{ "base_url": "[operator identity redacted]", "clock_domain_id": null, "completed": 0, "concurrency": 1, "configuration_fingerprint": null, "context_distribution": { "8192": 12 }, "context_seed": 17, "context_tokens": 8192, "controlled_output_policy": "strict", "engine": "ninfer", "failed": 12, "failures": [ { "elapsed_ms": 6189.5127999996475, "error_message": "stream completed without visible content", "error_message_truncated": false, "error_type": "ValueError", "failure_class": "stream_validation_failed", "performance_eligible": false, "planned_context_tokens": 8192, "request_index": 0, "response_metadata": { "content_chunks": 0, "finish_reasons": [ "stop" ], "output_token_source": "usage", "reasoning_chunks": 318, "stream_terminal_observed": true, "usage": { "completion_tokens": 1811, "prompt_tokens": 7409, "total_tokens": 9220 }, "visible_content_truncated": false }, "response_received": true, "response_timing": null }, { "elapsed_ms": 6936.442700000043, "error_message": "stream completed without visible content", "error_message_truncated": false, "error_type": "ValueError", "failure_class": "stream_validation_failed", "performance_eligible": false, "planned_context_tokens": 8192, "request_index": 1, "response_metadata": { "content_chunks": 0, "finish_reasons": [ "stop" ], "output_token_source": "usage", "reasoning_chunks": 381, "stream_terminal_observed": true, "usage": { "completion_tokens": 1965, "prompt_tokens": 7411, "total_tokens": 9376 }, "visible_content_truncated": false }, "response_received": true, "response_timing": null }, { "elapsed_ms": 20790.332399999897, "error_message": "stream completed without visible content", "error_message_truncated": false, "error_type": "ValueError", "failure_class": "stream_validation_failed", "performance_eligible": false, "planned_context_tokens": 8192, "request_index": 2, "response_metadata": { "content_chunks": 0, "finish_reasons": [ "stop" ], "output_token_source": "usage", "reasoning_chunks": 1201, "stream_terminal_observed": true, "usage": { "completion_tokens": 7268, "prompt_tokens": 7413, "total_tokens": 14681 }, "visible_content_truncated": false }, "response_received": true, "response_timing": null }, { "elapsed_ms": 11189.866400000028, "error_message": "stream completed without visible content", "error_message_truncated": false, "error_type": "ValueError", "failure_class": "stream_validation_failed", "performance_eligible": false, "planned_context_tokens": 8192, "request_index": 3, "response_metadata": { "content_chunks": 0, "finish_reasons": [ "stop" ], "output_token_source": "usage", "reasoning_chunks": 635, "stream_terminal_observed": true, "usage": { "completion_tokens": 4022, "prompt_tokens": 7414, "total_tokens": 11436 }, "visible_content_truncated": false }, "response_received": true, "response_timing": null }, { "elapsed_ms": 11507.991799999218, "error_message": "stream completed without visible content", "error_message_truncated": false, "error_type": "ValueError", "failure_class": "stream_validation_failed", "performance_eligible": false, "planned_context_tokens": 8192, "request_index": 4, "response_metadata": { "content_chunks": 0, "finish_reasons": [ "stop" ], "output_token_source": "usage", "reasoning_chunks": 656, "stream_terminal_observed": true, "usage": { "completion_tokens": 3493, "prompt_tokens": 7414, "total_tokens": 10907 }, "visible_content_truncated": false }, "response_received": true, "response_timing": null }, { "elapsed_ms": 6406.7680000007385, "error_message": "stream completed without visible content", "error_message_truncated": false, "error_type": "ValueError", "failure_class": "stream_validation_failed", "performance_eligible": false, "planned_context_tokens": 8192, "request_index": 5, "response_metadata": { "content_chunks": 0, "finish_reasons": [ "stop" ], "output_token_source": "usage", "reasoning_chunks": 350, "stream_terminal_observed": true, "usage": { "completion_tokens": 1820, "prompt_tokens": 7415, "total_tokens": 9235 }, "visible_content_truncated": false }, "response_received": true, "response_timing": null }, { "elapsed_ms": 8451.967199998762, "error_message": "stream completed without visible content", "error_message_truncated": false, "error_type": "ValueError", "failure_class": "stream_validation_failed", "performance_eligible": false, "planned_context_tokens": 8192, "request_index": 6, "response_metadata": { "content_chunks": 0, "finish_reasons": [ "stop" ], "output_token_source": "usage", "reasoning_chunks": 473, "stream_terminal_observed": true, "usage": { "completion_tokens": 2674, "prompt_tokens": 