{
  "base_url": "redacted://routed-capability",
  "clock_domain_id": null,
  "completed": 20,
  "concurrency": 4,
  "configuration_fingerprint": null,
  "context_distribution": {
    "8192": 20
  },
  "context_seed": 941314,
  "context_tokens": 8192,
  "controlled_output_policy": "observe",
  "engine": "vllm-exl3-v84",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-10-17T00:47:36Z",
  "finished_at_unix_ns": 1792198056218766522,
  "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": 397.0,
    "completion_tokens_mean": 234.7,
    "completion_tokens_mean_ci95_high": 261.6422057628625,
    "completion_tokens_mean_ci95_low": 207.7577942371375,
    "completion_tokens_min": 154.0,
    "completion_tokens_p25": 182.0,
    "completion_tokens_p50": 235.0,
    "completion_tokens_p75": 253.0,
    "completion_tokens_p90": 314.0,
    "completion_tokens_p95": 328.0,
    "completion_tokens_p99": 397.0,
    "completion_tokens_samples": 20,
    "completion_tokens_stddev": 61.47408525463992,
    "content_chunks": 1090,
    "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": 63.76374921220551,
    "decode_tok_s_mean": 34.56094399085048,
    "decode_tok_s_mean_ci95_high": 38.27318520492857,
    "decode_tok_s_mean_ci95_low": 30.84870277677239,
    "decode_tok_s_min": 24.79218156890132,
    "decode_tok_s_p25": 30.72690096237287,
    "decode_tok_s_p50": 32.099480512315225,
    "decode_tok_s_p75": 36.31373698817776,
    "decode_tok_s_p90": 42.23980766622943,
    "decode_tok_s_p95": 43.51736623164756,
    "decode_tok_s_p99": 63.76374921220551,
    "decode_tok_s_samples": 20,
    "decode_tok_s_stddev": 8.470228268933639,
    "e2e_max_ms": 15163.332936004736,
    "e2e_mean_ci95_high_ms": 9941.600440976805,
    "e2e_mean_ci95_low_ms": 7778.91736263536,
    "e2e_mean_ms": 8860.258901806083,
    "e2e_min_ms": 5247.489602072164,
    "e2e_p25_ms": 6853.114811005071,
    "e2e_p50_ms": 8885.95572905615,
    "e2e_p75_ms": 10085.828938055784,
    "e2e_p90_ms": 11456.717417924665,
    "e2e_p95_ms": 12284.365364001133,
    "e2e_p99_ms": 15163.332936004736,
    "e2e_samples": 20,
    "e2e_stddev_ms": 2467.299171918354,
    "effective_prefill_tok_s_max": 4861.380996410101,
    "effective_prefill_tok_s_mean": 4037.035300456161,
    "effective_prefill_tok_s_mean_ci95_high": 4539.963342827687,
    "effective_prefill_tok_s_mean_ci95_low": 3534.1072580846358,
    "effective_prefill_tok_s_min": 1281.3319751816034,
    "effective_prefill_tok_s_p25": 3627.75939621013,
    "effective_prefill_tok_s_p50": 4604.530362501544,
    "effective_prefill_tok_s_p75": 4682.727142416416,
    "effective_prefill_tok_s_p90": 4767.370246068335,
    "effective_prefill_tok_s_p95": 4798.912441714542,
    "effective_prefill_tok_s_p99": 4861.380996410101,
    "effective_prefill_tok_s_samples": 20,
    "effective_prefill_tok_s_stddev": 1147.531929115944,
    "generation_max_ms": 13897.629493032582,
    "generation_mean_ci95_high_ms": 8190.988483763735,
    "generation_mean_ci95_low_ms": 6022.3056627489,
    "generation_mean_ms": 7106.6470732563175,
    "generation_min_ms": 3622.175583965145,
    "generation_p25_ms": 5469.080980983563,
    "generation_p50_ms": 7547.887919005007,
    "generation_p75_ms": 8055.061953957193,
    "generation_p90_ms": 9004.856760031544,
    "generation_p95_ms": 10136.163905961439,
    "generation_p99_ms": 13897.629493032582,
    "generation_samples": 20,
    "generation_stddev_ms": 2474.143984400601,
    "mean_inter_token_latency_ms_max": 40.33529672331758,
    "mean_inter_token_latency_ms_mean": 30.17711245376713,
    "mean_inter_token_latency_ms_mean_ci95_high": 32.63989598096852,
    "mean_inter_token_latency_ms_mean_ci95_low": 27.71432892656574,
    "mean_inter_token_latency_ms_min": 15.682892119031518,
    "mean_inter_token_latency_ms_p25": 25.585869870711022,
    "mean_inter_token_latency_ms_p50": 30.740709232890772,
    "mean_inter_token_latency_ms_p75": 32.3839102426883,
    "mean_inter_token_latency_ms_p90": 36.78110481258992,
