{
  "clock_domain_id": null,
  "completed": 15,
  "concurrency": 8,
  "configuration_fingerprint": null,
  "context_distribution": {
    "32768": 16
  },
  "context_seed": 2801,
  "context_tokens": 32768,
  "controlled_output_policy": "strict",
  "engine": "vllm-exl3-v84",
  "failed": 1,
  "failures": [
    {
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": false,
        "observed_code_words": 3,
        "observed_extra_words": 150,
        "passed": false,
        "policy": "strict",
        "requested_words": 16
      },
      "elapsed_ms": 47932.64711700613,
      "error_message": "request canary validation failed",
      "error_message_truncated": false,
      "error_type": "ValueError",
      "failure_class": "request_canary_failed",
      "performance_eligible": false,
      "planned_context_tokens": 32768,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_2801_00015",
        "marker_at_start": false,
        "observed_marker_count": 0,
        "passed": false
      },
      "request_index": 15,
      "response_metadata": {
        "content_chunks": 217,
        "finish_reasons": [
          "stop"
        ],
        "output_token_source": "usage",
        "reasoning_chunks": 476,
        "stream_terminal_observed": true,
        "usage": {
          "completion_tokens": 697,
          "prompt_tokens": 25284,
          "total_tokens": 25981
        },
        "visible_content_truncated": false
      },
      "response_received": true,
      "response_timing": {
        "content_chunks": 217,
        "controlled_output": {
          "capture_complete": true,
          "exact_adherence": false,
          "observed_code_words": 3,
          "observed_extra_words": 150,
          "passed": false,
          "policy": "strict",
          "requested_words": 16
        },
        "decode_tok_s": 74.35084551591208,
        "e2e_ms": 47932.23928500083,
        "effective_prefill_tok_s": 655.5147246092794,
        "generation_ms": 9361.023336998187,
        "mean_inter_token_latency_ms": 13.44974617384797,
        "mean_time_per_output_token_ms": 13.44974617384797,
        "output_token_source": "usage",
        "output_tokens": 697,
        "planned_context_tokens": 32768,
        "prompt_tokens": 25284,
        "reasoning_chunks": 476,
        "request_canary": {
          "capture_complete": true,
          "foreign_marker_count": 0,
          "marker": "ANVIL_REQ_2801_00015",
          "marker_at_start": false,
          "observed_marker_count": 0,
          "passed": false
        },
        "request_index": 15,
        "time_to_first_output_ms": 38571.215948002646,
        "tpot_ms": 13.44974617384797,
        "ttft_ms": 45140.4202220001,
        "visible_content_truncated": false,
        "visible_generation_ms": 2791.8190630007302
      }
    }
  ],
  "finished_at": "2026-09-24T00:53:58Z",
  "finished_at_unix_ns": 1790211238489013630,
  "gpu": "dual-pro6000-maxq",
  "manifest": null,
  "max_context_tokens": 327680,
  "max_tokens": 8192,
  "measurement_protocol": "capacity-v4-reasoning",
  "metric_population": {
    "aggregate_throughput_eligible": false,
    "attempted_requests": 16,
    "excluded_requests": 1,
    "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": 15,
      "e2e": 15,
      "effective_prefill_tok_s": 15,
      "tpot": 15,
      "ttft": 15
    },
    "performance_eligible_requests": 15,
    "successful_requests": 15
  },
  "metrics": {
    "completion_tokens_max": 248.0,
    "completion_tokens_mean": 142.26666666666668,
    "completion_tokens_mean_ci95_high": 169.8446309457388,
    "completion_tokens_mean_ci95_low": 114.68870238759456,
    "completion_tokens_min": 70.0,
    "completion_tokens_p25": 117.0,
    "completion_tokens_p50": 126.0,
    "completion_tokens_p75": 178.0,
    "completion_tokens_p90": 236.0,
    "completion_tokens_p95": 248.0,
    "completion_tokens_p99": 248.0,
