{ "context": { "cap_tokens": 260864, "max_model_len": 262144, "targets": [ { "attempted_context_tokens": 32768, "chars_per_token": 3.0, "clamped_tokens": 32768, "e2e_ms": 4121.553182601929, "estimated_prompt_tokens": 32768, "output_tokens": 0, "status": "passed", "target_tokens": 32768, "ttft_ms": 4121.552228927612, "usage": { "completion_tokens": 256, "prompt_tokens": 27774, "total_tokens": 28030 } }, { "attempted_context_tokens": 131072, "chars_per_token": 3.539, "clamped_tokens": 131072, "e2e_ms": 24868.23797225952, "estimated_prompt_tokens": 131071, "output_tokens": 0, "status": "passed", "target_tokens": 131072, "ttft_ms": 24868.237495422363, "usage": { "completion_tokens": 256, "prompt_tokens": 152736, "total_tokens": 152992 } }, { "attempted_context_tokens": 240000, "chars_per_token": 3.037, "clamped_tokens": 240000, "e2e_ms": 55764.29343223572, "estimated_prompt_tokens": 239999, "output_tokens": 0, "status": "passed", "target_tokens": 240000, "ttft_ms": 55764.29319381714, "usage": { "completion_tokens": 256, "prompt_tokens": 243409, "total_tokens": 243665 } } ] }, "evaluation_protocol": { "budget_semantics": "visible_answer_tokens plus reasoning_headroom_tokens are sent as one max_tokens cap; the endpoint does not hard-partition the channels", "minimum_comparable_repetitions": 3, "minimum_pass_rate": 1.0, "reasoning_headroom_tokens": 2048, "records_finish_reason": true, "records_full_visible_answer": true, "records_reasoning_channel_metadata": true, "repetitions": 3, "version": 3, "visible_answer_tokens": 512 }, "failures": [ { "error": "pass rate below threshold", "failure_classes": [ "deterministic_check_failed" ], "pass_rate": 0.0, "prompt_id": "parallel_timeout_triage", "suite": "intelligence" } ], "identity": { "base_url": "http://127.0.0.1:39038/v1", "candidate_id": "google/gemma-4-26B-A4B-it@01e5b3ee840d3a9e0b0b493c593e85398a30ef75", "config_id": "heavy-26b-bf16-256k-thinking", "engine": "vllm-0.25.1", "gpu": "RTX-PRO-6000-REDACTED_GPU_UUID_COMPUTE_A", "model": "gemma4-heavy-lab", "started_at": "2026-07-16T16:06:22Z" }, "intelligence": { "checks": [ { "attempt_count": 3, "attempts": [ { "attempt": 1, "budget": { "legacy_max_tokens_as_visible": false, "max_completion_tokens": 2560, "reasoning_headroom_tokens": 2048, "visible_answer_tokens": 512 }, "content": "```diff\n--- app.py\n+++ app.py\n@@ -1,2 +1,2 @@\n-timeout = 30\n+timeout = 45\n retries = 2\n```", "content_excerpt": "```diff\n--- app.py\n+++ app.py\n@@ -1,2 +1,2 @@\n-timeout = 30\n+timeout = 45\n retries = 2\n```", "failure_class": null, "finish_reason": "stop", "latency_ms": 1926.496982574463, "reasoning_chars": 591, "reasoning_excerpt": "* Target file: `app.py`\n * Original content:\n ```python\n timeout = 30\n retries = 2\n ```\n * Goal: Change `timeout` to `45`, leave `retries` unchanged.\n * Ou", "reasoning_field": "reasoning", "reasoning_tokens": null, "status": "passed", "text_checks": [ { "name": "diff_shape", "passed": true }, { "name": "removes_old_timeout", "passed": true }, { "name": "adds_new_timeout", "passed": true } ], "usage": { "completion_tokens": 265, "prompt_tokens": 57, "prompt_tokens_details": null, "total_tokens": 322 } }, { "attempt": 2, "budget": { "legacy_max_tokens_as_visible": false, "max_completion_tokens": 2560, "reasoning_headroom_tokens": 2048, "visible_answer_tokens": 512 }, "content": "```diff\n--- app.py\n+++ app.py\n@@ -1,2 +1,2 @@\n-timeout = 30\n+timeout = 45\n retries = 2\n```", "content_excerpt": "```diff\n--- app.py\n+++ app.py\n@@ -1,2 +1,2 @@\n-timeout = 30\n+timeout = 45\n retries = 2\n```", "failure_class": null, "finish_reason": "stop", "latency_ms": 1934.0159893035889, "reasoning_chars": 591, "reasoning_excerpt": "* Target file: `app.py`\n * Original content:\n ```python\n timeout = 30\n retries = 2\n ```\n * Goal: Change `timeout` to `45`, leave `retries` unchanged.\n * Ou", "reasoning_field": "reasoning", "reasoning_tokens": null, "status": "passed", "text_checks": [ { "name": "diff_shape", "passed": true }, { "name": "removes_old_timeout", "passed": true }, { "name": "adds_new_timeout", "passed": true } ], "usage": { "completion_tokens": 265, "prompt_tokens": 57, "prompt_tokens_details": null, "total_tokens": 322 } }, { "attempt": 3, "budget": { "legacy_max_tokens_as_visible": false, "max_completion_tokens": 2560, "reasoning_headroom_tokens": 2048, "visible_answer_tokens": 512 }, "content": "```diff\n--- app.py\n+++ app.py\n@@ -1,2 +1,2 @@\n-timeout = 30\n+timeout = 45\n retries = 2\n```", "content_excerpt": "```diff\n--- app.py\n+++ app.py\n@@ -1,2 +1,2 @@\n-timeout = 30\n+timeout = 45\n retries = 2\n```", "failure_class": null, "finish_reason": "stop", "latency_ms": 1933.1200122833252, "reasoning_chars": 591, "reasoning_excerpt": "* Target file: `app.py`\n * Original content:\n ```python\n timeout = 30\n retries = 2\n ```\n * Goal: Change `timeout` to `45`, leave `retries` unchanged.