{
  "base_url": "http://127.0.0.1:8080/v1",
  "clock_domain_id": null,
  "completed": 16,
  "concurrency": 4,
  "configuration_fingerprint": null,
  "context_distribution": {
    "32768": 16
  },
  "context_seed": 0,
  "context_tokens": 32768,
  "controlled_output_policy": "strict",
  "engine": "llamacpp",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-09-26T14:20:19Z",
  "finished_at_unix_ns": 1790432419385210870,
  "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": 322.0,
    "completion_tokens_mean": 203.0,
    "completion_tokens_mean_ci95_high": 231.98602996847504,
    "completion_tokens_mean_ci95_low": 174.01397003152496,
    "completion_tokens_min": 134.0,
    "completion_tokens_p25": 158.0,
    "completion_tokens_p50": 170.0,
    "completion_tokens_p75": 247.0,
    "completion_tokens_p90": 287.0,
    "completion_tokens_p95": 322.0,
    "completion_tokens_p99": 322.0,
    "completion_tokens_samples": 16,
    "completion_tokens_stddev": 59.155163200969476,
    "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": 59.707597209425515,
    "decode_tok_s_mean": 9.397557864150953,
    "decode_tok_s_mean_ci95_high": 16.042374493566587,
    "decode_tok_s_mean_ci95_low": 2.7527412347353186,
    "decode_tok_s_min": 3.7207224530960232,
    "decode_tok_s_p25": 4.425477053085371,
    "decode_tok_s_p50": 5.735438982714025,
    "decode_tok_s_p75": 6.772009305752865,
    "decode_tok_s_p90": 10.034053566215704,
    "decode_tok_s_p95": 59.707597209425515,
    "decode_tok_s_p99": 59.707597209425515,
    "decode_tok_s_samples": 16,
    "decode_tok_s_stddev": 13.560850264113538,
    "e2e_max_ms": 90221.0741470335,
    "e2e_mean_ci95_high_ms": 62467.32184103422,
    "e2e_mean_ci95_low_ms": 43313.04922084854,
    "e2e_mean_ms": 52890.18553094138,
    "e2e_min_ms": 17599.185872997623,
    "e2e_p25_ms": 40321.716309990734,
    "e2e_p50_ms": 51448.243671038654,
    "e2e_p75_ms": 66231.96768999333,
    "e2e_p90_ms": 75669.9626739719,
    "e2e_p95_ms": 90221.0741470335,
    "e2e_p99_ms": 90221.0741470335,
    "e2e_samples": 16,
    "e2e_stddev_ms": 19545.176143046618,
    "effective_prefill_tok_s_max": 2467.961283848623,
    "effective_prefill_tok_s_mean": 1950.2183813764127,
    "effective_prefill_tok_s_mean_ci95_high": 2272.586210256717,
    "effective_prefill_tok_s_mean_ci95_low": 1627.850552496108,
    "effective_prefill_tok_s_min": 623.256323701409,
    "effective_prefill_tok_s_p25": 1233.3715929637374,
    "effective_prefill_tok_s_p50": 2122.922900989577,
    "effective_prefill_tok_s_p75": 2447.665481039272,
    "effective_prefill_tok_s_p90": 2466.9857033576186,
    "effective_prefill_tok_s_p95": 2467.961283848623,
    "effective_prefill_tok_s_p99": 2467.961283848623,
    "effective_prefill_tok_s_samples": 16,
    "effective_prefill_tok_s_stddev": 657.8935283271522,
    "generation_max_ms": 64625.80114399316,
    "generation_mean_ci95_high_ms": 43247.528104113175,
    "generation_mean_ci95_low_ms": 25668.765078406264,
    "generation_mean_ms": 34458.14659125972,
    "generation_min_ms": 5376.2002660078,
    "generation_p25_ms": 16742.984167998657,
    "generation_p50_ms": 28591.633408970665,
    "generation_p75_ms": 40167.47894702712,
    "generation_p90_ms": 61047.93173103826,
    "generation_p95_ms": 64625.80114399316,
    "generation_p99_ms": 64625.80114399316,
    "generation_samples": 16,
    "generation_stddev_ms": 17937.513291537667,
    "mean_inter_token_latency_ms_max": 268.76500803436636,
    "mean_inter_token_latency_ms_mean": 171.6191517151357,
    "mean_inter_token_latency_ms_mean_ci95_high": 204.4600111519698,
    "mean_inter_token_latency_ms_mean_ci95_low": 138.7782922783016,
    "mean_inter_token_latency_ms_min": 16.748287433046105,
    "mean_inter_token_latency_ms_p25": 113.13418163585344,
    "mean_inter_token_latency_ms_p50": 168.24907796476697,
    "mean_inter_token_latency_ms_p75": 218.09485374603443,
    "mean_inter_token_latency_ms_p90": 267.7540865396415,
    "mean_inter_token_latency_ms_p95": 268.76500803436636,
