{
  "base_url": "redacted://routed-capability",
  "clock_domain_id": null,
  "completed": 20,
  "concurrency": 4,
  "configuration_fingerprint": null,
  "context_distribution": {
    "32768": 20
  },
  "context_seed": 941324,
  "context_tokens": 32768,
  "controlled_output_policy": "observe",
  "engine": "vllm-exl3-v84",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-10-17T00:54:49Z",
  "finished_at_unix_ns": 1792198489177735126,
  "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": 4158.0,
    "completion_tokens_mean": 431.05,
    "completion_tokens_mean_ci95_high": 819.3081012240507,
    "completion_tokens_mean_ci95_low": 42.79189877594928,
    "completion_tokens_min": 98.0,
    "completion_tokens_p25": 148.0,
    "completion_tokens_p50": 208.0,
    "completion_tokens_p75": 319.0,
    "completion_tokens_p90": 403.0,
    "completion_tokens_p95": 618.0,
    "completion_tokens_p99": 4158.0,
    "completion_tokens_samples": 20,
    "completion_tokens_stddev": 885.8892930122162,
    "content_chunks": 1079,
    "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": 53.32078560536342,
    "decode_tok_s_mean": 28.03644629406777,
    "decode_tok_s_mean_ci95_high": 34.365825803337856,
    "decode_tok_s_mean_ci95_low": 21.707066784797682,
    "decode_tok_s_min": 9.332255325484384,
    "decode_tok_s_p25": 17.401111155669156,
    "decode_tok_s_p50": 23.47637160193797,
    "decode_tok_s_p75": 30.550095516515924,
    "decode_tok_s_p90": 52.79695726017829,
    "decode_tok_s_p95": 52.999860918104204,
    "decode_tok_s_p99": 53.32078560536342,
    "decode_tok_s_samples": 20,
    "decode_tok_s_stddev": 14.441757998083848,
    "e2e_max_ms": 129764.50541394297,
    "e2e_mean_ci95_high_ms": 39837.05837104328,
    "e2e_mean_ci95_low_ms": 17897.9470618712,
    "e2e_mean_ms": 28867.50271645724,
    "e2e_min_ms": 12451.198771013878,
    "e2e_p25_ms": 16705.242490046658,
    "e2e_p50_ms": 22222.37262793351,
    "e2e_p75_ms": 27644.7261699941,
    "e2e_p90_ms": 40181.905269040726,
    "e2e_p95_ms": 43077.05162407365,
    "e2e_p99_ms": 129764.50541394297,
    "e2e_samples": 20,
    "e2e_stddev_ms": 25029.257272062845,
    "effective_prefill_tok_s_max": 4184.257379401155,
    "effective_prefill_tok_s_mean": 2132.001412065711,
    "effective_prefill_tok_s_mean_ci95_high": 2396.5080418619023,
    "effective_prefill_tok_s_mean_ci95_low": 1867.4947822695199,
    "effective_prefill_tok_s_min": 1158.9900201955807,
    "effective_prefill_tok_s_p25": 1747.8269593302052,
    "effective_prefill_tok_s_p50": 2052.6471655970163,
    "effective_prefill_tok_s_p75": 2347.2013825432564,
    "effective_prefill_tok_s_p90": 2416.366402506288,
    "effective_prefill_tok_s_p95": 2525.4893358325585,
    "effective_prefill_tok_s_p99": 4184.257379401155,
    "effective_prefill_tok_s_samples": 20,
    "effective_prefill_tok_s_stddev": 603.5253109425053,
    "generation_max_ms": 118735.6421869481,
    "generation_mean_ci95_high_ms": 27216.72954592486,
    "generation_mean_ci95_low_ms": 5193.578691868064,
    "generation_mean_ms": 16205.154118896462,
    "generation_min_ms": 1987.9677089629695,
    "generation_p25_ms": 5574.35666793026,
    "generation_p50_ms": 8681.484902044758,
    "generation_p75_ms": 16180.440283031203,
    "generation_p90_ms": 19640.96206589602,
    "generation_p95_ms": 30759.299005032517,
    "generation_p99_ms": 118735.6421869481,
    "generation_samples": 20,
    "generation_stddev_ms": 25125.133871634465,
    "mean_inter_token_latency_ms_max": 107.1552336624583,
    "mean_inter_token_latency_ms_mean": 45.95168289072027,
    "mean_inter_token_latency_ms_mean_ci95_high": 56.639881498181055,
    "mean_inter_token_latency_ms_mean_ci95_low": 35.26348428325949,
    "mean_inter_token_latency_ms_min": 18.75441234870726,
    "mean_inter_token_latency_ms_p25": 28.56281986695889,
    "mean_inter_token_latency_ms_p50": 41.939540589588205,