    "mean_inter_token_latency_ms_p95": 37.23672602150069,
    "mean_inter_token_latency_ms_p99": 40.33529672331758,
    "mean_inter_token_latency_ms_samples": 20,
    "mean_inter_token_latency_ms_stddev": 5.619338143560215,
    "mean_time_per_output_token_max_ms": 40.33529672331758,
    "mean_time_per_output_token_mean_ci95_high_ms": 32.63989598096852,
    "mean_time_per_output_token_mean_ci95_low_ms": 27.71432892656574,
    "mean_time_per_output_token_mean_ms": 30.17711245376713,
    "mean_time_per_output_token_min_ms": 15.682892119031518,
    "mean_time_per_output_token_p25_ms": 25.585869870711022,
    "mean_time_per_output_token_p50_ms": 30.740709232890772,
    "mean_time_per_output_token_p75_ms": 32.3839102426883,
    "mean_time_per_output_token_p90_ms": 36.78110481258992,
    "mean_time_per_output_token_p95_ms": 37.23672602150069,
    "mean_time_per_output_token_p99_ms": 40.33529672331758,
    "mean_time_per_output_token_samples": 20,
    "mean_time_per_output_token_stddev_ms": 5.619338143560215,
    "output_token_sources": {
      "usage": 20
    },
    "output_tokens": 4694,
    "prefix_cache_hit_avg": null,
    "prompt_token_samples": 20,
    "prompt_tokens": 121473,
    "prompt_tokens_max": 6080.0,
    "prompt_tokens_mean": 6073.65,
    "prompt_tokens_mean_ci95_high": 6075.500016842028,
    "prompt_tokens_mean_ci95_low": 6071.799983157971,
    "prompt_tokens_min": 6067.0,
    "prompt_tokens_p25": 6070.0,
    "prompt_tokens_p50": 6073.0,
    "prompt_tokens_p75": 6077.0,
    "prompt_tokens_p90": 6079.0,
    "prompt_tokens_p95": 6079.0,
    "prompt_tokens_p99": 6080.0,
    "prompt_tokens_samples": 20,
    "prompt_tokens_stddev": 4.221187161525958,
    "reasoning_chunks": 3451,
    "request_canary_passed": 0,
    "request_canary_samples": 0,
    "throughput_performance_eligible": true,
    "throughput_tok_s": 102.30475662955082,
    "time_to_first_output_max_ms": 4736.477444996126,
    "time_to_first_output_mean_ci95_high_ms": 2175.699429762502,
    "time_to_first_output_mean_ci95_low_ms": 1331.524227337028,
    "time_to_first_output_mean_ms": 1753.611828549765,
    "time_to_first_output_min_ms": 1248.616392025724,
    "time_to_first_output_p25_ms": 1296.4060120284557,
    "time_to_first_output_p50_ms": 1317.4722399562597,
    "time_to_first_output_p75_ms": 1495.2365709468722,
    "time_to_first_output_p90_ms": 2760.6620910810307,
    "time_to_first_output_p95_ms": 3855.394067009911,
    "time_to_first_output_p99_ms": 4736.477444996126,
    "time_to_first_output_samples": 20,
    "time_to_first_output_stddev_ms": 963.0781313994927,
    "tpot_max_ms": 40.33529672331758,
    "tpot_mean_ci95_high_ms": 32.63989598096852,
    "tpot_mean_ci95_low_ms": 27.71432892656574,
    "tpot_mean_ms": 30.17711245376713,
    "tpot_min_ms": 15.682892119031518,
    "tpot_p25_ms": 25.585869870711022,
    "tpot_p50_ms": 30.740709232890772,
    "tpot_p75_ms": 32.3839102426883,
    "tpot_p90_ms": 36.78110481258992,
    "tpot_p95_ms": 37.23672602150069,
    "tpot_p99_ms": 40.33529672331758,
    "tpot_samples": 20,
    "tpot_stddev_ms": 5.619338143560215,
    "ttft_max_ms": 14042.72162693087,
    "ttft_mean_ci95_high_ms": 8310.438082108429,
    "ttft_mean_ci95_low_ms": 6077.405698320675,
    "ttft_mean_ms": 7193.921890214551,
    "ttft_min_ms": 3051.445860066451,
    "ttft_p25_ms": 5333.882624050602,
    "ttft_p50_ms": 6960.768657037988,
    "ttft_p75_ms": 7943.275690078735,
    "ttft_p90_ms": 10230.488458066247,
    "ttft_p95_ms": 10437.998352921568,
    "ttft_p99_ms": 14042.72162693087,
    "ttft_samples": 20,
    "ttft_stddev_ms": 2547.557248014192
  },
  "model": "llm.primary",
  "performance_eligible": true,
  "prompt_cache_mode": "unique",
  "request_canaries": false,
  "request_timings": [
    {
      "content_chunks": 48,
      "controlled_output": null,
      "decode_tok_s": 24.79218156890132,
      "e2e_ms": 8973.346492974088,
      "effective_prefill_tok_s": 3627.75939621013,
