{
  "base_url": "http://127.0.0.1:8080/v1",
  "clock_domain_id": null,
  "completed": 16,
  "concurrency": 1,
  "configuration_fingerprint": null,
  "context_distribution": {
    "32768": 16
  },
  "context_seed": 0,
  "context_tokens": 32768,
  "controlled_output_policy": "strict",
  "engine": "llamacpp",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-09-26T14:11:54Z",
  "finished_at_unix_ns": 1790431914652984373,
  "gpu": "RTX PRO 6000 Blackwell Max-Q",
  "manifest": null,
  "max_context_tokens": 327680,
  "max_tokens": 16384,
  "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": 269.0,
    "completion_tokens_mean": 181.1875,
    "completion_tokens_mean_ci95_high": 202.66584109632305,
    "completion_tokens_mean_ci95_low": 159.70915890367695,
    "completion_tokens_min": 124.0,
    "completion_tokens_p25": 152.0,
    "completion_tokens_p50": 167.0,
    "completion_tokens_p75": 196.0,
    "completion_tokens_p90": 247.0,
    "completion_tokens_p95": 269.0,
    "completion_tokens_p99": 269.0,
    "completion_tokens_samples": 16,
    "completion_tokens_stddev": 43.833349176169506,
    "content_chunks": 448,
    "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": 83.4889850950903,
    "decode_tok_s_mean": 82.34297717787672,
    "decode_tok_s_mean_ci95_high": 82.75830276314147,
    "decode_tok_s_mean_ci95_low": 81.92765159261197,
    "decode_tok_s_min": 80.51299697767419,
    "decode_tok_s_p25": 81.94303345920834,
    "decode_tok_s_p50": 82.16454082434856,
    "decode_tok_s_p75": 82.88346233046381,
    "decode_tok_s_p90": 83.27663240003099,
    "decode_tok_s_p95": 83.4889850950903,
    "decode_tok_s_p99": 83.4889850950903,
    "decode_tok_s_samples": 16,
    "decode_tok_s_stddev": 0.847603235234198,
    "e2e_max_ms": 17349.821961019188,
    "e2e_mean_ci95_high_ms": 15084.492575473223,
    "e2e_mean_ci95_low_ms": 13845.660933299268,
    "e2e_mean_ms": 14465.076754386246,
    "e2e_min_ms": 13057.056069024839,
    "e2e_p25_ms": 13450.418619962875,
    "e2e_p50_ms": 13733.92170900479,
    "e2e_p75_ms": 14983.891447016504,
    "e2e_p90_ms": 16289.924735028762,
    "e2e_p95_ms": 17349.821961019188,
    "e2e_p99_ms": 17349.821961019188,
    "e2e_samples": 16,
    "e2e_stddev_ms": 1264.113920585668,
    "effective_prefill_tok_s_max": 2638.91150458616,
    "effective_prefill_tok_s_mean": 2492.0667120121134,
    "effective_prefill_tok_s_mean_ci95_high": 2587.7200205028516,
    "effective_prefill_tok_s_mean_ci95_low": 2396.413403521375,
    "effective_prefill_tok_s_min": 2126.1451241464447,
    "effective_prefill_tok_s_p25": 2198.331129852536,
    "effective_prefill_tok_s_p50": 2587.8582391633527,
    "effective_prefill_tok_s_p75": 2595.5512010457583,
    "effective_prefill_tok_s_p90": 2628.2049920547684,
    "effective_prefill_tok_s_p95": 2638.91150458616,
    "effective_prefill_tok_s_p99": 2638.91150458616,
    "effective_prefill_tok_s_samples": 16,
    "effective_prefill_tok_s_stddev": 195.21083365456803,
    "generation_max_ms": 3237.0270200190134,
    "generation_mean_ci95_high_ms": 2452.5083997433962,
    "generation_mean_ci95_low_ms": 1926.2662767722084,
    "generation_mean_ms": 2189.387338257802,
    "generation_min_ms": 1491.3676650030538,
    "generation_p25_ms": 1838.908881007228,
    "generation_p50_ms": 2002.812084974721,
    "generation_p75_ms": 2374.765553046018,
    "generation_p90_ms": 3055.407316016499,
    "generation_p95_ms": 3237.0270200190134,
    "generation_p99_ms": 3237.0270200190134,
    "generation_samples": 16,
    "generation_stddev_ms": 536.9817581338651,
    "mean_inter_token_latency_ms_max": 12.4203549431565,
    "mean_inter_token_latency_ms_mean": 12.145541102830482,
    "mean_inter_token_latency_ms_mean_ci95_high": 12.207217772566379,
