{
  "base_url": "http://127.0.0.1:8080/v1",
  "clock_domain_id": null,
  "completed": 12,
  "concurrency": 1,
  "configuration_fingerprint": null,
  "context_distribution": {
    "2048": 12
  },
  "context_seed": 43,
  "context_tokens": 2048,
  "controlled_output_policy": "strict",
  "engine": "ninfer",
  "failed": 0,
  "failures": [],
  "finished_at": "2026-09-19T23:29:12Z",
  "finished_at_unix_ns": 1789860552325911600,
  "gpu": "RTX5090",
  "manifest": null,
  "max_context_tokens": 65536,
  "max_tokens": 512,
  "measurement_protocol": "capacity-v3",
  "metric_population": {
    "aggregate_throughput_eligible": true,
    "attempted_requests": 12,
    "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": 12,
      "e2e": 12,
      "effective_prefill_tok_s": 12,
      "tpot": 12,
      "ttft": 12
    },
    "performance_eligible_requests": 12,
    "successful_requests": 12
  },
  "metrics": {
    "completion_tokens_max": 33.0,
    "completion_tokens_mean": 33.0,
    "completion_tokens_mean_ci95_high": 33.0,
    "completion_tokens_mean_ci95_low": 33.0,
    "completion_tokens_min": 33.0,
    "completion_tokens_p25": 33.0,
    "completion_tokens_p50": 33.0,
    "completion_tokens_p75": 33.0,
    "completion_tokens_p90": 33.0,
    "completion_tokens_p95": 33.0,
    "completion_tokens_p99": 33.0,
    "completion_tokens_samples": 12,
    "completion_tokens_stddev": 0.0,
    "content_chunks": 111,
    "content_chunks_s": null,
    "controlled_output_exact_adherence": 12,
    "controlled_output_exact_adherence_rate": 1.0,
    "controlled_output_samples": 12,
    "controlled_output_unobservable": 0,
    "decode_tok_s_max": 183.97603242554607,
    "decode_tok_s_mean": 182.14706387338313,
    "decode_tok_s_mean_ci95_high": 182.48571847405984,
    "decode_tok_s_mean_ci95_low": 181.80840927270643,
    "decode_tok_s_min": 181.75611614649154,
    "decode_tok_s_p25": 181.82159102206356,
    "decode_tok_s_p50": 181.95795861789748,
    "decode_tok_s_p75": 182.1477035749886,
    "decode_tok_s_p90": 182.26659913821643,
    "decode_tok_s_p95": 183.97603242554607,
    "decode_tok_s_p99": 183.97603242554607,
    "decode_tok_s_samples": 12,
    "decode_tok_s_stddev": 0.598537729172447,
    "e2e_max_ms": 659.051000024192,
    "e2e_mean_ci95_high_ms": 498.5433483779971,
    "e2e_mean_ci95_low_ms": 428.92915161671914,
    "e2e_mean_ms": 463.73624999735813,
    "e2e_min_ms": 444.1402000375092,
    "e2e_p25_ms": 445.0199999846518,
    "e2e_p50_ms": 446.36210007593036,
    "e2e_p75_ms": 446.45889999810606,
    "e2e_p90_ms": 447.8696999140084,
    "e2e_p95_ms": 659.051000024192,
    "e2e_p99_ms": 659.051000024192,
    "e2e_samples": 12,
    "e2e_stddev_ms": 61.51802332583172,
    "effective_prefill_tok_s_max": 6518.972855781053,
    "effective_prefill_tok_s_mean": 6239.250427811009,
    "effective_prefill_tok_s_mean_ci95_high": 6709.808947361315,
    "effective_prefill_tok_s_mean_ci95_low": 5768.691908260703,
    "effective_prefill_tok_s_min": 3599.1443688590552,
    "effective_prefill_tok_s_p25": 6461.75095059095,
    "effective_prefill_tok_s_p50": 6475.443949081529,
    "effective_prefill_tok_s_p75": 6479.794248763132,
    "effective_prefill_tok_s_p90": 6507.062973856003,
    "effective_prefill_tok_s_p95": 6518.972855781053,
    "effective_prefill_tok_s_p99": 6518.972855781053,
    "effective_prefill_tok_s_samples": 12,
    "effective_prefill_tok_s_stddev": 831.664554893391,
    "generation_max_ms": 176.06010008603334,
