{ "base_url": "[operator identity redacted]", "clock_domain_id": null, "completed": 12, "concurrency": 1, "configuration_fingerprint": null, "context_distribution": { "8192": 12 }, "context_seed": 17, "context_tokens": 8192, "controlled_output_policy": "observe", "engine": "strata", "failed": 0, "failures": [], "finished_at": "2026-10-05T10:11:02Z", "finished_at_unix_ns": 1791195062159570600, "gpu": "rtx5090", "manifest": null, "max_context_tokens": 32768, "max_tokens": 8192, "measurement_protocol": "capacity-v4-reasoning", "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": 437.0, "completion_tokens_mean": 273.25, "completion_tokens_mean_ci95_high": 320.43586129404235, "completion_tokens_mean_ci95_low": 226.06413870595762, "completion_tokens_min": 144.0, "completion_tokens_p25": 201.0, "completion_tokens_p50": 266.0, "completion_tokens_p75": 309.0, "completion_tokens_p90": 376.0, "completion_tokens_p95": 437.0, "completion_tokens_p99": 437.0, "completion_tokens_samples": 12, "completion_tokens_stddev": 83.39623383691747, "content_chunks": 448, "content_chunks_s": null, "controlled_output_exact_adherence": 0, "controlled_output_exact_adherence_rate": null, "controlled_output_samples": 0, "controlled_output_unobservable": 0, "decode_tok_s_max": 97.57542153199611, "decode_tok_s_mean": 84.96997490337388, "decode_tok_s_mean_ci95_high": 88.59438637503679, "decode_tok_s_mean_ci95_low": 81.34556343171097, "decode_tok_s_min": 76.74325148662787, "decode_tok_s_p25": 79.48964053960528, "decode_tok_s_p50": 83.0969929066089, "decode_tok_s_p75": 89.17971314939088, "decode_tok_s_p90": 92.85065343195501, "decode_tok_s_p95": 97.57542153199611, "decode_tok_s_p99": 97.57542153199611, "decode_tok_s_samples": 12, "decode_tok_s_stddev": 6.405780424954739, "e2e_max_ms": 33371.91470000107, "e2e_mean_ci95_high_ms": 31703.609776753343, "e2e_mean_ci95_low_ms": 30034.95475658094, "e2e_mean_ms": 30869.28226666714, "e2e_min_ms": 27670.345200000156, "e2e_p25_ms": 30120.0276999989, "e2e_p50_ms": 30851.82150000037, "e2e_p75_ms": 31553.68339999768, "e2e_p90_ms": 32245.12990000221, "e2e_p95_ms": 33371.91470000107, "e2e_p99_ms": 33371.91470000107, "e2e_samples": 12, "e2e_stddev_ms": 1474.5894261446244, "effective_prefill_tok_s_max": 286.0874725437244, "effective_prefill_tok_s_mean": 269.1537966038481, "effective_prefill_tok_s_mean_ci95_high": 275.75867003815347, "effective_prefill_tok_s_mean_ci95_low": 262.5489231695427, "effective_prefill_tok_s_min": 255.44893639537503, "effective_prefill_tok_s_p25": 257.7945051293644, "effective_prefill_tok_s_p50": 264.07079093993497, "effective_prefill_tok_s_p75": 278.372175111899, "effective_prefill_tok_s_p90": 284.5760964173583, "effective_prefill_tok_s_p95": 286.0874725437244, "effective_prefill_tok_s_p99": 286.0874725437244, "effective_prefill_tok_s_samples": 12, "effective_prefill_tok_s_stddev": 11.673445271202928, "generation_max_ms": 5246.880599999713, "generation_mean_ci95_high_ms": 3845.212962883482, "generation_mean_ci95_low_ms": 2634.3986537830824, "generation_mean_ms": 3239.805808333282, "generation_min_ms": 1603.5037000001466, "generation_p25_ms": 2049.69649999839, "generation_p50_ms": 3317.1548999998777, "generation_p75_ms": 3505.194199999096, "generation_p90_ms": 4680.224600000656, "generation_p95_ms": 5246.880599999713, "generation_p99_ms": 5246.880599999713, "generation_samples": 12, "generation_stddev_ms": 1069.995868312908, "mean_inter_token_latency_ms_max": 13.030461710033814, "mean_inter_token_latency_ms_mean": 11.828594955874477, "mean_inter_token_latency_ms_mean_ci95_high": 12.31892060846487, "mean_inter_token_latency_ms_mean_ci95_low": 11.338269303284084, "mean_inter_token_latency_ms_min": 10.248482499991951, "mean_inter_token_latency_ms_p25": 11.103363773584551, "mean_inter_token_latency_ms_p50": 11.82615972695836, "mean_inter_token_latency_ms_p75": 12.480598933335083, "mean_inter_token_latency_ms_p90": 12.913333976844799, "mean_inter_token_latency_ms_p95": 13.030461710033814, "mean_inter_token_latency_ms_p99": 13.030461710033814, "mean_inter_token_latency_ms_samples": 12, "mean_inter_token_latency_ms_stddev": 0.8666009617764543, "mean_time_per_output_token_max_ms": 