{
  "clock_domain_id": null,
  "completed": 16,
  "concurrency": 8,
  "configuration_fingerprint": null,
  "context_distribution": {
    "32768": 16
  },
  "context_seed": 1801,
  "context_tokens": 32768,
  "controlled_output_policy": "strict",
  "engine": "vllm-exl3-v84",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-09-24T00:52:20Z",
  "finished_at_unix_ns": 1790211140285441858,
  "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": 295.0,
    "completion_tokens_mean": 143.4375,
    "completion_tokens_mean_ci95_high": 173.02417920281016,
    "completion_tokens_mean_ci95_low": 113.85082079718983,
    "completion_tokens_min": 69.0,
    "completion_tokens_p25": 101.0,
    "completion_tokens_p50": 125.0,
    "completion_tokens_p75": 158.0,
    "completion_tokens_p90": 240.0,
    "completion_tokens_p95": 295.0,
    "completion_tokens_p99": 295.0,
    "completion_tokens_samples": 16,
    "completion_tokens_stddev": 60.38097796491872,
    "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": 73.33923620462798,
    "decode_tok_s_mean": 16.627072446774942,
    "decode_tok_s_mean_ci95_high": 25.88215201006748,
    "decode_tok_s_mean_ci95_low": 7.371992883482406,
    "decode_tok_s_min": 6.533474462642345,
    "decode_tok_s_p25": 7.704462848418047,
    "decode_tok_s_p50": 9.900292539137299,
    "decode_tok_s_p75": 11.45064646831096,
    "decode_tok_s_p90": 54.31026782118108,
    "decode_tok_s_p95": 73.33923620462798,
    "decode_tok_s_p99": 73.33923620462798,
    "decode_tok_s_samples": 16,
    "decode_tok_s_stddev": 18.88791747610722,
    "e2e_max_ms": 66451.73877000343,
    "e2e_mean_ci95_high_ms": 47023.25111575753,
    "e2e_mean_ci95_low_ms": 34631.30521249408,
    "e2e_mean_ms": 40827.278164125804,
    "e2e_min_ms": 21842.08695700363,
    "e2e_p25_ms": 31049.272210999334,
    "e2e_p50_ms": 38027.73242299736,
    "e2e_p75_ms": 44630.03300499986,
    "e2e_p90_ms": 58006.820093003626,
    "e2e_p95_ms": 66451.73877000343,
    "e2e_p99_ms": 66451.73877000343,
    "e2e_samples": 16,
    "e2e_stddev_ms": 12644.84275843209,
    "effective_prefill_tok_s_max": 4266.533817885676,
    "effective_prefill_tok_s_mean": 1194.2711366501353,
    "effective_prefill_tok_s_mean_ci95_high": 1643.6540773373158,
    "effective_prefill_tok_s_mean_ci95_low": 744.8881959629549,
    "effective_prefill_tok_s_min": 592.9858797463509,
    "effective_prefill_tok_s_p25": 761.2409545618334,
    "effective_prefill_tok_s_p50": 859.6187187196485,
    "effective_prefill_tok_s_p75": 937.9909991184481,
    "effective_prefill_tok_s_p90": 2262.865691869645,
    "effective_prefill_tok_s_p95": 4266.533817885676,
    "effective_prefill_tok_s_p99": 4266.533817885676,
    "effective_prefill_tok_s_samples": 16,
    "effective_prefill_tok_s_stddev": 917.1080422187355,
    "generation_max_ms": 30130.90463000117,
    "generation_mean_ci95_high_ms": 17683.61754617993,
    "generation_mean_ci95_low_ms": 9742.281499197334,
    "generation_mean_ms": 13712.949522688632,
    "generation_min_ms": 1636.2319300023955,
    "generation_p25_ms": 6558.201066000038,
    "generation_p50_ms": 11418.472527999256,
    "generation_p75_ms": 16830.76955300203,
    "generation_p90_ms": 24140.70079800149,
    "generation_p95_ms": 30130.90463000117,
    "generation_p99_ms": 30130.90463000117,
    "generation_samples": 16,
    "generation_stddev_ms": 8103.404129574077,
    "mean_inter_token_latency_ms_max": 153.05791822067798,
    "mean_inter_token_latency_ms_mean": 95.58549479762358,
    "mean_inter_token_latency_ms_mean_ci95_high": 114.84061084201392,
    "mean_inter_token_latency_ms_mean_ci95_low": 76.33037875323325,
    "mean_inter_token_latency_ms_min": 13.635266083353295,
    "mean_inter_token_latency_ms_p25": 77.47832333695007,
    "mean_inter_token_latency_ms_p50": 94.41669238055181,
    "mean_inter_token_latency_ms_p75": 117.92095169804801,
    "mean_inter_token_latency_ms_p90": 140.689957579583,
    "mean_inter_token_latency_ms_p95": 153.05791822067798,
    "mean_inter_token_latency_ms_p99": 153.05791822067798,
    "mean_inter_token_latency_ms_samples": 16,
    "mean_inter_token_latency_ms_stddev": 39.296155192633336,
    "mean_time_per_output_token_max_ms": 153.05791822067798,