    "completion_tokens_samples": 15,
    "completion_tokens_stddev": 54.494385905688446,
    "content_chunks": 390,
    "content_chunks_s": null,
    "controlled_output_exact_adherence": 15,
    "controlled_output_exact_adherence_rate": 1.0,
    "controlled_output_samples": 15,
    "controlled_output_unobservable": 0,
    "decode_tok_s_max": 26.39724845289074,
    "decode_tok_s_mean": 12.400222549653277,
    "decode_tok_s_mean_ci95_high": 15.514906314301708,
    "decode_tok_s_mean_ci95_low": 9.285538785004846,
    "decode_tok_s_min": 4.644252882797655,
    "decode_tok_s_p25": 8.00197458488931,
    "decode_tok_s_p50": 10.265791293554306,
    "decode_tok_s_p75": 16.66089904026122,
    "decode_tok_s_p90": 21.576411392955357,
    "decode_tok_s_p95": 26.39724845289074,
    "decode_tok_s_p99": 26.39724845289074,
    "decode_tok_s_samples": 15,
    "decode_tok_s_stddev": 6.154652218972449,
    "e2e_max_ms": 56302.18415000127,
    "e2e_mean_ci95_high_ms": 45268.293179813,
    "e2e_mean_ci95_low_ms": 34605.917907384195,
    "e2e_mean_ms": 39937.1055435986,
    "e2e_min_ms": 20773.16264699766,
    "e2e_p25_ms": 31434.396501004812,
    "e2e_p50_ms": 40229.41991399421,
    "e2e_p75_ms": 49724.865858996054,
    "e2e_p90_ms": 53535.65713900025,
    "e2e_p95_ms": 56302.18415000127,
    "e2e_p99_ms": 56302.18415000127,
    "e2e_samples": 15,
    "e2e_stddev_ms": 10534.49027069656,
    "effective_prefill_tok_s_max": 4273.261178686488,
    "effective_prefill_tok_s_mean": 1225.641795255502,
    "effective_prefill_tok_s_mean_ci95_high": 1703.1515384465142,
    "effective_prefill_tok_s_mean_ci95_low": 748.1320520644899,
    "effective_prefill_tok_s_min": 595.646423943746,
    "effective_prefill_tok_s_p25": 758.9720346588722,
    "effective_prefill_tok_s_p50": 915.6285255711504,
    "effective_prefill_tok_s_p75": 1252.2088174095845,
    "effective_prefill_tok_s_p90": 2264.1251555330973,
    "effective_prefill_tok_s_p95": 4273.261178686488,
    "effective_prefill_tok_s_p99": 4273.261178686488,
    "effective_prefill_tok_s_samples": 15,
    "effective_prefill_tok_s_stddev": 943.5649403217196,
    "generation_max_ms": 28069.83387400396,
    "generation_mean_ci95_high_ms": 16461.63147922735,
    "generation_mean_ci95_low_ms": 10011.46762423883,
    "generation_mean_ms": 13236.549551733091,
    "generation_min_ms": 4201.453945002868,
    "generation_p25_ms": 8203.403095001704,
    "generation_p50_ms": 12150.833421997959,
    "generation_p75_ms": 16653.91044699936,
    "generation_p90_ms": 22835.85875100107,
    "generation_p95_ms": 28069.83387400396,
    "generation_p99_ms": 28069.83387400396,
    "generation_samples": 15,
    "generation_stddev_ms": 6372.800303744789,
    "mean_inter_token_latency_ms_max": 215.31988572457087,
    "mean_inter_token_latency_ms_mean": 98.7134352876961,
    "mean_inter_token_latency_ms_mean_ci95_high": 121.38520491887384,
    "mean_inter_token_latency_ms_mean_ci95_low": 76.04166565651836,
    "mean_inter_token_latency_ms_min": 37.88273621717156,
    "mean_inter_token_latency_ms_p25": 60.020770642898114,
    "mean_inter_token_latency_ms_p50": 97.41090300831274,
    "mean_inter_token_latency_ms_p75": 124.96915472443142,
    "mean_inter_token_latency_ms_p90": 135.36976883199532,
    "mean_inter_token_latency_ms_p95": 215.31988572457087,
    "mean_inter_token_latency_ms_p99": 215.31988572457087,
    "mean_inter_token_latency_ms_samples": 15,
    "mean_inter_token_latency_ms_stddev": 44.799686842144105,
    "mean_time_per_output_token_max_ms": 215.31988572457087,
    "mean_time_per_output_token_mean_ci95_high_ms": 121.38520491887384,
    "mean_time_per_output_token_mean_ci95_low_ms": 76.04166565651836,
    "mean_time_per_output_token_mean_ms": 98.7134352876961,
    "mean_time_per_output_token_min_ms": 37.88273621717156,
    "mean_time_per_output_token_p25_ms": 60.020770642898114,
