{
  "clock_domain_id": null,
  "completed": 16,
  "concurrency": 8,
  "configuration_fingerprint": null,
  "context_distribution": {
    "32768": 16
  },
  "context_seed": 801,
  "context_tokens": 32768,
  "controlled_output_policy": "strict",
  "engine": "vllm-exl3-v84",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-09-24T00:50:48Z",
  "finished_at_unix_ns": 1790211048774874871,
  "gpu": "dual-pro6000-maxq",
  "manifest": null,
  "max_context_tokens": 327680,
  "max_tokens": 8192,
  "measurement_protocol": "capacity-v4-reasoning",
  "metric_population": {
    "aggregate_throughput_eligible": true,
    "attempted_requests": 16,
    "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": 16,
      "e2e": 16,
      "effective_prefill_tok_s": 16,
      "tpot": 16,
      "ttft": 16
    },
    "performance_eligible_requests": 16,
    "successful_requests": 16
  },
  "metrics": {
    "completion_tokens_max": 362.0,
    "completion_tokens_mean": 144.125,
    "completion_tokens_mean_ci95_high": 187.01455752861062,
    "completion_tokens_mean_ci95_low": 101.23544247138938,
    "completion_tokens_min": 68.0,
    "completion_tokens_p25": 98.0,
    "completion_tokens_p50": 121.0,
    "completion_tokens_p75": 131.0,
    "completion_tokens_p90": 348.0,
    "completion_tokens_p95": 362.0,
    "completion_tokens_p99": 362.0,
    "completion_tokens_samples": 16,
    "completion_tokens_stddev": 87.5297092420625,
    "content_chunks": 416,
    "content_chunks_s": null,
    "controlled_output_exact_adherence": 16,
    "controlled_output_exact_adherence_rate": 1.0,
    "controlled_output_samples": 16,
    "controlled_output_unobservable": 0,
    "decode_tok_s_max": 61.973047398007736,
    "decode_tok_s_mean": 13.037061352368676,
    "decode_tok_s_mean_ci95_high": 19.573882862378333,
    "decode_tok_s_mean_ci95_low": 6.500239842359018,
    "decode_tok_s_min": 4.8824777590949235,
    "decode_tok_s_p25": 7.511141744119188,
    "decode_tok_s_p50": 9.855026859512417,
    "decode_tok_s_p75": 11.04448064197634,
    "decode_tok_s_p90": 16.894386747247474,
    "decode_tok_s_p95": 61.973047398007736,
    "decode_tok_s_p99": 61.973047398007736,
    "decode_tok_s_samples": 16,
    "decode_tok_s_stddev": 13.3404520612442,
    "e2e_max_ms": 61049.60514100094,
    "e2e_mean_ci95_high_ms": 47218.07381193705,
    "e2e_mean_ci95_low_ms": 35849.53426531481,
    "e2e_mean_ms": 41533.80403862593,
    "e2e_min_ms": 20882.704404000833,
    "e2e_p25_ms": 31871.98584500584,
    "e2e_p50_ms": 38253.444895999564,
    "e2e_p75_ms": 46522.36179200554,
    "e2e_p90_ms": 60098.57282500161,
    "e2e_p95_ms": 61049.60514100094,
    "e2e_p99_ms": 61049.60514100094,
    "e2e_samples": 16,
    "e2e_stddev_ms": 11600.5505577778,
    "effective_prefill_tok_s_max": 4262.823962969075,
    "effective_prefill_tok_s_mean": 1217.78613586788,
    "effective_prefill_tok_s_mean_ci95_high": 1661.73823798907,
    "effective_prefill_tok_s_mean_ci95_low": 773.8340337466897,
    "effective_prefill_tok_s_min": 581.114964605957,
    "effective_prefill_tok_s_p25": 869.0880863490091,
    "effective_prefill_tok_s_p50": 900.2468158669527,
    "effective_prefill_tok_s_p75": 965.7865324419954,
    "effective_prefill_tok_s_p90": 2261.774371786411,
    "effective_prefill_tok_s_p95": 4262.823962969075,
    "effective_prefill_tok_s_p99": 4262.823962969075,
    "effective_prefill_tok_s_samples": 16,
    "effective_prefill_tok_s_stddev": 906.0246982065107,
    "generation_max_ms": 49871.65551199723,
    "generation_mean_ci95_high_ms": 20795.215012688834,
    "generation_mean_ci95_low_ms": 9742.796796561914,
    "generation_mean_ms": 15269.005904625374,
    "generation_min_ms": 1081.1151430025348,
    "generation_p25_ms": 10381.549442005053,
    "generation_p50_ms": 11863.2499239975,
    "generation_p75_ms": 17073.26410099631,
    "generation_p90_ms": 31008.310978002555,
    "generation_p95_ms": 49871.65551199723,
    "generation_p99_ms": 49871.65551199723,
    "generation_samples": 16,
    "generation_stddev_ms": 11277.97777155808,
    "mean_inter_token_latency_ms_max": 204.81404101375207,
    "mean_inter_token_latency_ms_mean": 105.22234342361456,
    "mean_inter_token_latency_ms_mean_ci95_high": 125.58509709606922,
    "mean_inter_token_latency_ms_mean_ci95_low": 84.8595897511599,
    "mean_inter_token_latency_ms_min": 16.136046910485593,