    "mean_inter_token_latency_ms_p75": 51.4659950340192,
    "mean_inter_token_latency_ms_p90": 81.9636540262237,
    "mean_inter_token_latency_ms_p95": 86.64191035358323,
    "mean_inter_token_latency_ms_p99": 107.1552336624583,
    "mean_inter_token_latency_ms_samples": 20,
    "mean_inter_token_latency_ms_stddev": 24.38728432989888,
    "mean_time_per_output_token_max_ms": 107.1552336624583,
    "mean_time_per_output_token_mean_ci95_high_ms": 56.639881498181055,
    "mean_time_per_output_token_mean_ci95_low_ms": 35.26348428325949,
    "mean_time_per_output_token_mean_ms": 45.95168289072027,
    "mean_time_per_output_token_min_ms": 18.75441234870726,
    "mean_time_per_output_token_p25_ms": 28.56281986695889,
    "mean_time_per_output_token_p50_ms": 41.939540589588205,
    "mean_time_per_output_token_p75_ms": 51.4659950340192,
    "mean_time_per_output_token_p90_ms": 81.9636540262237,
    "mean_time_per_output_token_p95_ms": 86.64191035358323,
    "mean_time_per_output_token_p99_ms": 107.1552336624583,
    "mean_time_per_output_token_samples": 20,
    "mean_time_per_output_token_stddev_ms": 24.38728432989888,
    "output_token_sources": {
      "usage": 20
    },
    "output_tokens": 8621,
    "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": 7252,
    "request_canary_passed": 0,
    "request_canary_samples": 0,
    "throughput_performance_eligible": true,
    "throughput_tok_s": 48.760151045486914,
    "time_to_first_output_max_ms": 21813.82027408108,
    "time_to_first_output_mean_ci95_high_ms": 14135.446873110252,
    "time_to_first_output_mean_ci95_low_ms": 11189.250322011301,
    "time_to_first_output_mean_ms": 12662.348597560776,
    "time_to_first_output_min_ms": 6041.932344902307,
    "time_to_first_output_p25_ms": 10750.911605078727,
    "time_to_first_output_p50_ms": 12010.134132928215,
    "time_to_first_output_p75_ms": 13161.536928033456,
    "time_to_first_output_p90_ms": 15221.36882704217,
    "time_to_first_output_p95_ms": 19082.92610105127,
    "time_to_first_output_p99_ms": 21813.82027408108,
    "time_to_first_output_samples": 20,
    "time_to_first_output_stddev_ms": 3361.171307822797,
    "tpot_max_ms": 107.1552336624583,
    "tpot_mean_ci95_high_ms": 56.639881498181055,
    "tpot_mean_ci95_low_ms": 35.26348428325949,
    "tpot_mean_ms": 45.95168289072027,
    "tpot_min_ms": 18.75441234870726,
    "tpot_p25_ms": 28.56281986695889,
    "tpot_p50_ms": 41.939540589588205,
    "tpot_p75_ms": 51.4659950340192,
    "tpot_p90_ms": 81.9636540262237,
    "tpot_p95_ms": 86.64191035358323,
    "tpot_p99_ms": 107.1552336624583,
    "tpot_samples": 20,
    "tpot_stddev_ms": 24.38728432989888,
    "ttft_max_ms": 128863.96210291423,
    "ttft_mean_ci95_high_ms": 37724.38727725747,
    "ttft_mean_ci95_low_ms": 15573.759433834508,
    "ttft_mean_ms": 26649.07335554599,
    "ttft_min_ms": 11634.730325080454,
    "ttft_p25_ms": 15658.205341082066,
    "ttft_p50_ms": 20243.38181805797,
    "ttft_p75_ms": 26748.817190062255,
    "ttft_p90_ms": 34117.96212999616,
    "ttft_p95_ms": 42156.62638202775,
    "ttft_p99_ms": 128863.96210291423,
    "ttft_samples": 20,
    "ttft_stddev_ms": 25270.566123568016
  },
  "model": "llm.primary",
  "performance_eligible": true,
  "prompt_cache_mode": "unique",
  "request_canaries": false,
  "request_timings": [
    {
      "content_chunks": 39,
      "controlled_output": null,
      "decode_tok_s": 9.332255325484384,
      "e2e_ms": 22222.37262793351,
      "effective_prefill_tok_s": 4184.257379401155,
      "generation_ms": 16180.440283031203,
      "mean_inter_token_latency_ms": 107.1552336624583,
      "mean_time_per_output_token_ms": 107.1552336624583,
      "output_token_source": "usage",
      "output_tokens": 152,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25281,
      "reasoning_chunks": 110,
      "request_canary": null,
      "request_index": 0,