    "mean_time_per_output_token_p50_ms": 97.41090300831274,
    "mean_time_per_output_token_p75_ms": 124.96915472443142,
    "mean_time_per_output_token_p90_ms": 135.36976883199532,
    "mean_time_per_output_token_p95_ms": 215.31988572457087,
    "mean_time_per_output_token_p99_ms": 215.31988572457087,
    "mean_time_per_output_token_samples": 15,
    "mean_time_per_output_token_stddev_ms": 44.799686842144105,
    "output_token_sources": {
      "usage": 15
    },
    "output_tokens": 2134,
    "prefix_cache_hit_avg": null,
    "prompt_token_samples": 15,
    "prompt_tokens": 379235,
    "prompt_tokens_max": 25284.0,
    "prompt_tokens_mean": 25282.333333333332,
    "prompt_tokens_mean_ci95_high": 25282.928035873356,
    "prompt_tokens_mean_ci95_low": 25281.73863079331,
    "prompt_tokens_min": 25280.0,
    "prompt_tokens_p25": 25281.0,
    "prompt_tokens_p50": 25283.0,
    "prompt_tokens_p75": 25283.0,
    "prompt_tokens_p90": 25284.0,
    "prompt_tokens_p95": 25284.0,
    "prompt_tokens_p99": 25284.0,
    "prompt_tokens_samples": 15,
    "prompt_tokens_stddev": 1.1751393027860062,
    "reasoning_chunks": 1594,
    "request_canary_passed": 15,
    "request_canary_samples": 15,
    "throughput_performance_eligible": false,
    "throughput_tok_s": null,
    "time_to_first_output_max_ms": 42446.32215300226,
    "time_to_first_output_mean_ci95_high_ms": 31697.904225915838,
    "time_to_first_output_mean_ci95_low_ms": 21703.207757815166,
    "time_to_first_output_mean_ms": 26700.5559918655,
    "time_to_first_output_min_ms": 5916.090532002272,
    "time_to_first_output_p25_ms": 20188.32613900304,
    "time_to_first_output_p50_ms": 27612.726442996063,
    "time_to_first_output_p75_ms": 33313.48039900331,
    "time_to_first_output_p90_ms": 38328.15711799776,
    "time_to_first_output_p95_ms": 42446.32215300226,
    "time_to_first_output_p99_ms": 42446.32215300226,
    "time_to_first_output_samples": 15,
    "time_to_first_output_stddev_ms": 9874.819635548976,
    "tpot_max_ms": 215.31988572457087,
    "tpot_mean_ci95_high_ms": 121.38520491887384,
    "tpot_mean_ci95_low_ms": 76.04166565651836,
    "tpot_mean_ms": 98.7134352876961,
    "tpot_min_ms": 37.88273621717156,
    "tpot_p25_ms": 60.020770642898114,
    "tpot_p50_ms": 97.41090300831274,
    "tpot_p75_ms": 124.96915472443142,
    "tpot_p90_ms": 135.36976883199532,
    "tpot_p95_ms": 215.31988572457087,
    "tpot_p99_ms": 215.31988572457087,
    "tpot_samples": 15,
    "tpot_stddev_ms": 44.799686842144105,
    "ttft_max_ms": 52821.59195699933,
    "ttft_mean_ci95_high_ms": 43739.55814664175,
    "ttft_mean_ci95_low_ms": 33022.51467042274,
    "ttft_mean_ms": 38381.03640853224,
    "ttft_min_ms": 20292.45143099979,
    "ttft_p25_ms": 26794.5232580023,
    "ttft_p50_ms": 39757.57481799519,
    "ttft_p75_ms": 48894.95666199946,
    "ttft_p90_ms": 50444.17855500069,
    "ttft_p95_ms": 52821.59195699933,
    "ttft_p99_ms": 52821.59195699933,
    "ttft_samples": 15,
    "ttft_stddev_ms": 10588.50278162681
  },
  "model": "glm53-flash-exl3-runtime-dcp1-candidate",
  "performance_eligible": false,
  "prompt_cache_mode": "unique",
  "request_canaries": true,
  "request_timings": [
    {
      "content_chunks": 26,
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": true,
        "observed_code_words": 16,
        "observed_extra_words": 0,
        "passed": true,
        "policy": "strict",
        "requested_words": 16
      },
      "decode_tok_s": 7.387173728877973,
      "e2e_ms": 42958.70574899891,
      "effective_prefill_tok_s": 970.9463239128678,
      "generation_ms": 16921.221103999414,
      "mean_inter_token_latency_ms": 135.36976883199532,
      "mean_time_per_output_token_ms": 135.36976883199532,
      "output_token_source": "usage",
      "output_tokens": 126,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25281,
