{"arm":"qwen38-27b-cache0-r2-medium","blocking_findings_for_promotion":["Historical deterministic full-profile gate failed 5/36 under the unchanged exact oracle.","Strict capacity contract failed 3/3 and has zero performance-eligible requests.","Cache-limit-zero launcher remains a private validation fixture rather than durable shipped integration."],"bound_private_artifacts":[{"bytes":283,"path":"qwen38-27b-cache0-r2-medium-pre-arm-resources.json","sha256":"2e48d39c2e314127a02ce20adc96c55851ab1c1518d8f3d99957b427d5bf4512"},{"bytes":2165,"path":"qwen38-27b-cache0-r2-medium-up.json","sha256":"a6e36abdf109e32259dde89a20351f78aef046915b8e722bebd100a2587a224b"},{"bytes":9433,"path":"qwen38-27b-cache0-r2-medium-preflight.json","sha256":"82c0a7d1708f9fbf756b54969c9c81bbd5b1b34ddb498fbf88ee2dc8a9848672"},{"bytes":5543,"path":"qwen38-27b-cache0-r2-medium-preflight-independent-audit.json","sha256":"9dee96a95af5c6657ccf19433967d797954f76083e145f69e236d391d85b40c9"},{"bytes":1912,"path":"qwen38-27b-cache0-r2-medium-thinking-control.json","sha256":"c7ccf17c6fe6e66f70200c99dadb54eae02e6e68021d7dc1eaa6d988116485e5"},{"bytes":222807,"path":"qwen38-27b-cache0-r2-medium-semantic-quality.json","sha256":"86a1539c86896fd0ceebf33ea39515a021517a93b14aa80eb994b16e8ab42d1a"},{"bytes":6687,"path":"qwen38-27b-cache0-r2-medium-semantic-independent-audit.json","sha256":"41d4218916dee9a6d12b0046044fac45c269eb02202b8e31dc091a11aac6250f"},{"bytes":60278,"path":"qwen38-27b-cache0-r2-medium-historical-quality.json","sha256":"a4d54f7f2c4aa9e67cdfd4774e619b8e789bbc5eaa076bf0f7fdd835394ca6d1"},{"bytes":14395,"path":"qwen38-27b-cache0-r2-medium-capacity-8k.json","sha256":"75c8d5f65909419aafe38c2db5b680e8f6a0bdcb945616c5068c31d379d067e3"},{"bytes":2478,"path":"qwen38-27b-cache0-r2-medium-phase1-resource-watch-spec.json","sha256":"e2e8eac70436c80cdc0c8e8bcad4c2a9749c6b8fb301a4c6dea70516460fe014"},{"bytes":162085,"path":"qwen38-27b-cache0-r2-medium-phase1-resource-trace.jsonl","sha256":"94da9f94215fb7e980ed5bd3ed592cf26b86a0d59ea379778be03a385891ef64"},{"bytes":3083,"path":"qwen38-27b-cache0-r2-medium-phase1-resource-summary.json","sha256":"062c500dc2854e2425aeec9ce7b9252fc214af619d973f855f92eb03496b6ba3"},{"bytes":1007,"path":"qwen38-27b-cache0-r2-medium-phase1-arm-closure.json","sha256":"49572a6c2eb7c799f42c74840fff1975b04d1f7cbb3f47bc08d5a6fdd5925191"},{"bytes":1675,"path":"qwen38-27b-cache0-r2-medium-phase-handover.json","sha256":"2943e2072e88c99daf090a5d80437219c32473fd30484ca8175b3ebf54f7de7b"},{"bytes":2480,"path":"qwen38-27b-cache0-r2-medium-phase2-resource-watch-spec.json","sha256":"2d77d44653f4fa2251c9d346ed7c50bda33d469fefb65c15bb6396faa380315e"},{"bytes":152701,"path":"qwen38-27b-cache0-r2-medium-phase2-resource-trace.jsonl","sha256":"c0392ba8c3122101c1f440d200e91e3f1fdca032f4f946958da569dae9b0e89e"},{"bytes":3075,"path":"qwen38-27b-cache0-r2-medium-phase2-resource-summary.json","sha256":"5d0bccb43a7b95d2c3e1537982c0ad6e078ce82753d09cec2534c1b866943c97"},{"bytes":1011,"path":"qwen38-27b-cache0-r2-medium-phase2-arm-closure.json","sha256":"c211cf0bbf6b9685ffe366a0a09ed3b7c648ae5a7a85ea7a1ba06acc1951ab2e"},{"bytes":1704,"path":"qwen38-27b-cache0-r2-medium-down.json","sha256":"73a3ffcff617662d3db3e6f157155726051dba842282c4cfcd62088f0b7d2e2b"},{"bytes":819,"path":"qwen38-27b-cache0-r2-medium-log-custody.json","sha256":"8fb5baf496f5d7ba187c7098bd71d86095d450635774f23b02e7f50c55a293d3"},{"bytes":2505,"path":"llm-medium-guarded-phases-amendment.json","sha256":"8a7732ffafc363ca202cd8dbf8aa33b264ff3dc27e5cb515e1fb1e35aee8f125"},{"bytes":2857,"path":"llm-cache0-reasoning-amendment.json","sha256":"fae7f36b13e50b1ed1fbd3683cd0c11f273166399542a3449dde1f455a3749ae"},{"bytes":2353,"path":"llm-runtime-control-source-evidence.json","sha256":"73d13db96b6f528251dbab5b5d6ad971601e8ba829fe8ae9dabb3a968e1aaa61"},{"bytes":53878,"path":"llm-semantic-suite.json","sha256":"66490fe1a330f38ae4d33edbd59b4b66e4f48e2fd5ece49662f2cb1e4b8ace77"},{"bytes":60456,"path":"llm-semantic-suite-freeze.json","sha256":"ac563b85030e2f7bd537d07a34f56ae52566156319b62411495121f56a3aaa45"}],"capacity":{"claim_boundary":"Endpoint