{
  "base_url": "http://127.0.0.1:30001/v1",
  "clock_domain_id": null,
  "completed": 8,
  "concurrency": 4,
  "configuration_fingerprint": null,
  "context_distribution": {
    "32768": 8
  },
  "context_seed": 1910,
  "context_tokens": 32768,
  "controlled_output_policy": "strict",
  "engine": "vllm-v84",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-09-19T11:23:21Z",
  "finished_at_unix_ns": 1789817001986633382,
  "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": true,
    "attempted_requests": 8,
    "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": 8,
      "e2e": 8,
      "effective_prefill_tok_s": 8,
      "tpot": 8,
      "ttft": 8
    },
    "performance_eligible_requests": 8,
    "successful_requests": 8
  },
  "metrics": {
    "completion_tokens_max": 183.0,
    "completion_tokens_mean": 141.625,
    "completion_tokens_mean_ci95_high": 165.67835371627,
    "completion_tokens_mean_ci95_low": 117.57164628373,
    "completion_tokens_min": 100.0,
    "completion_tokens_p25": 113.0,
    "completion_tokens_p50": 124.0,
    "completion_tokens_p75": 182.0,
    "completion_tokens_p90": 183.0,
    "completion_tokens_p95": 183.0,
    "completion_tokens_p99": 183.0,
    "completion_tokens_samples": 8,
    "completion_tokens_stddev": 34.710794945006455,
    "content_chunks": 216,
    "content_chunks_s": null,
    "controlled_output_exact_adherence": 8,
    "controlled_output_exact_adherence_rate": 1.0,
    "controlled_output_samples": 8,
    "controlled_output_unobservable": 0,
    "decode_tok_s_max": 52.80673985524007,
    "decode_tok_s_mean": 15.572424309079366,
    "decode_tok_s_mean_ci95_high": 26.705288393939924,
    "decode_tok_s_mean_ci95_low": 4.439560224218807,
    "decode_tok_s_min": 5.072052785394339,
    "decode_tok_s_p25": 5.670074881363838,
    "decode_tok_s_p50": 8.926274592735142,
    "decode_tok_s_p75": 12.51642179952517,
    "decode_tok_s_p90": 52.80673985524007,
    "decode_tok_s_p95": 52.80673985524007,
    "decode_tok_s_p99": 52.80673985524007,
    "decode_tok_s_samples": 8,
    "decode_tok_s_stddev": 16.06555854782259,
    "e2e_max_ms": 41797.5598910125,
    "e2e_mean_ci95_high_ms": 35864.49049547991,
    "e2e_mean_ci95_low_ms": 22066.648393540163,
    "e2e_mean_ms": 28965.569444510038,
    "e2e_min_ms": 15408.510389039293,
    "e2e_p25_ms": 17391.354557010345,
    "e2e_p50_ms": 26781.01452696137,
    "e2e_p75_ms": 34433.878888958134,
    "e2e_p90_ms": 41797.5598910125,
    "e2e_p95_ms": 41797.5598910125,
    "e2e_p99_ms": 41797.5598910125,
    "e2e_samples": 8,
    "e2e_stddev_ms": 9955.660934717189,
    "effective_prefill_tok_s_max": 3887.72818986517,
    "effective_prefill_tok_s_mean": 2333.032414993296,
    "effective_prefill_tok_s_mean_ci95_high": 3101.285138262411,
    "effective_prefill_tok_s_mean_ci95_low": 1564.7796917241808,
    "effective_prefill_tok_s_min": 855.4691860894085,
    "effective_prefill_tok_s_p25": 1061.0021518394312,
    "effective_prefill_tok_s_p50": 1930.9942639344656,
    "effective_prefill_tok_s_p75": 2724.1031790776715,
    "effective_prefill_tok_s_p90": 3887.72818986517,
    "effective_prefill_tok_s_p95": 3887.72818986517,
    "effective_prefill_tok_s_p99": 3887.72818986517,
    "effective_prefill_tok_s_samples": 8,
    "effective_prefill_tok_s_stddev": 1108.6463475278026,
    "generation_max_ms": 24250.536262989044,
    "generation_mean_ci95_high_ms": 20166.412963537754,
    "generation_mean_ci95_low_ms": 9731.365075212647,
    "generation_mean_ms": 14948.889019375201,
    "generation_min_ms": 1874.7606890974566,
    "generation_p25_ms": 7990.7700079493225,
    "generation_p50_ms": 14461.001946008764,
    "generation_p75_ms": 20277.21622493118,
    "generation_p90_ms": 24250.536262989044,
    "generation_p95_ms": 24250.536262989044,
    "generation_p99_ms": 24250.536262989044,
    "generation_samples": 8,
    "generation_stddev_ms": 7529.278697796988,