      "reasoning_chunks": 95,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_2801_00000",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 0,
      "time_to_first_output_ms": 26037.484644999495,
      "tpot_ms": 135.36976883199532,
      "ttft_ms": 42477.62761199556,
      "visible_content_truncated": false,
      "visible_generation_ms": 481.078137003351
    },
    {
      "content_chunks": 26,
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": true,
        "observed_code_words": 16,
        "observed_extra_words": 0,
        "passed": true,
        "policy": "strict",
        "requested_words": 16
      },
      "decode_tok_s": 10.265791293554306,
      "e2e_ms": 50017.46547899529,
      "effective_prefill_tok_s": 659.6195043285378,
      "generation_ms": 11689.30836099753,
      "mean_inter_token_latency_ms": 97.41090300831274,
      "mean_time_per_output_token_ms": 97.41090300831274,
      "output_token_source": "usage",
      "output_tokens": 121,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25282,
      "reasoning_chunks": 90,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_2801_00001",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 1,
      "time_to_first_output_ms": 38328.15711799776,
      "tpot_ms": 97.41090300831274,
      "ttft_ms": 48915.20876199502,
      "visible_content_truncated": false,
      "visible_generation_ms": 1102.256717000273
    },
    {
      "content_chunks": 26,
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": true,
        "observed_code_words": 16,
        "observed_extra_words": 0,
        "passed": true,
        "policy": "strict",
        "requested_words": 16
      },
      "decode_tok_s": 8.00197458488931,
      "e2e_ms": 27037.868168998102,
      "effective_prefill_tok_s": 2264.1251555330973,
      "generation_ms": 15871.08265000279,
      "mean_inter_token_latency_ms": 124.96915472443142,
      "mean_time_per_output_token_ms": 124.96915472443142,
      "output_token_source": "usage",
      "output_tokens": 128,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25283,
      "reasoning_chunks": 93,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_2801_00002",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 2,
      "time_to_first_output_ms": 11166.785518995312,
      "tpot_ms": 124.96915472443142,
      "ttft_ms": 26559.603043999232,
      "visible_content_truncated": false,
      "visible_generation_ms": 478.2651249988703
    },
    {
      "content_chunks": 26,
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": true,
        "observed_code_words": 16,
        "observed_extra_words": 0,
        "passed": true,
        "policy": "strict",
        "requested_words": 16
      },
      "decode_tok_s": 8.94938610706,
      "e2e_ms": 29821.396195002308,
      "effective_prefill_tok_s": 1583.5825812960318,
      "generation_ms": 13855.698985003983,
      "mean_inter_token_latency_ms": 111.7395079435805,
      "mean_time_per_output_token_ms": 111.7395079435805,
      "output_token_source": "usage",
      "output_tokens": 125,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25283,
      "reasoning_chunks": 93,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_2801_00003",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 3,
      "time_to_first_output_ms": 15965.697209998325,
      "tpot_ms": 111.7395079435805,
      "ttft_ms": 26724.251785999513,
      "visible_content_truncated": false,
      "visible_generation_ms": 3097.1444090027944
    },
    {
      "content_chunks": 26,
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": true,
        "observed_code_words": 16,
        "observed_extra_words": 0,
        "passed": true,
        "policy": "strict",
        "requested_words": 16
      },
      "decode_tok_s": 14.70074294008659,
      "e2e_ms": 49724.865858996054,
      "effective_prefill_tok_s": 595.646423943746,