    "mean_inter_token_latency_ms_p25": 85.8955982770154,
    "mean_inter_token_latency_ms_p50": 97.03542451580257,
    "mean_inter_token_latency_ms_p75": 121.00776409276955,
    "mean_inter_token_latency_ms_p90": 145.92533419654967,
    "mean_inter_token_latency_ms_p95": 204.81404101375207,
    "mean_inter_token_latency_ms_p99": 204.81404101375207,
    "mean_inter_token_latency_ms_samples": 16,
    "mean_inter_token_latency_ms_stddev": 41.55664014786666,
    "mean_time_per_output_token_max_ms": 204.81404101375207,
    "mean_time_per_output_token_mean_ci95_high_ms": 125.58509709606922,
    "mean_time_per_output_token_mean_ci95_low_ms": 84.8595897511599,
    "mean_time_per_output_token_mean_ms": 105.22234342361456,
    "mean_time_per_output_token_min_ms": 16.136046910485593,
    "mean_time_per_output_token_p25_ms": 85.8955982770154,
    "mean_time_per_output_token_p50_ms": 97.03542451580257,
    "mean_time_per_output_token_p75_ms": 121.00776409276955,
    "mean_time_per_output_token_p90_ms": 145.92533419654967,
    "mean_time_per_output_token_p95_ms": 204.81404101375207,
    "mean_time_per_output_token_p99_ms": 204.81404101375207,
    "mean_time_per_output_token_samples": 16,
    "mean_time_per_output_token_stddev_ms": 41.55664014786666,
    "output_token_sources": {
      "usage": 16
    },
    "output_tokens": 2306,
    "prefix_cache_hit_avg": null,
    "prompt_token_samples": 16,
    "prompt_tokens": 404528,
    "prompt_tokens_max": 25285.0,
    "prompt_tokens_mean": 25283.0,
    "prompt_tokens_mean_ci95_high": 25283.619806421393,
    "prompt_tokens_mean_ci95_low": 25282.380193578607,
    "prompt_tokens_min": 25281.0,
    "prompt_tokens_p25": 25282.0,
    "prompt_tokens_p50": 25282.0,
    "prompt_tokens_p75": 25284.0,
    "prompt_tokens_p90": 25285.0,
    "prompt_tokens_p95": 25285.0,
    "prompt_tokens_p99": 25285.0,
    "prompt_tokens_samples": 16,
    "prompt_tokens_stddev": 1.2649110640673518,
    "reasoning_chunks": 1758,
    "request_canary_passed": 16,
    "request_canary_samples": 16,
    "throughput_performance_eligible": true,
    "throughput_tok_s": 25.072575374045783,
    "time_to_first_output_max_ms": 43506.02125199657,
    "time_to_first_output_mean_ci95_high_ms": 30811.619575072662,
    "time_to_first_output_mean_ci95_low_ms": 21717.976692928452,
    "time_to_first_output_mean_ms": 26264.798134000557,
    "time_to_first_output_min_ms": 5931.279409996932,
    "time_to_first_output_p25_ms": 20224.147628003266,
    "time_to_first_output_p50_ms": 27871.89545399451,
    "time_to_first_output_p75_ms": 28977.186834999884,
    "time_to_first_output_p90_ms": 37824.738911003806,
    "time_to_first_output_p95_ms": 43506.02125199657,
    "time_to_first_output_p99_ms": 43506.02125199657,
    "time_to_first_output_samples": 16,
    "time_to_first_output_stddev_ms": 9279.2274307594,
    "tpot_max_ms": 204.81404101375207,
    "tpot_mean_ci95_high_ms": 125.58509709606922,
    "tpot_mean_ci95_low_ms": 84.8595897511599,
    "tpot_mean_ms": 105.22234342361456,
    "tpot_min_ms": 16.136046910485593,
    "tpot_p25_ms": 85.8955982770154,
    "tpot_p50_ms": 97.03542451580257,
    "tpot_p75_ms": 121.00776409276955,
    "tpot_p90_ms": 145.92533419654967,
    "tpot_p95_ms": 204.81404101375207,
    "tpot_p99_ms": 204.81404101375207,
    "tpot_samples": 16,
    "tpot_stddev_ms": 41.55664014786666,
    "ttft_max_ms": 59763.6070869994,
    "ttft_mean_ci95_high_ms": 44853.71432704411,
    "ttft_mean_ci95_low_ms": 33977.49616683272,
    "ttft_mean_ms": 39415.605246938416,
    "ttft_min_ms": 20400.55191200372,
    "ttft_p25_ms": 29682.62065899762,
    "ttft_p50_ms": 36433.52954099828,
    "ttft_p75_ms": 43834.38006000506,
    "ttft_p90_ms": 55593.14832000382,
    "ttft_p95_ms": 59763.6070869994,
    "ttft_p99_ms": 59763.6070869994,
    "ttft_samples": 16,
    "ttft_stddev_ms": 11098.181796134068
  },
  "model": "glm53-flash-exl3-runtime-dcp1-candidate",
  "performance_eligible": true,
  "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": 6.957860059739243,
      "e2e_ms": 61049.60514100094,
      "effective_prefill_tok_s": 2261.774371786411,
      "generation_ms": 49871.65551199723,
      "mean_inter_token_latency_ms": 143.72235017866637,
      "mean_time_per_output_token_ms": 143.72235017866637,
      "output_token_source": "usage",
      "output_tokens": 348,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25282,