      "time_to_first_output_ms": 6041.932344902307,
      "tpot_ms": 107.1552336624583,
      "ttft_ms": 16667.002298985608,
      "visible_content_truncated": false,
      "visible_generation_ms": 5555.370328947902
    },
    {
      "content_chunks": 67,
      "controlled_output": null,
      "decode_tok_s": 11.541758439062894,
      "e2e_ms": 40181.905269040726,
      "effective_prefill_tok_s": 1158.9900201955807,
      "generation_ms": 18368.084994959645,
      "mean_inter_token_latency_ms": 86.64191035358323,
      "mean_time_per_output_token_ms": 86.64191035358323,
      "output_token_source": "usage",
      "output_tokens": 213,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25282,
      "reasoning_chunks": 140,
      "request_canary": null,
      "request_index": 1,
      "time_to_first_output_ms": 21813.82027408108,
      "tpot_ms": 86.64191035358323,
      "ttft_ms": 34117.96212999616,
      "visible_content_truncated": false,
      "visible_generation_ms": 6063.943139044568
    },
    {
      "content_chunks": 45,
      "controlled_output": null,
      "decode_tok_s": 12.200529757739414,
      "e2e_ms": 27504.964138031937,
      "effective_prefill_tok_s": 1680.2591262528963,
      "generation_ms": 12458.475411986,
      "mean_inter_token_latency_ms": 81.9636540262237,
      "mean_time_per_output_token_ms": 81.9636540262237,
      "output_token_source": "usage",
      "output_tokens": 153,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25282,
      "reasoning_chunks": 102,
      "request_canary": null,
      "request_index": 2,
      "time_to_first_output_ms": 15046.488726045936,
      "tpot_ms": 81.9636540262237,
      "ttft_ms": 21838.035342050716,
      "visible_content_truncated": false,
      "visible_generation_ms": 5666.928795981221
    },
    {
      "content_chunks": 44,
      "controlled_output": null,
      "decode_tok_s": 17.401111155669156,
      "e2e_ms": 16629.41253988538,
      "effective_prefill_tok_s": 2286.826976979265,
      "generation_ms": 5574.35666793026,
      "mean_inter_token_latency_ms": 57.46759451474495,
      "mean_time_per_output_token_ms": 57.46759451474495,
      "output_token_source": "usage",
      "output_tokens": 98,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25281,
      "reasoning_chunks": 51,
      "request_canary": null,
      "request_index": 3,
      "time_to_first_output_ms": 11055.055871955119,
      "tpot_ms": 57.46759451474495,
      "ttft_ms": 15799.547603935935,
      "visible_content_truncated": false,
      "visible_generation_ms": 829.8649359494448
    },
    {
      "content_chunks": 46,
      "controlled_output": null,
      "decode_tok_s": 14.255920817961542,
      "e2e_ms": 30391.873670974746,
      "effective_prefill_tok_s": 2351.614535464769,
      "generation_ms": 19640.96206589602,
      "mean_inter_token_latency_ms": 70.14629309248578,
      "mean_time_per_output_token_ms": 70.14629309248578,
      "output_token_source": "usage",
      "output_tokens": 281,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25282,
      "reasoning_chunks": 230,
      "request_canary": null,
      "request_index": 4,
      "time_to_first_output_ms": 10750.911605078727,
      "tpot_ms": 70.14629309248578,
      "ttft_ms": 29510.276112006977,
      "visible_content_truncated": false,
      "visible_generation_ms": 881.5975589677691
    },
    {
      "content_chunks": 43,
      "controlled_output": null,
      "decode_tok_s": 53.32078560536342,
      "e2e_ms": 12451.198771013878,
      "effective_prefill_tok_s": 2416.366402506288,
      "generation_ms": 1987.9677089629695,
      "mean_inter_token_latency_ms": 18.75441234870726,
      "mean_time_per_output_token_ms": 18.75441234870726,
      "output_token_source": "usage",
      "output_tokens": 107,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25283,
      "reasoning_chunks": 60,
