{
  "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": "2026-09-26T14:16:45Z",
  "finished_at_unix_ns": 1790432205074055260,
  "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": 281.0,
    "completion_tokens_mean": 177.5,
    "completion_tokens_mean_ci95_high": 202.7686749421756,
    "completion_tokens_mean_ci95_low": 152.2313250578244,
    "completion_tokens_min": 115.0,
    "completion_tokens_p25": 140.0,
    "completion_tokens_p50": 159.0,
    "completion_tokens_p75": 194.0,
    "completion_tokens_p90": 248.0,
    "completion_tokens_p95": 281.0,
    "completion_tokens_p99": 281.0,
    "completion_tokens_samples": 16,
    "completion_tokens_stddev": 51.56872437178695,
    "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": 54.555346170115946,
    "decode_tok_s_mean": 8.26497523639511,
    "decode_tok_s_mean_ci95_high": 14.368165521764475,
    "decode_tok_s_mean_ci95_low": 2.1617849510257443,
    "decode_tok_s_min": 3.2266454262278463,
    "decode_tok_s_p25": 3.769012927646546,
    "decode_tok_s_p50": 4.975982547642799,
    "decode_tok_s_p75": 5.51814022120115,
    "decode_tok_s_p90": 9.773648094834487,
    "decode_tok_s_p95": 54.555346170115946,
    "decode_tok_s_p99": 54.555346170115946,
    "decode_tok_s_samples": 16,
    "decode_tok_s_stddev": 12.455490378304827,
    "e2e_max_ms": 86737.24037199281,
    "e2e_mean_ci95_high_ms": 61710.44874440913,
    "e2e_mean_ci95_low_ms": 43196.94266447123,
    "e2e_mean_ms": 52453.69570444018,
    "e2e_min_ms": 16819.264965015464,
    "e2e_p25_ms": 40356.22199601494,
    "e2e_p50_ms": 53125.057200028095,
    "e2e_p75_ms": 60064.09513298422,
    "e2e_p90_ms": 78983.10199205298,
    "e2e_p95_ms": 86737.24037199281,
    "e2e_p99_ms": 86737.24037199281,
    "e2e_samples": 16,
    "e2e_stddev_ms": 18891.33273463051,
    "effective_prefill_tok_s_max": 2486.331517218204,
    "effective_prefill_tok_s_mean": 1969.5023600400846,
    "effective_prefill_tok_s_mean_ci95_high": 2297.3184141212355,
    "effective_prefill_tok_s_mean_ci95_low": 1641.6863059589336,
    "effective_prefill_tok_s_min": 656.7667151898277,
    "effective_prefill_tok_s_p25": 1244.020749057291,
    "effective_prefill_tok_s_p50": 2378.683174506314,
    "effective_prefill_tok_s_p75": 2447.7460199386664,
    "effective_prefill_tok_s_p90": 2475.3526550036377,
    "effective_prefill_tok_s_p95": 2486.331517218204,
    "effective_prefill_tok_s_p99": 2486.331517218204,
    "effective_prefill_tok_s_samples": 16,
    "effective_prefill_tok_s_stddev": 669.0123552676552,
    "generation_max_ms": 65003.756873018574,
    "generation_mean_ci95_high_ms": 41417.50721657958,
    "generation_mean_ci95_low_ms": 26460.173445675864,
    "generation_mean_ms": 33938.84033112772,
    "generation_min_ms": 4527.5122850434855,
    "generation_p25_ms": 24417.30390797602,
    "generation_p50_ms": 29291.394585976377,
    "generation_p75_ms": 41227.566134009976,
    "generation_p90_ms": 55675.8283090312,
    "generation_p95_ms": 65003.756873018574,
    "generation_p99_ms": 65003.756873018574,
    "generation_samples": 16,
    "generation_stddev_ms": 15262.585480513999,
    "mean_inter_token_latency_ms_max": 309.91939550329323,
    "mean_inter_token_latency_ms_mean": 197.72913878413948,
    "mean_inter_token_latency_ms_mean_ci95_high": 233.52048998225413,
    "mean_inter_token_latency_ms_mean_ci95_low": 161.93778758602483,
    "mean_inter_token_latency_ms_min": 18.3300092511882,
    "mean_inter_token_latency_ms_p25": 154.16523466303357,
    "mean_inter_token_latency_ms_p50": 199.2887364892811,
    "mean_inter_token_latency_ms_p75": 233.9375314162775,
    "mean_inter_token_latency_ms_p90": 286.83143071754347,
    "mean_inter_token_latency_ms_p95": 309.91939550329323,
    "mean_inter_token_latency_ms_p99": 309.91939550329323,
    "mean_inter_token_latency_ms_samples": 16,
    "mean_inter_token_latency_ms_stddev": 73.04357387370338,
    "mean_time_per_output_token_max_ms": 309.91939550329323,
    "mean_time_per_output_token_mean_ci95_high_ms": 233.52048998225413,
    "mean_time_per_output_token_mean_ci95_low_ms": 161.93778758602483,
    "mean_time_per_output_token_mean_ms": 197.72913878413948,