      "reasoning_chunks": 297,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_801_00000",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 0,
      "time_to_first_output_ms": 11177.949629003706,
      "tpot_ms": 143.72235017866637,
      "ttft_ms": 55593.14832000382,
      "visible_content_truncated": false,
      "visible_generation_ms": 5456.4568209971185
    },
    {
      "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": 10.903310780417234,
      "e2e_ms": 31871.98584500584,
      "effective_prefill_tok_s": 1250.0897671946086,
      "generation_ms": 11647.838217002572,
      "mean_inter_token_latency_ms": 91.71526155120135,
      "mean_time_per_output_token_ms": 91.71526155120135,
      "output_token_source": "usage",
      "output_tokens": 128,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25282,
      "reasoning_chunks": 97,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_801_00001",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 1,
      "time_to_first_output_ms": 20224.147628003266,
      "tpot_ms": 91.71526155120135,
      "ttft_ms": 27076.72457100125,
      "visible_content_truncated": false,
      "visible_generation_ms": 4795.261274004588
    },
    {
      "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": 11.04448064197634,
      "e2e_ms": 27296.60466799396,
      "effective_prefill_tok_s": 1582.2279473731019,
      "generation_ms": 11317.870350998419,
      "mean_inter_token_latency_ms": 90.54296280798735,
      "mean_time_per_output_token_ms": 90.54296280798735,
      "output_token_source": "usage",
      "output_tokens": 126,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25282,
      "reasoning_chunks": 95,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_801_00002",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 2,
      "time_to_first_output_ms": 15978.734316995542,
      "tpot_ms": 90.54296280798735,
      "ttft_ms": 26816.753515995515,
      "visible_content_truncated": false,
      "visible_generation_ms": 479.85115199844586
    },
    {
      "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": 4.8824777590949235,
      "e2e_ms": 20882.704404000833,
      "effective_prefill_tok_s": 4262.823962969075,
      "generation_ms": 14951.4249940039,
      "mean_inter_token_latency_ms": 204.81404101375207,
      "mean_time_per_output_token_ms": 204.81404101375207,
      "output_token_source": "usage",
      "output_tokens": 74,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25284,
      "reasoning_chunks": 45,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_801_00003",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 3,
      "time_to_first_output_ms": 5931.279409996932,
      "tpot_ms": 204.81404101375207,
      "ttft_ms": 20400.55191200372,
      "visible_content_truncated": false,
      "visible_generation_ms": 482.1524919971125
    },
    {
      "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": 9.463276071164687,
      "e2e_ms": 38226.26639499504,
      "effective_prefill_tok_s": 965.7865324419954,
      "generation_ms": 12046.568138001021,
      "mean_inter_token_latency_ms": 105.67165033334228,
      "mean_time_per_output_token_ms": 105.67165033334228,
      "output_token_source": "usage",
      "output_tokens": 115,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25284,
      "reasoning_chunks": 85,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_801_00004",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 4,
      "time_to_first_output_ms": 26179.698256994016,
      "tpot_ms": 105.67165033334228,
      "ttft_ms": 36433.52954099828,
      "visible_content_truncated": false,
      "visible_generation_ms": 1792.7368539967574
    },
    {
      "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": 8.68227514887359,
      "e2e_ms": 49687.988835001306,
      "effective_prefill_tok_s": 668.4249707443396,
      "generation_ms": 11863.2499239975,
      "mean_inter_token_latency_ms": 115.17718372813107,
      "mean_time_per_output_token_ms": 115.17718372813107,
      "output_token_source": "usage",
      "output_tokens": 104,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25283,
      "reasoning_chunks": 74,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_801_00005",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 5,
