{
  "base_url": "redacted://routed-capability",
  "clock_domain_id": null,
  "completed": 20,
  "concurrency": 4,
  "configuration_fingerprint": null,
  "context_distribution": {
    "32768": 20
  },
  "context_seed": 941224,
  "context_tokens": 32768,
  "controlled_output_policy": "observe",
  "engine": "vllm-exl3-v84",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-10-17T00:37:51Z",
  "finished_at_unix_ns": 1792197471874630438,
  "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": 643.0,
    "completion_tokens_mean": 267.15,
    "completion_tokens_mean_ci95_high": 329.8281224502995,
    "completion_tokens_mean_ci95_low": 204.47187754970048,
    "completion_tokens_min": 135.0,
    "completion_tokens_p25": 173.0,
    "completion_tokens_p50": 227.0,
    "completion_tokens_p75": 271.0,
    "completion_tokens_p90": 492.0,
    "completion_tokens_p95": 500.0,
    "completion_tokens_p99": 643.0,
    "completion_tokens_samples": 20,
    "completion_tokens_stddev": 143.01279847033197,
    "content_chunks": 1077,
    "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": 58.38647235884231,
    "decode_tok_s_mean": 18.70668622988355,
    "decode_tok_s_mean_ci95_high": 24.628325670655432,
    "decode_tok_s_mean_ci95_low": 12.785046789111671,
    "decode_tok_s_min": 10.610881969639557,
    "decode_tok_s_p25": 12.382992658129277,
    "decode_tok_s_p50": 13.422126259078633,
    "decode_tok_s_p75": 15.029061646539368,
    "decode_tok_s_p90": 31.45245183421057,
    "decode_tok_s_p95": 53.4924010405968,
    "decode_tok_s_p99": 58.38647235884231,
    "decode_tok_s_samples": 20,
    "decode_tok_s_stddev": 13.51141666103037,
    "e2e_max_ms": 57483.25485596433,
    "e2e_mean_ci95_high_ms": 29073.886345401603,
    "e2e_mean_ci95_low_ms": 19706.32132766283,
    "e2e_mean_ms": 24390.103836532217,
    "e2e_min_ms": 7733.983907033689,
    "e2e_p25_ms": 17974.21182191465,
    "e2e_p50_ms": 23523.790008970536,
    "e2e_p75_ms": 27954.243918997236,
    "e2e_p90_ms": 35073.81568197161,
    "e2e_p95_ms": 38421.83968098834,
    "e2e_p99_ms": 57483.25485596433,
    "e2e_samples": 20,
    "e2e_stddev_ms": 10686.996001690266,
    "effective_prefill_tok_s_max": 4893.69700071049,
    "effective_prefill_tok_s_mean": 4061.782528726988,
    "effective_prefill_tok_s_mean_ci95_high": 4607.500775784001,
    "effective_prefill_tok_s_mean_ci95_low": 3516.0642816699747,
    "effective_prefill_tok_s_min": 1278.1611541977293,
    "effective_prefill_tok_s_p25": 2506.983286048959,
    "effective_prefill_tok_s_p50": 4825.771777737633,
    "effective_prefill_tok_s_p75": 4846.43251238086,
    "effective_prefill_tok_s_p90": 4885.020011656863,
    "effective_prefill_tok_s_p95": 4885.901180220048,
    "effective_prefill_tok_s_p99": 4893.69700071049,
    "effective_prefill_tok_s_samples": 20,
    "effective_prefill_tok_s_stddev": 1245.1664254913335,
    "generation_max_ms": 37703.27647298109,
    "generation_mean_ci95_high_ms": 20772.36654655384,
    "generation_mean_ci95_low_ms": 13304.312778736186,
    "generation_mean_ms": 17038.339662645012,
    "generation_min_ms": 2505.0287030171603,
    "generation_p25_ms": 12436.412121984176,
    "generation_p50_ms": 12881.540936068632,
    "generation_p75_ms": 20411.763234995306,
    "generation_p90_ms": 24913.61806495115,
    "generation_p95_ms": 33202.339023933746,
    "generation_p99_ms": 37703.27647298109,
    "generation_samples": 20,
    "generation_stddev_ms": 8519.936675746787,
    "mean_inter_token_latency_ms_max": 94.24287282256606,
    "mean_inter_token_latency_ms_mean": 67.227990731685,
    "mean_inter_token_latency_ms_mean_ci95_high": 76.64475980313438,
    "mean_inter_token_latency_ms_mean_ci95_low": 57.811221660235624,
    "mean_inter_token_latency_ms_min": 17.127254989032668,
    "mean_inter_token_latency_ms_p25": 58.81620552315359,
    "mean_inter_token_latency_ms_p50": 74.15597559275041,
    "mean_inter_token_latency_ms_p75": 80.58187687324721,
    "mean_inter_token_latency_ms_p90": 81.36733232308016,