    "mean_time_per_output_token_min_ms": 18.3300092511882,
    "mean_time_per_output_token_p25_ms": 154.16523466303357,
    "mean_time_per_output_token_p50_ms": 199.2887364892811,
    "mean_time_per_output_token_p75_ms": 233.9375314162775,
    "mean_time_per_output_token_p90_ms": 286.83143071754347,
    "mean_time_per_output_token_p95_ms": 309.91939550329323,
    "mean_time_per_output_token_p99_ms": 309.91939550329323,
    "mean_time_per_output_token_samples": 16,
    "mean_time_per_output_token_stddev_ms": 73.04357387370338,
    "output_token_sources": {
      "usage": 16
    },
    "output_tokens": 2840,
    "prefix_cache_hit_avg": 0.0,
    "prompt_token_samples": 16,
    "prompt_tokens": 486393,
    "prompt_tokens_max": 30413.0,
    "prompt_tokens_mean": 30399.5625,
    "prompt_tokens_mean_ci95_high": 30403.901029656845,
    "prompt_tokens_mean_ci95_low": 30395.223970343155,
    "prompt_tokens_min": 30383.0,
    "prompt_tokens_p25": 30392.0,
    "prompt_tokens_p50": 30400.0,
    "prompt_tokens_p75": 30406.0,
    "prompt_tokens_p90": 30411.0,
    "prompt_tokens_p95": 30413.0,
    "prompt_tokens_p99": 30413.0,
    "prompt_tokens_samples": 16,
    "prompt_tokens_stddev": 8.854142156828821,
    "reasoning_chunks": 2344,
    "request_canary_passed": 16,
    "request_canary_samples": 16,
    "throughput_performance_eligible": true,
    "throughput_tok_s": 13.413587840186631,
    "time_to_first_output_max_ms": 46266.047436976805,
    "time_to_first_output_mean_ci95_high_ms": 23511.059353553363,
    "time_to_first_output_mean_ci95_low_ms": 13518.651393071552,
    "time_to_first_output_mean_ms": 18514.855373312457,
    "time_to_first_output_min_ms": 12226.848989957944,
    "time_to_first_output_p25_ms": 12324.79655201314,
    "time_to_first_output_p50_ms": 12515.08991594892,
    "time_to_first_output_p75_ms": 24410.838487034198,
    "time_to_first_output_p90_ms": 36229.03792501893,
    "time_to_first_output_p95_ms": 46266.047436976805,
    "time_to_first_output_p99_ms": 46266.047436976805,
    "time_to_first_output_samples": 16,
    "time_to_first_output_stddev_ms": 10196.334653552867,
    "tpot_max_ms": 309.91939550329323,
    "tpot_mean_ci95_high_ms": 233.52048998225413,
    "tpot_mean_ci95_low_ms": 161.93778758602483,
    "tpot_mean_ms": 197.72913878413948,
    "tpot_min_ms": 18.3300092511882,
    "tpot_p25_ms": 154.16523466303357,
    "tpot_p50_ms": 199.2887364892811,
    "tpot_p75_ms": 233.9375314162775,
    "tpot_p90_ms": 286.83143071754347,
    "tpot_p95_ms": 309.91939550329323,
    "tpot_p99_ms": 309.91939550329323,
    "tpot_samples": 16,
    "tpot_stddev_ms": 73.04357387370338,
    "ttft_max_ms": 85870.28065900085,
    "ttft_mean_ci95_high_ms": 57641.75567947331,
    "ttft_mean_ci95_low_ms": 39769.713342527946,
    "ttft_mean_ms": 48705.73451100063,
    "ttft_min_ms": 16493.239582981914,
    "ttft_p25_ms": 30722.005745978095,
    "ttft_p50_ms": 48297.363799007144,
    "ttft_p75_ms": 57665.035797981545,
    "ttft_p90_ms": 73814.34904702473,
    "ttft_p95_ms": 85870.28065900085,
    "ttft_p99_ms": 85870.28065900085,
    "ttft_samples": 16,
    "ttft_stddev_ms": 18236.777894842206
  },
  "model": "swift15-flash-next-iq3-327k-c4-medium-r3",
  "performance_eligible": true,
  "prompt_cache_mode": "unique",
  "request_canaries": true,
  "request_timings": [
    {
      "content_chunks": 28,
      "controlled_output": {
        "capture_complete": true,
        "exact_adherence": true,
        "observed_code_words": 16,
        "observed_extra_words": 0,
        "passed": true,
        "policy": "strict",
        "requested_words": 16
      },
      "decode_tok_s": 4.668820129758937,
      "e2e_ms": 48840.52998601692,
      "effective_prefill_tok_s": 1244.020749057291,
      "generation_ms": 24417.30390797602,
      "mean_inter_token_latency_ms": 214.18687638575457,
      "mean_time_per_output_token_ms": 214.18687638575457,
      "output_token_source": "usage",
      "output_tokens": 115,
      "planned_context_tokens": 32768,
      "prompt_tokens": 30383,
      "reasoning_chunks": 84,
      "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": 24423.226078040898,