    "mean_inter_token_latency_ms_max": 197.158831406415,
    "mean_inter_token_latency_ms_mean": 109.11706295153081,
    "mean_inter_token_latency_ms_mean_ci95_high": 151.9074900028667,
    "mean_inter_token_latency_ms_mean_ci95_low": 66.32663590019493,
    "mean_inter_token_latency_ms_min": 18.936976657550066,
    "mean_inter_token_latency_ms_p25": 43.905329714007266,
    "mean_inter_token_latency_ms_p50": 102.0571773031926,
    "mean_inter_token_latency_ms_p75": 142.58981737334074,
    "mean_inter_token_latency_ms_p90": 197.158831406415,
    "mean_inter_token_latency_ms_p95": 197.158831406415,
    "mean_inter_token_latency_ms_p99": 197.158831406415,
    "mean_inter_token_latency_ms_samples": 8,
    "mean_inter_token_latency_ms_stddev": 61.749798240546724,
    "mean_time_per_output_token_max_ms": 197.158831406415,
    "mean_time_per_output_token_mean_ci95_high_ms": 151.9074900028667,
    "mean_time_per_output_token_mean_ci95_low_ms": 66.32663590019493,
    "mean_time_per_output_token_mean_ms": 109.11706295153081,
    "mean_time_per_output_token_min_ms": 18.936976657550066,
    "mean_time_per_output_token_p25_ms": 43.905329714007266,
    "mean_time_per_output_token_p50_ms": 102.0571773031926,
    "mean_time_per_output_token_p75_ms": 142.58981737334074,
    "mean_time_per_output_token_p90_ms": 197.158831406415,
    "mean_time_per_output_token_p95_ms": 197.158831406415,
    "mean_time_per_output_token_p99_ms": 197.158831406415,
    "mean_time_per_output_token_samples": 8,
    "mean_time_per_output_token_stddev_ms": 61.749798240546724,
    "output_token_sources": {
      "usage": 8
    },
    "output_tokens": 1133,
    "prefix_cache_hit_avg": null,
    "prompt_token_samples": 8,
    "prompt_tokens": 202271,
    "prompt_tokens_max": 25285.0,
    "prompt_tokens_mean": 25283.875,
    "prompt_tokens_mean_ci95_high": 25284.655272388336,
    "prompt_tokens_mean_ci95_low": 25283.094727611664,
    "prompt_tokens_min": 25282.0,
    "prompt_tokens_p25": 25283.0,
    "prompt_tokens_p50": 25284.0,
    "prompt_tokens_p75": 25285.0,
    "prompt_tokens_p90": 25285.0,
    "prompt_tokens_p95": 25285.0,
    "prompt_tokens_p99": 25285.0,
    "prompt_tokens_samples": 8,
    "prompt_tokens_stddev": 1.1259916264596033,
    "reasoning_chunks": 826,
    "request_canary_passed": 8,
    "request_canary_samples": 8,
    "throughput_performance_eligible": true,
    "throughput_tok_s": 19.406947928442612,
    "time_to_first_output_max_ms": 29556.879910058342,
    "time_to_first_output_mean_ci95_high_ms": 19810.156121250853,
    "time_to_first_output_mean_ci95_low_ms": 8223.20472901882,
    "time_to_first_output_mean_ms": 14016.680425134837,
    "time_to_first_output_min_ms": 6503.798302030191,
    "time_to_first_output_p25_ms": 6931.828566011973,
    "time_to_first_output_p50_ms": 9400.584549061023,
    "time_to_first_output_p75_ms": 13533.749699941836,
    "time_to_first_output_p90_ms": 29556.879910058342,
    "time_to_first_output_p95_ms": 29556.879910058342,
    "time_to_first_output_p99_ms": 29556.879910058342,
    "time_to_first_output_samples": 8,
    "time_to_first_output_stddev_ms": 8360.420308904264,
    "tpot_max_ms": 197.158831406415,
    "tpot_mean_ci95_high_ms": 151.9074900028667,
    "tpot_mean_ci95_low_ms": 66.32663590019493,
    "tpot_mean_ms": 109.11706295153081,
    "tpot_min_ms": 18.936976657550066,
    "tpot_p25_ms": 43.905329714007266,
    "tpot_p50_ms": 102.0571773031926,
    "tpot_p75_ms": 142.58981737334074,
    "tpot_p90_ms": 197.158831406415,
    "tpot_p95_ms": 197.158831406415,
    "tpot_p99_ms": 197.158831406415,
    "tpot_samples": 8,
    "tpot_stddev_ms": 61.749798240546724,
    "ttft_max_ms": 31198.09624296613,
    "ttft_mean_ci95_high_ms": 30287.0270821183,
    "ttft_mean_ci95_low_ms": 21062.106298628634,
    "ttft_mean_ms": 25674.566690373467,
    "ttft_min_ms": 14893.61618401017,