    "mean_inter_token_latency_ms_mean_ci95_low": 12.083864433094584,
    "mean_inter_token_latency_ms_min": 11.977627933326099,
    "mean_inter_token_latency_ms_p25": 12.020339968587923,
    "mean_inter_token_latency_ms_p50": 12.124940365878487,
    "mean_inter_token_latency_ms_p75": 12.180338211758674,
    "mean_inter_token_latency_ms_p90": 12.363545524885168,
    "mean_inter_token_latency_ms_p95": 12.4203549431565,
    "mean_inter_token_latency_ms_p99": 12.4203549431565,
    "mean_inter_token_latency_ms_samples": 16,
    "mean_inter_token_latency_ms_stddev": 0.1258707545630545,
    "mean_time_per_output_token_max_ms": 12.4203549431565,
    "mean_time_per_output_token_mean_ci95_high_ms": 12.207217772566379,
    "mean_time_per_output_token_mean_ci95_low_ms": 12.083864433094584,
    "mean_time_per_output_token_mean_ms": 12.145541102830482,
    "mean_time_per_output_token_min_ms": 11.977627933326099,
    "mean_time_per_output_token_p25_ms": 12.020339968587923,
    "mean_time_per_output_token_p50_ms": 12.124940365878487,
    "mean_time_per_output_token_p75_ms": 12.180338211758674,
    "mean_time_per_output_token_p90_ms": 12.363545524885168,
    "mean_time_per_output_token_p95_ms": 12.4203549431565,
    "mean_time_per_output_token_p99_ms": 12.4203549431565,
    "mean_time_per_output_token_samples": 16,
    "mean_time_per_output_token_stddev_ms": 0.1258707545630545,
    "output_token_sources": {
      "usage": 16
    },
    "output_tokens": 2899,
    "prefix_cache_hit_avg": 0.0,
    "prompt_token_samples": 16,
    "prompt_tokens": 486393,
    "prompt_tokens_max": 30413.0,
    "prompt_tokens_mean": 30399.5625,
    "prompt_tokens_mean_ci95_high": 30403.901029656845,
    "prompt_tokens_mean_ci95_low": 30395.223970343155,
    "prompt_tokens_min": 30383.0,
    "prompt_tokens_p25": 30392.0,
    "prompt_tokens_p50": 30400.0,
    "prompt_tokens_p75": 30406.0,
    "prompt_tokens_p90": 30411.0,
    "prompt_tokens_p95": 30413.0,
    "prompt_tokens_p99": 30413.0,
    "prompt_tokens_samples": 16,
    "prompt_tokens_stddev": 8.854142156828821,
    "reasoning_chunks": 2403,
    "request_canary_passed": 16,
    "request_canary_samples": 16,
    "throughput_performance_eligible": true,
    "throughput_tok_s": 12.525452058020356,
    "time_to_first_output_max_ms": 14294.41464500269,
    "time_to_first_output_mean_ci95_high_ms": 12790.598546564495,
    "time_to_first_output_mean_ci95_low_ms": 11760.780285692392,
    "time_to_first_output_mean_ms": 12275.689416128444,
    "time_to_first_output_min_ms": 11518.006552010775,
    "time_to_first_output_p25_ms": 11617.793901008554,
    "time_to_first_output_p50_ms": 11746.86442699749,
    "time_to_first_output_p75_ms": 11802.223214006517,
    "time_to_first_output_p90_ms": 14052.122995024547,
    "time_to_first_output_p95_ms": 14294.41464500269,
    "time_to_first_output_p99_ms": 14294.41464500269,
    "time_to_first_output_samples": 16,
    "time_to_first_output_stddev_ms": 1050.834960073576,
    "tpot_max_ms": 12.4203549431565,
    "tpot_mean_ci95_high_ms": 12.207217772566379,
    "tpot_mean_ci95_low_ms": 12.083864433094584,
    "tpot_mean_ms": 12.145541102830482,
    "tpot_min_ms": 11.977627933326099,
    "tpot_p25_ms": 12.020339968587923,
    "tpot_p50_ms": 12.124940365878487,
    "tpot_p75_ms": 12.180338211758674,
    "tpot_p90_ms": 12.363545524885168,
    "tpot_p95_ms": 12.4203549431565,
    "tpot_p99_ms": 12.4203549431565,
    "tpot_samples": 16,
    "tpot_stddev_ms": 0.1258707545630545,
    "ttft_max_ms": 17023.880088003352,
    "ttft_mean_ci95_high_ms": 14758.198147655794,
    "ttft_mean_ci95_low_ms": 13518.704400105189,
    "ttft_mean_ms": 14138.451273880491,
    "ttft_min_ms": 12730.925826006569,
    "ttft_p25_ms": 13120.715104974806,