      "generation_ms": 7300.688706920482,
      "mean_inter_token_latency_ms": 40.33529672331758,
      "mean_time_per_output_token_ms": 40.33529672331758,
      "output_token_source": "usage",
      "output_tokens": 182,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6068,
      "reasoning_chunks": 130,
      "request_canary": null,
      "request_index": 0,
      "time_to_first_output_ms": 1672.6577860536054,
      "tpot_ms": 40.33529672331758,
      "ttft_ms": 6989.418815006502,
      "visible_content_truncated": false,
      "visible_generation_ms": 1983.9276779675856
    },
    {
      "content_chunks": 38,
      "controlled_output": null,
      "decode_tok_s": 42.23980766622943,
      "e2e_ms": 7477.569650975056,
      "effective_prefill_tok_s": 1573.639398346982,
      "generation_ms": 3622.175583965145,
      "mean_inter_token_latency_ms": 23.674350221994413,
      "mean_time_per_output_token_ms": 23.674350221994413,
      "output_token_source": "usage",
      "output_tokens": 154,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6067,
      "reasoning_chunks": 109,
      "request_canary": null,
      "request_index": 1,
      "time_to_first_output_ms": 3855.394067009911,
      "tpot_ms": 23.674350221994413,
      "ttft_ms": 6769.409943022765,
      "visible_content_truncated": false,
      "visible_generation_ms": 708.1597079522908
    },
    {
      "content_chunks": 58,
      "controlled_output": null,
      "decode_tok_s": 27.18787282479092,
      "e2e_ms": 10815.724045038223,
      "effective_prefill_tok_s": 2198.023445029402,
      "generation_ms": 8055.061953957193,
      "mean_inter_token_latency_ms": 36.78110481258992,
      "mean_time_per_output_token_ms": 36.78110481258992,
      "output_token_source": "usage",
      "output_tokens": 220,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6068,
      "reasoning_chunks": 153,
      "request_canary": null,
      "request_index": 2,
      "time_to_first_output_ms": 2760.6620910810307,
      "tpot_ms": 36.78110481258992,
      "ttft_ms": 7550.107427989133,
      "visible_content_truncated": false,
      "visible_generation_ms": 3265.6166170490906
    },
    {
      "content_chunks": 52,
      "controlled_output": null,
      "decode_tok_s": 31.002050177614034,
      "e2e_ms": 12284.365364001133,
      "effective_prefill_tok_s": 1281.3319751816034,
      "generation_ms": 7547.887919005007,
      "mean_inter_token_latency_ms": 32.25593127779917,
      "mean_time_per_output_token_ms": 32.25593127779917,
      "output_token_source": "usage",
      "output_tokens": 235,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6069,
      "reasoning_chunks": 170,
      "request_canary": null,
      "request_index": 3,
      "time_to_first_output_ms": 4736.477444996126,
      "tpot_ms": 32.25593127779917,
      "ttft_ms": 10230.488458066247,
      "visible_content_truncated": false,
      "visible_generation_ms": 2053.876905934885
    },
    {
      "content_chunks": 56,
      "controlled_output": null,
      "decode_tok_s": 30.91768652689884,
      "e2e_ms": 8885.95572905615,
      "effective_prefill_tok_s": 4607.307703273904,
      "generation_ms": 7568.48348909989,
      "mean_inter_token_latency_ms": 32.343946534614915,
      "mean_time_per_output_token_ms": 32.343946534614915,
      "output_token_source": "usage",
      "output_tokens": 235,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6070,
      "reasoning_chunks": 165,
      "request_canary": null,
      "request_index": 4,
      "time_to_first_output_ms": 1317.4722399562597,
      "tpot_ms": 32.343946534614915,
      "ttft_ms": 7853.151484043337,
      "visible_content_truncated": false,
      "visible_generation_ms": 1032.8042450128123
    },
    {
      "content_chunks": 47,
      "controlled_output": null,
      "decode_tok_s": 31.196831909740737,
      "e2e_ms": 8941.705250996165,
      "effective_prefill_tok_s": 4861.380996410101,
      "generation_ms": 7693.088858970441,