      "time_to_first_output_ms": 37824.738911003806,
      "tpot_ms": 115.17718372813107,
      "ttft_ms": 44222.56395400473,
      "visible_content_truncated": false,
      "visible_generation_ms": 5465.424880996579
    },
    {
      "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": 8.263932545959271,
      "e2e_ms": 44315.65777800279,
      "effective_prefill_tok_s": 776.0474549568756,
      "generation_ms": 11737.753116998647,
      "mean_inter_token_latency_ms": 121.00776409276955,
      "mean_time_per_output_token_ms": 121.00776409276955,
      "output_token_source": "usage",
      "output_tokens": 98,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25282,
      "reasoning_chunks": 68,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_801_00006",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 6,
      "time_to_first_output_ms": 32577.90466100414,
      "tpot_ms": 121.00776409276955,
      "ttft_ms": 43834.38006000506,
      "visible_content_truncated": false,
      "visible_generation_ms": 481.27771799772745
    },
    {
      "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": 9.855026859512417,
      "e2e_ms": 55682.54818699643,
      "effective_prefill_tok_s": 581.114964605957,
      "generation_ms": 12176.526934999856,
      "mean_inter_token_latency_ms": 101.47105779166547,
      "mean_time_per_output_token_ms": 101.47105779166547,
      "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_801_00007",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 7,
      "time_to_first_output_ms": 43506.02125199657,
      "tpot_ms": 101.47105779166547,
      "ttft_ms": 55201.06858800136,
      "visible_content_truncated": false,
      "visible_generation_ms": 481.4795989950653
    },
    {
      "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": 6.852819666344437,
      "e2e_ms": 45792.59992000152,
      "effective_prefill_tok_s": 880.2780175462879,
      "generation_ms": 17073.26410099631,
      "mean_inter_token_latency_ms": 145.92533419654967,
      "mean_time_per_output_token_ms": 145.92533419654967,
      "output_token_source": "usage",
      "output_tokens": 118,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25281,
      "reasoning_chunks": 87,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_801_00008",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 8,
      "time_to_first_output_ms": 28719.33581900521,
      "tpot_ms": 145.92533419654967,
      "ttft_ms": 43347.818398004165,
      "visible_content_truncated": false,
      "visible_generation_ms": 2444.7815219973563
    },
    {
      "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": 7.511141744119188,
      "e2e_ms": 44986.44408799737,
      "effective_prefill_tok_s": 913.4059124618504,
      "generation_ms": 17307.621720996394,
      "mean_inter_token_latency_ms": 133.13555169997227,
      "mean_time_per_output_token_ms": 133.13555169997227,
      "output_token_source": "usage",
      "output_tokens": 131,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25282,
      "reasoning_chunks": 96,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_801_00009",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 9,
      "time_to_first_output_ms": 27678.822367000976,
      "tpot_ms": 133.13555169997227,
      "ttft_ms": 41875.12266599515,
      "visible_content_truncated": false,
      "visible_generation_ms": 3111.32142200222
    },
    {
      "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": 11.642040105186481,
      "e2e_ms": 60098.57282500161,
      "effective_prefill_tok_s": 869.0880863490091,
      "generation_ms": 31008.310978002555,
      "mean_inter_token_latency_ms": 85.8955982770154,
      "mean_time_per_output_token_ms": 85.8955982770154,
      "output_token_source": "usage",
      "output_tokens": 362,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25282,
      "reasoning_chunks": 311,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_801_00010",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 10,
      "time_to_first_output_ms": 29090.261846999056,
      "tpot_ms": 85.8955982770154,
      "ttft_ms": 59763.6070869994,
      "visible_content_truncated": false,
      "visible_generation_ms": 334.96573800221086
    },
    {
      "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": 10.305514764220424,
      "e2e_ms": 46522.36179200554,