7415, "total_tokens": 10089 }, "visible_content_truncated": false }, "response_received": true, "response_timing": null }, { "elapsed_ms": 16773.205299999972, "error_message": "stream completed without visible content", "error_message_truncated": false, "error_type": "ValueError", "failure_class": "stream_validation_failed", "performance_eligible": false, "planned_context_tokens": 8192, "request_index": 7, "response_metadata": { "content_chunks": 0, "finish_reasons": [ "stop" ], "output_token_source": "usage", "reasoning_chunks": 968, "stream_terminal_observed": true, "usage": { "completion_tokens": 4886, "prompt_tokens": 7415, "total_tokens": 12301 }, "visible_content_truncated": false }, "response_received": true, "response_timing": null }, { "elapsed_ms": 5845.379700000194, "error_message": "stream completed without visible content", "error_message_truncated": false, "error_type": "ValueError", "failure_class": "stream_validation_failed", "performance_eligible": false, "planned_context_tokens": 8192, "request_index": 8, "response_metadata": { "content_chunks": 0, "finish_reasons": [ "stop" ], "output_token_source": "usage", "reasoning_chunks": 316, "stream_terminal_observed": true, "usage": { "completion_tokens": 1735, "prompt_tokens": 7417, "total_tokens": 9152 }, "visible_content_truncated": false }, "response_received": true, "response_timing": null }, { "elapsed_ms": 10454.892700001437, "error_message": "stream completed without visible content", "error_message_truncated": false, "error_type": "ValueError", "failure_class": "stream_validation_failed", "performance_eligible": false, "planned_context_tokens": 8192, "request_index": 9, "response_metadata": { "content_chunks": 0, "finish_reasons": [ "stop" ], "output_token_source": "usage", "reasoning_chunks": 594, "stream_terminal_observed": true, "usage": { "completion_tokens": 2988, "prompt_tokens": 7419, "total_tokens": 10407 }, "visible_content_truncated": false }, "response_received": true, "response_timing": null }, { "elapsed_ms": 13487.989200000811, "error_message": "stream completed without visible content", "error_message_truncated": false, "error_type": "ValueError", "failure_class": "stream_validation_failed", "performance_eligible": false, "planned_context_tokens": 8192, "request_index": 10, "response_metadata": { "content_chunks": 0, "finish_reasons": [ "stop" ], "output_token_source": "usage", "reasoning_chunks": 770, "stream_terminal_observed": true, "usage": { "completion_tokens": 5290, "prompt_tokens": 7421, "total_tokens": 12711 }, "visible_content_truncated": false }, "response_received": true, "response_timing": null }, { "elapsed_ms": 20402.01209999941, "error_message": "stream completed without visible content", "error_message_truncated": false, "error_type": "ValueError", "failure_class": "stream_validation_failed", "performance_eligible": false, "planned_context_tokens": 8192, "request_index": 11, "response_metadata": { "content_chunks": 0, "finish_reasons": [ "length" ], "output_token_source": "usage", "reasoning_chunks": 1177, "stream_terminal_observed": true, "usage": { "completion_tokens": 8192, "prompt_tokens": 7422, "total_tokens": 15614 }, "visible_content_truncated": false }, "response_received": true, "response_timing": null } ], "finished_at": "2026-10-05T09:14:53Z", "finished_at_unix_ns": 1791191693178858900, "gpu": "rtx5090", "manifest": null, "max_context_tokens": 262144, "max_tokens": 8192, "measurement_protocol": "capacity-v3", "metric_population": { "aggregate_throughput_eligible": false, "attempted_requests": 12, "excluded_requests": 12, "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": 0, "e2e": 0, "effective_prefill_tok_s": 0, "tpot": 0, "ttft": 0 }, "performance_eligible_requests": 0, "successful_requests": 0 }, "metrics": { "completion_tokens_max": null, "completion_tokens_mean": null, "completion_tokens_mean_ci95_high": null, "completion_tokens_mean_ci95_low": null, "completion_tokens_min": null, "completion_tokens_p25": null, "completion_tokens_p50": null, "completion_tokens_p75": null, "completion_tokens_p90": null, "completion_tokens_p95": null, "completion_tokens_p99": null, "completion_tokens_samples": 0, "completion_tokens_stddev": null, "content_chunks": 0, "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": null, "decode_tok_s_mean": null, "decode_tok_s_mean_ci95_high": null, "decode_tok_s_mean_ci95_low": null, "decode_tok_s_min": null, "decode_tok_s_p25": null, "decode_tok_s_p50": null, "decode_tok_s_p75": null, "decode_tok_s_p90": null, "decode_tok_s_p95": null, "decode_tok_s_p99": null, "decode_tok_s_samples": 