{
  "base_url": "http://127.0.0.1:30000/v1",
  "clock_domain_id": null,
  "completed": 6,
  "concurrency": 4,
  "configuration_fingerprint": null,
  "context_distribution": {
    "8192": 8
  },
  "context_seed": 1909,
  "context_tokens": 8192,
  "controlled_output_policy": "strict",
  "engine": "vllm-v84",
  "failed": 2,
  "failures": [
    {
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": false,
        "observed_code_words": 65,
        "observed_extra_words": 0,
        "passed": false,
        "policy": "strict",
        "requested_words": 64
      },
      "elapsed_ms": 28366.036563995294,
      "error_message": "controlled output validation failed",
      "error_message_truncated": false,
      "error_type": "ValueError",
      "failure_class": "controlled_output_failed",
      "performance_eligible": false,
      "planned_context_tokens": 8192,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_1909_00000",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 0,
      "response_metadata": {
        "content_chunks": 75,
        "finish_reasons": [
          "stop"
        ],
        "output_token_source": "usage",
        "reasoning_chunks": 787,
        "stream_terminal_observed": true,
        "usage": {
          "completion_tokens": 940,
          "prompt_tokens": 6061,
          "total_tokens": 7001
        },
        "visible_content_truncated": false
      },
      "response_received": true,
      "response_timing": {
        "content_chunks": 75,
        "controlled_output": {
          "capture_complete": true,
          "exact_adherence": false,
          "observed_code_words": 65,
          "observed_extra_words": 0,
          "passed": false,
          "policy": "strict",
          "requested_words": 64
        },
        "decode_tok_s": 34.83383493429907,
        "e2e_ms": 28365.864873980172,
        "effective_prefill_tok_s": 4300.670130408416,
        "generation_ms": 26956.54962398112,
        "mean_inter_token_latency_ms": 28.707720579319616,
        "mean_time_per_output_token_ms": 28.707720579319616,
        "output_token_source": "usage",
        "output_tokens": 940,
        "planned_context_tokens": 8192,
        "prompt_tokens": 6061,
        "reasoning_chunks": 787,
        "request_canary": {
          "capture_complete": true,
          "foreign_marker_count": 0,
          "marker": "ANVIL_REQ_1909_00000",
          "marker_at_start": true,
          "observed_marker_count": 1,
          "passed": true
        },
        "request_index": 0,
        "time_to_first_output_ms": 1409.3152499990538,
        "tpot_ms": 28.707720579319616,
        "ttft_ms": 27089.871203061193,
        "visible_content_truncated": false,
        "visible_generation_ms": 1275.9936709189788
      }
    },
    {
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": false,
        "observed_code_words": 63,
        "observed_extra_words": 0,
        "passed": false,
        "policy": "strict",
        "requested_words": 64
      },
      "elapsed_ms": 8157.922583981417,
      "error_message": "controlled output validation failed",
      "error_message_truncated": false,
      "error_type": "ValueError",
      "failure_class": "controlled_output_failed",
      "performance_eligible": false,
      "planned_context_tokens": 8192,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_1909_00003",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 3,
      "response_metadata": {
        "content_chunks": 73,
        "finish_reasons": [
          "stop"
        ],
        "output_token_source": "usage",
        "reasoning_chunks": 62,
        "stream_terminal_observed": true,
        "usage": {
          "completion_tokens": 139,
          "prompt_tokens": 6061,
          "total_tokens": 6200
        },
        "visible_content_truncated": false
      },
      "response_received": true,
      "response_timing": {
        "content_chunks": 73,
        "controlled_output": {
          "capture_complete": true,
          "exact_adherence": false,