    "mean_inter_token_latency_ms_p95": 83.28814238707542,
    "mean_inter_token_latency_ms_p99": 94.24287282256606,
    "mean_inter_token_latency_ms_samples": 20,
    "mean_inter_token_latency_ms_stddev": 21.486260992018753,
    "mean_time_per_output_token_max_ms": 94.24287282256606,
    "mean_time_per_output_token_mean_ci95_high_ms": 76.64475980313438,
    "mean_time_per_output_token_mean_ci95_low_ms": 57.811221660235624,
    "mean_time_per_output_token_mean_ms": 67.227990731685,
    "mean_time_per_output_token_min_ms": 17.127254989032668,
    "mean_time_per_output_token_p25_ms": 58.81620552315359,
    "mean_time_per_output_token_p50_ms": 74.15597559275041,
    "mean_time_per_output_token_p75_ms": 80.58187687324721,
    "mean_time_per_output_token_p90_ms": 81.36733232308016,
    "mean_time_per_output_token_p95_ms": 83.28814238707542,
    "mean_time_per_output_token_p99_ms": 94.24287282256606,
    "mean_time_per_output_token_samples": 20,
    "mean_time_per_output_token_stddev_ms": 21.486260992018753,
    "output_token_sources": {
      "usage": 20
    },
    "output_tokens": 5343,
    "prefix_cache_hit_avg": null,
    "prompt_token_samples": 20,
    "prompt_tokens": 505696,
    "prompt_tokens_max": 25288.0,
    "prompt_tokens_mean": 25284.8,
    "prompt_tokens_mean_ci95_high": 25285.77067139,
    "prompt_tokens_mean_ci95_low": 25283.829328609998,
    "prompt_tokens_min": 25281.0,
    "prompt_tokens_p25": 25282.0,
    "prompt_tokens_p50": 25285.0,
    "prompt_tokens_p75": 25286.0,
    "prompt_tokens_p90": 25287.0,
    "prompt_tokens_p95": 25287.0,
    "prompt_tokens_p99": 25288.0,
    "prompt_tokens_samples": 20,
    "prompt_tokens_stddev": 2.2147828692435603,
    "reasoning_chunks": 4140,
    "request_canary_passed": 0,
    "request_canary_samples": 0,
    "throughput_performance_eligible": true,
    "throughput_tok_s": 42.97067033622411,
    "time_to_first_output_max_ms": 19779.978382983245,
    "time_to_first_output_mean_ci95_high_ms": 9134.161298513693,
    "time_to_first_output_mean_ci95_low_ms": 5569.367049260716,
    "time_to_first_output_mean_ms": 7351.764173887204,
    "time_to_first_output_min_ms": 5167.054682038724,
    "time_to_first_output_p25_ms": 5198.520465986803,
    "time_to_first_output_p50_ms": 5232.2686850093305,
    "time_to_first_output_p75_ms": 5988.578251097351,
    "time_to_first_output_p90_ms": 11053.851780947298,
    "time_to_first_output_p95_ms": 15687.516699894331,
    "time_to_first_output_p99_ms": 19779.978382983245,
    "time_to_first_output_samples": 20,
    "time_to_first_output_stddev_ms": 4066.899115882645,
    "tpot_max_ms": 94.24287282256606,
    "tpot_mean_ci95_high_ms": 76.64475980313438,
    "tpot_mean_ci95_low_ms": 57.811221660235624,
    "tpot_mean_ms": 67.227990731685,
    "tpot_min_ms": 17.127254989032668,
    "tpot_p25_ms": 58.81620552315359,
    "tpot_p50_ms": 74.15597559275041,
    "tpot_p75_ms": 80.58187687324721,
    "tpot_p90_ms": 81.36733232308016,
    "tpot_p95_ms": 83.28814238707542,
    "tpot_p99_ms": 94.24287282256606,
    "tpot_samples": 20,
    "tpot_stddev_ms": 21.486260992018753,
    "ttft_max_ms": 55244.93750894908,
    "ttft_mean_ci95_high_ms": 26786.45814704042,
    "ttft_mean_ci95_low_ms": 17479.195966731906,
    "ttft_mean_ms": 22132.827056886163,
    "ttft_min_ms": 6862.008529016748,
    "ttft_p25_ms": 16651.472870027646,
    "ttft_p50_ms": 21267.867935006507,
    "ttft_p75_ms": 26707.24966202397,
    "ttft_p90_ms": 29474.54334993381,
    "ttft_p95_ms": 35690.542233060114,
    "ttft_p99_ms": 55244.93750894908,
    "ttft_samples": 20,
    "ttft_stddev_ms": 10618.199448766702
  },
  "model": "llm.primary",
  "performance_eligible": true,
  "prompt_cache_mode": "unique",
  "request_canaries": false,
  "request_timings": [
    {
      "content_chunks": 30,
      "controlled_output": null,
      "decode_tok_s": 12.409738253838006,
      "e2e_ms": 21851.823281962425,
      "effective_prefill_tok_s": 2287.0760799936706,