    "generation_mean_ci95_high_ms": 176.0079042545213,
    "generation_mean_ci95_low_ms": 175.35997909090133,
    "generation_mean_ms": 175.6839416727113,
    "generation_min_ms": 173.93570009153336,
    "generation_p25_ms": 175.61989999376237,
    "generation_p50_ms": 175.8463999722153,
    "generation_p75_ms": 175.8902999572456,
    "generation_p90_ms": 176.05440004263073,
    "generation_p95_ms": 176.06010008603334,
    "generation_p99_ms": 176.06010008603334,
    "generation_samples": 12,
    "generation_stddev_ms": 0.5725710729041457,
    "mean_inter_token_latency_ms_max": 5.501878127688542,
    "mean_inter_token_latency_ms_mean": 5.4901231772722285,
    "mean_inter_token_latency_ms_mean_ci95_high": 5.50024700795379,
    "mean_inter_token_latency_ms_mean_ci95_low": 5.479999346590667,
    "mean_inter_token_latency_ms_min": 5.435490627860418,
    "mean_inter_token_latency_ms_p25": 5.488121874805074,
    "mean_inter_token_latency_ms_p50": 5.495199999131728,
    "mean_inter_token_latency_ms_p75": 5.496571873663925,
    "mean_inter_token_latency_ms_p90": 5.50170000133221,
    "mean_inter_token_latency_ms_p95": 5.501878127688542,
    "mean_inter_token_latency_ms_p99": 5.501878127688542,
    "mean_inter_token_latency_ms_samples": 12,
    "mean_inter_token_latency_ms_stddev": 0.017892846028254552,
    "mean_time_per_output_token_max_ms": 5.501878127688542,
    "mean_time_per_output_token_mean_ci95_high_ms": 5.50024700795379,
    "mean_time_per_output_token_mean_ci95_low_ms": 5.479999346590667,
    "mean_time_per_output_token_mean_ms": 5.4901231772722285,
    "mean_time_per_output_token_min_ms": 5.435490627860418,
    "mean_time_per_output_token_p25_ms": 5.488121874805074,
    "mean_time_per_output_token_p50_ms": 5.495199999131728,
    "mean_time_per_output_token_p75_ms": 5.496571873663925,
    "mean_time_per_output_token_p90_ms": 5.50170000133221,
    "mean_time_per_output_token_p95_ms": 5.501878127688542,
    "mean_time_per_output_token_p99_ms": 5.501878127688542,
    "mean_time_per_output_token_samples": 12,
    "mean_time_per_output_token_stddev_ms": 0.017892846028254552,
    "output_token_sources": {
      "usage": 12
    },
    "output_tokens": 396,
    "prefix_cache_hit_avg": null,
    "prompt_token_samples": 12,
    "prompt_tokens": 20999,
    "prompt_tokens_max": 1756.0,
    "prompt_tokens_mean": 1749.9166666666667,
    "prompt_tokens_mean_ci95_high": 1751.956104844403,
    "prompt_tokens_mean_ci95_low": 1747.8772284889305,
    "prompt_tokens_min": 1746.0,
    "prompt_tokens_p25": 1746.0,
    "prompt_tokens_p50": 1749.0,
    "prompt_tokens_p75": 1752.0,
    "prompt_tokens_p90": 1756.0,
    "prompt_tokens_p95": 1756.0,
    "prompt_tokens_p99": 1756.0,
    "prompt_tokens_samples": 12,
    "prompt_tokens_stddev": 3.6045005538110604,
    "reasoning_chunks": 0,
    "request_canary_passed": 0,
    "request_canary_samples": 0,
    "throughput_performance_eligible": true,
    "throughput_tok_s": 71.12841567246136,
    "time_to_first_output_max_ms": 485.1152999326587,
    "time_to_first_output_mean_ci95_high_ms": 323.1703097695208,
    "time_to_first_output_mean_ci95_low_ms": 252.93430687977286,
    "time_to_first_output_mean_ms": 288.05230832464684,
    "time_to_first_output_min_ms": 268.2938000652939,
    "time_to_first_output_p25_ms": 269.45300004445016,
    "time_to_first_output_p50_ms": 270.3077000332996,
    "time_to_first_output_p75_ms": 270.58929996564984,
    "time_to_first_output_p90_ms": 271.9793999567628,