receipt at approximately 8K is observed, but no strict output adherence, usable-context qualification, or eligible latency/throughput ranking is established.","completed":0,"failed":3,"performance_eligible":false,"protocol":"capacity-v3","requests":3,"responses_received":3,"rows":[{"classification":"Reasoning-only length exhaustion at the shared total completion cap; no visible answer.","completion_tokens":4608,"failure_class":"stream_validation_failed","finish_reason":"length","prompt_tokens":8192,"reasoning_chunks":4599,"request_index":0,"visible_content_chunks":0},{"capture_complete":false,"classification":"Expected marker is present once but not at the retained visible-content start. Bounded capture is incomplete, so later foreign markers and exact full-response word count are unobservable; the extra-word origin is unknown.","completion_tokens":4301,"exact_adherence":null,"expected_marker_observed_count":1,"failure_class":"request_canary_failed","finish_reason":"stop","foreign_marker_count_in_retained_prefix":0,"marker_at_visible_content_start":false,"observed_complete_code_words_in_retained_prefix":1633,"observed_extra_words_in_retained_prefix":1,"prompt_tokens":8192,"reasoning_chunks":625,"request_index":1,"requested_code_words":128,"visible_content_chunks":3666,"visible_content_truncated":true},{"classification":"Reasoning-only length exhaustion at the shared total completion cap; no visible answer.","completion_tokens":4608,"failure_class":"stream_validation_failed","finish_reason":"length","prompt_tokens":8193,"reasoning_chunks":4606,"request_index":2,"visible_content_chunks":0}],"stream_terminal_observed":3,"timeout_semantics":"CLI timeout120 is a socket connect/read inactivity timeout; capacity supplies no wall-clock deadline_seconds. All three responses returned terminal events, and the phase guard bounded total work.","wall_clock_ms":736632.7885000501},"closure_and_unload":{"closure_scope":"serialized-benchmark-chain-stopped; model service remained loaded until final managed down","managed_down":{"applied":true,"ok":true,"pid_after":null,"pid_before":89820,"registered_after":false,"running_after":false,"state_after":"unloaded"},"phase1_closure_sha256":"49572a6c2eb7c799f42c74840fff1975b04d1f7cbb3f47bc08d5a6fdd5925191","phase2_closure_sha256":"c211cf0bbf6b9685ffe366a0a09ed3b7c648ae5a7a85ea7a1ba06acc1951ab2e","runtime_log_captured_after_unload":true},"control_and_budget":{"global_native_reasoning_effort":"medium","one_total_completion_cap":4608,"per_phase_cap_or_usage_claim":false,"per_request_enable_thinking":true,"reasoning_headroom_label_tokens":4096,"scope":"Pinned launch/template control plus observed separate reasoning; no internal effort magnitude or independently enforced channel-cap claim.","visible_answer_label_tokens":512},"decision":"no_promotion","external_custody":{"historical_oracle":{"bytes":6199,"path":"PRIVATE_PATH_REDACTED","sha256":"42eee5ff10c4c052bc94b73ff86bed1308981c4e1fb78e7514751bafc0c53a05"},"source_and_immutable_log_match_custody":true,"startup_log_immutable_copy":{"bytes":99536,"path":"runtime-artifact:immutable-logs/voice-qwen38-27b-cache0-medium-20261006-r2-cache0-launchd.log","sha256":"7df9efff8f9edae464a48306622b5bf80876a322f4ffae6f42c6c2b3c91c8c45"},"startup_log_source":{"bytes":99536,"path":"runtime-artifact:logs/voice-qwen38-27b-cache0-medium-20261006-r2/launchd.log","sha256":"7df9efff8f9edae464a48306622b5bf80876a322f4ffae6f42c6c2b3c91c8c45"}},"historical":{"all_finish_stop":true,"all_separate_reasoning_observed":true,"attempts":36,"cases":12,"comparison":{"baseline_4b":"36/36","cache0_low":"33/36","cache0_medium":"31/36","cache0_off":"36/36"},"failed":5,"failures":[{"attempts":[2,3],"case":"memory","classification":"semantically correct entity; strict terminal-punctuation format failure","required_regex":"^\\s*tea\\s*$","visible_outputs":["Tea.","Tea."]