{ "base_url": "http://127.0.0.1:8081/v1", "clock_domain_id": null, "completed": 0, "concurrency": 1, "configuration_fingerprint": null, "context_distribution": { "4096": 5 }, "context_seed": 0, "context_tokens": 4096, "controlled_output_policy": "strict", "engine": "llama.cpp", "failed": 5, "failures": [ { "controlled_output": { "capture_complete": true, "exact_adherence": false, "observed_code_words": 500, "observed_extra_words": 0, "passed": false, "policy": "strict", "requested_words": 128 }, "elapsed_ms": 4491.92200001562, "error_message": "controlled output validation failed", "error_message_truncated": false, "error_type": "ValueError", "failure_class": "controlled_output_failed", "performance_eligible": false, "planned_context_tokens": 4096, "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, "response_metadata": { "content_chunks": 512, "finish_reasons": [ "length" ], "output_token_source": "usage", "reasoning_chunks": 0, "stream_terminal_observed": true, "usage": { "completion_tokens": 512, "prompt_tokens": 3609, "total_tokens": 4121 }, "visible_content_truncated": false }, "response_received": true, "response_timing": { "content_chunks": 512, "controlled_output": { "capture_complete": true, "exact_adherence": false, "observed_code_words": 500, "observed_extra_words": 0, "passed": false, "policy": "strict", "requested_words": 128 }, "decode_tok_s": 419.96671332160145, "e2e_ms": 4491.475699993316, "effective_prefill_tok_s": 1102.0814467848588, "generation_ms": 1216.763100004755, "mean_inter_token_latency_ms": 2.381141095899716, "mean_time_per_output_token_ms": 2.381141095899716, "output_token_source": "usage", "output_tokens": 512, "planned_context_tokens": 4096, "prompt_tokens": 3609, "reasoning_chunks": 0, "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": 3274.712599988561, "tpot_ms": 2.381141095899716, "ttft_ms": 3274.7137999976985, "visible_content_truncated": false, "visible_generation_ms": 1216.7618999956176 } }, { "controlled_output": { "capture_complete": true, "exact_adherence": false, "observed_code_words": 500, "observed_extra_words": 0, "passed": false, "policy": "strict", "requested_words": 128 }, "elapsed_ms": 2755.686499993317, "error_message": "controlled output validation failed", "error_message_truncated": false, "error_type": "ValueError", "failure_class": "controlled_output_failed", "performance_eligible": false, "planned_context_tokens": 4096, "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, "response_metadata": { "content_chunks": 512, "finish_reasons": [ "length" ], "output_token_source": "usage", "reasoning_chunks": 0, "stream_terminal_observed": true, "usage": { "completion_tokens": 512, "prompt_tokens": 3611, "total_tokens": 4123 }, "visible_content_truncated": false }, "response_received": true, "response_timing": { "content_chunks": 512, "controlled_output": { "capture_complete": true, "exact_adherence": false, "observed_code_words": 500, "observed_extra_words": 0, "passed": false, "policy": "strict", "requested_words": 128 }, "decode_tok_s": 600.5041885249409, "e2e_ms": 2755.429299955722, "effective_prefill_tok_s": 1896.0579060510477, "generation_ms": 850.9515999467112, "mean_inter_token_latency_ms": 1.6652673188781042, "mean_time_per_output_token_ms": 1.6652673188781042, "output_token_source": "usage", "output_tokens": 512, "planned_context_tokens": 4096, "prompt_tokens": 3611, "reasoning_chunks": 0, "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": 1904.4777000090107, "tpot_ms": 1.6652673188781042, "ttft_ms": 1904.478799959179, "visible_content_truncated": false, "visible_generation_ms": 850.9504999965429 } }, { "controlled_output": { "capture_complete": true, "exact_adherence": false, "observed_code_words": 500, "observed_extra_words": 0, "passed": false, "policy": "strict", "requested_words": 128 }, "elapsed_ms": 2713.231200003065, "error_message": "controlled output validation failed", "error_message_truncated": false, "error_type": "ValueError", "failure_class": "controlled_output_failed", "performance_eligible": false, "planned_context_tokens": 4096, "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, "response_metadata": { "content_chunks": 512, "finish_reasons": [ "length" ], "output_token_source": "usage", "reasoning_chunks": 0, "stream_terminal_observed": true, "usage": { "completion_tokens": 512, "prompt_tokens": 3611, "total_tokens": 4123 }, "visible_content_truncated": false }, "response_received": true, "response_timing": { "content_chunks": 512, "controlled_output": { "capture_complete": true, "exact_adherence": false, "observed_code_words": 500, "observed_extra_words": 0, "passed": false, "policy": "strict", "requested_words": 128 }, "decode_tok_s": 594.8056818498994, "e2e_ms": 2712.998700037133, "effective_prefill_tok_s": 1947.7914224419146, "generation_ms": 