    "mean_time_per_output_token_mean_ci95_high_ms": 114.84061084201392,
    "mean_time_per_output_token_mean_ci95_low_ms": 76.33037875323325,
    "mean_time_per_output_token_mean_ms": 95.58549479762358,
    "mean_time_per_output_token_min_ms": 13.635266083353295,
    "mean_time_per_output_token_p25_ms": 77.47832333695007,
    "mean_time_per_output_token_p50_ms": 94.41669238055181,
    "mean_time_per_output_token_p75_ms": 117.92095169804801,
    "mean_time_per_output_token_p90_ms": 140.689957579583,
    "mean_time_per_output_token_p95_ms": 153.05791822067798,
    "mean_time_per_output_token_p99_ms": 153.05791822067798,
    "mean_time_per_output_token_samples": 16,
    "mean_time_per_output_token_stddev_ms": 39.296155192633336,
    "output_token_sources": {
      "usage": 16
    },
    "output_tokens": 2295,
    "prefix_cache_hit_avg": null,
    "prompt_token_samples": 16,
    "prompt_tokens": 404519,
    "prompt_tokens_max": 25284.0,
    "prompt_tokens_mean": 25282.4375,
    "prompt_tokens_mean_ci95_high": 25283.0300759445,
    "prompt_tokens_mean_ci95_low": 25281.8449240555,
    "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": 16,
    "prompt_tokens_stddev": 1.2093386622447824,
    "reasoning_chunks": 1749,
    "request_canary_passed": 16,
    "request_canary_samples": 16,
    "throughput_performance_eligible": true,
    "throughput_tok_s": 25.138436274025043,
    "time_to_first_output_max_ms": 42631.70652699773,
    "time_to_first_output_mean_ci95_high_ms": 31815.237604497008,
    "time_to_first_output_mean_ci95_low_ms": 22413.41967837733,
    "time_to_first_output_mean_ms": 27114.32864143717,
    "time_to_first_output_min_ms": 5925.887636003608,
    "time_to_first_output_p25_ms": 20211.648542994226,
    "time_to_first_output_p50_ms": 28016.856768997968,
    "time_to_first_output_p75_ms": 31748.05298499996,
    "time_to_first_output_p90_ms": 38385.276691005856,
    "time_to_first_output_p95_ms": 42631.70652699773,
    "time_to_first_output_p99_ms": 42631.70652699773,
    "time_to_first_output_samples": 16,
    "time_to_first_output_stddev_ms": 9593.691761346608,
    "tpot_max_ms": 153.05791822067798,
    "tpot_mean_ci95_high_ms": 114.84061084201392,
    "tpot_mean_ci95_low_ms": 76.33037875323325,
    "tpot_mean_ms": 95.58549479762358,
    "tpot_min_ms": 13.635266083353295,
    "tpot_p25_ms": 77.47832333695007,
    "tpot_p50_ms": 94.41669238055181,
    "tpot_p75_ms": 117.92095169804801,
    "tpot_p90_ms": 140.689957579583,
    "tpot_p95_ms": 153.05791822067798,
    "tpot_p99_ms": 153.05791822067798,
    "tpot_samples": 16,
    "tpot_stddev_ms": 39.296155192633336,
    "ttft_max_ms": 65597.25788400101,
    "ttft_mean_ci95_high_ms": 44950.86709525038,
    "ttft_mean_ci95_low_ms": 32457.3413278755,
    "ttft_mean_ms": 38704.10421156294,
    "ttft_min_ms": 21367.37956700381,
    "ttft_p25_ms": 26407.091046996356,
    "ttft_p50_ms": 37548.40470400086,
    "ttft_p75_ms": 43511.57836600032,
    "ttft_p90_ms": 55946.830451001006,
    "ttft_p95_ms": 65597.25788400101,
    "ttft_p99_ms": 65597.25788400101,
    "ttft_samples": 16,
    "ttft_stddev_ms": 12748.495680994776
  },
  "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": 7.367458725491413,
      "e2e_ms": 43931.51925100392,
      "effective_prefill_tok_s": 932.8524222288928,
      "generation_ms": 16830.76955300203,
      "mean_inter_token_latency_ms": 135.7320125242099,
      "mean_time_per_output_token_ms": 135.7320125242099,
      "output_token_source": "usage",
      "output_tokens": 125,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25281,
      "reasoning_chunks": 94,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_1801_00000",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 0,
      "time_to_first_output_ms": 27100.74969800189,
      "tpot_ms": 135.7320125242099,
      "ttft_ms": 43458.946051003295,
      "visible_content_truncated": false,
      "visible_generation_ms": 472.57320000062464
    },
    {
      "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": 11.278965096151678,
      "e2e_ms": 31648.868083997513,
      "effective_prefill_tok_s": 1250.8628351725056,
      "generation_ms": 11437.219541003287,