      "mean_inter_token_latency_ms": 32.05453691237684,
      "mean_time_per_output_token_ms": 32.05453691237684,
      "output_token_source": "usage",
      "output_tokens": 241,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6070,
      "reasoning_chunks": 187,
      "request_canary": null,
      "request_index": 5,
      "time_to_first_output_ms": 1248.616392025724,
      "tpot_ms": 32.05453691237684,
      "ttft_ms": 6960.768657037988,
      "visible_content_truncated": false,
      "visible_generation_ms": 1980.9365939581767
    },
    {
      "content_chunks": 55,
      "controlled_output": null,
      "decode_tok_s": 30.8795322277606,
      "e2e_ms": 11456.717417924665,
      "effective_prefill_tok_s": 4598.071903177135,
      "generation_ms": 10136.163905961439,
      "mean_inter_token_latency_ms": 32.3839102426883,
      "mean_time_per_output_token_ms": 32.3839102426883,
      "output_token_source": "usage",
      "output_tokens": 314,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6072,
      "reasoning_chunks": 247,
      "request_canary": null,
      "request_index": 6,
      "time_to_first_output_ms": 1320.553511963226,
      "tpot_ms": 32.3839102426883,
      "ttft_ms": 10437.998352921568,
      "visible_content_truncated": false,
      "visible_generation_ms": 1018.7190650030971
    },
    {
      "content_chunks": 49,
      "controlled_output": null,
      "decode_tok_s": 33.24484024398761,
      "e2e_ms": 6911.150405998342,
      "effective_prefill_tok_s": 4720.819449503253,
      "generation_ms": 5624.933031038381,
      "mean_inter_token_latency_ms": 30.07985578095391,
      "mean_time_per_output_token_ms": 30.07985578095391,
      "output_token_source": "usage",
      "output_tokens": 188,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6072,
      "reasoning_chunks": 128,
      "request_canary": null,
      "request_index": 7,
      "time_to_first_output_ms": 1286.2173749599606,
      "tpot_ms": 30.07985578095391,
      "ttft_ms": 3789.3475060118362,
      "visible_content_truncated": false,
      "visible_generation_ms": 3121.8028999865055
    },
    {
      "content_chunks": 48,
      "controlled_output": null,
      "decode_tok_s": 30.72690096237287,
      "e2e_ms": 8846.79322107695,
      "effective_prefill_tok_s": 4683.71786590166,
      "generation_ms": 7550.387209048495,
      "mean_inter_token_latency_ms": 32.544772452795236,
      "mean_time_per_output_token_ms": 32.544772452795236,
      "output_token_source": "usage",
      "output_tokens": 233,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6072,
      "reasoning_chunks": 179,
      "request_canary": null,
      "request_index": 8,
      "time_to_first_output_ms": 1296.4060120284557,
      "tpot_ms": 32.544772452795236,
      "ttft_ms": 7958.1037040334195,
      "visible_content_truncated": false,
      "visible_generation_ms": 888.6895170435309
    },
    {
      "content_chunks": 73,
      "controlled_output": null,
      "decode_tok_s": 32.099480512315225,
      "e2e_ms": 9311.38116796501,
      "effective_prefill_tok_s": 4767.370246068335,
      "generation_ms": 8037.513251998462,
      "mean_inter_token_latency_ms": 31.15315213952892,
      "mean_time_per_output_token_ms": 31.15315213952892,
      "output_token_source": "usage",
      "output_tokens": 259,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6073,
      "reasoning_chunks": 178,
      "request_canary": null,
      "request_index": 9,
      "time_to_first_output_ms": 1273.867915966548,
      "tpot_ms": 31.15315213952892,
      "ttft_ms": 6845.212684944272,
      "visible_content_truncated": false,
      "visible_generation_ms": 2466.1684830207378
    },
    {
      "content_chunks": 60,
      "controlled_output": null,
      "decode_tok_s": 33.1069328274053,
      "e2e_ms": 9725.26381991338,
      "effective_prefill_tok_s": 4678.643189962914,
      "generation_ms": 8427.237927913666,
      "mean_inter_token_latency_ms": 30.20515386348984,