      "generation_ms": 7278.543705993798,
      "mean_inter_token_latency_ms": 68.02377295321307,
      "mean_time_per_output_token_ms": 68.02377295321307,
      "output_token_source": "usage",
      "output_tokens": 108,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25283,
      "reasoning_chunks": 77,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_2801_00004",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 4,
      "time_to_first_output_ms": 42446.32215300226,
      "tpot_ms": 68.02377295321307,
      "ttft_ms": 48894.95666199946,
      "visible_content_truncated": false,
      "visible_generation_ms": 829.9091969965957
    },
    {
      "content_chunks": 26,
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": true,
        "observed_code_words": 16,
        "observed_extra_words": 0,
        "passed": true,
        "policy": "strict",
        "requested_words": 16
      },
      "decode_tok_s": 7.866020441079235,
      "e2e_ms": 49315.35527000233,
      "effective_prefill_tok_s": 774.0625112271293,
      "generation_ms": 16653.91044699936,
      "mean_inter_token_latency_ms": 127.12908738167452,
      "mean_time_per_output_token_ms": 127.12908738167452,
      "output_token_source": "usage",
      "output_tokens": 132,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25282,
      "reasoning_chunks": 101,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_2801_00005",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 5,
      "time_to_first_output_ms": 32661.444823002967,
      "tpot_ms": 127.12908738167452,
      "ttft_ms": 48857.82692700013,
      "visible_content_truncated": false,
      "visible_generation_ms": 457.52834300219547
    },
    {
      "content_chunks": 26,
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": true,
        "observed_code_words": 16,
        "observed_extra_words": 0,
        "passed": true,
        "policy": "strict",
        "requested_words": 16
      },
      "decode_tok_s": 4.644252882797655,
      "e2e_ms": 20773.16264699766,
      "effective_prefill_tok_s": 4273.261178686488,
      "generation_ms": 14857.07211499539,
      "mean_inter_token_latency_ms": 215.31988572457087,
      "mean_time_per_output_token_ms": 215.31988572457087,
      "output_token_source": "usage",
      "output_tokens": 70,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25281,
      "reasoning_chunks": 41,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_2801_00006",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 6,
      "time_to_first_output_ms": 5916.090532002272,
      "tpot_ms": 215.31988572457087,
      "ttft_ms": 20292.45143099979,
      "visible_content_truncated": false,
      "visible_generation_ms": 480.7112159978715
    },
    {
      "content_chunks": 26,
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": true,
        "observed_code_words": 16,
        "observed_extra_words": 0,
        "passed": true,
        "policy": "strict",
        "requested_words": 16
      },
      "decode_tok_s": 10.3147140526473,
      "e2e_ms": 31434.396501004812,
      "effective_prefill_tok_s": 1252.2088174095845,
      "generation_ms": 11246.070362001774,
      "mean_inter_token_latency_ms": 96.94888243104978,
      "mean_time_per_output_token_ms": 96.94888243104978,
      "output_token_source": "usage",
      "output_tokens": 117,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25280,
      "reasoning_chunks": 86,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_2801_00007",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 7,
      "time_to_first_output_ms": 20188.32613900304,
      "tpot_ms": 96.94888243104978,
      "ttft_ms": 26794.5232580023,
      "visible_content_truncated": false,
      "visible_generation_ms": 4639.873243002512
    },
    {
      "content_chunks": 26,
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": true,
