "generation_ms": 2019.8808549903333,
      "mean_inter_token_latency_ms": 15.302127689320708,
      "mean_time_per_output_token_ms": 15.302127689320708,
      "output_token_source": "usage",
      "output_tokens": 133,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6078,
      "reasoning_chunks": 85,
      "request_canary": null,
      "request_index": 19,
      "time_to_first_output_ms": 1364.8482980206609,
      "tpot_ms": 15.302127689320708,
      "ttft_ms": 2709.3968259869143,
      "visible_content_truncated": false,
      "visible_generation_ms": 675.3323270240799
    }
  ],
  "requests": 20,
  "response_words": 0,
  "responses_received": 20,
  "run_id": "benchmark-20261017T003501Z",
  "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": "2026-10-17T00:35:01Z",
  "started_at_unix_ns": 1792197301349192773,
  "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": 45977.208545897156
}