          "observed_code_words": 63,
          "observed_extra_words": 0,
          "passed": false,
          "policy": "strict",
          "requested_words": 64
        },
        "decode_tok_s": 52.55920568313009,
        "e2e_ms": 8157.716154004447,
        "effective_prefill_tok_s": 1095.6045185781866,
        "generation_ms": 2625.6104559870437,
        "mean_inter_token_latency_ms": 19.026162724543795,
        "mean_time_per_output_token_ms": 19.026162724543795,
        "output_token_source": "usage",
        "output_tokens": 139,
        "planned_context_tokens": 8192,
        "prompt_tokens": 6061,
        "reasoning_chunks": 62,
        "request_canary": {
          "capture_complete": true,
          "foreign_marker_count": 0,
          "marker": "ANVIL_REQ_1909_00003",
          "marker_at_start": true,
          "observed_marker_count": 1,
          "passed": true
        },
        "request_index": 3,
        "time_to_first_output_ms": 5532.1056980174035,
        "tpot_ms": 19.026162724543795,
        "ttft_ms": 6767.517467029393,
        "visible_content_truncated": false,
        "visible_generation_ms": 1390.1986869750544
      }
    }
  ],
  "finished_at": "2026-09-19T11:14:53Z",
  "finished_at_unix_ns": 1789816493980006448,
  "gpu": "2x-rtx-pro-6000-max-q",
  "manifest": null,
  "max_context_tokens": 327680,
  "max_tokens": 4096,
  "measurement_protocol": "capacity-v4-reasoning",
  "metric_population": {
    "aggregate_throughput_eligible": false,
    "attempted_requests": 8,
    "excluded_requests": 2,
    "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": 6,
      "e2e": 6,
      "effective_prefill_tok_s": 6,
      "tpot": 6,
      "ttft": 6
    },
    "performance_eligible_requests": 6,
    "successful_requests": 6
  },
  "metrics": {
    "completion_tokens_max": 979.0,
    "completion_tokens_mean": 488.6666666666667,
    "completion_tokens_mean_ci95_high": 720.0732384354429,
    "completion_tokens_mean_ci95_low": 257.26009489789044,
    "completion_tokens_min": 128.0,
    "completion_tokens_p25": 344.0,
    "completion_tokens_p50": 427.0,
    "completion_tokens_p75": 627.0,
    "completion_tokens_p90": 979.0,
    "completion_tokens_p95": 979.0,
    "completion_tokens_p99": 979.0,
    "completion_tokens_samples": 6,
    "completion_tokens_stddev": 289.19797140828405,
    "content_chunks": 444,
    "content_chunks_s": null,
    "controlled_output_exact_adherence": 6,
    "controlled_output_exact_adherence_rate": 1.0,
    "controlled_output_samples": 6,
    "controlled_output_unobservable": 0,
    "decode_tok_s_max": 57.20521084594358,
    "decode_tok_s_mean": 45.23452399657581,
    "decode_tok_s_mean_ci95_high": 52.16893830782217,
    "decode_tok_s_mean_ci95_low": 38.30010968532945,
    "decode_tok_s_min": 35.33334326707485,
    "decode_tok_s_p25": 38.19650362749884,
    "decode_tok_s_p50": 42.96263812885195,
    "decode_tok_s_p75": 54.01605087173826,
    "decode_tok_s_p90": 57.20521084594358,
    "decode_tok_s_p95": 57.20521084594358,
    "decode_tok_s_p99": 57.20521084594358,
    "decode_tok_s_samples": 6,
    "decode_tok_s_stddev": 8.666212616125932,
    "e2e_max_ms": 19414.546150015667,
    "e2e_mean_ci95_high_ms": 17608.100079702628,
    "e2e_mean_ci95_low_ms": 8612.186192918776,
    "e2e_mean_ms": 13110.143136310702,
    "e2e_min_ms": 3774.6014470467344,
    "e2e_p25_ms": 9272.955111926422,
    "e2e_p50_ms": 14858.949607936665,
    "e2e_p75_ms": 15992.981267045252,
    "e2e_p90_ms": 19414.546150015667,
    "e2e_p95_ms": 19414.546150015667,
    "e2e_p99_ms": 19414.546150015667,
    "e2e_samples": 6,
    "e2e_stddev_ms": 5621.275202203519,
    "effective_prefill_tok_s_max": 4261.762815893672,
    "effective_prefill_tok_s_mean": 3167.0284115146737,
    "effective_prefill_tok_s_mean_ci95_high": 4170.366469142878,
    "effective_prefill_tok_s_mean_ci95_low": 2163.6903538864694,
    "effective_prefill_tok_s_min": 1445.169409245411,
    "effective_prefill_tok_s_p25": 2162.8282257321084,