      "mean_time_per_output_token_ms": 30.20515386348984,
      "output_token_source": "usage",
      "output_tokens": 280,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6073,
      "reasoning_chunks": 209,
      "request_canary": null,
      "request_index": 10,
      "time_to_first_output_ms": 1298.025891999714,
      "tpot_ms": 30.20515386348984,
      "ttft_ms": 7493.334827944636,
      "visible_content_truncated": false,
      "visible_generation_ms": 2231.9289919687435
    },
    {
      "content_chunks": 51,
      "controlled_output": null,
      "decode_tok_s": 28.49406801343568,
      "e2e_ms": 15163.332936004736,
      "effective_prefill_tok_s": 4798.912441714542,
      "generation_ms": 13897.629493032582,
      "mean_inter_token_latency_ms": 35.0950239723045,
      "mean_time_per_output_token_ms": 35.0950239723045,
      "output_token_source": "usage",
      "output_tokens": 397,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6074,
      "reasoning_chunks": 338,
      "request_canary": null,
      "request_index": 11,
      "time_to_first_output_ms": 1265.7034429721534,
      "tpot_ms": 35.0950239723045,
      "ttft_ms": 14042.72162693087,
      "visible_content_truncated": false,
      "visible_generation_ms": 1120.6113090738654
    },
    {
      "content_chunks": 60,
      "controlled_output": null,
      "decode_tok_s": 34.0322073943603,
      "e2e_ms": 7247.170562040992,
      "effective_prefill_tok_s": 4530.944922473851,
      "generation_ms": 5906.169931055047,
      "mean_inter_token_latency_ms": 29.38393000524899,
      "mean_time_per_output_token_ms": 29.38393000524899,
      "output_token_source": "usage",
      "output_tokens": 202,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6076,
      "reasoning_chunks": 138,
      "request_canary": null,
      "request_index": 12,
      "time_to_first_output_ms": 1341.0006309859455,
      "tpot_ms": 29.38393000524899,
      "ttft_ms": 6144.551579025574,
      "visible_content_truncated": false,
      "visible_generation_ms": 1102.618983015418
    },
    {
      "content_chunks": 56,
      "controlled_output": null,
      "decode_tok_s": 32.53015382384405,
      "e2e_ms": 6853.114811005071,
      "effective_prefill_tok_s": 4604.530362501544,
      "generation_ms": 5533.327661920339,
      "mean_inter_token_latency_ms": 30.740709232890772,
      "mean_time_per_output_token_ms": 30.740709232890772,
      "output_token_source": "usage",
      "output_tokens": 181,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6077,
      "reasoning_chunks": 122,
      "request_canary": null,
      "request_index": 13,
      "time_to_first_output_ms": 1319.787149084732,
      "tpot_ms": 30.740709232890772,
      "ttft_ms": 4696.5753780677915,
      "visible_content_truncated": false,
      "visible_generation_ms": 2156.5394329372793
    },
    {
      "content_chunks": 58,
      "controlled_output": null,
      "decode_tok_s": 39.08407277349333,
      "e2e_ms": 5247.489602072164,
      "effective_prefill_tok_s": 4648.634374558567,
      "generation_ms": 3940.2239600894973,
      "mean_inter_token_latency_ms": 25.585869870711022,
      "mean_time_per_output_token_ms": 25.585869870711022,
      "output_token_source": "usage",
      "output_tokens": 155,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6077,
      "reasoning_chunks": 93,
      "request_canary": null,
      "request_index": 14,
      "time_to_first_output_ms": 1307.2656419826671,
      "tpot_ms": 25.585869870711022,
      "ttft_ms": 3051.445860066451,
      "visible_content_truncated": false,
      "visible_generation_ms": 2196.0437420057133
    },
    {
      "content_chunks": 54,
      "controlled_output": null,
      "decode_tok_s": 26.855207394511396,
      "e2e_ms": 10085.828938055784,
      "effective_prefill_tok_s": 4682.727142416416,
      "generation_ms": 8787.867341074161,
      "mean_inter_token_latency_ms": 37.23672602150069,