13.030461710033814, "mean_time_per_output_token_mean_ci95_high_ms": 12.31892060846487, "mean_time_per_output_token_mean_ci95_low_ms": 11.338269303284084, "mean_time_per_output_token_mean_ms": 11.828594955874477, "mean_time_per_output_token_min_ms": 10.248482499991951, "mean_time_per_output_token_p25_ms": 11.103363773584551, "mean_time_per_output_token_p50_ms": 11.82615972695836, "mean_time_per_output_token_p75_ms": 12.480598933335083, "mean_time_per_output_token_p90_ms": 12.913333976844799, "mean_time_per_output_token_p95_ms": 13.030461710033814, "mean_time_per_output_token_p99_ms": 13.030461710033814, "mean_time_per_output_token_samples": 12, "mean_time_per_output_token_stddev_ms": 0.8666009617764543, "output_token_sources": { "usage": 12 }, "output_tokens": 3279, "prefix_cache_hit_avg": 0.0, "prompt_token_samples": 12, "prompt_tokens": 89085, "prompt_tokens_max": 7433.0, "prompt_tokens_mean": 7423.75, "prompt_tokens_mean_ci95_high": 7427.21377312225, "prompt_tokens_mean_ci95_low": 7420.28622687775, "prompt_tokens_min": 7415.0, "prompt_tokens_p25": 7418.0, "prompt_tokens_p50": 7422.0, "prompt_tokens_p75": 7427.0, "prompt_tokens_p90": 7433.0, "prompt_tokens_p95": 7433.0, "prompt_tokens_p99": 7433.0, "prompt_tokens_samples": 12, "prompt_tokens_stddev": 6.1218684016624145, "reasoning_chunks": 2795, "request_canary_passed": 12, "request_canary_samples": 12, "throughput_performance_eligible": true, "throughput_tok_s": 8.85175147610154, "time_to_first_output_max_ms": 29086.047900000267, "time_to_first_output_mean_ci95_high_ms": 28308.195732251934, "time_to_first_output_mean_ci95_low_ms": 26950.757184415783, "time_to_first_output_mean_ms": 27629.47645833386, "time_to_first_output_min_ms": 25943.11430000016, "time_to_first_output_p25_ms": 26264.85870000033, "time_to_first_output_p50_ms": 27842.63550000105, "time_to_first_output_p75_ms": 28773.446899998817, "time_to_first_output_p90_ms": 28833.04280000084, "time_to_first_output_p95_ms": 29086.047900000267, "time_to_first_output_p99_ms": 29086.047900000267, "time_to_first_output_samples": 12, "time_to_first_output_stddev_ms": 1199.5676188799634, "tpot_max_ms": 13.030461710033814, "tpot_mean_ci95_high_ms": 12.31892060846487, "tpot_mean_ci95_low_ms": 11.338269303284084, "tpot_mean_ms": 11.828594955874477, "tpot_min_ms": 10.248482499991951, "tpot_p25_ms": 11.103363773584551, "tpot_p50_ms": 11.82615972695836, "tpot_p75_ms": 12.480598933335083, "tpot_p90_ms": 12.913333976844799, "tpot_p95_ms": 13.030461710033814, "tpot_p99_ms": 13.030461710033814, "tpot_samples": 12, "tpot_stddev_ms": 0.8666009617764543, "ttft_max_ms": 32935.10580000293, "ttft_mean_ci95_high_ms": 31283.005821265717, "ttft_mean_ci95_low_ms": 29632.172078736447, "ttft_mean_ms": 30457.58895000108, "ttft_min_ms": 27364.927000002353, "ttft_p25_ms": 29724.924800000736, "ttft_p50_ms": 30405.80200000113, "ttft_p75_ms": 31041.311299999506, "ttft_p90_ms": 31872.39590000172, "ttft_p95_ms": 32935.10580000293, "ttft_p99_ms": 32935.10580000293, "ttft_samples": 12, "ttft_stddev_ms": 1458.8407739335582 }, "model": "qwen38-flash-next-strata-iq4xs-32k-c1-r36-m50", "performance_eligible": true, "prompt_cache_mode": "unique", "request_canaries": true, "request_timings": [ { "content_chunks": 37, "controlled_output": null, "decode_tok_s": 76.74325148662787, "e2e_ms": 30324.183299999277, "effective_prefill_tok_s": 276.48320271698645, "generation_ms": 3505.194199999096, "mean_inter_token_latency_ms": 13.030461710033814, "mean_time_per_output_token_ms": 13.030461710033814, "output_token_source": "usage", "output_tokens": 270, "planned_context_tokens": 8192, "prompt_tokens": 7415, "reasoning_chunks": 230, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_17_00000", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 0, "time_to_first_output_ms": 26818.98910000018, "tpot_ms": 13.030461710033814, "ttft_ms": 29813.372399999935, "visible_content_truncated": false, "visible_generation_ms": 510.8108999993419 }, { "content_chunks": 45, "controlled_output": null, "decode_tok_s": 90.06279721997845, "e2e_ms": 30785.026900000958, "effective_prefill_tok_s": 266.39001182196705, "generation_ms": 2942.391399999906, "mean_inter_token_latency_ms": 11.103363773584551, "mean_time_per_output_token_ms": 