    "effective_prefill_tok_s_p50": 2614.52681389472,
    "effective_prefill_tok_s_p75": 4259.913758206635,
    "effective_prefill_tok_s_p90": 4261.762815893672,
    "effective_prefill_tok_s_p95": 4261.762815893672,
    "effective_prefill_tok_s_p99": 4261.762815893672,
    "effective_prefill_tok_s_samples": 6,
    "effective_prefill_tok_s_stddev": 1253.911367706267,
    "generation_max_ms": 17096.34464303963,
    "generation_mean_ci95_high_ms": 15013.921332567437,
    "generation_mean_ci95_low_ms": 6678.714488378288,
    "generation_mean_ms": 10846.317910472862,
    "generation_min_ms": 2351.1529989773408,
    "generation_p25_ms": 7850.156354950741,
    "generation_p50_ms": 11152.853259933181,
    "generation_p75_ms": 14570.799821987748,
    "generation_p90_ms": 17096.34464303963,
    "generation_p95_ms": 17096.34464303963,
    "generation_p99_ms": 17096.34464303963,
    "generation_samples": 6,
    "generation_stddev_ms": 5208.419303269762,
    "mean_inter_token_latency_ms_max": 28.30187883555994,
    "mean_inter_token_latency_ms_mean": 22.77316959293408,
    "mean_inter_token_latency_ms_mean_ci95_high": 26.141451735722892,
    "mean_inter_token_latency_ms_mean_ci95_low": 19.40488745014527,
    "mean_inter_token_latency_ms_min": 17.480924992883057,
    "mean_inter_token_latency_ms_p25": 18.51301573997906,
    "mean_inter_token_latency_ms_p50": 22.88675322143073,
    "mean_inter_token_latency_ms_p75": 26.180406713458172,
    "mean_inter_token_latency_ms_p90": 28.30187883555994,
    "mean_inter_token_latency_ms_p95": 28.30187883555994,
    "mean_inter_token_latency_ms_p99": 28.30187883555994,
    "mean_inter_token_latency_ms_samples": 6,
    "mean_inter_token_latency_ms_stddev": 4.209475795694354,
    "mean_time_per_output_token_max_ms": 28.30187883555994,
    "mean_time_per_output_token_mean_ci95_high_ms": 26.141451735722892,
    "mean_time_per_output_token_mean_ci95_low_ms": 19.40488745014527,
    "mean_time_per_output_token_mean_ms": 22.77316959293408,
    "mean_time_per_output_token_min_ms": 17.480924992883057,
    "mean_time_per_output_token_p25_ms": 18.51301573997906,
    "mean_time_per_output_token_p50_ms": 22.88675322143073,
    "mean_time_per_output_token_p75_ms": 26.180406713458172,
    "mean_time_per_output_token_p90_ms": 28.30187883555994,
    "mean_time_per_output_token_p95_ms": 28.30187883555994,
    "mean_time_per_output_token_p99_ms": 28.30187883555994,
    "mean_time_per_output_token_samples": 6,
    "mean_time_per_output_token_stddev_ms": 4.209475795694354,
    "output_token_sources": {
      "usage": 6
    },
    "output_tokens": 2932,
    "prefix_cache_hit_avg": null,
    "prompt_token_samples": 6,
    "prompt_tokens": 36366,
    "prompt_tokens_max": 6061.0,
    "prompt_tokens_mean": 6061.0,
    "prompt_tokens_mean_ci95_high": 6061.0,
    "prompt_tokens_mean_ci95_low": 6061.0,
    "prompt_tokens_min": 6061.0,
    "prompt_tokens_p25": 6061.0,
    "prompt_tokens_p50": 6061.0,
    "prompt_tokens_p75": 6061.0,
    "prompt_tokens_p90": 6061.0,
    "prompt_tokens_p95": 6061.0,
    "prompt_tokens_p99": 6061.0,
    "prompt_tokens_samples": 6,
    "prompt_tokens_stddev": 0.0,
    "reasoning_chunks": 2153,
    "request_canary_passed": 6,
    "request_canary_samples": 6,
    "throughput_performance_eligible": false,
    "throughput_tok_s": null,
    "time_to_first_output_max_ms": 4193.971973960288,
    "time_to_first_output_mean_ci95_high_ms": 3150.5567689185777,
    "time_to_first_output_mean_ci95_low_ms": 1377.0936827571009,
    "time_to_first_output_mean_ms": 2263.8252258378393,
    "time_to_first_output_min_ms": 1422.1814450575039,
    "time_to_first_output_p25_ms": 1422.7987569756806,
    "time_to_first_output_p50_ms": 1423.4484480693936,
    "time_to_first_output_p75_ms": 2802.3492239881307,
    "time_to_first_output_p90_ms": 4193.971973960288,
    "time_to_first_output_p95_ms": 4193.971973960288,
    "time_to_first_output_p99_ms": 4193.971973960288,
    "time_to_first_output_samples": 6,
    "time_to_first_output_stddev_ms": 1108.1835813155963,