    "mean_inter_token_latency_ms_p99": 268.76500803436636,
    "mean_inter_token_latency_ms_samples": 16,
    "mean_inter_token_latency_ms_stddev": 67.02216211598801,
    "mean_time_per_output_token_max_ms": 268.76500803436636,
    "mean_time_per_output_token_mean_ci95_high_ms": 204.4600111519698,
    "mean_time_per_output_token_mean_ci95_low_ms": 138.7782922783016,
    "mean_time_per_output_token_mean_ms": 171.6191517151357,
    "mean_time_per_output_token_min_ms": 16.748287433046105,
    "mean_time_per_output_token_p25_ms": 113.13418163585344,
    "mean_time_per_output_token_p50_ms": 168.24907796476697,
    "mean_time_per_output_token_p75_ms": 218.09485374603443,
    "mean_time_per_output_token_p90_ms": 267.7540865396415,
    "mean_time_per_output_token_p95_ms": 268.76500803436636,
    "mean_time_per_output_token_p99_ms": 268.76500803436636,
    "mean_time_per_output_token_samples": 16,
    "mean_time_per_output_token_stddev_ms": 67.02216211598801,
    "output_token_sources": {
      "usage": 16
    },
    "output_tokens": 3248,
    "prefix_cache_hit_avg": 0.0,
    "prompt_token_samples": 16,
    "prompt_tokens": 476731,
    "prompt_tokens_max": 29810.0,
    "prompt_tokens_mean": 29795.6875,
    "prompt_tokens_mean_ci95_high": 29801.08408499269,
    "prompt_tokens_mean_ci95_low": 29790.29091500731,
    "prompt_tokens_min": 29779.0,
    "prompt_tokens_p25": 29785.0,
    "prompt_tokens_p50": 29794.0,
    "prompt_tokens_p75": 29806.0,
    "prompt_tokens_p90": 29810.0,
    "prompt_tokens_p95": 29810.0,
    "prompt_tokens_p99": 29810.0,
    "prompt_tokens_samples": 16,
    "prompt_tokens_stddev": 11.013438760593049,
    "reasoning_chunks": 2752,
    "request_canary_passed": 16,
    "request_canary_samples": 16,
    "throughput_performance_eligible": true,
    "throughput_tok_s": 15.1624751998207,
    "time_to_first_output_max_ms": 47779.69972795108,
    "time_to_first_output_mean_ci95_high_ms": 23624.43176583701,
    "time_to_first_output_mean_ci95_low_ms": 13239.646113526313,
    "time_to_first_output_mean_ms": 18432.03893968166,
    "time_to_first_output_min_ms": 12076.275901985355,
    "time_to_first_output_p25_ms": 12144.443960976787,
    "time_to_first_output_p50_ms": 12290.353915013839,
    "time_to_first_output_p75_ms": 19653.2639900106,
    "time_to_first_output_p90_ms": 37680.53749395767,
    "time_to_first_output_p95_ms": 47779.69972795108,
    "time_to_first_output_p99_ms": 47779.69972795108,
    "time_to_first_output_samples": 16,
    "time_to_first_output_stddev_ms": 10596.72005337826,
    "tpot_max_ms": 268.76500803436636,
    "tpot_mean_ci95_high_ms": 204.4600111519698,
    "tpot_mean_ci95_low_ms": 138.7782922783016,
    "tpot_mean_ms": 171.6191517151357,
    "tpot_min_ms": 16.748287433046105,
    "tpot_p25_ms": 113.13418163585344,
    "tpot_p50_ms": 168.24907796476697,
    "tpot_p75_ms": 218.09485374603443,
    "tpot_p90_ms": 267.7540865396415,
    "tpot_p95_ms": 268.76500803436636,
    "tpot_p99_ms": 268.76500803436636,
    "tpot_samples": 16,
    "tpot_stddev_ms": 67.02216211598801,
    "ttft_max_ms": 89372.11955501698,
    "ttft_mean_ci95_high_ms": 56472.59232334899,
    "ttft_mean_ci95_low_ms": 38326.363780775966,
    "ttft_mean_ms": 47399.47805206248,
    "ttft_min_ms": 16328.819906979334,
    "ttft_p25_ms": 27907.396355003584,
    "ttft_p50_ms": 50576.29493804416,
    "ttft_p75_ms": 54880.80775900744,
    "ttft_p90_ms": 64729.353005008306,
    "ttft_p95_ms": 89372.11955501698,
    "ttft_p99_ms": 89372.11955501698,
    "ttft_samples": 16,
    "ttft_stddev_ms": 18516.559737319407
  },
  "model": "swift15-flash-next-iq3-327k-c4-medium-r3",
  "performance_eligible": true,
  "prompt_cache_mode": "shared",
  "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": 5.629204726533411,
      "e2e_ms": 75669.9626739719,
      "effective_prefill_tok_s": 623.256323701409,
      "generation_ms": 27890.26294602081,
      "mean_inter_token_latency_ms": 177.64498691733002,
      "mean_time_per_output_token_ms": 177.64498691733002,
      "output_token_source": "usage",