11.103363773584551, "output_token_source": "usage", "output_tokens": 266, "planned_context_tokens": 8192, "prompt_tokens": 7417, "reasoning_chunks": 218, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_17_00001", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 1, "time_to_first_output_ms": 27842.63550000105, "tpot_ms": 11.103363773584551, "ttft_ms": 30322.291100001166, "visible_content_truncated": false, "visible_generation_ms": 462.7357999997912 }, { "content_chunks": 36, "controlled_output": null, "decode_tok_s": 89.17971314939088, "e2e_ms": 27670.345200000156, "effective_prefill_tok_s": 284.5760964173583, "generation_ms": 1603.5037000001466, "mean_inter_token_latency_ms": 11.213312587413613, "mean_time_per_output_token_ms": 11.213312587413613, "output_token_source": "usage", "output_tokens": 144, "planned_context_tokens": 8192, "prompt_tokens": 7418, "reasoning_chunks": 105, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_17_00002", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 2, "time_to_first_output_ms": 26066.84150000001, "tpot_ms": 11.213312587413613, "ttft_ms": 27364.927000002353, "visible_content_truncated": false, "visible_generation_ms": 305.41819999780273 }, { "content_chunks": 28, "controlled_output": null, "decode_tok_s": 92.85065343195501, "e2e_ms": 32140.75910000247, "effective_prefill_tok_s": 257.4279034819432, "generation_ms": 3317.1548999998777, "mean_inter_token_latency_ms": 10.769983441558045, "mean_time_per_output_token_ms": 10.769983441558045, "output_token_source": "usage", "output_tokens": 309, "planned_context_tokens": 8192, "prompt_tokens": 7420, "reasoning_chunks": 278, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_17_00003", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 3, "time_to_first_output_ms": 28823.604200002592, "tpot_ms": 10.769983441558045, "ttft_ms": 31872.39590000172, "visible_content_truncated": false, "visible_generation_ms": 268.3632000007492 }, { "content_chunks": 44, "controlled_output": null, "decode_tok_s": 84.55830320982786, "e2e_ms": 30120.0276999989, "effective_prefill_tok_s": 278.372175111899, "generation_ms": 3465.0647999987996, "mean_inter_token_latency_ms": 11.82615972695836, "mean_time_per_output_token_ms": 11.82615972695836, "output_token_source": "usage", "output_tokens": 294, "planned_context_tokens": 8192, "prompt_tokens": 7420, "reasoning_chunks": 247, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_17_00004", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 4, "time_to_first_output_ms": 26654.9629000001, "tpot_ms": 11.82615972695836, "ttft_ms": 29724.924800000736, "visible_content_truncated": false, "visible_generation_ms": 395.1028999981645 }, { "content_chunks": 38, "controlled_output": null, "decode_tok_s": 77.43933532526327, "e2e_ms": 29287.667800002964, "effective_prefill_tok_s": 286.0874725437244, "generation_ms": 3344.553500002803, "mean_inter_token_latency_ms": 12.913333976844799, "mean_time_per_output_token_ms": 12.913333976844799, "output_token_source": "usage", "output_tokens": 260, "planned_context_tokens": 8192, "prompt_tokens": 7422, "reasoning_chunks": 219, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_17_00005", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 5, "time_to_first_output_ms": 25943.11430000016, "tpot_ms": 12.913333976844799, "ttft_ms": 28883.999900001072, "visible_content_truncated": false, "visible_generation_ms": 403.6679000018921 }, { "content_chunks": 44, "controlled_output": null, "decode_tok_s": 80.12435984374498, "e2e_ms": 30945.083300000988, "effective_prefill_tok_s": 282.65904967537125, "generation_ms": 4680.224600000656, "mean_inter_token_latency_ms": 12.480598933335083, "mean_time_per_output_token_ms": 12.480598933335083, "output_token_source": "usage", "output_tokens": 376, "planned_context_tokens": 8192, "prompt_tokens": 7424, "reasoning_chunks": 329, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_17_00006", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 6, "time_to_first_output_ms": 26264.85870000033, "tpot_ms": 12.480598933335083, "ttft_ms": 30405.80200000113, "visible_content_truncated": false, "visible_generation_ms": 539.2812999998569 }, { "content_chunks": 40, "controlled_output": null, "decode_tok_s": 82.8238540752425, "e2e_ms": 32245.12990000221, "effective_prefill_tok_s": 262.2069628864737, "generation_ms": 3923.9903000016056, "mean_inter_token_latency_ms": 12.073816307697248, "mean_time_per_output_token_ms": 12.073816307697248, "output_token_source": "usage", "output_tokens": 326, "planned_context_tokens": 8192, "prompt_tokens": 7426, "reasoning_chunks": 283, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_17_00007", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 7, "time_to_first_output_ms": 28321.139600000606, "tpot_ms": 12.073816307697248, "ttft_ms": 31766.22640000278, "visible_content_truncated": false, "visible_generation_ms": 478.9034999994328 }, { "content_chunks": 39, "controlled_output": null, "decode_tok_s": 83.0969929066089, "e2e_ms": 33371.91470000107, "effective_prefill_tok_s": 264.07079093993497, "generation_ms": 5246.880599999713, "mean_inter_token_latency_ms": 12.034129816513103, "mean_time_per_output_token_ms": 12.034129816513103, "output_token_source": "usage", "output_tokens": 437, "planned_context_tokens": 8192, "prompt_tokens": 7427, "reasoning_chunks": 395, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_17_00008", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 8, "time_to_first_output_ms": 28125.034100001358, "tpot_ms": 12.034129816513103, "ttft_ms": 32935.10580000293, "visible_content_truncated": false, "visible_generation_ms": 436.808899998141 }, { "content_chunks": 27, "controlled_output": null, "decode_tok_s": 97.57542153199611, "e2e_ms": 31135.744399998657, "effective_prefill_tok_s": 255.44893639537503, "generation_ms": 2049.69649999839, "mean_inter_token_latency_ms": 10.248482499991951, "mean_time_per_output_token_ms": 10.248482499991951, "output_token_source": "usage", "output_tokens": 201, "planned_context_tokens": 8192, "prompt_tokens": 7430, "reasoning_chunks": 171, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_17_00009", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 9, "time_to_first_output_ms": 29086.047900000267, "tpot_ms": 10.248482499991951, "ttft_ms": 30909.002399999736, "visible_content_truncated": false, "visible_generation_ms": 226.74199999892153 }, { "content_chunks": 35, "controlled_output": null, "decode_tok_s": 79.48964053960528, "e2e_ms": 31553.68339999768, "effective_prefill_tok_s": 258.3284521257794, "generation_ms": 2780.236499998864, "mean_inter_token_latency_ms": 12.580255656103457, "mean_time_per_output_token_ms": 12.580255656103457, "output_token_source": "usage", "output_tokens": 222, "planned_context_tokens": 8192, "prompt_tokens": 7433, "reasoning_chunks": 184, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_17_00010", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 10, "time_to_first_output_ms": 28773.446899998817, "tpot_ms": 12.580255656103457, "ttft_ms": 31041.311299999506, "visible_content_truncated": false, "visible_generation_ms": 512.3720999981742 }, { "content_chunks": 35, "controlled_output": null, "decode_tok_s": 85.69537612024553, "e2e_ms": 30851.82150000037, "effective_prefill_tok_s": 257.7945051293644, "generation_ms": 2018.7786999995296, "mean_inter_token_latency_ms": 11.669241040459708, "mean_time_per_output_token_ms": 11.669241040459708, "output_token_source": "usage", "output_tokens": 174, "planned_context_tokens": 8192, "prompt_tokens": 7433, "reasoning_chunks": 136, "request_canary": { "capture_complete": true, "foreign_marker_count": 0, "marker": "ANVIL_REQ_17_00011", "marker_at_start": true, "observed_marker_count": 1, "passed": true }, "request_index": 11, "time_to_first_output_ms": 28833.04280000084, "tpot_ms": 11.669241040459708, "ttft_ms": 30451.708399999916, "visible_content_truncated": false, "visible_generation_ms": 400.1131000004534 } ], "requests": 12, "response_words": 0, "responses_received": 12, "run_id": "benchmark-20261005T100451Z", "sampling": { "temperature": { "effective_request": 0.9, "requested": 0.9, "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": "2026-10-05T10:04:51Z", "started_at_unix_ns": 1791194691724404300, "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": 370435.16290000116 }