    "time_to_first_output_p95_ms": 485.1152999326587,
    "time_to_first_output_p99_ms": 485.1152999326587,
    "time_to_first_output_samples": 12,
    "time_to_first_output_stddev_ms": 62.06751302326425,
    "tpot_max_ms": 5.501878127688542,
    "tpot_mean_ci95_high_ms": 5.50024700795379,
    "tpot_mean_ci95_low_ms": 5.479999346590667,
    "tpot_mean_ms": 5.4901231772722285,
    "tpot_min_ms": 5.435490627860418,
    "tpot_p25_ms": 5.488121874805074,
    "tpot_p50_ms": 5.495199999131728,
    "tpot_p75_ms": 5.496571873663925,
    "tpot_p90_ms": 5.50170000133221,
    "tpot_p95_ms": 5.501878127688542,
    "tpot_p99_ms": 5.501878127688542,
    "tpot_samples": 12,
    "tpot_stddev_ms": 0.017892846028254552,
    "ttft_max_ms": 485.1163999410346,
    "ttft_mean_ci95_high_ms": 323.1710131919816,
    "ttft_mean_ci95_low_ms": 252.934836780634,
    "ttft_mean_ms": 288.0529249863078,
    "ttft_min_ms": 268.2942000683397,
    "ttft_p25_ms": 269.4539000513032,
    "ttft_p50_ms": 270.3082000371069,
    "ttft_p75_ms": 270.58989997021854,
    "ttft_p90_ms": 271.98009996209294,
    "ttft_p95_ms": 485.1163999410346,
    "ttft_p99_ms": 485.1163999410346,
    "ttft_samples": 12,
    "ttft_stddev_ms": 62.06766636419634
  },
  "model": "qwen38-huihui-ninfer-schemafix-mtp3-64k",
  "performance_eligible": true,
  "prompt_cache_mode": "unique",
  "request_canaries": false,
  "request_timings": [
    {
      "content_chunks": 9,
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": true,
        "observed_code_words": 32,
        "observed_extra_words": 0,
        "passed": true,
        "policy": "strict",
        "requested_words": 32
      },
      "decode_tok_s": 183.97603242554607,
      "e2e_ms": 659.051000024192,
      "effective_prefill_tok_s": 3599.1443688590552,
      "generation_ms": 173.93570009153336,
      "mean_inter_token_latency_ms": 5.435490627860418,
      "mean_time_per_output_token_ms": 5.435490627860418,
      "output_token_source": "usage",
      "output_tokens": 33,
      "planned_context_tokens": 2048,
      "prompt_tokens": 1746,
      "reasoning_chunks": 0,
      "request_canary": null,
      "request_index": 0,
      "time_to_first_output_ms": 485.1152999326587,
      "tpot_ms": 5.435490627860418,
      "ttft_ms": 485.1163999410346,
      "visible_content_truncated": false,
      "visible_generation_ms": 173.93460008315742
    },
    {
      "content_chunks": 9,
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": true,
        "observed_code_words": 32,
        "observed_extra_words": 0,
        "passed": true,
        "policy": "strict",
        "requested_words": 32
      },
      "decode_tok_s": 182.26659913821643,
      "e2e_ms": 445.0199999846518,
      "effective_prefill_tok_s": 6479.794248763132,
      "generation_ms": 175.56699994020164,
      "mean_inter_token_latency_ms": 5.486468748131301,
      "mean_time_per_output_token_ms": 5.486468748131301,
      "output_token_source": "usage",
      "output_tokens": 33,
      "planned_context_tokens": 2048,
      "prompt_tokens": 1746,
      "reasoning_chunks": 0,
      "request_canary": null,
      "request_index": 1,
      "time_to_first_output_ms": 269.45300004445016,
      "tpot_ms": 5.486468748131301,
      "ttft_ms": 269.4539000513032,
      "visible_content_truncated": false,
      "visible_generation_ms": 175.5660999333486
    },
    {
      "content_chunks": 10,
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": true,
        "observed_code_words": 32,