      "generation_ms": 10797.971501015127,
      "mean_inter_token_latency_ms": 80.58187687324721,
      "mean_time_per_output_token_ms": 80.58187687324721,
      "output_token_source": "usage",
      "output_tokens": 135,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25281,
      "reasoning_chunks": 102,
      "request_canary": null,
      "request_index": 0,
      "time_to_first_output_ms": 11053.851780947298,
      "tpot_ms": 80.58187687324721,
      "ttft_ms": 21267.867935006507,
      "visible_content_truncated": false,
      "visible_generation_ms": 583.9553469559178
    },
    {
      "content_chunks": 57,
      "controlled_output": null,
      "decode_tok_s": 10.610881969639557,
      "e2e_ms": 28324.139110045508,
      "effective_prefill_tok_s": 4221.703205659424,
      "generation_ms": 22335.560858948156,
      "mean_inter_token_latency_ms": 94.24287282256606,
      "mean_time_per_output_token_ms": 94.24287282256606,
      "output_token_source": "usage",
      "output_tokens": 238,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25282,
      "reasoning_chunks": 177,
      "request_canary": null,
      "request_index": 1,
      "time_to_first_output_ms": 5988.578251097351,
      "tpot_ms": 94.24287282256606,
      "ttft_ms": 27237.819071044214,
      "visible_content_truncated": false,
      "visible_generation_ms": 1086.3200390012935
    },
    {
      "content_chunks": 52,
      "controlled_output": null,
      "decode_tok_s": 13.02274088438601,
      "e2e_ms": 57483.25485596433,
      "effective_prefill_tok_s": 1278.1611541977293,
      "generation_ms": 37703.27647298109,
      "mean_inter_token_latency_ms": 76.78875045413663,
      "mean_time_per_output_token_ms": 76.78875045413663,
      "output_token_source": "usage",
      "output_tokens": 492,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25282,
      "reasoning_chunks": 432,
      "request_canary": null,
      "request_index": 2,
      "time_to_first_output_ms": 19779.978382983245,
      "tpot_ms": 76.78875045413663,
      "ttft_ms": 55244.93750894908,
      "visible_content_truncated": false,
      "visible_generation_ms": 2238.317347015254
    },
    {
      "content_chunks": 53,
      "controlled_output": null,
      "decode_tok_s": 17.002116867371527,
      "e2e_ms": 28391.817092895508,
      "effective_prefill_tok_s": 1611.5361330688042,
      "generation_ms": 12704.300393001176,
      "mean_inter_token_latency_ms": 58.81620552315359,
      "mean_time_per_output_token_ms": 58.81620552315359,
      "output_token_source": "usage",
      "output_tokens": 217,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25281,
      "reasoning_chunks": 157,
      "request_canary": null,
      "request_index": 3,
      "time_to_first_output_ms": 15687.516699894331,
      "tpot_ms": 58.81620552315359,
      "ttft_ms": 27383.074898971245,
      "visible_content_truncated": false,
      "visible_generation_ms": 1008.7421939242631
    },
    {
      "content_chunks": 56,
      "controlled_output": null,
      "decode_tok_s": 14.067084413166546,
      "e2e_ms": 17974.21182191465,
      "effective_prefill_tok_s": 4816.1041137057255,
      "generation_ms": 12724.740588921122,
      "mean_inter_token_latency_ms": 71.08793625095599,
      "mean_time_per_output_token_ms": 71.08793625095599,
      "output_token_source": "usage",
      "output_tokens": 180,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25282,
      "reasoning_chunks": 120,
      "request_canary": null,
      "request_index": 4,
      "time_to_first_output_ms": 5249.471232993528,
      "tpot_ms": 71.08793625095599,
      "ttft_ms": 16914.16984598618,
      "visible_content_truncated": false,
      "visible_generation_ms": 1060.0419759284705
    },
    {
      "content_chunks": 48,
      "controlled_output": null,
      "decode_tok_s": 12.006511027134868,
      "e2e_ms": 17274.770312011242,
      "effective_prefill_tok_s": 4311.345326384558,
      "generation_ms": 11410.475507029332,