      "mean_inter_token_latency_ms": 88.66061659692471,
      "mean_time_per_output_token_ms": 88.66061659692471,
      "output_token_source": "usage",
      "output_tokens": 130,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25282,
      "reasoning_chunks": 99,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_1801_00001",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 1,
      "time_to_first_output_ms": 20211.648542994226,
      "tpot_ms": 88.66061659692471,
      "ttft_ms": 26407.091046996356,
      "visible_content_truncated": false,
      "visible_generation_ms": 5241.777037001157
    },
    {
      "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": 10.591347512677563,
      "e2e_ms": 21842.08695700363,
      "effective_prefill_tok_s": 2262.865691869645,
      "generation_ms": 10669.086239002354,
      "mean_inter_token_latency_ms": 94.41669238055181,
      "mean_time_per_output_token_ms": 94.41669238055181,
      "output_token_source": "usage",
      "output_tokens": 114,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25283,
      "reasoning_chunks": 84,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_1801_00002",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 2,
      "time_to_first_output_ms": 11173.000718001276,
      "tpot_ms": 94.41669238055181,
      "ttft_ms": 21367.37956700381,
      "visible_content_truncated": false,
      "visible_generation_ms": 474.7073899998213
    },
    {
      "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": 7.107827859243882,
      "e2e_ms": 28014.21097599814,
      "effective_prefill_tok_s": 4266.533817885676,
      "generation_ms": 22088.32333999453,
      "mean_inter_token_latency_ms": 140.689957579583,
      "mean_time_per_output_token_ms": 140.689957579583,
      "output_token_source": "usage",
      "output_tokens": 158,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25283,
      "reasoning_chunks": 111,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_1801_00003",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 3,
      "time_to_first_output_ms": 5925.887636003608,
      "tpot_ms": 140.689957579583,
      "ttft_ms": 23242.111915998976,
      "visible_content_truncated": false,
      "visible_generation_ms": 4772.099059999164
    },
    {
      "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": 11.386417663543378,
      "e2e_ms": 27311.44497400237,
      "effective_prefill_tok_s": 1581.9516318516694,
      "generation_ms": 11329.287561005913,
      "mean_inter_token_latency_ms": 87.82393458144118,
      "mean_time_per_output_token_ms": 87.82393458144118,
      "output_token_source": "usage",
      "output_tokens": 130,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25283,
      "reasoning_chunks": 96,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_1801_00004",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 4,
      "time_to_first_output_ms": 15982.157412996457,
      "tpot_ms": 87.82393458144118,
      "ttft_ms": 22252.217228000518,
      "visible_content_truncated": false,
      "visible_generation_ms": 5059.227746001852
    },
    {
      "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": 8.67016142108692,
      "e2e_ms": 44630.03300499986,
      "effective_prefill_tok_s": 761.2409545618334,
      "generation_ms": 11418.472527999256,
      "mean_inter_token_latency_ms": 115.33810634342683,
      "mean_time_per_output_token_ms": 115.33810634342683,
      "output_token_source": "usage",
      "output_tokens": 100,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25282,
      "reasoning_chunks": 70,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_1801_00005",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 5,
      "time_to_first_output_ms": 33211.5604770006,
      "tpot_ms": 115.33810634342683,
      "ttft_ms": 43511.57836600032,
      "visible_content_truncated": false,
      "visible_generation_ms": 1118.4546389995376
    },
    {
      "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": 7.704462848418047,
      "e2e_ms": 54479.8445330016,
      "effective_prefill_tok_s": 658.6118996485872,
      "generation_ms": 16094.56784199574,
      "mean_inter_token_latency_ms": 129.79490195157854,