    "ttft_p50_ms": 13408.619838010054,
    "ttft_p75_ms": 14657.136532012373,
    "ttft_p90_ms": 15964.504885021597,
    "ttft_p95_ms": 17023.880088003352,
    "ttft_p99_ms": 17023.880088003352,
    "ttft_samples": 16,
    "ttft_stddev_ms": 1264.789538316943
  },
  "model": "swift15-flash-next-iq3-327k-c4-medium-r3",
  "performance_eligible": true,
  "prompt_cache_mode": "unique",
  "request_canaries": true,
  "request_timings": [
    {
      "content_chunks": 28,
      "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": 80.88294720856685,
      "e2e_ms": 16289.924735028762,
      "effective_prefill_tok_s": 2162.1643940035074,
      "generation_ms": 2237.8017400042154,
      "mean_inter_token_latency_ms": 12.363545524885168,
      "mean_time_per_output_token_ms": 12.363545524885168,
      "output_token_source": "usage",
      "output_tokens": 182,
      "planned_context_tokens": 32768,
      "prompt_tokens": 30383,
      "reasoning_chunks": 151,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_0_00000",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 0,
      "time_to_first_output_ms": 14052.122995024547,
      "tpot_ms": 12.363545524885168,
      "ttft_ms": 15964.504885021597,
      "visible_content_truncated": false,
      "visible_generation_ms": 325.4198500071652
    },
    {
      "content_chunks": 28,
      "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": 82.16454082434856,
      "e2e_ms": 15866.981115017552,
      "effective_prefill_tok_s": 2198.331129852536,
      "generation_ms": 2044.677647005301,
      "mean_inter_token_latency_ms": 12.170700279793458,
      "mean_time_per_output_token_ms": 12.170700279793458,
      "output_token_source": "usage",
      "output_tokens": 169,
      "planned_context_tokens": 32768,
      "prompt_tokens": 30386,
      "reasoning_chunks": 138,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_0_00001",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 1,
      "time_to_first_output_ms": 13822.303468012251,
      "tpot_ms": 12.170700279793458,
      "ttft_ms": 15541.205663001165,
      "visible_content_truncated": false,
      "visible_generation_ms": 325.77545201638713
    },
    {
      "content_chunks": 28,
      "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": 83.27663240003099,
      "e2e_ms": 15787.515946023632,
      "effective_prefill_tok_s": 2180.2581553512955,
      "generation_ms": 1849.2582560284063,
      "mean_inter_token_latency_ms": 12.008170493690951,
      "mean_time_per_output_token_ms": 12.008170493690951,
      "output_token_source": "usage",
      "output_tokens": 155,
      "planned_context_tokens": 32768,
      "prompt_tokens": 30389,
      "reasoning_chunks": 124,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_0_00002",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 2,
      "time_to_first_output_ms": 13938.257689995226,
      "tpot_ms": 12.008170493690951,
      "ttft_ms": 15462.088995031081,
      "visible_content_truncated": false,
      "visible_generation_ms": 325.4269509925507
    },
    {
      "content_chunks": 28,
      "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": 80.51299697767419,
      "e2e_ms": 17349.821961019188,
      "effective_prefill_tok_s": 2126.1451241464447,
      "generation_ms": 3055.407316016499,
      "mean_inter_token_latency_ms": 12.4203549431565,
      "mean_time_per_output_token_ms": 12.4203549431565,
      "output_token_source": "usage",
      "output_tokens": 247,
      "planned_context_tokens": 32768,
      "prompt_tokens": 30392,
      "reasoning_chunks": 216,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_0_00003",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 3,
      "time_to_first_output_ms": 14294.41464500269,