    "ttft_p25_ms": 17007.84266507253,
    "ttft_p50_ms": 26289.9997580098,
    "ttft_p75_ms": 30881.501365918666,
    "ttft_p90_ms": 31198.09624296613,
    "ttft_p95_ms": 31198.09624296613,
    "ttft_p99_ms": 31198.09624296613,
    "ttft_samples": 8,
    "ttft_stddev_ms": 6656.126573381902
  },
  "model": "glm53-flash-exl3-4bpw-v84-fp8-tp2-c4-327k-r10-vision8-apc-nospec",
  "performance_eligible": true,
  "prompt_cache_mode": "unique",
  "request_canaries": true,
  "request_timings": [
    {
      "content_chunks": 27,
      "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": 5.072052785394339,
      "e2e_ms": 31182.364829001017,
      "effective_prefill_tok_s": 3647.378142611199,
      "generation_ms": 24250.536262989044,
      "mean_inter_token_latency_ms": 197.158831406415,
      "mean_time_per_output_token_ms": 197.158831406415,
      "output_token_source": "usage",
      "output_tokens": 124,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25283,
      "reasoning_chunks": 92,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_1910_00000",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 0,
      "time_to_first_output_ms": 6931.828566011973,
      "tpot_ms": 197.158831406415,
      "ttft_ms": 30673.785450984724,
      "visible_content_truncated": false,
      "visible_generation_ms": 508.57937801629305
    },
    {
      "content_chunks": 27,
      "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": 5.670074881363838,
      "e2e_ms": 34433.878888958134,
      "effective_prefill_tok_s": 1930.9942639344656,
      "generation_ms": 21340.106177027337,
      "mean_inter_token_latency_ms": 176.36451385973007,
      "mean_time_per_output_token_ms": 176.36451385973007,
      "output_token_source": "usage",
      "output_tokens": 122,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25284,
      "reasoning_chunks": 90,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_1910_00001",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 1,
      "time_to_first_output_ms": 13093.772711930797,
      "tpot_ms": 176.36451385973007,
      "ttft_ms": 30881.501365918666,
      "visible_content_truncated": false,
      "visible_generation_ms": 3552.3775230394676
    },
    {
      "content_chunks": 27,
      "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": 7.0131235064402615,
      "e2e_ms": 41797.5598910125,
      "effective_prefill_tok_s": 1061.0021518394312,
      "generation_ms": 17966.316989040934,
      "mean_inter_token_latency_ms": 142.58981737334074,
      "mean_time_per_output_token_ms": 142.58981737334074,
      "output_token_source": "usage",
      "output_tokens": 127,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25285,
      "reasoning_chunks": 94,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_1910_00002",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 2,
      "time_to_first_output_ms": 23831.242901971564,
      "tpot_ms": 142.58981737334074,
      "ttft_ms": 31198.09624296613,
      "visible_content_truncated": false,
      "visible_generation_ms": 10599.463648046367
    },
    {
      "content_chunks": 27,
      "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": 9.798428943700733,
      "e2e_ms": 40987.283768015914,
      "effective_prefill_tok_s": 855.4691860894085,
      "generation_ms": 11430.403857957572,
      "mean_inter_token_latency_ms": 102.0571773031926,
      "mean_time_per_output_token_ms": 102.0571773031926,
      "output_token_source": "usage",
      "output_tokens": 113,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25285,
      "reasoning_chunks": 81,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_1910_00003",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 3,
      "time_to_first_output_ms": 29556.879910058342,
      "tpot_ms": 102.0571773031926,
      "ttft_ms": 31161.688882042654,
      "visible_content_truncated": false,