      "request_canary": null,
      "request_index": 5,
      "time_to_first_output_ms": 10463.231062050909,
      "tpot_ms": 18.75441234870726,
      "ttft_ms": 11634.730325080454,
      "visible_content_truncated": false,
      "visible_generation_ms": 816.4684459334239
    },
    {
      "content_chunks": 48,
      "controlled_output": null,
      "decode_tok_s": 20.05897468271474,
      "e2e_ms": 43077.05162407365,
      "effective_prefill_tok_s": 2052.6471655970163,
      "generation_ms": 30759.299005032517,
      "mean_inter_token_latency_ms": 49.85299676666534,
      "mean_time_per_output_token_ms": 49.85299676666534,
      "output_token_source": "usage",
      "output_tokens": 618,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25284,
      "reasoning_chunks": 549,
      "request_canary": null,
      "request_index": 6,
      "time_to_first_output_ms": 12317.75261904113,
      "tpot_ms": 49.85299676666534,
      "ttft_ms": 42156.62638202775,
      "visible_content_truncated": false,
      "visible_generation_ms": 920.4252420458943
    },
    {
      "content_chunks": 69,
      "controlled_output": null,
      "decode_tok_s": 19.430305376180844,
      "e2e_ms": 18223.64609106444,
      "effective_prefill_tok_s": 2372.364086292901,
      "generation_ms": 7565.501270000823,
      "mean_inter_token_latency_ms": 51.4659950340192,
      "mean_time_per_output_token_ms": 51.4659950340192,
      "output_token_source": "usage",
      "output_tokens": 148,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25285,
      "reasoning_chunks": 75,
      "request_canary": null,
      "request_index": 7,
      "time_to_first_output_ms": 10658.144821063615,
      "tpot_ms": 51.4659950340192,
      "ttft_ms": 12119.2340409616,
      "visible_content_truncated": false,
      "visible_generation_ms": 6104.412050102837
    },
    {
      "content_chunks": 67,
      "controlled_output": null,
      "decode_tok_s": 29.514103237196103,
      "e2e_ms": 22784.644116996787,
      "effective_prefill_tok_s": 2105.305379619038,
      "generation_ms": 10774.509984068573,
      "mean_inter_token_latency_ms": 33.88210686814016,
      "mean_time_per_output_token_ms": 33.88210686814016,
      "output_token_source": "usage",
      "output_tokens": 319,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25285,
      "reasoning_chunks": 245,
      "request_canary": null,
      "request_index": 8,
      "time_to_first_output_ms": 12010.134132928215,
      "tpot_ms": 33.88210686814016,
      "ttft_ms": 21502.49453599099,
      "visible_content_truncated": false,
      "visible_generation_ms": 1282.1495810057968
    },
    {
      "content_chunks": 61,
      "controlled_output": null,
      "decode_tok_s": 35.010548841390396,
      "e2e_ms": 129764.50541394297,
      "effective_prefill_tok_s": 2292.6206880606696,
      "generation_ms": 118735.6421869481,
      "mean_inter_token_latency_ms": 28.56281986695889,
      "mean_time_per_output_token_ms": 28.56281986695889,
      "output_token_source": "usage",
      "output_tokens": 4158,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25285,
      "reasoning_chunks": 3923,
      "request_canary": null,
      "request_index": 9,
      "time_to_first_output_ms": 11028.863226994872,
      "tpot_ms": 28.56281986695889,
      "ttft_ms": 128863.96210291423,
      "visible_content_truncated": false,
      "visible_generation_ms": 900.5433110287413
    },
    {
      "content_chunks": 55,
      "controlled_output": null,
      "decode_tok_s": 21.589724823995738,
      "e2e_ms": 23280.75825504493,
      "effective_prefill_tok_s": 1661.2172195103165,
      "generation_ms": 8059.38942800276,
      "mean_inter_token_latency_ms": 46.31833004599287,
      "mean_time_per_output_token_ms": 46.31833004599287,
      "output_token_source": "usage",
      "output_tokens": 175,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25286,