      "tpot_ms": 214.18687638575457,
      "ttft_ms": 48006.142334023025,
      "visible_content_truncated": false,
      "visible_generation_ms": 834.3876519938931
    },
    {
      "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": 4.274645431820691,
      "e2e_ms": 86737.24037199281,
      "effective_prefill_tok_s": 656.7667151898277,
      "generation_ms": 40471.192935016006,
      "mean_inter_token_latency_ms": 233.9375314162775,
      "mean_time_per_output_token_ms": 233.9375314162775,
      "output_token_source": "usage",
      "output_tokens": 174,
      "planned_context_tokens": 32768,
      "prompt_tokens": 30386,
      "reasoning_chunks": 143,
      "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": 46266.047436976805,
      "tpot_ms": 233.9375314162775,
      "ttft_ms": 85870.28065900085,
      "visible_content_truncated": false,
      "visible_generation_ms": 866.9597129919566
    },
    {
      "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": 3.2266454262278463,
      "e2e_ms": 55854.351932008285,
      "effective_prefill_tok_s": 2378.683174506314,
      "generation_ms": 43078.79597495776,
      "mean_inter_token_latency_ms": 309.91939550329323,
      "mean_time_per_output_token_ms": 309.91939550329323,
      "output_token_source": "usage",
      "output_tokens": 140,
      "planned_context_tokens": 32768,
      "prompt_tokens": 30389,
      "reasoning_chunks": 109,
      "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": 12775.555957050528,
      "tpot_ms": 309.91939550329323,
      "ttft_ms": 48297.363799007144,
      "visible_content_truncated": false,
      "visible_generation_ms": 7556.988133001141
    },
    {
      "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": 3.732130386393828,
      "e2e_ms": 73473.1815130217,
      "effective_prefill_tok_s": 838.8850971670985,
      "generation_ms": 37244.14358800277,
      "mean_inter_token_latency_ms": 267.94347905037966,
      "mean_time_per_output_token_ms": 267.94347905037966,
      "output_token_source": "usage",
      "output_tokens": 140,
      "planned_context_tokens": 32768,
      "prompt_tokens": 30392,
      "reasoning_chunks": 109,
      "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": 36229.03792501893,
      "tpot_ms": 267.94347905037966,
      "ttft_ms": 57665.035797981545,
      "visible_content_truncated": false,
      "visible_generation_ms": 15808.145715040155
    },
    {
      "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": 6.48654673789294,
      "e2e_ms": 53702.233073010575,
      "effective_prefill_tok_s": 1245.1436281528913,
      "generation_ms": 29291.394585976377,
      "mean_inter_token_latency_ms": 154.16523466303357,
      "mean_time_per_output_token_ms": 154.16523466303357,
      "output_token_source": "usage",
      "output_tokens": 191,
      "planned_context_tokens": 32768,
      "prompt_tokens": 30395,
      "reasoning_chunks": 160,
      "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": 24410.838487034198,
      "tpot_ms": 154.16523466303357,
      "ttft_ms": 52843.89792900765,
      "visible_content_truncated": false,
      "visible_generation_ms": 858.3351440029219
    },
    {
      "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": 5.355729933527622,
      "e2e_ms": 60064.09513298422,
      "effective_prefill_tok_s": 2068.930996763514,
      "generation_ms": 45371.966662991326,
      "mean_inter_token_latency_ms": 186.71591219337995,
      "mean_time_per_output_token_ms": 186.71591219337995,
      "output_token_source": "usage",
      "output_tokens": 244,
      "planned_context_tokens": 32768,
      "prompt_tokens": 30397,
      "reasoning_chunks": 213,
      "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": 14692.128469992895,
      "tpot_ms": 186.71591219337995,
      "ttft_ms": 59226.14471899578,
      "visible_content_truncated": false,
      "visible_generation_ms": 837.9504139884375
    },
    {
      "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": 5.0291124264168525,
      "e2e_ms": 68000.62486104434,
      "effective_prefill_tok_s": 2466.4910184691535,
      "generation_ms": 55675.8283090312,