        "observed_extra_words": 0,
        "passed": true,
        "policy": "strict",
        "requested_words": 32
      },
      "decode_tok_s": 182.21169697247618,
      "e2e_ms": 445.82529994659126,
      "effective_prefill_tok_s": 6461.75095059095,
      "generation_ms": 175.61989999376237,
      "mean_inter_token_latency_ms": 5.488121874805074,
      "mean_time_per_output_token_ms": 5.488121874805074,
      "output_token_source": "usage",
      "output_tokens": 33,
      "planned_context_tokens": 2048,
      "prompt_tokens": 1746,
      "reasoning_chunks": 0,
      "request_canary": null,
      "request_index": 2,
      "time_to_first_output_ms": 270.2053999528289,
      "tpot_ms": 5.488121874805074,
      "ttft_ms": 270.20589995663613,
      "visible_content_truncated": false,
      "visible_generation_ms": 175.61939998995513
    },
    {
      "content_chunks": 9,
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": true,
        "observed_code_words": 32,
        "observed_extra_words": 0,
        "passed": true,
        "policy": "strict",
        "requested_words": 32
      },
      "decode_tok_s": 181.95795861789748,
      "e2e_ms": 444.34230006299913,
      "effective_prefill_tok_s": 6507.062973856003,
      "generation_ms": 175.86479999590665,
      "mean_inter_token_latency_ms": 5.495774999872083,
      "mean_time_per_output_token_ms": 5.495774999872083,
      "output_token_source": "usage",
      "output_tokens": 33,
      "planned_context_tokens": 2048,
      "prompt_tokens": 1747,
      "reasoning_chunks": 0,
      "request_canary": null,
      "request_index": 3,
      "time_to_first_output_ms": 268.4775000670925,
      "tpot_ms": 5.495774999872083,
      "ttft_ms": 268.47789995372295,
      "visible_content_truncated": false,
      "visible_generation_ms": 175.86440010927618
    },
    {
      "content_chunks": 9,
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": true,
        "observed_code_words": 32,
        "observed_extra_words": 0,
        "passed": true,
        "policy": "strict",
        "requested_words": 32
      },
      "decode_tok_s": 182.1477035749886,
      "e2e_ms": 445.5998999765143,
      "effective_prefill_tok_s": 6476.033674406641,
      "generation_ms": 175.68159999791533,
      "mean_inter_token_latency_ms": 5.490049999934854,
      "mean_time_per_output_token_ms": 5.490049999934854,
      "output_token_source": "usage",
      "output_tokens": 33,
      "planned_context_tokens": 2048,
      "prompt_tokens": 1748,
      "reasoning_chunks": 0,
      "request_canary": null,
      "request_index": 4,
      "time_to_first_output_ms": 269.91829997859895,
      "tpot_ms": 5.490049999934854,
      "ttft_ms": 269.91889998316765,
      "visible_content_truncated": false,
      "visible_generation_ms": 175.68099999334663
    },
    {
      "content_chunks": 9,
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": true,
        "observed_code_words": 32,
        "observed_extra_words": 0,
        "passed": true,
        "policy": "strict",
        "requested_words": 32
      },
      "decode_tok_s": 181.9769981361926,
      "e2e_ms": 444.1402000375092,
      "effective_prefill_tok_s": 6518.972855781053,
      "generation_ms": 175.8463999722153,
      "mean_inter_token_latency_ms": 5.495199999131728,
      "mean_time_per_output_token_ms": 5.495199999131728,
      "output_token_source": "usage",
      "output_tokens": 33,
      "planned_context_tokens": 2048,
      "prompt_tokens": 1749,