      "mean_inter_token_latency_ms": 83.28814238707542,
      "mean_time_per_output_token_ms": 83.28814238707542,
      "output_token_source": "usage",
      "output_tokens": 138,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25283,
      "reasoning_chunks": 86,
      "request_canary": null,
      "request_index": 5,
      "time_to_first_output_ms": 5864.29480498191,
      "tpot_ms": 83.28814238707542,
      "ttft_ms": 11563.596170046367,
      "visible_content_truncated": false,
      "visible_generation_ms": 5711.174141964875
    },
    {
      "content_chunks": 61,
      "controlled_output": null,
      "decode_tok_s": 13.740592129346547,
      "e2e_ms": 22966.969163971953,
      "effective_prefill_tok_s": 2506.983286048959,
      "generation_ms": 12881.540936068632,
      "mean_inter_token_latency_ms": 72.777067435416,
      "mean_time_per_output_token_ms": 72.777067435416,
      "output_token_source": "usage",
      "output_tokens": 178,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25284,
      "reasoning_chunks": 114,
      "request_canary": null,
      "request_index": 6,
      "time_to_first_output_ms": 10085.428227903321,
      "tpot_ms": 72.777067435416,
      "ttft_ms": 17032.880869926885,
      "visible_content_truncated": false,
      "visible_generation_ms": 5934.088294045068
    },
    {
      "content_chunks": 44,
      "controlled_output": null,
      "decode_tok_s": 12.289944520109897,
      "e2e_ms": 23831.63956797216,
      "effective_prefill_tok_s": 4863.883900320532,
      "generation_ms": 18633.119101985358,
      "mean_inter_token_latency_ms": 81.36733232308016,
      "mean_time_per_output_token_ms": 81.36733232308016,
      "output_token_source": "usage",
      "output_tokens": 230,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25285,
      "reasoning_chunks": 179,
      "request_canary": null,
      "request_index": 7,
      "time_to_first_output_ms": 5198.520465986803,
      "tpot_ms": 81.36733232308016,
      "ttft_ms": 22970.132844988257,
      "visible_content_truncated": false,
      "visible_generation_ms": 861.5067229839042
    },
    {
      "content_chunks": 46,
      "controlled_output": null,
      "decode_tok_s": 13.594524144336159,
      "e2e_ms": 18038.847516989335,
      "effective_prefill_tok_s": 4825.771777737633,
      "generation_ms": 12799.271100084297,
      "mean_inter_token_latency_ms": 73.55902931082929,
      "mean_time_per_output_token_ms": 73.55902931082929,
      "output_token_source": "usage",
      "output_tokens": 175,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25285,
      "reasoning_chunks": 126,
      "request_canary": null,
      "request_index": 8,
      "time_to_first_output_ms": 5239.576416905038,
      "tpot_ms": 73.55902931082929,
      "ttft_ms": 17156.82068094611,
      "visible_content_truncated": false,
      "visible_generation_ms": 882.0268360432237
    },
    {
      "content_chunks": 48,
      "controlled_output": null,
      "decode_tok_s": 13.422126259078633,
      "e2e_ms": 24078.102182946168,
      "effective_prefill_tok_s": 4835.875095571208,
      "generation_ms": 18849.472514004447,
      "mean_inter_token_latency_ms": 74.50384392887133,
      "mean_time_per_output_token_ms": 74.50384392887133,
      "output_token_source": "usage",
      "output_tokens": 254,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25285,
      "reasoning_chunks": 194,
      "request_canary": null,
      "request_index": 9,
      "time_to_first_output_ms": 5228.629668941721,
      "tpot_ms": 74.50384392887133,
      "ttft_ms": 23156.29034896847,
      "visible_content_truncated": false,
      "visible_generation_ms": 921.8118339776993
    },
    {
      "content_chunks": 58,
      "controlled_output": null,
      "decode_tok_s": 12.317089245645228,
      "e2e_ms": 23523.790008970536,
      "effective_prefill_tok_s": 4885.901180220048,
      "generation_ms": 18348.49090501666,