      "mean_inter_token_latency_ms": 198.84224396082573,
      "mean_time_per_output_token_ms": 198.84224396082573,
      "output_token_source": "usage",
      "output_tokens": 281,
      "planned_context_tokens": 32768,
      "prompt_tokens": 30399,
      "reasoning_chunks": 250,
      "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": 12324.79655201314,
      "tpot_ms": 198.84224396082573,
      "ttft_ms": 67176.43010301981,
      "visible_content_truncated": false,
      "visible_generation_ms": 824.1947580245323
    },
    {
      "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": 9.773648094834487,
      "e2e_ms": 29211.295164946932,
      "effective_prefill_tok_s": 2486.331517218204,
      "generation_ms": 16984.44617498899,
      "mean_inter_token_latency_ms": 102.31594081318669,
      "mean_time_per_output_token_ms": 102.31594081318669,
      "output_token_source": "usage",
      "output_tokens": 167,
      "planned_context_tokens": 32768,
      "prompt_tokens": 30400,
      "reasoning_chunks": 136,
      "request_canary": {
        "capture_complete": true,
        "foreign_marker_count": 0,
        "marker": "ANVIL_REQ_0_00007",
        "marker_at_start": true,
        "observed_marker_count": 1,
        "passed": true
      },
      "request_index": 7,
      "time_to_first_output_ms": 12226.848989957944,
      "tpot_ms": 102.31594081318669,
      "ttft_ms": 28375.112326000817,
      "visible_content_truncated": false,
      "visible_generation_ms": 836.1828389461152
    },
    {
      "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": 4.681333828260795,
      "e2e_ms": 53697.40897597512,
      "effective_prefill_tok_s": 2437.9617598459686,
      "generation_ms": 41227.566134009976,
      "mean_inter_token_latency_ms": 213.61433230056983,
      "mean_time_per_output_token_ms": 213.61433230056983,
      "output_token_source": "usage",
      "output_tokens": 194,
      "planned_context_tokens": 32768,
      "prompt_tokens": 30401,
      "reasoning_chunks": 163,
      "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": 12469.842841965146,
      "tpot_ms": 213.61433230056983,
      "ttft_ms": 52864.775548980106,
      "visible_content_truncated": false,
      "visible_generation_ms": 832.6334269950166
    },
    {
      "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": 3.769012927646546,
      "e2e_ms": 78983.10199205298,
      "effective_prefill_tok_s": 2174.7084531596347,
      "generation_ms": 65003.756873018574,
      "mean_inter_token_latency_ms": 265.3214566245656,
      "mean_time_per_output_token_ms": 265.3214566245656,
      "output_token_source": "usage",
      "output_tokens": 246,
      "planned_context_tokens": 32768,
      "prompt_tokens": 30401,
      "reasoning_chunks": 215,
      "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": 13979.345119034406,
      "tpot_ms": 265.3214566245656,
      "ttft_ms": 73814.34904702473,
      "visible_content_truncated": false,
      "visible_generation_ms": 5168.752945028245
    },
    {
      "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": 5.51814022120115,
      "e2e_ms": 53125.057200028095,
      "effective_prefill_tok_s": 1241.3736939840041,
      "generation_ms": 28632.835278986022,
      "mean_inter_token_latency_ms": 181.2204764492786,
      "mean_time_per_output_token_ms": 181.2204764492786,
      "output_token_source": "usage",
      "output_tokens": 159,
      "planned_context_tokens": 32768,
      "prompt_tokens": 30404,
      "reasoning_chunks": 128,
      "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": 24492.221921042074,
      "tpot_ms": 181.2204764492786,
      "ttft_ms": 40291.64257703815,
      "visible_content_truncated": false,
      "visible_generation_ms": 12833.414622989949
    },
    {
      "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": 4.975982547642799,
      "e2e_ms": 40356.22199601494,
      "effective_prefill_tok_s": 2447.7460199386664,
      "generation_ms": 27934.181575023104,
      "mean_inter_token_latency_ms": 200.96533507210867,
      "mean_time_per_output_token_ms": 200.96533507210867,
      "output_token_source": "usage",