      "effective_prefill_tok_s": 900.2468158669527,
      "generation_ms": 18436.73065800249,
      "mean_inter_token_latency_ms": 97.03542451580257,
      "mean_time_per_output_token_ms": 97.03542451580257,
      "output_token_source": "usage",
      "output_tokens": 191,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25284,
      "reasoning_chunks": 143,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_801_00011",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 11,
      "time_to_first_output_ms": 28085.63113400305,
      "tpot_ms": 97.03542451580257,
      "ttft_ms": 42755.336936003005,
      "visible_content_truncated": false,
      "visible_generation_ms": 3767.0248560025357
    },
    {
      "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": 16.894386747247474,
      "e2e_ms": 34859.997711006145,
      "effective_prefill_tok_s": 914.8054245282194,
      "generation_ms": 7221.333442001196,
      "mean_inter_token_latency_ms": 59.19125772132128,
      "mean_time_per_output_token_ms": 59.19125772132128,
      "output_token_source": "usage",
      "output_tokens": 123,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25284,
      "reasoning_chunks": 92,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_801_00012",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 12,
      "time_to_first_output_ms": 27638.66426900495,
      "tpot_ms": 59.19125772132128,
      "ttft_ms": 34381.34541400359,
      "visible_content_truncated": false,
      "visible_generation_ms": 478.6522970025544
    },
    {
      "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": 12.522215563892965,
      "e2e_ms": 38253.444895999564,
      "effective_prefill_tok_s": 907.1862386157247,
      "generation_ms": 10381.549442005053,
      "mean_inter_token_latency_ms": 79.8580726308081,
      "mean_time_per_output_token_ms": 79.8580726308081,
      "output_token_source": "usage",
      "output_tokens": 131,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25285,
      "reasoning_chunks": 100,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_801_00013",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 13,
      "time_to_first_output_ms": 27871.89545399451,
      "tpot_ms": 79.8580726308081,
      "ttft_ms": 34770.307761995355,
      "visible_content_truncated": false,
      "visible_generation_ms": 3483.1371340042097
    },
    {
      "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": 10.839175782142407,
      "e2e_ms": 34955.780155003595,
      "effective_prefill_tok_s": 878.6947278433552,
      "generation_ms": 6181.281801000296,
      "mean_inter_token_latency_ms": 92.25793732836262,
      "mean_time_per_output_token_ms": 92.25793732836262,
      "output_token_source": "usage",
      "output_tokens": 68,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25284,
      "reasoning_chunks": 39,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_801_00014",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 14,
      "time_to_first_output_ms": 28774.4983540033,
      "tpot_ms": 92.25793732836262,
      "ttft_ms": 34494.804567002575,
      "visible_content_truncated": false,
      "visible_generation_ms": 460.9755880010198
    },
    {
      "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": 61.973047398007736,
      "e2e_ms": 30058.30197800242,
      "effective_prefill_tok_s": 872.5829786023154,
      "generation_ms": 1081.1151430025348,
      "mean_inter_token_latency_ms": 16.136046910485593,
      "mean_time_per_output_token_ms": 16.136046910485593,
      "output_token_source": "usage",
      "output_tokens": 68,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25285,
      "reasoning_chunks": 39,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_801_00015",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 15,
      "time_to_first_output_ms": 28977.186834999884,
      "tpot_ms": 16.136046910485593,
      "ttft_ms": 29682.62065899762,
      "visible_content_truncated": false,
      "visible_generation_ms": 375.68131900479784
    }
  ],
  "requests": 16,
  "response_words": 16,
  "responses_received": 16,
  "run_id": "benchmark-20260924T004916Z",
  "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": "2026-09-24T00:49:16Z",
  "started_at_unix_ns": 1790210956801868059,
  "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": 91973.00100200664
}