      "mean_inter_token_latency_ms": 81.18801285405603,
      "mean_time_per_output_token_ms": 81.18801285405603,
      "output_token_source": "usage",
      "output_tokens": 227,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25286,
      "reasoning_chunks": 162,
      "request_canary": null,
      "request_index": 10,
      "time_to_first_output_ms": 5175.299103953876,
      "tpot_ms": 81.18801285405603,
      "ttft_ms": 17652.277055894956,
      "visible_content_truncated": false,
      "visible_generation_ms": 5871.51295307558
    },
    {
      "content_chunks": 49,
      "controlled_output": null,
      "decode_tok_s": 12.923049535878333,
      "e2e_ms": 23845.400225021876,
      "effective_prefill_tok_s": 4291.036917621431,
      "generation_ms": 17952.418998000212,
      "mean_inter_token_latency_ms": 77.38111637069056,
      "mean_time_per_output_token_ms": 77.38111637069056,
      "output_token_source": "usage",
      "output_tokens": 233,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25287,
      "reasoning_chunks": 180,
      "request_canary": null,
      "request_index": 11,
      "time_to_first_output_ms": 5892.981227021664,
      "tpot_ms": 77.38111637069056,
      "ttft_ms": 22906.1605889583,
      "visible_content_truncated": false,
      "visible_generation_ms": 939.2396360635757
    },
    {
      "content_chunks": 44,
      "controlled_output": null,
      "decode_tok_s": 12.523271376586054,
      "e2e_ms": 35073.81568197161,
      "effective_prefill_tok_s": 2488.8295437914494,
      "generation_ms": 24913.61806495115,
      "mean_inter_token_latency_ms": 79.8513399517665,
      "mean_time_per_output_token_ms": 79.8513399517665,
      "output_token_source": "usage",
      "output_tokens": 313,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25287,
      "reasoning_chunks": 258,
      "request_canary": null,
      "request_index": 12,
      "time_to_first_output_ms": 10160.197617020458,
      "tpot_ms": 79.8513399517665,
      "ttft_ms": 29474.54334993381,
      "visible_content_truncated": false,
      "visible_generation_ms": 5599.272332037799
    },
    {
      "content_chunks": 53,
      "controlled_output": null,
      "decode_tok_s": 13.485089933841575,
      "e2e_ms": 17987.0964869624,
      "effective_prefill_tok_s": 4832.702890898064,
      "generation_ms": 12754.82780195307,
      "mean_inter_token_latency_ms": 74.15597559275041,
      "mean_time_per_output_token_ms": 74.15597559275041,
      "output_token_source": "usage",
      "output_tokens": 173,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25286,
      "reasoning_chunks": 113,
      "request_canary": null,
      "request_index": 13,
      "time_to_first_output_ms": 5232.2686850093305,
      "tpot_ms": 74.15597559275041,
      "ttft_ms": 13213.846625993028,
      "visible_content_truncated": false,
      "visible_generation_ms": 4773.249860969372
    },
    {
      "content_chunks": 78,
      "controlled_output": null,
      "decode_tok_s": 15.029061646539368,
      "e2e_ms": 38421.83968098834,
      "effective_prefill_tok_s": 4844.524727825034,
      "generation_ms": 33202.339023933746,
      "mean_inter_token_latency_ms": 66.53775355497744,
      "mean_time_per_output_token_ms": 66.53775355497744,
      "output_token_source": "usage",
      "output_tokens": 500,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25286,
      "reasoning_chunks": 405,
      "request_canary": null,
      "request_index": 14,
      "time_to_first_output_ms": 5219.500657054596,
      "tpot_ms": 66.53775355497744,
      "ttft_ms": 35690.542233060114,
      "visible_content_truncated": false,
      "visible_generation_ms": 2731.2974479282275
    },
    {
      "content_chunks": 50,
      "controlled_output": null,
      "decode_tok_s": 12.382992658129277,
      "e2e_ms": 17603.4668040229,
      "effective_prefill_tok_s": 4893.69700071049,
      "generation_ms": 12436.412121984176,
      "mean_inter_token_latency_ms": 80.75592287002712,