      "reasoning_chunks": 0,
      "request_canary": null,
      "request_index": 5,
      "time_to_first_output_ms": 268.2938000652939,
      "tpot_ms": 5.495199999131728,
      "ttft_ms": 268.2942000683397,
      "visible_content_truncated": false,
      "visible_generation_ms": 175.8459999691695
    },
    {
      "content_chunks": 9,
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": true,
        "observed_code_words": 32,
        "observed_extra_words": 0,
        "passed": true,
        "policy": "strict",
        "requested_words": 32
      },
      "decode_tok_s": 182.0096020595931,
      "e2e_ms": 446.40420004725456,
      "effective_prefill_tok_s": 6467.365857490134,
      "generation_ms": 175.81490008160472,
      "mean_inter_token_latency_ms": 5.4942156275501475,
      "mean_time_per_output_token_ms": 5.4942156275501475,
      "output_token_source": "usage",
      "output_tokens": 33,
      "planned_context_tokens": 2048,
      "prompt_tokens": 1750,
      "reasoning_chunks": 0,
      "request_canary": null,
      "request_index": 6,
      "time_to_first_output_ms": 270.58929996564984,
      "tpot_ms": 5.4942156275501475,
      "ttft_ms": 270.58989997021854,
      "visible_content_truncated": false,
      "visible_generation_ms": 175.81430007703602
    },
    {
      "content_chunks": 9,
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": true,
        "observed_code_words": 32,
        "observed_extra_words": 0,
        "passed": true,
        "policy": "strict",
        "requested_words": 32
      },
      "decode_tok_s": 181.7620007920924,
      "e2e_ms": 446.36210007593036,
      "effective_prefill_tok_s": 6477.802888279881,
      "generation_ms": 176.05440004263073,
      "mean_inter_token_latency_ms": 5.50170000133221,
      "mean_time_per_output_token_ms": 5.50170000133221,
      "output_token_source": "usage",
      "output_tokens": 33,
      "planned_context_tokens": 2048,
      "prompt_tokens": 1751,
      "reasoning_chunks": 0,
      "request_canary": null,
      "request_index": 7,
      "time_to_first_output_ms": 270.3077000332996,
      "tpot_ms": 5.50170000133221,
      "ttft_ms": 270.3082000371069,
      "visible_content_truncated": false,
      "visible_generation_ms": 176.05390003882349
    },
    {
      "content_chunks": 10,
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": true,
        "observed_code_words": 32,
        "observed_extra_words": 0,
        "passed": true,
        "policy": "strict",
        "requested_words": 32
      },
      "decode_tok_s": 181.94688860644058,
      "e2e_ms": 446.43609994091094,
      "effective_prefill_tok_s": 6475.443949081529,
      "generation_ms": 175.8754999609664,
      "mean_inter_token_latency_ms": 5.4961093737802,
      "mean_time_per_output_token_ms": 5.4961093737802,
      "output_token_source": "usage",
      "output_tokens": 33,
      "planned_context_tokens": 2048,
      "prompt_tokens": 1752,
      "reasoning_chunks": 0,
      "request_canary": null,
      "request_index": 8,
      "time_to_first_output_ms": 270.5605999799445,
      "tpot_ms": 5.4961093737802,
      "ttft_ms": 270.5609999829903,
      "visible_content_truncated": false,
      "visible_generation_ms": 175.8750999579206
    },
    {
      "content_chunks": 10,
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": true,
        "observed_code_words": 32,
        "observed_extra_words": 0,
        "passed": true,
        "policy": "strict",