      "output_tokens": 158,
      "planned_context_tokens": 32768,
      "prompt_tokens": 29779,
      "reasoning_chunks": 127,
      "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": 47779.69972795108,
      "tpot_ms": 177.64498691733002,
      "ttft_ms": 58332.81241595978,
      "visible_content_truncated": false,
      "visible_generation_ms": 17337.150258012116
    },
    {
      "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": 4.425477053085371,
      "e2e_ms": 90221.0741470335,
      "effective_prefill_tok_s": 1163.5351573105881,
      "generation_ms": 64625.80114399316,
      "mean_inter_token_latency_ms": 225.96433966431175,
      "mean_time_per_output_token_ms": 225.96433966431175,
      "output_token_source": "usage",
      "output_tokens": 287,
      "planned_context_tokens": 32768,
      "prompt_tokens": 29781,
      "reasoning_chunks": 256,
      "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": 25595.273003040347,
      "tpot_ms": 225.96433966431175,
      "ttft_ms": 89372.11955501698,
      "visible_content_truncated": false,
      "visible_generation_ms": 848.9545920165256
    },
    {
      "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": 9.756038709127074,
      "e2e_ms": 51313.11972998083,
      "effective_prefill_tok_s": 790.4080456595373,
      "generation_ms": 13632.582236023154,
      "mean_inter_token_latency_ms": 102.50061831596356,
      "mean_time_per_output_token_ms": 102.50061831596356,
      "output_token_source": "usage",
      "output_tokens": 134,
      "planned_context_tokens": 32768,
      "prompt_tokens": 29783,
      "reasoning_chunks": 103,
      "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": 37680.53749395767,
      "tpot_ms": 102.50061831596356,
      "ttft_ms": 50489.17845299002,
      "visible_content_truncated": false,
      "visible_generation_ms": 823.9412769908085
    },
    {
      "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": 4.253335048156181,
      "e2e_ms": 53025.39634302957,
      "effective_prefill_tok_s": 2092.7736302728013,
      "generation_ms": 38793.08780800784,
      "mean_inter_token_latency_ms": 235.1096230788354,
      "mean_time_per_output_token_ms": 235.1096230788354,
      "output_token_source": "usage",
      "output_tokens": 166,
      "planned_context_tokens": 32768,
      "prompt_tokens": 29785,
      "reasoning_chunks": 135,
      "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": 14232.30853502173,
      "tpot_ms": 235.1096230788354,
      "ttft_ms": 50576.29493804416,
      "visible_content_truncated": false,
      "visible_generation_ms": 2449.101404985413
    },
    {
      "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": 5.94356897580984,
      "e2e_ms": 52584.15582001908,
      "effective_prefill_tok_s": 1233.3715929637374,
      "generation_ms": 28434.09417604562,
      "mean_inter_token_latency_ms": 168.24907796476697,
      "mean_time_per_output_token_ms": 168.24907796476697,
      "output_token_source": "usage",
      "output_tokens": 170,
      "planned_context_tokens": 32768,
      "prompt_tokens": 29786,
      "reasoning_chunks": 139,
      "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": 24150.06164397346,
      "tpot_ms": 168.24907796476697,
      "ttft_ms": 51730.96237401478,
      "visible_content_truncated": false,
      "visible_generation_ms": 853.1934460042976
    },
    {
      "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": 3.7347702622344405,
      "e2e_ms": 75536.55871300725,
      "effective_prefill_tok_s": 2056.026429355399,
      "generation_ms": 61047.93173103826,
      "mean_inter_token_latency_ms": 267.7540865396415,
      "mean_time_per_output_token_ms": 267.7540865396415,
      "output_token_source": "usage",
      "output_tokens": 229,
      "planned_context_tokens": 32768,
      "prompt_tokens": 29789,
      "reasoning_chunks": 198,
      "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": 14488.626981968991,
      "tpot_ms": 267.7540865396415,