      "mean_time_per_output_token_ms": 129.79490195157854,
      "output_token_source": "usage",
      "output_tokens": 125,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25281,
      "reasoning_chunks": 94,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_1801_00006",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 6,
      "time_to_first_output_ms": 38385.276691005856,
      "tpot_ms": 129.79490195157854,
      "ttft_ms": 49410.132860000886,
      "visible_content_truncated": false,
      "visible_generation_ms": 5069.711673000711
    },
    {
      "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": 8.480257202813547,
      "e2e_ms": 66451.73877000343,
      "effective_prefill_tok_s": 592.9858797463509,
      "generation_ms": 23820.0322430057,
      "mean_inter_token_latency_ms": 117.92095169804801,
      "mean_time_per_output_token_ms": 117.92095169804801,
      "output_token_source": "usage",
      "output_tokens": 203,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25280,
      "reasoning_chunks": 172,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_1801_00007",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 7,
      "time_to_first_output_ms": 42631.70652699773,
      "tpot_ms": 117.92095169804801,
      "ttft_ms": 65597.25788400101,
      "visible_content_truncated": false,
      "visible_generation_ms": 854.4808860024204
    },
    {
      "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": 6.533474462642345,
      "e2e_ms": 38399.780826002825,
      "effective_prefill_tok_s": 903.1559856075778,
      "generation_ms": 10407.938439006102,
      "mean_inter_token_latency_ms": 153.05791822067798,
      "mean_time_per_output_token_ms": 153.05791822067798,
      "output_token_source": "usage",
      "output_tokens": 69,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25281,
      "reasoning_chunks": 40,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_1801_00008",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 8,
      "time_to_first_output_ms": 27991.842386996723,
      "tpot_ms": 153.05791822067798,
      "ttft_ms": 37922.464894996665,
      "visible_content_truncated": false,
      "visible_generation_ms": 477.3159310061601
    },
    {
      "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": 9.75742360245188,
      "e2e_ms": 57084.25495400297,
      "effective_prefill_tok_s": 937.9909991184481,
      "generation_ms": 30130.90463000117,
      "mean_inter_token_latency_ms": 102.486070170072,
      "mean_time_per_output_token_ms": 102.486070170072,
      "output_token_source": "usage",
      "output_tokens": 295,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25282,
      "reasoning_chunks": 244,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_1801_00009",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 9,
      "time_to_first_output_ms": 26953.350324001804,
      "tpot_ms": 102.486070170072,
      "ttft_ms": 55946.830451001006,
      "visible_content_truncated": false,
      "visible_generation_ms": 1137.4245030019665
    },
    {
      "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": 11.45064646831096,
      "e2e_ms": 38027.73242299736,
      "effective_prefill_tok_s": 796.568913598618,
      "generation_ms": 6287.854593996599,
      "mean_inter_token_latency_ms": 87.33131380550832,
      "mean_time_per_output_token_ms": 87.33131380550832,
      "output_token_source": "usage",
      "output_tokens": 73,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25283,
      "reasoning_chunks": 44,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_1801_00010",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 10,
      "time_to_first_output_ms": 31739.87782900076,
      "tpot_ms": 87.33131380550832,
      "ttft_ms": 37548.40470400086,
      "visible_content_truncated": false,
      "visible_generation_ms": 479.32771899650106
    },
    {
      "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": 9.900292539137299,
      "e2e_ms": 58006.820093003626,
      "effective_prefill_tok_s": 746.5573418602819,
      "generation_ms": 24140.70079800149,
      "mean_inter_token_latency_ms": 101.00711630962967,