859.1041000327095, "mean_inter_token_latency_ms": 1.6812213307880812, "mean_time_per_output_token_ms": 1.6812213307880812, "output_token_source": "usage", "output_tokens": 512, "planned_context_tokens": 4096, "prompt_tokens": 3611, "reasoning_chunks": 0, "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": 1853.8946000044234, "tpot_ms": 1.6812213307880812, "ttft_ms": 1853.8960000150837, "visible_content_truncated": false, "visible_generation_ms": 859.1027000220492 } }, { "controlled_output": { "capture_complete": true, "exact_adherence": false, "observed_code_words": 500, "observed_extra_words": 0, "passed": false, "policy": "strict", "requested_words": 128 }, "elapsed_ms": 2768.4341000276618, "error_message": "controlled output validation failed", "error_message_truncated": false, "error_type": "ValueError", "failure_class": "controlled_output_failed", "performance_eligible": false, "planned_context_tokens": 4096, "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, "response_metadata": { "content_chunks": 512, "finish_reasons": [ "length" ], "output_token_source": "usage", "reasoning_chunks": 0, "stream_terminal_observed": true, "usage": { "completion_tokens": 512, "prompt_tokens": 3613, "total_tokens": 4125 }, "visible_content_truncated": false }, "response_received": true, "response_timing": { "content_chunks": 512, "controlled_output": { "capture_complete": true, "exact_adherence": false, "observed_code_words": 500, "observed_extra_words": 0, "passed": false, "policy": "strict", "requested_words": 128 }, "decode_tok_s": 590.9075403176097, "e2e_ms": 2768.206399981864, "effective_prefill_tok_s": 1898.147396597982, "generation_ms": 864.7714999970049, "mean_inter_token_latency_ms": 1.6923121330665458, "mean_time_per_output_token_ms": 1.6923121330665458, "output_token_source": "usage", "output_tokens": 512, "planned_context_tokens": 4096, "prompt_tokens": 3613, "reasoning_chunks": 0, "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": 1903.434899984859, "tpot_ms": 1.6923121330665458, "ttft_ms": 1903.4363999962807, "visible_content_truncated": false, "visible_generation_ms": 864.7699999855831 } }, { "controlled_output": { "capture_complete": true, "exact_adherence": false, "observed_code_words": 500, "observed_extra_words": 0, "passed": false, "policy": "strict", "requested_words": 128 }, "elapsed_ms": 2901.7437999718823, "error_message": "controlled output validation failed", "error_message_truncated": false, "error_type": "ValueError", "failure_class": "controlled_output_failed", "performance_eligible": false, "planned_context_tokens": 4096, "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, "response_metadata": { "content_chunks": 512, "finish_reasons": [ "length" ], "output_token_source": "usage", "reasoning_chunks": 0, "stream_terminal_observed": true, "usage": { "completion_tokens": 512, "prompt_tokens": 3615, "total_tokens": 4127 }, "visible_content_truncated": false }, "response_received": true, "response_timing": { "content_chunks": 512, "controlled_output": { "capture_complete": true, "exact_adherence": false, "observed_code_words": 500, "observed_extra_words": 0, "passed": false, "policy": "strict", "requested_words": 128 }, "decode_tok_s": 601.7029252819101, "e2e_ms": 2901.508199982345, "effective_prefill_tok_s": 1761.4796701908626, "generation_ms": 849.2562999599613, "mean_inter_token_latency_ms": 1.661949706379572, "mean_time_per_output_token_ms": 1.661949706379572, "output_token_source": "usage", "output_tokens": 512, "planned_context_tokens": 4096, "prompt_tokens": 3615, "reasoning_chunks": 0, "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": 2052.251900022384, "tpot_ms": 1.661949706379572, "ttft_ms": 2052.254299982451, "visible_content_truncated": false, "visible_generation_ms": 849.2538999998942 } } ], "finished_at": "2026-09-12T18:32:27Z", "finished_at_unix_ns": 1789237947206212500, "gpu": "rtx5090-primary", "manifest": null, "max_context_tokens": 65536, "max_tokens": 512, "measurement_protocol": "capacity-v3", "metric_population": { "aggregate_throughput_eligible": false, "attempted_requests": 5, "excluded_requests": 5, "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": "signal38-27b-q6k-mtp3-64k", "performance_eligible": false, "prompt_cache_mode": "unique", "request_canaries": true, "request_timings": [], "requests": 5, "response_words": 128, "responses_received": 5, "run_id": "benchmark-20260912T183211Z", "schema": "anvil-serving.benchmark/v1", "serve_flags": { "no_thinking": false, "reasoning_effort": null, "shared_prefix_burst": false, "thinking_mode": "disabled" }, "shared_prefix_tokens": 0, "source_recipe": null, "started_at": "2026-09-12T18:32:11Z", "started_at_unix_ns": 1789237931574585800, "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": 15631.928400020115 }