        "observed_code_words": 16,
        "observed_extra_words": 0,
        "passed": true,
        "policy": "strict",
        "requested_words": 16
      },
      "decode_tok_s": 16.66089904026122,
      "e2e_ms": 31652.02763799607,
      "effective_prefill_tok_s": 920.9643588051134,
      "generation_ms": 4201.453945002868,
      "mean_inter_token_latency_ms": 60.020770642898114,
      "mean_time_per_output_token_ms": 60.020770642898114,
      "output_token_source": "usage",
      "output_tokens": 71,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25281,
      "reasoning_chunks": 42,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_2801_00008",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 8,
      "time_to_first_output_ms": 27450.573692993203,
      "tpot_ms": 60.020770642898114,
      "ttft_ms": 28246.712499996647,
      "visible_content_truncated": false,
      "visible_generation_ms": 3405.3151379994233
    },
    {
      "content_chunks": 26,
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": true,
        "observed_code_words": 16,
        "observed_extra_words": 0,
        "passed": true,
        "policy": "strict",
        "requested_words": 16
      },
      "decode_tok_s": 8.799482074197513,
      "e2e_ms": 56302.18415000127,
      "effective_prefill_tok_s": 895.4975321871929,
      "generation_ms": 28069.83387400396,
      "mean_inter_token_latency_ms": 113.64305212147353,
      "mean_time_per_output_token_ms": 113.64305212147353,
      "output_token_source": "usage",
      "output_tokens": 248,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25282,
      "reasoning_chunks": 200,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_2801_00009",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 9,
      "time_to_first_output_ms": 28232.350275997305,
      "tpot_ms": 113.64305212147353,
      "ttft_ms": 52821.59195699933,
      "visible_content_truncated": false,
      "visible_generation_ms": 3480.5921930019394
    },
    {
      "content_chunks": 26,
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": true,
        "observed_code_words": 16,
        "observed_extra_words": 0,
        "passed": true,
        "policy": "strict",
        "requested_words": 16
      },
      "decode_tok_s": 10.29083261384677,
      "e2e_ms": 53535.65713900025,
      "effective_prefill_tok_s": 823.5558970277585,
      "generation_ms": 22835.85875100107,
      "mean_inter_token_latency_ms": 97.17386702553647,
      "mean_time_per_output_token_ms": 97.17386702553647,
      "output_token_source": "usage",
      "output_tokens": 236,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25283,
      "reasoning_chunks": 188,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_2801_00010",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 10,
      "time_to_first_output_ms": 30699.79838799918,
      "tpot_ms": 97.17386702553647,
      "ttft_ms": 50444.17855500069,
      "visible_content_truncated": false,
      "visible_generation_ms": 3091.47858399956
    },
    {
      "content_chunks": 26,
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": true,
        "observed_code_words": 16,
        "observed_extra_words": 0,
        "passed": true,
        "policy": "strict",
        "requested_words": 16
      },
      "decode_tok_s": 20.025646214378245,
      "e2e_ms": 40229.41991399421,
      "effective_prefill_tok_s": 746.0851666528364,
      "generation_ms": 6341.867754999839,
      "mean_inter_token_latency_ms": 49.93596657480188,
      "mean_time_per_output_token_ms": 49.93596657480188,
      "output_token_source": "usage",
      "output_tokens": 128,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25283,
      "reasoning_chunks": 93,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_2801_00011",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 11,
      "time_to_first_output_ms": 33887.55215899437,