      "output_tokens": 140,
      "planned_context_tokens": 32768,
      "prompt_tokens": 30406,
      "reasoning_chunks": 109,
      "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": 12422.040420991834,
      "tpot_ms": 200.96533507210867,
      "ttft_ms": 30722.005745978095,
      "visible_content_truncated": false,
      "visible_generation_ms": 9634.216250036843
    },
    {
      "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": 3.4863682738616864,
      "e2e_ms": 52097.82735496992,
      "effective_prefill_tok_s": 2429.6269706580433,
      "generation_ms": 39582.737439021,
      "mean_inter_token_latency_ms": 286.83143071754347,
      "mean_time_per_output_token_ms": 286.83143071754347,
      "output_token_source": "usage",
      "output_tokens": 139,
      "planned_context_tokens": 32768,
      "prompt_tokens": 30407,
      "reasoning_chunks": 108,
      "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": 12515.08991594892,
      "tpot_ms": 286.83143071754347,
      "ttft_ms": 51270.02828498371,
      "visible_content_truncated": false,
      "visible_generation_ms": 827.799069986213
    },
    {
      "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": 5.017845050434077,
      "e2e_ms": 40384.426080971025,
      "effective_prefill_tok_s": 2475.3526550036377,
      "generation_ms": 28099.711844988633,
      "mean_inter_token_latency_ms": 199.2887364892811,
      "mean_time_per_output_token_ms": 199.2887364892811,
      "output_token_source": "usage",
      "output_tokens": 142,
      "planned_context_tokens": 32768,
      "prompt_tokens": 30409,
      "reasoning_chunks": 111,
      "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": 12284.714235982392,
      "tpot_ms": 199.2887364892811,
      "ttft_ms": 39371.268623974174,
      "visible_content_truncated": false,
      "visible_generation_ms": 1013.1574569968507
    },
    {
      "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": 7.688296196285539,
      "e2e_ms": 27912.27067098953,
      "effective_prefill_tok_s": 2445.7546595409535,
      "generation_ms": 15478.0717290123,
      "mean_inter_token_latency_ms": 130.06782965556553,
      "mean_time_per_output_token_ms": 130.06782965556553,
      "output_token_source": "usage",
      "output_tokens": 120,
      "planned_context_tokens": 32768,
      "prompt_tokens": 30411,
      "reasoning_chunks": 89,
      "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": 12434.198941977229,
      "tpot_ms": 130.06782965556553,
      "ttft_ms": 27004.035098012537,
      "visible_content_truncated": false,
      "visible_generation_ms": 908.2355729769915
    },
    {
      "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": 54.555346170115946,
      "e2e_ms": 16819.264965015464,
      "effective_prefill_tok_s": 2474.260651986152,
      "generation_ms": 4527.5122850434855,
      "mean_inter_token_latency_ms": 18.3300092511882,
      "mean_time_per_output_token_ms": 18.3300092511882,
      "output_token_source": "usage",
      "output_tokens": 248,
      "planned_context_tokens": 32768,
      "prompt_tokens": 30413,
      "reasoning_chunks": 217,
      "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": 12291.752679971978,
      "tpot_ms": 18.3300092511882,
      "ttft_ms": 16493.239582981914,
      "visible_content_truncated": false,
      "visible_generation_ms": 326.02538203354925
    }
  ],
  "requests": 16,
  "response_words": 16,
  "responses_received": 16,
  "run_id": "benchmark-20260926T141313Z",
  "sampling": {
    "temperature": {
      "effective_request": 1.0,
      "requested": 1.0,
      "sent": true
    },
    "top_p": {
      "effective_request": 0.95,
      "requested": 0.95,
      "sent": true
    }
  },
  "schema": "anvil-serving.benchmark/v1",
  "serve_flags": {
    "no_thinking": false,
    "reasoning_effort": null,
    "shared_prefix_burst": false,
    "thinking_mode": "default"
  },
  "shared_prefix_tokens": 0,
  "source_recipe": null,
  "started_at": "2026-09-26T14:13:13Z",
  "started_at_unix_ns": 1790431993348445801,
  "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": 211725.60494899517
}