      "tpot_ms": 12.4203549431565,
      "ttft_ms": 17023.880088003352,
      "visible_content_truncated": false,
      "visible_generation_ms": 325.9418730158359
    },
    {
      "content_chunks": 28,
      "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": 82.1139108955157,
      "e2e_ms": 13356.915433018003,
      "effective_prefill_tok_s": 2638.91150458616,
      "generation_ms": 1838.908881007228,
      "mean_inter_token_latency_ms": 12.178204509981642,
      "mean_time_per_output_token_ms": 12.178204509981642,
      "output_token_source": "usage",
      "output_tokens": 152,
      "planned_context_tokens": 32768,
      "prompt_tokens": 30395,
      "reasoning_chunks": 121,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_0_00004",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 4,
      "time_to_first_output_ms": 11518.006552010775,
      "tpot_ms": 12.178204509981642,
      "ttft_ms": 13031.407771981321,
      "visible_content_truncated": false,
      "visible_generation_ms": 325.5076610366814
    },
    {
      "content_chunks": 28,
      "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": 82.47463243729918,
      "e2e_ms": 13057.056069024839,
      "effective_prefill_tok_s": 2628.2049920547684,
      "generation_ms": 1491.3676650030538,
      "mean_inter_token_latency_ms": 12.124940365878487,
      "mean_time_per_output_token_ms": 12.124940365878487,
      "output_token_source": "usage",
      "output_tokens": 124,
      "planned_context_tokens": 32768,
      "prompt_tokens": 30397,
      "reasoning_chunks": 93,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_0_00005",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 5,
      "time_to_first_output_ms": 11565.688404021785,
      "tpot_ms": 12.124940365878487,
      "ttft_ms": 12730.925826006569,
      "visible_content_truncated": false,
      "visible_generation_ms": 326.13024301826954
    },
    {
      "content_chunks": 28,
      "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": 82.11336893862268,
      "e2e_ms": 13992.559454054572,
      "effective_prefill_tok_s": 2616.5897122138676,
      "generation_ms": 2374.765553046018,
      "mean_inter_token_latency_ms": 12.178284887415476,
      "mean_time_per_output_token_ms": 12.178284887415476,
      "output_token_source": "usage",
      "output_tokens": 196,
      "planned_context_tokens": 32768,
      "prompt_tokens": 30399,
      "reasoning_chunks": 165,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_0_00006",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 6,
      "time_to_first_output_ms": 11617.793901008554,
      "tpot_ms": 12.178284887415476,
      "ttft_ms": 13666.890432010405,
      "visible_content_truncated": false,
      "visible_generation_ms": 325.6690220441669
    },
    {
      "content_chunks": 28,
      "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": 82.88346233046381,
      "e2e_ms": 13616.798781964462,
      "effective_prefill_tok_s": 2617.5335647559723,
      "generation_ms": 2002.812084974721,
      "mean_inter_token_latency_ms": 12.065133042016392,
      "mean_time_per_output_token_ms": 12.065133042016392,
      "output_token_source": "usage",
      "output_tokens": 167,
      "planned_context_tokens": 32768,
      "prompt_tokens": 30400,
      "reasoning_chunks": 136,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_0_00007",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 7,
      "time_to_first_output_ms": 11613.986696989741,
      "tpot_ms": 12.065133042016392,
      "ttft_ms": 13291.271750989836,
      "visible_content_truncated": false,
      "visible_generation_ms": 325.52703097462654
    },
    {
      "content_chunks": 28,