      "tpot_ms": 49.93596657480188,
      "ttft_ms": 39757.57481799519,
      "visible_content_truncated": false,
      "visible_generation_ms": 471.84509599901503
    },
    {
      "content_chunks": 26,
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": true,
        "observed_code_words": 16,
        "observed_extra_words": 0,
        "passed": true,
        "policy": "strict",
        "requested_words": 16
      },
      "decode_tok_s": 21.576411392955357,
      "e2e_ms": 34804.953475999355,
      "effective_prefill_tok_s": 950.4709175921257,
      "generation_ms": 8203.403095001704,
      "mean_inter_token_latency_ms": 46.346910141252565,
      "mean_time_per_output_token_ms": 46.346910141252565,
      "output_token_source": "usage",
      "output_tokens": 178,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25284,
      "reasoning_chunks": 127,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_2801_00012",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 12,
      "time_to_first_output_ms": 26601.55038099765,
      "tpot_ms": 46.346910141252565,
      "ttft_ms": 34324.06170899776,
      "visible_content_truncated": false,
      "visible_generation_ms": 480.8917670015944
    },
    {
      "content_chunks": 26,
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": true,
        "observed_code_words": 16,
        "observed_extra_words": 0,
        "passed": true,
        "policy": "strict",
        "requested_words": 16
      },
      "decode_tok_s": 10.122762425276926,
      "e2e_ms": 39763.55986499402,
      "effective_prefill_tok_s": 915.6285255711504,
      "generation_ms": 12150.833421997959,
      "mean_inter_token_latency_ms": 98.78726359347934,
      "mean_time_per_output_token_ms": 98.78726359347934,
      "output_token_source": "usage",
      "output_tokens": 124,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25283,
      "reasoning_chunks": 93,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_2801_00013",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 13,
      "time_to_first_output_ms": 27612.726442996063,
      "tpot_ms": 98.78726359347934,
      "ttft_ms": 39299.88185399998,
      "visible_content_truncated": false,
      "visible_generation_ms": 463.6780109940446
    },
    {
      "content_chunks": 26,
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": true,
        "observed_code_words": 16,
        "observed_extra_words": 0,
        "passed": true,
        "policy": "strict",
        "requested_words": 16
      },
      "decode_tok_s": 26.39724845289074,
      "e2e_ms": 41685.565102998225,
      "effective_prefill_tok_s": 758.9720346588722,
      "generation_ms": 8372.084703994915,
      "mean_inter_token_latency_ms": 37.88273621717156,
      "mean_time_per_output_token_ms": 37.88273621717156,
      "output_token_source": "usage",
      "output_tokens": 222,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25284,
      "reasoning_chunks": 175,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_2801_00014",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 14,
      "time_to_first_output_ms": 33313.48039900331,
      "tpot_ms": 37.88273621717156,
      "ttft_ms": 41305.095253002946,
      "visible_content_truncated": false,
      "visible_generation_ms": 380.4698499952792
    }
  ],
  "requests": 16,
  "response_words": 16,
  "responses_received": 16,
  "run_id": "benchmark-20260924T005220Z",
  "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": {
    "no_thinking": false,
    "reasoning_effort": "max",
    "shared_prefix_burst": false,
    "thinking_mode": "default"
  },
  "shared_prefix_tokens": 0,
  "source_recipe": null,
  "started_at": "2026-09-24T00:52:20Z",
  "started_at_unix_ns": 1790211140536983515,
  "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"
  },
  "wall_clock_ms": 97952.026965002
}