    "tpot_max_ms": 28.30187883555994,
    "tpot_mean_ci95_high_ms": 26.141451735722892,
    "tpot_mean_ci95_low_ms": 19.40488745014527,
    "tpot_mean_ms": 22.77316959293408,
    "tpot_min_ms": 17.480924992883057,
    "tpot_p25_ms": 18.51301573997906,
    "tpot_p50_ms": 22.88675322143073,
    "tpot_p75_ms": 26.180406713458172,
    "tpot_p90_ms": 28.30187883555994,
    "tpot_p95_ms": 28.30187883555994,
    "tpot_p99_ms": 28.30187883555994,
    "tpot_samples": 6,
    "tpot_stddev_ms": 4.209475795694354,
    "ttft_max_ms": 18419.819155009463,
    "ttft_mean_ci95_high_ms": 16248.802828815384,
    "ttft_mean_ci95_low_ms": 7176.33163950382,
    "ttft_mean_ms": 11712.567234159602,
    "ttft_min_ms": 2416.566329076886,
    "ttft_p25_ms": 7877.685210900381,
    "ttft_p50_ms": 13453.5202489933,
    "ttft_p75_ms": 14597.232555039227,
    "ttft_p90_ms": 18419.819155009463,
    "ttft_p95_ms": 18419.819155009463,
    "ttft_p99_ms": 18419.819155009463,
    "ttft_samples": 6,
    "ttft_stddev_ms": 5669.113550998615
  },
  "model": "glm53-flash-exl3-4bpw-v84-fp8-tp2-c4-327k-r9-vision8-nospec",
  "performance_eligible": false,
  "prompt_cache_mode": "shared",
  "request_canaries": true,
  "request_timings": [
    {
      "content_chunks": 74,
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": true,
        "observed_code_words": 64,
        "observed_extra_words": 0,
        "passed": true,
        "policy": "strict",
        "requested_words": 64
      },
      "decode_tok_s": 35.33334326707485,
      "e2e_ms": 14858.949607936665,
      "effective_prefill_tok_s": 2162.8282257321084,
      "generation_ms": 12056.600383948535,
      "mean_inter_token_latency_ms": 28.30187883555994,
      "mean_time_per_output_token_ms": 28.30187883555994,
      "output_token_source": "usage",
      "output_tokens": 427,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6061,
      "reasoning_chunks": 281,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_1909_00001",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 1,
      "time_to_first_output_ms": 2802.3492239881307,
      "tpot_ms": 28.30187883555994,
      "ttft_ms": 13453.5202489933,
      "visible_content_truncated": false,
      "visible_generation_ms": 1405.4293589433655
    },
    {
      "content_chunks": 74,
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": true,
        "observed_code_words": 64,
        "observed_extra_words": 0,
        "passed": true,
        "policy": "strict",
        "requested_words": 64
      },
      "decode_tok_s": 38.19650362749884,
      "e2e_ms": 15346.825233893469,
      "effective_prefill_tok_s": 1445.169409245411,
      "generation_ms": 11152.853259933181,
      "mean_inter_token_latency_ms": 26.180406713458172,
      "mean_time_per_output_token_ms": 26.180406713458172,
      "output_token_source": "usage",
      "output_tokens": 427,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6061,
      "reasoning_chunks": 282,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_1909_00002",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 2,
      "time_to_first_output_ms": 4193.971973960288,
      "tpot_ms": 26.180406713458172,
      "ttft_ms": 13510.579905938357,
      "visible_content_truncated": false,
      "visible_generation_ms": 1836.2453279551119
    },
    {
      "content_chunks": 74,
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": true,
        "observed_code_words": 64,
        "observed_extra_words": 0,
        "passed": true,
        "policy": "strict",
        "requested_words": 64
      },
      "decode_tok_s": 42.96263812885195,
      "e2e_ms": 15992.981267045252,
      "effective_prefill_tok_s": 4261.762815893672,
      "generation_ms": 14570.799821987748,
      "mean_inter_token_latency_ms": 23.276038054293526,
      "mean_time_per_output_token_ms": 23.276038054293526,
      "output_token_source": "usage",
      "output_tokens": 627,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6061,