      "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": 82.09952651681039,
      "e2e_ms": 13396.63447800558,
      "effective_prefill_tok_s": 2592.1884606838425,
      "generation_ms": 1668.7063350109383,
      "mean_inter_token_latency_ms": 12.180338211758674,
      "mean_time_per_output_token_ms": 12.180338211758674,
      "output_token_source": "usage",
      "output_tokens": 138,
      "planned_context_tokens": 32768,
      "prompt_tokens": 30401,
      "reasoning_chunks": 107,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_0_00008",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 8,
      "time_to_first_output_ms": 11727.928142994642,
      "tpot_ms": 12.180338211758674,
      "ttft_ms": 13071.220837009605,
      "visible_content_truncated": false,
      "visible_generation_ms": 325.4136409959756
    },
    {
      "content_chunks": 28,
      "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": 82.79201821380714,
      "e2e_ms": 14983.891447016504,
      "effective_prefill_tok_s": 2588.009778177931,
      "generation_ms": 3237.0270200190134,
      "mean_inter_token_latency_ms": 12.078459029921692,
      "mean_time_per_output_token_ms": 12.078459029921692,
      "output_token_source": "usage",
      "output_tokens": 269,
      "planned_context_tokens": 32768,
      "prompt_tokens": 30401,
      "reasoning_chunks": 238,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_0_00009",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 9,
      "time_to_first_output_ms": 11746.86442699749,
      "tpot_ms": 12.078459029921692,
      "ttft_ms": 14657.136532012373,
      "visible_content_truncated": false,
      "visible_generation_ms": 326.75491500413045
    },
    {
      "content_chunks": 28,
      "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": 83.24269473387005,
      "e2e_ms": 13450.418619962875,
      "effective_prefill_tok_s": 2586.1270534899377,
      "generation_ms": 1693.842329957988,
      "mean_inter_token_latency_ms": 12.01306616991481,
      "mean_time_per_output_token_ms": 12.01306616991481,
      "output_token_source": "usage",
      "output_tokens": 142,
      "planned_context_tokens": 32768,
      "prompt_tokens": 30404,
      "reasoning_chunks": 111,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_0_00010",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 10,
      "time_to_first_output_ms": 11756.576290004887,
      "tpot_ms": 12.01306616991481,
      "ttft_ms": 13120.715104974806,
      "visible_content_truncated": false,
      "visible_generation_ms": 329.7035149880685
    },
    {
      "content_chunks": 28,
      "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": 83.19232256435706,
      "e2e_ms": 13660.718450963032,
      "effective_prefill_tok_s": 2587.8582391633527,
      "generation_ms": 1911.2340550054796,
      "mean_inter_token_latency_ms": 12.020339968587923,
      "mean_time_per_output_token_ms": 12.020339968587923,
      "output_token_source": "usage",
      "output_tokens": 160,
      "planned_context_tokens": 32768,
      "prompt_tokens": 30406,
      "reasoning_chunks": 129,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_0_00011",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 11,
      "time_to_first_output_ms": 11749.484395957552,
      "tpot_ms": 12.020339968587923,
      "ttft_ms": 13334.718047990464,
      "visible_content_truncated": false,
      "visible_generation_ms": 326.0004029725678
    },
    {
      "content_chunks": 28,
      "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": 82.71928013169729,
      "e2e_ms": 13733.92170900479,
      "effective_prefill_tok_s": 2595.5512010457583,
      "generation_ms": 2018.8763699843548,
      "mean_inter_token_latency_ms": 12.089080059786555,