      "reasoning_chunks": 115,
      "request_canary": null,
      "request_index": 10,
      "time_to_first_output_ms": 15221.36882704217,
      "tpot_ms": 46.31833004599287,
      "ttft_ms": 17451.097680954263,
      "visible_content_truncated": false,
      "visible_generation_ms": 5829.660574090667
    },
    {
      "content_chunks": 68,
      "controlled_output": null,
      "decode_tok_s": 23.40814745504389,
      "e2e_ms": 29943.234015023336,
      "effective_prefill_tok_s": 1980.230376802461,
      "generation_ms": 17173.507675994188,
      "mean_inter_token_latency_ms": 42.72016834824425,
      "mean_time_per_output_token_ms": 42.72016834824425,
      "output_token_source": "usage",
      "output_tokens": 403,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25287,
      "reasoning_chunks": 325,
      "request_canary": null,
      "request_index": 11,
      "time_to_first_output_ms": 12769.726339029148,
      "tpot_ms": 42.72016834824425,
      "ttft_ms": 28624.08246099949,
      "visible_content_truncated": false,
      "visible_generation_ms": 1319.151554023847
    },
    {
      "content_chunks": 49,
      "controlled_output": null,
      "decode_tok_s": 52.999860918104204,
      "e2e_ms": 14886.47375407163,
      "effective_prefill_tok_s": 2347.2013825432564,
      "generation_ms": 4113.21834102273,
      "mean_inter_token_latency_ms": 18.86797404138867,
      "mean_time_per_output_token_ms": 18.86797404138867,
      "output_token_source": "usage",
      "output_tokens": 219,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25287,
      "reasoning_chunks": 166,
      "request_canary": null,
      "request_index": 12,
      "time_to_first_output_ms": 10773.2554130489,
      "tpot_ms": 18.86797404138867,
      "ttft_ms": 13937.879542005248,
      "visible_content_truncated": false,
      "visible_generation_ms": 948.5942120663822
    },
    {
      "content_chunks": 54,
      "controlled_output": null,
      "decode_tok_s": 52.722460923782705,
      "e2e_ms": 16705.242490046658,
      "effective_prefill_tok_s": 1747.8269593302052,
      "generation_ms": 2238.1352829979733,
      "mean_inter_token_latency_ms": 18.967248160999773,
      "mean_time_per_output_token_ms": 18.967248160999773,
      "output_token_source": "usage",
      "output_tokens": 119,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25286,
      "reasoning_chunks": 62,
      "request_canary": null,
      "request_index": 13,
      "time_to_first_output_ms": 14467.107207048684,
      "tpot_ms": 18.967248160999773,
      "ttft_ms": 15658.205341082066,
      "visible_content_truncated": false,
      "visible_generation_ms": 1047.0371489645913
    },
    {
      "content_chunks": 53,
      "controlled_output": null,
      "decode_tok_s": 27.034767949200056,
      "e2e_ms": 21373.184996074997,
      "effective_prefill_tok_s": 1921.2041981314512,
      "generation_ms": 8211.64806804154,
      "mean_inter_token_latency_ms": 36.98940571189883,
      "mean_time_per_output_token_ms": 36.98940571189883,
      "output_token_source": "usage",
      "output_tokens": 223,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25286,
      "reasoning_chunks": 161,
      "request_canary": null,
      "request_index": 14,
      "time_to_first_output_ms": 13161.536928033456,
      "tpot_ms": 36.98940571189883,
      "ttft_ms": 20348.15921401605,
      "visible_content_truncated": false,
      "visible_generation_ms": 1025.0257820589468
    },
    {
      "content_chunks": 52,
      "controlled_output": null,
      "decode_tok_s": 30.550095516515924,
      "e2e_ms": 21239.777880022302,
      "effective_prefill_tok_s": 2525.4893358325585,
      "generation_ms": 11227.460804977454,
      "mean_inter_token_latency_ms": 32.73312188040074,
      "mean_time_per_output_token_ms": 32.73312188040074,
      "output_token_source": "usage",