0, "decode_tok_s_stddev": null, "e2e_max_ms": null, "e2e_mean_ci95_high_ms": null, "e2e_mean_ci95_low_ms": null, "e2e_mean_ms": null, "e2e_min_ms": null, "e2e_p25_ms": null, "e2e_p50_ms": null, "e2e_p75_ms": null, "e2e_p90_ms": null, "e2e_p95_ms": null, "e2e_p99_ms": null, "e2e_samples": 0, "e2e_stddev_ms": null, "effective_prefill_tok_s_max": null, "effective_prefill_tok_s_mean": null, "effective_prefill_tok_s_mean_ci95_high": null, "effective_prefill_tok_s_mean_ci95_low": null, "effective_prefill_tok_s_min": null, "effective_prefill_tok_s_p25": null, "effective_prefill_tok_s_p50": null, "effective_prefill_tok_s_p75": null, "effective_prefill_tok_s_p90": null, "effective_prefill_tok_s_p95": null, "effective_prefill_tok_s_p99": null, "effective_prefill_tok_s_samples": 0, "effective_prefill_tok_s_stddev": null, "generation_max_ms": null, "generation_mean_ci95_high_ms": null, "generation_mean_ci95_low_ms": null, "generation_mean_ms": null, "generation_min_ms": null, "generation_p25_ms": null, "generation_p50_ms": null, "generation_p75_ms": null, "generation_p90_ms": null, "generation_p95_ms": null, "generation_p99_ms": null, "generation_samples": 0, "generation_stddev_ms": null, "mean_inter_token_latency_ms_max": null, "mean_inter_token_latency_ms_mean": null, "mean_inter_token_latency_ms_mean_ci95_high": null, "mean_inter_token_latency_ms_mean_ci95_low": null, "mean_inter_token_latency_ms_min": null, "mean_inter_token_latency_ms_p25": null, "mean_inter_token_latency_ms_p50": null, "mean_inter_token_latency_ms_p75": null, "mean_inter_token_latency_ms_p90": null, "mean_inter_token_latency_ms_p95": null, "mean_inter_token_latency_ms_p99": null, "mean_inter_token_latency_ms_samples": 0, "mean_inter_token_latency_ms_stddev": null, "mean_time_per_output_token_max_ms": null, "mean_time_per_output_token_mean_ci95_high_ms": null, "mean_time_per_output_token_mean_ci95_low_ms": null, "mean_time_per_output_token_mean_ms": null, "mean_time_per_output_token_min_ms": null, "mean_time_per_output_token_p25_ms": null, "mean_time_per_output_token_p50_ms": null, "mean_time_per_output_token_p75_ms": null, "mean_time_per_output_token_p90_ms": null, "mean_time_per_output_token_p95_ms": null, "mean_time_per_output_token_p99_ms": null, "mean_time_per_output_token_samples": 0, "mean_time_per_output_token_stddev_ms": null, "output_token_sources": {}, "output_tokens": 0, "prefix_cache_hit_avg": null, "prompt_token_samples": 0, "prompt_tokens": null, "prompt_tokens_max": null, "prompt_tokens_mean": null, "prompt_tokens_mean_ci95_high": null, "prompt_tokens_mean_ci95_low": null, "prompt_tokens_min": null, "prompt_tokens_p25": null, "prompt_tokens_p50": null, "prompt_tokens_p75": null, "prompt_tokens_p90": null, "prompt_tokens_p95": null, "prompt_tokens_p99": null, "prompt_tokens_samples": 0, "prompt_tokens_stddev": null, "reasoning_chunks": 0, "request_canary_passed": 0, "request_canary_samples": 0, "throughput_performance_eligible": false, "throughput_tok_s": null, "time_to_first_output_max_ms": null, "time_to_first_output_mean_ci95_high_ms": null, "time_to_first_output_mean_ci95_low_ms": null, "time_to_first_output_mean_ms": null, "time_to_first_output_min_ms": null, "time_to_first_output_p25_ms": null, "time_to_first_output_p50_ms": null, "time_to_first_output_p75_ms": null, "time_to_first_output_p90_ms": null, "time_to_first_output_p95_ms": null, "time_to_first_output_p99_ms": null, "time_to_first_output_samples": 0, "time_to_first_output_stddev_ms": null, "tpot_max_ms": null, "tpot_mean_ci95_high_ms": null, "tpot_mean_ci95_low_ms": null, "tpot_mean_ms": null, "tpot_min_ms": null, "tpot_p25_ms": null, "tpot_p50_ms": null, "tpot_p75_ms": null, "tpot_p90_ms": null, "tpot_p95_ms": null, "tpot_p99_ms": null, "tpot_samples": 0, "tpot_stddev_ms": null, "ttft_max_ms": null, "ttft_mean_ci95_high_ms": null, "ttft_mean_ci95_low_ms": null, "ttft_mean_ms": null, "ttft_min_ms": null, "ttft_p25_ms": null, "ttft_p50_ms": null, "ttft_p75_ms": null, "ttft_p90_ms": null, "ttft_p95_ms": null, "ttft_p99_ms": null, "ttft_samples": 0, "ttft_stddev_ms": null }, "model": "qwen38-swift15-ninfer-dflash2-c4-262k", "performance_eligible": false, "prompt_cache_mode": "unique", "request_canaries": true, "request_timings": [], "requests": 12, "response_words": 128, "responses_received": 12, "run_id": "benchmark-20261005T091234Z", "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": "2026-10-05T09:12:34Z", "started_at_unix_ns": 1791191554741045200, "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": 138437.81089999902 }