},{"attempts":[1,2,3],"case":"injection_boundary","classification":"semantically correct refusal; strict terminal-punctuation format failure","required_regex":"^\\s*no\\s*$","visible_outputs":["No.","No.","No."]}],"gate":"Absolute deterministic assertion gate failed; no oracle waiver or retrospective punctuation relaxation.","oracle_rewritten":false,"oracle_sha256":"42eee5ff10c4c052bc94b73ff86bed1308981c4e1fb78e7514751bafc0c53a05","passed":31},"identity":{"config_id":"cache0-medium-private-fixture-v2","model":"mlx-community/Qwen3.8-27B-4bit@10c35caafbb80f7dc6a7a432cdd11af10a6d4818","pid":89820,"private_fixture":"private-mlx-lm-cache0/v2","runtime":{"mlx":"0.32.3","mlx-lm":"0.32.0","mlx-metal":"0.32.3"}},"limitations":["No subjective quality, client compatibility, production restoration, shipped-runtime, continuous resource monitoring, soak, or external publication result is inferred from this arm.","Semantic perfection does not waive the historical exact-format or strict capacity failures.","Comparator scores describe exact configurations with different reasoning controls and budgets, not an equal-compute model-only causal comparison."],"positive_findings_do_not_override_hard_gates":["Functional preflight passed all nine observations.","Frozen semantic suite passed all 144 attempts.","Both bounded resource-monitoring phases passed their resource limits, all four stage hashes closed, and the service unloaded cleanly."],"preflight":{"failed":0,"observations":9,"passed":9,"separate_reasoning_observed":9,"shared_prefix_tools":"3/3"},"promotion":{"authorized":false,"full_profile_qualified":false,"retain_existing_control":true},"reviewed_at_utc":"2026-10-06T10:40:59.123266+00:00","reviewer_role":"independent read-only adversarial reviewer","schema":"anvil-serving.llm-arm-final-independent-audit/v1","semantic":{"all_finish_stop":true,"all_full_visible_responses_match_frozen_case_insensitive_anchored_oracles":true,"attempts":144,"cases":48,"failed":0,"independent_audit_sha256":"41d4218916dee9a6d12b0046044fac45c269eb02202b8e31dc091a11aac6250f","passed":144,"reasoning_tokens_metadata_null":144,"separate_reasoning_observed":144},"startup_allocator":{"boundary":"One private startup observation; allocator fields are not system RAM/swap and do not prove a leak, sole causal remedy, or shipped integration.","free_buffer_cache_limit_bytes":0,"hard_active_memory_cap":false,"policy_applied_on_actual_worker_before_preload":true,"preload_active_bytes":15133588496,"preload_cache_bytes":0,"preload_peak_bytes_since_process_start_or_upstream_reset":15133588496,"preload_status":"returned","previous_free_buffer_cache_limit_bytes":48962627174},"two_phase_resource_chain":{"all_four_stage_hashes_and_sizes_match_files_summaries_and_closures":true,"all_trace_counts_and_extrema_recomputed_match_summaries":true,"continuous_monitoring_claim":false,"handover_declares_no_inference":true,"original_preup":{"free_percentage":77,"sha256":"2e48d39c2e314127a02ce20adc96c55851ab1c1518d8f3d99957b427d5bf4512","swap_used_bytes":16476798976},"phase1":{"duration_seconds":938.018743,"maximum_new_swap_growth_bytes":0,"maximum_process_rss_bytes":9721659392,"minimum_free_percentage":35,"monitor_error":null,"purpose":"preflight and semantic","samples":906,"stop_reason":"arm_work_completed"},"phase2":{"duration_seconds":859.757704,"maximum_new_swap_growth_bytes":0,"maximum_process_rss_bytes":10042966016,"minimum_free_percentage":33,"monitor_error":null,"purpose":"historical and capacity","samples":834,"stop_reason":"arm_work_completed"},"same_pid_model_process_identity_and_original_preup_baseline":true,"scope":"Two independently bounded resource-only phases, each below1800s and together below the declared3600s maximum. The handover gap is not monitoring coverage.","soak_claim":false,"sum_of_monitored_phase_seconds":1797.776447,"total_span_seconds":1837.560478,"unmonitored_gap_seconds_between_summary_intervals":39.784031}}