      "reasoning_chunks": 480,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_1909_00004",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 4,
      "time_to_first_output_ms": 1422.1814450575039,
      "tpot_ms": 23.276038054293526,
      "ttft_ms": 14597.232555039227,
      "visible_content_truncated": false,
      "visible_generation_ms": 1395.748712006025
    },
    {
      "content_chunks": 74,
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": true,
        "observed_code_words": 64,
        "observed_extra_words": 0,
        "passed": true,
        "policy": "strict",
        "requested_words": 64
      },
      "decode_tok_s": 43.6933972383474,
      "e2e_ms": 9272.955111926422,
      "effective_prefill_tok_s": 4259.913758206635,
      "generation_ms": 7850.156354950741,
      "mean_inter_token_latency_ms": 22.88675322143073,
      "mean_time_per_output_token_ms": 22.88675322143073,
      "output_token_source": "usage",
      "output_tokens": 344,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6061,
      "reasoning_chunks": 254,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_1909_00005",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 5,
      "time_to_first_output_ms": 1422.7987569756806,
      "tpot_ms": 22.88675322143073,
      "ttft_ms": 7877.685210900381,
      "visible_content_truncated": false,
      "visible_generation_ms": 1395.2699010260403
    },
    {
      "content_chunks": 74,
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": true,
        "observed_code_words": 64,
        "observed_extra_words": 0,
        "passed": true,
        "policy": "strict",
        "requested_words": 64
      },
      "decode_tok_s": 57.20521084594358,
      "e2e_ms": 19414.546150015667,
      "effective_prefill_tok_s": 2614.52681389472,
      "generation_ms": 17096.34464303963,
      "mean_inter_token_latency_ms": 17.480924992883057,
      "mean_time_per_output_token_ms": 17.480924992883057,
      "output_token_source": "usage",
      "output_tokens": 979,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6061,
      "reasoning_chunks": 806,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_1909_00006",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 6,
      "time_to_first_output_ms": 2318.201506976038,
      "tpot_ms": 17.480924992883057,
      "ttft_ms": 18419.819155009463,
      "visible_content_truncated": false,
      "visible_generation_ms": 994.7269950062037
    },
    {
      "content_chunks": 74,
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": true,
        "observed_code_words": 64,
        "observed_extra_words": 0,
        "passed": true,
        "policy": "strict",
        "requested_words": 64
      },
      "decode_tok_s": 54.01605087173826,
      "e2e_ms": 3774.6014470467344,
      "effective_prefill_tok_s": 4257.969446115497,
      "generation_ms": 2351.1529989773408,
      "mean_inter_token_latency_ms": 18.51301573997906,
      "mean_time_per_output_token_ms": 18.51301573997906,
      "output_token_source": "usage",
      "output_tokens": 128,
      "planned_context_tokens": 8192,
      "prompt_tokens": 6061,
      "reasoning_chunks": 50,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_1909_00007",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 7,
      "time_to_first_output_ms": 1423.4484480693936,
      "tpot_ms": 18.51301573997906,
      "ttft_ms": 2416.566329076886,
      "visible_content_truncated": false,
      "visible_generation_ms": 1358.0351179698482
    }
  ],
  "requests": 8,
  "response_words": 64,
  "responses_received": 8,
  "run_id": "benchmark-20260919T111419Z",
  "sampling": {
    "temperature": {
      "effective_request": 0.0,
      "requested": 0.0,
      "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": "max",
    "shared_prefix_burst": false,
    "thinking_mode": "default"
  },
  "shared_prefix_tokens": 8000,
  "source_recipe": null,
  "started_at": "2026-09-19T11:14:19Z",
  "started_at_unix_ns": 1789816459217255479,
  "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": 34762.74745899718
}