      "mean_time_per_output_token_ms": 12.089080059786555,
      "output_token_source": "usage",
      "output_tokens": 168,
      "planned_context_tokens": 32768,
      "prompt_tokens": 30407,
      "reasoning_chunks": 137,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_0_00012",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 12,
      "time_to_first_output_ms": 11715.045339020435,
      "tpot_ms": 12.089080059786555,
      "ttft_ms": 13408.619838010054,
      "visible_content_truncated": false,
      "visible_generation_ms": 325.3018709947355
    },
    {
      "content_chunks": 28,
      "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": 81.58728211866494,
      "e2e_ms": 13594.678049033973,
      "effective_prefill_tok_s": 2592.049645191378,
      "generation_ms": 1863.035464019049,
      "mean_inter_token_latency_ms": 12.256812263283217,
      "mean_time_per_output_token_ms": 12.256812263283217,
      "output_token_source": "usage",
      "output_tokens": 153,
      "planned_context_tokens": 32768,
      "prompt_tokens": 30409,
      "reasoning_chunks": 122,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_0_00013",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 13,
      "time_to_first_output_ms": 11731.642585014924,
      "tpot_ms": 12.256812263283217,
      "ttft_ms": 13258.364521025214,
      "visible_content_truncated": false,
      "visible_generation_ms": 336.313528008759
    },
    {
      "content_chunks": 28,
      "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": 81.94303345920834,
      "e2e_ms": 14626.537903037388,
      "effective_prefill_tok_s": 2586.2570622317953,
      "generation_ms": 2867.8459910443053,
      "mean_inter_token_latency_ms": 12.20359996189066,
      "mean_time_per_output_token_ms": 12.20359996189066,
      "output_token_source": "usage",
      "output_tokens": 236,
      "planned_context_tokens": 32768,
      "prompt_tokens": 30411,
      "reasoning_chunks": 205,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_0_00014",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 14,
      "time_to_first_output_ms": 11758.691911993083,
      "tpot_ms": 12.20359996189066,
      "ttft_ms": 14300.851432024501,
      "visible_content_truncated": false,
      "visible_generation_ms": 325.6864710128866
    },
    {
      "content_chunks": 28,
      "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": 83.4889850950903,
      "e2e_ms": 14676.853918004781,
      "effective_prefill_tok_s": 2576.887375245266,
      "generation_ms": 2874.630703998264,
      "mean_inter_token_latency_ms": 11.977627933326099,
      "mean_time_per_output_token_ms": 11.977627933326099,
      "output_token_source": "usage",
      "output_tokens": 241,
      "planned_context_tokens": 32768,
      "prompt_tokens": 30413,
      "reasoning_chunks": 210,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_0_00015",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 15,
      "time_to_first_output_ms": 11802.223214006517,
      "tpot_ms": 11.977627933326099,
      "ttft_ms": 14351.41865699552,
      "visible_content_truncated": false,
      "visible_generation_ms": 325.435261009261
    }
  ],
  "requests": 16,
  "response_words": 16,
  "responses_received": 16,
  "run_id": "benchmark-20260926T140803Z",
  "sampling": {
    "temperature": {
      "effective_request": 1.0,
      "requested": 1.0,
      "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": null,
    "shared_prefix_burst": false,
    "thinking_mode": "default"
  },
  "shared_prefix_tokens": 0,
  "source_recipe": null,
  "started_at": "2026-09-26T14:08:03Z",
  "started_at_unix_ns": 1790431683204246238,
  "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": 231448.73227499193
}