        "requested_words": 32
      },
      "decode_tok_s": 181.93157898859903,
      "e2e_ms": 447.8696999140084,
      "effective_prefill_tok_s": 6441.664332955068,
      "generation_ms": 175.8902999572456,
      "mean_inter_token_latency_ms": 5.496571873663925,
      "mean_time_per_output_token_ms": 5.496571873663925,
      "output_token_source": "usage",
      "output_tokens": 33,
      "planned_context_tokens": 2048,
      "prompt_tokens": 1752,
      "reasoning_chunks": 0,
      "request_canary": null,
      "request_index": 9,
      "time_to_first_output_ms": 271.9793999567628,
      "tpot_ms": 5.496571873663925,
      "ttft_ms": 271.98009996209294,
      "visible_content_truncated": false,
      "visible_generation_ms": 175.88959995191544
    },
    {
      "content_chunks": 9,
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": true,
        "observed_code_words": 32,
        "observed_extra_words": 0,
        "passed": true,
        "policy": "strict",
        "requested_words": 32
      },
      "decode_tok_s": 181.75611614649154,
      "e2e_ms": 446.45889999810606,
      "effective_prefill_tok_s": 6494.111662370578,
      "generation_ms": 176.06010008603334,
      "mean_inter_token_latency_ms": 5.501878127688542,
      "mean_time_per_output_token_ms": 5.501878127688542,
      "output_token_source": "usage",
      "output_tokens": 33,
      "planned_context_tokens": 2048,
      "prompt_tokens": 1756,
      "reasoning_chunks": 0,
      "request_canary": null,
      "request_index": 10,
      "time_to_first_output_ms": 270.3987999120727,
      "tpot_ms": 5.501878127688542,
      "ttft_ms": 270.39929991587996,
      "visible_content_truncated": false,
      "visible_generation_ms": 176.0596000822261
    },
    {
      "content_chunks": 9,
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": true,
        "observed_code_words": 32,
        "observed_extra_words": 0,
        "passed": true,
        "policy": "strict",
        "requested_words": 32
      },
      "decode_tok_s": 181.82159102206356,
      "e2e_ms": 447.3252999596298,
      "effective_prefill_tok_s": 6471.857371298084,
      "generation_ms": 175.99669995252043,
      "mean_inter_token_latency_ms": 5.499896873516263,
      "mean_time_per_output_token_ms": 5.499896873516263,
      "output_token_source": "usage",
      "output_tokens": 33,
      "planned_context_tokens": 2048,
      "prompt_tokens": 1756,
      "reasoning_chunks": 0,
      "request_canary": null,
      "request_index": 11,
      "time_to_first_output_ms": 271.32860000710934,
      "tpot_ms": 5.499896873516263,
      "ttft_ms": 271.32940001320094,
      "visible_content_truncated": false,
      "visible_generation_ms": 175.99589994642884
    }
  ],
  "requests": 12,
  "response_words": 32,
  "responses_received": 12,
  "run_id": "benchmark-20260919T232906Z",
  "sampling": {
    "temperature": {
      "effective_request": 0.0,
      "requested": null,
      "sent": true
    },
    "top_p": {
      "effective_request": null,
      "requested": null,
      "sent": false
    }
  },
  "schema": "anvil-serving.benchmark/v1",
  "serve_flags": {
    "no_thinking": false,
    "reasoning_effort": null,
    "shared_prefix_burst": false,
    "thinking_mode": "unsupported"
  },
  "shared_prefix_tokens": 0,
  "source_recipe": null,
  "started_at": "2026-09-19T23:29:06Z",
  "started_at_unix_ns": 1789860546758603100,
  "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": 5567.395200021565
}