      "ttft_ms": 64729.353005008306,
      "visible_content_truncated": false,
      "visible_generation_ms": 10807.205707998946
    },
    {
      "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": 5.735438982714025,
      "e2e_ms": 40321.716309990734,
      "effective_prefill_tok_s": 2466.9857033576186,
      "generation_ms": 28245.44040800538,
      "mean_inter_token_latency_ms": 174.35457041978628,
      "mean_time_per_output_token_ms": 174.35457041978628,
      "output_token_source": "usage",
      "output_tokens": 163,
      "planned_context_tokens": 32768,
      "prompt_tokens": 29792,
      "reasoning_chunks": 132,
      "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": 12076.275901985355,
      "tpot_ms": 174.35457041978628,
      "ttft_ms": 27907.396355003584,
      "visible_content_truncated": false,
      "visible_generation_ms": 12414.31995498715
    },
    {
      "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": 4.804581562106524,
      "e2e_ms": 71816.86965900008,
      "effective_prefill_tok_s": 2424.1775465557416,
      "generation_ms": 59526.51574398624,
      "mean_inter_token_latency_ms": 208.13467043351832,
      "mean_time_per_output_token_ms": 208.13467043351832,
      "output_token_source": "usage",
      "output_tokens": 287,
      "planned_context_tokens": 32768,
      "prompt_tokens": 29794,
      "reasoning_chunks": 256,
      "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": 12290.353915013839,
      "tpot_ms": 208.13467043351832,
      "ttft_ms": 54880.80775900744,
      "visible_content_truncated": false,
      "visible_generation_ms": 16936.061899992637
    },
    {
      "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": 6.772009305752865,
      "e2e_ms": 41263.97279696539,
      "effective_prefill_tok_s": 2447.665481039272,
      "generation_ms": 29090.3321459773,
      "mean_inter_token_latency_ms": 147.6666606394787,
      "mean_time_per_output_token_ms": 147.6666606394787,
      "output_token_source": "usage",
      "output_tokens": 198,
      "planned_context_tokens": 32768,
      "prompt_tokens": 29797,
      "reasoning_chunks": 167,
      "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": 12173.64065098809,
      "tpot_ms": 147.6666606394787,
      "ttft_ms": 40429.427813971415,
      "visible_content_truncated": false,
      "visible_generation_ms": 834.5449829939753
    },
    {
      "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": 4.585160918856311,
      "e2e_ms": 66231.96768999333,
      "effective_prefill_tok_s": 2453.7969869806343,
      "generation_ms": 54087.52372901654,
      "mean_inter_token_latency_ms": 218.09485374603443,
      "mean_time_per_output_token_ms": 218.09485374603443,
      "output_token_source": "usage",
      "output_tokens": 249,
      "planned_context_tokens": 32768,
      "prompt_tokens": 29800,
      "reasoning_chunks": 218,
      "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": 12144.443960976787,
      "tpot_ms": 218.09485374603443,
      "ttft_ms": 57690.4152649804,
      "visible_content_truncated": false,
      "visible_generation_ms": 8541.552425012924
    },
    {
      "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": 10.034053566215704,
      "e2e_ms": 28840.24007100379,
      "effective_prefill_tok_s": 2463.616562215279,
      "generation_ms": 16742.984167998657,
      "mean_inter_token_latency_ms": 99.66062004761105,
      "mean_time_per_output_token_ms": 99.66062004761105,
      "output_token_source": "usage",
      "output_tokens": 169,
      "planned_context_tokens": 32768,
      "prompt_tokens": 29803,
      "reasoning_chunks": 138,
      "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": 12097.255903005134,
      "tpot_ms": 99.66062004761105,
      "ttft_ms": 16328.819906979334,
      "visible_content_truncated": false,
      "visible_generation_ms": 12511.420164024457
    },
    {
      "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": 3.7207224530960232,
      "e2e_ms": 51448.243671038654,