      "mean_time_per_output_token_ms": 80.75592287002712,
      "output_token_source": "usage",
      "output_tokens": 155,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25286,
      "reasoning_chunks": 101,
      "request_canary": null,
      "request_index": 15,
      "time_to_first_output_ms": 5167.054682038724,
      "tpot_ms": 80.75592287002712,
      "ttft_ms": 16651.472870027646,
      "visible_content_truncated": false,
      "visible_generation_ms": 951.9939339952543
    },
    {
      "content_chunks": 69,
      "controlled_output": null,
      "decode_tok_s": 19.975584498993772,
      "e2e_ms": 27954.243918997236,
      "effective_prefill_tok_s": 4885.020011656863,
      "generation_ms": 22777.806577971205,
      "mean_inter_token_latency_ms": 50.061113358178474,
      "mean_time_per_output_token_ms": 50.061113358178474,
      "output_token_source": "usage",
      "output_tokens": 456,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25287,
      "reasoning_chunks": 379,
      "request_canary": null,
      "request_index": 16,
      "time_to_first_output_ms": 5176.4373410260305,
      "tpot_ms": 50.061113358178474,
      "ttft_ms": 26707.24966202397,
      "visible_content_truncated": false,
      "visible_generation_ms": 1246.9942569732666
    },
    {
      "content_chunks": 80,
      "controlled_output": null,
      "decode_tok_s": 31.45245183421057,
      "e2e_ms": 25600.647573010065,
      "effective_prefill_tok_s": 4873.109198975573,
      "generation_ms": 20411.763234995306,
      "mean_inter_token_latency_ms": 31.794023730522284,
      "mean_time_per_output_token_ms": 31.794023730522284,
      "output_token_source": "usage",
      "output_tokens": 643,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25286,
      "reasoning_chunks": 556,
      "request_canary": null,
      "request_index": 17,
      "time_to_first_output_ms": 5188.884338014759,
      "tpot_ms": 31.794023730522284,
      "ttft_ms": 24551.93562398199,
      "visible_content_truncated": false,
      "visible_generation_ms": 1048.7119490280747
    },
    {
      "content_chunks": 46,
      "controlled_output": null,
      "decode_tok_s": 53.4924010405968,
      "e2e_ms": 7733.983907033689,
      "effective_prefill_tok_s": 4835.956517771703,
      "generation_ms": 2505.0287030171603,
      "mean_inter_token_latency_ms": 18.69424405236687,
      "mean_time_per_output_token_ms": 18.69424405236687,
      "output_token_source": "usage",
      "output_tokens": 135,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25287,
      "reasoning_chunks": 86,
      "request_canary": null,
      "request_index": 18,
      "time_to_first_output_ms": 5228.955204016529,
      "tpot_ms": 18.69424405236687,
      "ttft_ms": 6862.008529016748,
      "visible_content_truncated": false,
      "visible_generation_ms": 871.9753780169412
    },
    {
      "content_chunks": 55,
      "controlled_output": null,
      "decode_tok_s": 58.38647235884231,
      "e2e_ms": 9842.217536992393,
      "effective_prefill_tok_s": 4846.43251238086,
      "generation_ms": 4624.35884703882,
      "mean_inter_token_latency_ms": 17.127254989032668,
      "mean_time_per_output_token_ms": 17.127254989032668,
      "output_token_source": "usage",
      "output_tokens": 271,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25288,
      "reasoning_chunks": 213,
      "request_canary": null,
      "request_index": 19,
      "time_to_first_output_ms": 5217.858689953573,
      "tpot_ms": 17.127254989032668,
      "ttft_ms": 9018.914423999377,
      "visible_content_truncated": false,
      "visible_generation_ms": 823.3031129930168
    }
  ],
  "requests": 20,
  "response_words": 0,
  "responses_received": 20,
  "run_id": "benchmark-20261017T003547Z",
  "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": "2026-10-17T00:35:47Z",
  "started_at_unix_ns": 1792197347534002110,
  "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": 124340.62485396862
}