      "mean_time_per_output_token_ms": 37.23672602150069,
      "output_token_source": "usage",
      "output_tokens": 237,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6078,
      "reasoning_chunks": 180,
      "request_canary": null,
      "request_index": 15,
      "time_to_first_output_ms": 1297.9615969816223,
      "tpot_ms": 37.23672602150069,
      "ttft_ms": 7943.275690078735,
      "visible_content_truncated": false,
      "visible_generation_ms": 2142.553247977048
    },
    {
      "content_chunks": 54,
      "controlled_output": null,
      "decode_tok_s": 36.31373698817776,
      "e2e_ms": 10500.093330978416,
      "effective_prefill_tok_s": 4065.5773929809766,
      "generation_ms": 9004.856760031544,
      "mean_inter_token_latency_ms": 27.537788256977198,
      "mean_time_per_output_token_ms": 27.537788256977198,
      "output_token_source": "usage",
      "output_tokens": 328,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6079,
      "reasoning_chunks": 267,
      "request_canary": null,
      "request_index": 16,
      "time_to_first_output_ms": 1495.2365709468722,
      "tpot_ms": 27.537788256977198,
      "ttft_ms": 9606.17577796802,
      "visible_content_truncated": false,
      "visible_generation_ms": 893.9175530103967
    },
    {
      "content_chunks": 58,
      "controlled_output": null,
      "decode_tok_s": 39.234200537317236,
      "e2e_ms": 6417.003188049421,
      "effective_prefill_tok_s": 2599.4809128377597,
      "generation_ms": 4078.0746850650758,
      "mean_inter_token_latency_ms": 25.487966781656723,
      "mean_time_per_output_token_ms": 25.487966781656723,
      "output_token_source": "usage",
      "output_tokens": 161,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6080,
      "reasoning_chunks": 99,
      "request_canary": null,
      "request_index": 17,
      "time_to_first_output_ms": 2338.928502984345,
      "tpot_ms": 25.487966781656723,
      "ttft_ms": 5333.882624050602,
      "visible_content_truncated": false,
      "visible_generation_ms": 1083.1205639988184
    },
    {
      "content_chunks": 73,
      "controlled_output": null,
      "decode_tok_s": 43.51736623164756,
      "e2e_ms": 6773.798963986337,
      "effective_prefill_tok_s": 4659.244433811926,
      "generation_ms": 5469.080980983563,
      "mean_inter_token_latency_ms": 22.97933185287211,
      "mean_time_per_output_token_ms": 22.97933185287211,
      "output_token_source": "usage",
      "output_tokens": 239,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6079,
      "reasoning_chunks": 160,
      "request_canary": null,
      "request_index": 18,
      "time_to_first_output_ms": 1304.7179830027744,
      "tpot_ms": 22.97933185287211,
      "ttft_ms": 5448.3238810207695,
      "visible_content_truncated": false,
      "visible_generation_ms": 1325.4750829655677
    },
    {
      "content_chunks": 42,
      "controlled_output": null,
      "decode_tok_s": 63.76374921220551,
      "e2e_ms": 5287.3731380095705,
      "effective_prefill_tok_s": 4552.58845676223,
      "generation_ms": 3952.0888139959425,
      "mean_inter_token_latency_ms": 15.682892119031518,
      "mean_time_per_output_token_ms": 15.682892119031518,
      "output_token_source": "usage",
      "output_tokens": 253,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6079,
      "reasoning_chunks": 199,
      "request_canary": null,
      "request_index": 19,
      "time_to_first_output_ms": 1335.284324013628,
      "tpot_ms": 15.682892119031518,
      "ttft_ms": 4734.143526060507,
      "visible_content_truncated": false,
      "visible_generation_ms": 553.2296119490638
    }
  ],
  "requests": 20,
  "response_words": 0,
  "responses_received": 20,
  "run_id": "benchmark-20261017T004650Z",
  "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": "2026-10-17T00:46:50Z",
  "started_at_unix_ns": 1792198010336242511,
  "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": 45882.51958799083
}