{"base_url":"REDACTED_LOOPBACK_ENDPOINT","clock_domain_id":null,"completed":0,"concurrency":4,"configuration_fingerprint":null,"context_distribution":{"32768":4},"context_seed":0,"context_tokens":32768,"controlled_output_policy":"strict","engine":"vllm-00a33e231420","failed":4,"failures":[{"controlled_output":{"capture_complete":true,"exact_adherence":false,"observed_code_words":189,"observed_extra_words":0,"passed":false,"policy":"strict","requested_words":128},"elapsed_ms":6374.2115829954855,"error_message":"controlled output validation failed","error_message_truncated":false,"error_type":"ValueError","failure_class":"controlled_output_failed","performance_eligible":false,"planned_context_tokens":32768,"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":194,"finish_reasons":["stop"],"output_token_source":"usage","reasoning_chunks":36,"stream_terminal_observed":true,"usage":{"completion_tokens":236,"prompt_tokens":25280,"total_tokens":25516},"visible_content_truncated":false},"response_received":true,"response_timing":{"content_chunks":194,"controlled_output":{"capture_complete":true,"exact_adherence":false,"observed_code_words":189,"observed_extra_words":0,"passed":false,"policy":"strict","requested_words":128},"decode_tok_s":65.61429913759335,"e2e_ms":6370.797639989178,"effective_prefill_tok_s":9063.331730232056,"generation_ms":3581.536388999666,"mean_inter_token_latency_ms":15.240580378721983,"mean_time_per_output_token_ms":15.240580378721983,"output_token_source":"usage","output_tokens":236,"planned_context_tokens":32768,"prompt_tokens":25280,"reasoning_chunks":36,"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":2789.2612509895116,"tpot_ms":15.240580378721983,"ttft_ms":3483.726748992922,"visible_content_truncated":false,"visible_generation_ms":2887.070890996256}},{"controlled_output":{"capture_complete":true,"exact_adherence":false,"observed_code_words":186,"observed_extra_words":0,"passed":false,"policy":"strict","requested_words":128},"elapsed_ms":11300.83131599531,"error_message":"controlled output validation failed","error_message_truncated":false,"error_type":"ValueError","failure_class":"controlled_output_failed","performance_eligible":false,"planned_context_tokens":32768,"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":193,"finish_reasons":["stop"],"output_token_source":"usage","reasoning_chunks":56,"stream_terminal_observed":true,"usage":{"completion_tokens":254,"prompt_tokens":25281,"total_tokens":25535},"visible_content_truncated":false},"response_received":true,"response_timing":{"content_chunks":193,"controlled_output":{"capture_complete":true,"exact_adherence":false,"observed_code_words":186,"observed_extra_words":0,"passed":false,"policy":"strict","requested_words":128},"decode_tok_s":67.70669321348495,"e2e_ms":11298.245597994537,"effective_prefill_tok_s":3343.3666964268186,"generation_ms":3736.705899995286,"mean_inter_token_latency_ms":14.769588537530774,"mean_time_per_output_token_ms":14.769588537530774,"output_token_source":"usage","output_tokens":254,"planned_context_tokens":32768,"prompt_tokens":25281,"reasoning_chunks":56,"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":7561.539697999251,"tpot_ms":14.769588537530774,"ttft_ms":8436.798504000762,"visible_content_truncated":false,"visible_generation_ms":2861.4470939937746}},{"controlled_output":{"capture_complete":true,"exact_adherence":false,"observed_code_words":192,"observed_extra_words":0,"passed":false,"policy":"strict","requested_words":128},"elapsed_ms":15872.525091996067,"error_message":"controlled output validation failed","error_message_truncated":false,"error_type":"ValueError","failure_class":"controlled_output_failed","performance_eligible":false,"planned_context_tokens":32768,"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":197,"finish_reasons":["stop"],"output_token_source":"usage","reasoning_chunks":54,"stream_terminal_observed":true,"usage":{"completion_tokens":258,"prompt_tokens":25281,"total_tokens":25539},"visible_content_truncated":false},"response_received":true,"response_timing":{"content_chunks":197,"controlled_output":{"capture_complete":true,"exact_adherence":false,"observed_code_words":192,"observed_extra_words":0,"passed":false,"policy":"strict","requested_words":128},"decode_tok_s":68.17739891396867,"e2e_ms":15869.910444002016,"effective_prefill_tok_s":2089.281382213098,"generation_ms":3769.5776620093966,"mean_inter_token_latency_ms":14.667617361904266,"mean_time_per_output_token_ms":14.667617361904266,"output_token_source":"usage","output_tokens":258,"planned_context_tokens":32768,"prompt_tokens":25281,"reasoning_chunks":54,"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":12100.33278199262,"tpot_ms":14.667617361904266,"ttft_ms":12967.040435003582,"visible_content_truncated":false,"visible_generation_ms":2902.870008998434}},{"controlled_output":{"capture_complete":false,"exact_adherence":null,"observed_code_words":1635,"observed_extra_words":0,"passed":false,"policy":"strict","requested_words":128},"elapsed_ms":88040.09105899604,"error_message":"request canary validation failed","error_message_truncated":false,"error_type":"ValueError","failure_class":"request_canary_failed","performance_eligible":false,"planned_context_tokens":32768,"request_canary":{"capture_complete":false,"foreign_marker_count":0,"marker":"ANVIL_REQ_0_00003","marker_at_start":true,"observed_marker_count":1,"passed":false},"request_index":3,"response_metadata":{"content_chunks":8072,"finish_reasons":["length"],"output_token_source":"usage","reasoning_chunks":24,"stream_terminal_observed":true,"usage":{"completion_tokens":8192,"prompt_tokens":25281,"total_tokens":33473},"visible_content_truncated":true},"response_received":true,"response_timing":{"content_chunks":8072,"controlled_output":{"capture_complete":false,"exact_adherence":null,"observed_code_words":1635,"observed_extra_words":0,"passed":false,"policy":"strict","requested_words":128},"decode_tok_s":114.51035107700919,"e2e_ms":88037.29832699173,"effective_prefill_tok_s":1531.5648326728224,"generation_ms":71530.6513599935,"mean_inter_token_latency_ms":8.732834984738554,"mean_time_per_output_token_ms":8.732834984738554,"output_token_source":"usage","output_tokens":8192,"planned_context_tokens":32768,"prompt_tokens":25281,"reasoning_chunks":24,"request_canary":{"capture_complete":false,"foreign_marker_count":0,"marker":"ANVIL_REQ_0_00003","marker_at_start":true,"observed_marker_count":1,"passed":false},"request_index":3,"time_to_first_output_ms":16506.646966998233,"tpot_ms":8.732834984738554,"ttft_ms":16713.062598995748,"visible_content_truncated":true,"visible_generation_ms":71324.23572799598}}],"finished_at":"2026-10-05T09:11:58Z","finished_at_unix_ns":1791191518098717043,"gpu":"dual-rtx-pro-6000-maxq","manifest":null,"max_context_tokens":327680,"max_tokens":8192,"measurement_protocol":"capacity-v3","metric_population":{"aggregate_throughput_eligible":false,"attempted_requests":4,"excluded_requests":4,"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":"glm53-csf-qad-tp2-dcp2-c4-327k-r1","performance_eligible":false,"prompt_cache_mode":"unique","request_canaries":true,"request_timings":[],"requests":4,"response_words":128,"responses_received":4,"run_id":"benchmark-20261005T091030Z","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":"enabled"},"shared_prefix_tokens":0,"source_recipe":null,"started_at":"2026-10-05T09:10:30Z","started_at_unix_ns":1791191430047095281,"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":88051.61723899073}