      "visible_generation_ms": 9825.59488597326
    },
    {
      "content_chunks": 27,
      "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": 8.926274592735142,
      "e2e_ms": 26781.01452696137,
      "effective_prefill_tok_s": 3887.72818986517,
      "generation_ms": 20277.21622493118,
      "mean_inter_token_latency_ms": 112.02881892227171,
      "mean_time_per_output_token_ms": 112.02881892227171,
      "output_token_source": "usage",
      "output_tokens": 182,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25285,
      "reasoning_chunks": 133,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_1910_00004",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 4,
      "time_to_first_output_ms": 6503.798302030191,
      "tpot_ms": 112.02881892227171,
      "ttft_ms": 26289.9997580098,
      "visible_content_truncated": false,
      "visible_generation_ms": 491.0147689515725
    },
    {
      "content_chunks": 27,
      "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": 12.51642179952517,
      "e2e_ms": 23742.588706081733,
      "effective_prefill_tok_s": 2724.1031790776715,
      "generation_ms": 14461.001946008764,
      "mean_inter_token_latency_ms": 79.89503837573903,
      "mean_time_per_output_token_ms": 79.89503837573903,
      "output_token_source": "usage",
      "output_tokens": 182,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25284,
      "reasoning_chunks": 133,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_1910_00005",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 5,
      "time_to_first_output_ms": 9281.586760072969,
      "tpot_ms": 79.89503837573903,
      "ttft_ms": 23290.002973983064,
      "visible_content_truncated": false,
      "visible_generation_ms": 452.5857320986688
    },
    {
      "content_chunks": 27,
      "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": 22.776278108235378,
      "e2e_ms": 17391.354557010345,
      "effective_prefill_tok_s": 2689.5136007819206,
      "generation_ms": 7990.7700079493225,
      "mean_inter_token_latency_ms": 43.905329714007266,
      "mean_time_per_output_token_ms": 43.905329714007266,
      "output_token_source": "usage",
      "output_tokens": 183,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25283,
      "reasoning_chunks": 134,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_1910_00006",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 6,
      "time_to_first_output_ms": 9400.584549061023,
      "tpot_ms": 43.905329714007266,
      "ttft_ms": 17007.84266507253,
      "visible_content_truncated": false,
      "visible_generation_ms": 383.51189193781465
    },
    {
      "content_chunks": 27,
      "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": 52.80673985524007,
      "e2e_ms": 15408.510389039293,
      "effective_prefill_tok_s": 1868.0706057471016,
      "generation_ms": 1874.7606890974566,
      "mean_inter_token_latency_ms": 18.936976657550066,
      "mean_time_per_output_token_ms": 18.936976657550066,
      "output_token_source": "usage",
      "output_tokens": 100,
      "planned_context_tokens": 32768,
      "prompt_tokens": 25282,
      "reasoning_chunks": 69,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_1910_00007",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 7,
      "time_to_first_output_ms": 13533.749699941836,
      "tpot_ms": 18.936976657550066,
      "ttft_ms": 14893.61618401017,
      "visible_content_truncated": false,
      "visible_generation_ms": 514.8942050291225
    }
  ],
  "requests": 8,
  "response_words": 16,
  "responses_received": 8,
  "run_id": "benchmark-20260919T112223Z",
  "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": 0,
  "source_recipe": null,
  "started_at": "2026-09-19T11:22:23Z",
  "started_at_unix_ns": 1789816943605478210,
  "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": 58381.153192021884
}