      "effective_prefill_tok_s": 2441.4032707670412,
      "generation_ms": 39239.69117301749,
      "mean_inter_token_latency_ms": 268.76500803436636,
      "mean_time_per_output_token_ms": 268.76500803436636,
      "output_token_source": "usage",
      "output_tokens": 147,
      "planned_context_tokens": 32768,
      "prompt_tokens": 29806,
      "reasoning_chunks": 116,
      "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": 12208.552498021163,
      "tpot_ms": 268.76500803436636,
      "ttft_ms": 50628.11299500754,
      "visible_content_truncated": false,
      "visible_generation_ms": 820.1306760311127
    },
    {
      "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": 6.124357476465597,
      "e2e_ms": 54208.02652701968,
      "effective_prefill_tok_s": 2122.922900989577,
      "generation_ms": 40167.47894702712,
      "mean_inter_token_latency_ms": 163.28243474401268,
      "mean_time_per_output_token_ms": 163.28243474401268,
      "output_token_source": "usage",
      "output_tokens": 247,
      "planned_context_tokens": 32768,
      "prompt_tokens": 29807,
      "reasoning_chunks": 216,
      "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": 14040.547579992563,
      "tpot_ms": 163.28243474401268,
      "ttft_ms": 53492.27298103506,
      "visible_content_truncated": false,
      "visible_generation_ms": 715.7535459846258
    },
    {
      "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": 6.295547981652729,
      "e2e_ms": 48244.897398981266,
      "effective_prefill_tok_s": 1516.7455143914708,
      "generation_ms": 28591.633408970665,
      "mean_inter_token_latency_ms": 158.8424078276148,
      "mean_time_per_output_token_ms": 158.8424078276148,
      "output_token_source": "usage",
      "output_tokens": 181,
      "planned_context_tokens": 32768,
      "prompt_tokens": 29809,
      "reasoning_chunks": 150,
      "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": 19653.2639900106,
      "tpot_ms": 158.8424078276148,
      "ttft_ms": 47430.60947797494,
      "visible_content_truncated": false,
      "visible_generation_ms": 814.2879210063256
    },
    {
      "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": 8.839061595183619,
      "e2e_ms": 27917.581071029417,
      "effective_prefill_tok_s": 2467.961283848623,
      "generation_ms": 15838.785429019481,
      "mean_inter_token_latency_ms": 113.13418163585344,
      "mean_time_per_output_token_ms": 113.13418163585344,
      "output_token_source": "usage",
      "output_tokens": 141,
      "planned_context_tokens": 32768,
      "prompt_tokens": 29810,
      "reasoning_chunks": 110,
      "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": 12078.795642009936,
      "tpot_ms": 113.13418163585344,
      "ttft_ms": 27099.008924036752,
      "visible_content_truncated": false,
      "visible_generation_ms": 818.5721469926648
    },
    {
      "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": 59.707597209425515,
      "e2e_ms": 17599.185872997623,
      "effective_prefill_tok_s": 2438.8476726138733,
      "generation_ms": 5376.2002660078,
      "mean_inter_token_latency_ms": 16.748287433046105,
      "mean_time_per_output_token_ms": 16.748287433046105,
      "output_token_source": "usage",
      "output_tokens": 322,
      "planned_context_tokens": 32768,
      "prompt_tokens": 29810,
      "reasoning_chunks": 291,
      "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": 12222.985606989823,
      "tpot_ms": 16.748287433046105,
      "ttft_ms": 17274.05661396915,
      "visible_content_truncated": false,
      "visible_generation_ms": 325.129259028472
    }
  ],
  "requests": 16,
  "response_words": 16,
  "responses_received": 16,
  "run_id": "benchmark-20260926T141645Z",
  "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": 8000,
  "source_recipe": null,
  "started_at": "2026-09-26T14:16:45Z",
  "started_at_unix_ns": 1790432205172161652,
  "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": 214213.04616797715
}