      "mean_time_per_output_token_ms": 101.00711630962967,
      "output_token_source": "usage",
      "output_tokens": 240,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25283,
      "reasoning_chunks": 208,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_1801_00011",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 11,
      "time_to_first_output_ms": 33866.119295002136,
      "tpot_ms": 101.00711630962967,
      "ttft_ms": 52563.35715900059,
      "visible_content_truncated": false,
      "visible_generation_ms": 5443.462934003037
    },
    {
      "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": 12.906835833953721,
      "e2e_ms": 42272.86826299678,
      "effective_prefill_tok_s": 902.4566962835743,
      "generation_ms": 14256.011493998813,
      "mean_inter_token_latency_ms": 77.47832333695007,
      "mean_time_per_output_token_ms": 77.47832333695007,
      "output_token_source": "usage",
      "output_tokens": 185,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25284,
      "reasoning_chunks": 137,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_1801_00012",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 12,
      "time_to_first_output_ms": 28016.856768997968,
      "tpot_ms": 77.47832333695007,
      "ttft_ms": 40096.98699600267,
      "visible_content_truncated": false,
      "visible_generation_ms": 2175.881266994111
    },
    {
      "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": 54.31026782118108,
      "e2e_ms": 34049.64355000266,
      "effective_prefill_tok_s": 796.363796291555,
      "generation_ms": 2301.590565002698,
      "mean_inter_token_latency_ms": 18.412724520021584,
      "mean_time_per_output_token_ms": 18.412724520021584,
      "output_token_source": "usage",
      "output_tokens": 126,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25283,
      "reasoning_chunks": 95,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_1801_00013",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 13,
      "time_to_first_output_ms": 31748.05298499996,
      "tpot_ms": 18.412724520021584,
      "ttft_ms": 33568.92717500159,
      "visible_content_truncated": false,
      "visible_generation_ms": 480.7163750010659
    },
    {
      "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": 15.248083886667379,
      "e2e_ms": 36036.3317559968,
      "effective_prefill_tok_s": 857.7206019573007,
      "generation_ms": 6558.201066000038,
      "mean_inter_token_latency_ms": 65.58201066000038,
      "mean_time_per_output_token_ms": 65.58201066000038,
      "output_token_source": "usage",
      "output_tokens": 101,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25284,
      "reasoning_chunks": 71,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_1801_00014",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 14,
      "time_to_first_output_ms": 29478.13068999676,
      "tpot_ms": 65.58201066000038,
      "ttft_ms": 35657.65696299786,
      "visible_content_truncated": false,
      "visible_generation_ms": 378.674792998936
    },
    {
      "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": 73.33923620462798,
      "e2e_ms": 31049.272210999334,
      "effective_prefill_tok_s": 859.6187187196485,
      "generation_ms": 1636.2319300023955,
      "mean_inter_token_latency_ms": 13.635266083353295,
      "mean_time_per_output_token_ms": 13.635266083353295,
      "output_token_source": "usage",
      "output_tokens": 121,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25284,
      "reasoning_chunks": 90,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_1801_00015",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 15,
      "time_to_first_output_ms": 29413.04028099694,
      "tpot_ms": 13.635266083353295,
      "ttft_ms": 30714.32412300055,
      "visible_content_truncated": false,
      "visible_generation_ms": 334.9480879987823
    }
  ],
  "requests": 16,
  "response_words": 16,
  "responses_received": 16,
  "run_id": "benchmark-20260924T005048Z",
  "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": "2026-09-24T00:50:48Z",
  "started_at_unix_ns": 1790211048990975551,
  "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": 91294.46139700303
}