      "output_tokens": 344,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25286,
      "reasoning_chunks": 284,
      "request_canary": null,
      "request_index": 15,
      "time_to_first_output_ms": 10012.317075044848,
      "tpot_ms": 32.73312188040074,
      "ttft_ms": 20243.38181805797,
      "visible_content_truncated": false,
      "visible_generation_ms": 996.3960619643331
    },
    {
      "content_chunks": 47,
      "controlled_output": null,
      "decode_tok_s": 23.47637160193797,
      "e2e_ms": 27644.7261699941,
      "effective_prefill_tok_s": 1325.111246886134,
      "generation_ms": 8561.80006894283,
      "mean_inter_token_latency_ms": 42.59602024349667,
      "mean_time_per_output_token_ms": 42.59602024349667,
      "output_token_source": "usage",
      "output_tokens": 202,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25287,
      "reasoning_chunks": 149,
      "request_canary": null,
      "request_index": 16,
      "time_to_first_output_ms": 19082.92610105127,
      "tpot_ms": 42.59602024349667,
      "ttft_ms": 26748.817190062255,
      "visible_content_truncated": false,
      "visible_generation_ms": 895.9089799318463
    },
    {
      "content_chunks": 46,
      "controlled_output": null,
      "decode_tok_s": 23.843847260650662,
      "e2e_ms": 21109.503448940814,
      "effective_prefill_tok_s": 2034.596255596615,
      "generation_ms": 8681.484902044758,
      "mean_inter_token_latency_ms": 41.939540589588205,
      "mean_time_per_output_token_ms": 41.939540589588205,
      "output_token_source": "usage",
      "output_tokens": 208,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25286,
      "reasoning_chunks": 158,
      "request_canary": null,
      "request_index": 17,
      "time_to_first_output_ms": 12428.018546896055,
      "tpot_ms": 41.939540589588205,
      "ttft_ms": 20219.148436910473,
      "visible_content_truncated": false,
      "visible_generation_ms": 890.3550120303407
    },
    {
      "content_chunks": 64,
      "controlled_output": null,
      "decode_tok_s": 30.240398933183013,
      "e2e_ms": 22641.058138920926,
      "effective_prefill_tok_s": 2180.62243901754,
      "generation_ms": 11044.827839010395,
      "mean_inter_token_latency_ms": 33.06834682338442,
      "mean_time_per_output_token_ms": 33.06834682338442,
      "output_token_source": "usage",
      "output_tokens": 335,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25287,
      "reasoning_chunks": 267,
      "request_canary": null,
      "request_index": 18,
      "time_to_first_output_ms": 11596.23029991053,
      "tpot_ms": 33.06834682338442,
      "ttft_ms": 21439.8572439095,
      "visible_content_truncated": false,
      "visible_generation_ms": 1201.200895011425
    },
    {
      "content_chunks": 62,
      "controlled_output": null,
      "decode_tok_s": 52.79695726017829,
      "e2e_ms": 15294.520918047056,
      "effective_prefill_tok_s": 2015.2770672941056,
      "generation_ms": 2746.3703880785033,
      "mean_inter_token_latency_ms": 18.94048543502416,
      "mean_time_per_output_token_ms": 18.94048543502416,
      "output_token_source": "usage",
      "output_tokens": 146,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25288,
      "reasoning_chunks": 80,
      "request_canary": null,
      "request_index": 19,
      "time_to_first_output_ms": 12548.150529968552,
      "tpot_ms": 18.94048543502416,
      "ttft_ms": 14100.96730897203,
      "visible_content_truncated": false,
      "visible_generation_ms": 1193.5536090750247
    }
  ],
  "requests": 20,
  "response_words": 0,
  "responses_received": 20,
  "run_id": "benchmark-20261017T005152Z",
  "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": "2026-10-17T00:51:52Z",
  "started_at_unix_ns": 1792198312373518682,
  "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": 176804.21030602884
}