\n * Ou", "reasoning_field": "reasoning", "reasoning_tokens": null, "status": "passed", "text_checks": [ { "name": "diff_shape", "passed": true }, { "name": "removes_old_timeout", "passed": true }, { "name": "adds_new_timeout", "passed": true } ], "usage": { "completion_tokens": 265, "prompt_tokens": 57, "prompt_tokens_details": null, "total_tokens": 322 } } ], "id": "unified_diff_timeout_edit", "pass_count": 3, "pass_rate": 1.0, "required_pass_rate": 1.0, "status": "passed", "validator": "deterministic_text_checks" }, { "attempt_count": 3, "attempts": [ { "attempt": 1, "budget": { "legacy_max_tokens_as_visible": false, "max_completion_tokens": 2560, "reasoning_headroom_tokens": 2048, "visible_answer_tokens": 512 }, "content": "The total latency (3000 ms) exceeds the 2500 ms timeout due to the LLM bottleneck, which can be mitigated by implementing LLM streaming to the TTS engine.", "content_excerpt": "The total latency (3000 ms) exceeds the 2500 ms timeout due to the LLM bottleneck, which can be mitigated by implementing LLM streaming to the TTS engine.", "failure_class": "deterministic_check_failed", "finish_reason": "stop", "latency_ms": 4527.6124477386475, "reasoning_chars": 2045, "reasoning_excerpt": "\n* Components: STT (Speech-to-Text), LLM (Large Language Model), TTS (Text-to-Speech).\n* Constraint: Total turn timeout = 2500 ms.\n* Actual Latency: STT (550 ms) + LLM (1800 ms) + TTS (650 ms) =", "reasoning_field": "reasoning", "reasoning_tokens": null, "status": "failed", "text_checks": [ { "name": "identifies_budget_overrun", "passed": true }, { "name": "offers_latency_fix", "passed": false } ], "usage": { "completion_tokens": 627, "prompt_tokens": 81, "prompt_tokens_details": null, "total_tokens": 708 } }, { "attempt": 2, "budget": { "legacy_max_tokens_as_visible": false, "max_completion_tokens": 2560, "reasoning_headroom_tokens": 2048, "visible_answer_tokens": 512 }, "content": "The total latency exceeds the 2500ms timeout due to the LLM bottleneck, which can be mitigated by implementing LLM token streaming to the TTS engine.", "content_excerpt": "The total latency exceeds the 2500ms timeout due to the LLM bottleneck, which can be mitigated by implementing LLM token streaming to the TTS engine.", "failure_class": "deterministic_check_failed", "finish_reason": "stop", "latency_ms": 4130.256175994873, "reasoning_chars": 1904, "reasoning_excerpt": "\n* Components: STT (Speech-to-Text), LLM (Large Language Model), TTS (Text-to-Speech).\n* Total Turn Timeout: 2500 ms.\n* Actual Latency: 550 (STT) + 1800 (LLM) + 650 (TTS) = 3000 ms.\n* Constrai", "reasoning_field": "reasoning", "reasoning_tokens": null, "status": "failed", "text_checks": [ { "name": "identifies_budget_overrun", "passed": true }, { "name": "offers_latency_fix", "passed": false } ], "usage": { "completion_tokens": 572, "prompt_tokens": 81, "prompt_tokens_details": null, "total_tokens": 653 } }, { "attempt": 3, "budget": { "legacy_max_tokens_as_visible": false, "max_completion_tokens": 2560, "reasoning_headroom_tokens": 2048, "visible_answer_tokens": 512 }, "content": "The total latency exceeds the 2500ms timeout due to the LLM bottleneck, which can be mitigated by implementing LLM token streaming to the TTS engine.", "content_excerpt": "The total latency exceeds the 2500ms timeout due to the LLM bottleneck, which can be mitigated by implementing LLM token streaming to the TTS engine.", "failure_class": "deterministic_check_failed", "finish_reason": "stop", "latency_ms": 4140.666246414185, "reasoning_chars": 1904, "reasoning_excerpt": "\n* Components: STT (Speech-to-Text), LLM (Large Language Model), TTS (Text-to-Speech).\n* Total Turn Timeout: 2500 ms.\n* Actual Latency: 550 (STT) + 1800 (LLM) + 650 (TTS) = 3000 ms.\n* Constrai", "reasoning_field": "reasoning", "reasoning_tokens": null, "status": "failed", "text_checks": [ { "name": "identifies_budget_overrun", "passed": true }, { "name": "offers_latency_fix", "passed": false } ], "usage": { "completion_tokens": 572, "prompt_tokens": 81, "prompt_tokens_details": null, "total_tokens": 653 } } ], "error": "pass rate below threshold", "id": "parallel_timeout_triage", "pass_count": 0, "pass_rate": 0.0, "required_pass_rate": 1.0, "status": "failed", "validator": "deterministic_text_checks" } ], "status": "failed" }, "run_id": "fast-bakeoff-20260716T160622Z", "schema": "anvil-serving.fast-tier-bakeoff/v1", "score_inputs": { "e2e_p50_ms": 24868.23797225952, "intelligence_pass_rate": 0.5, "operational_fit_notes": [ "endpoint was already loaded; benchmark did not start or stop serves" ], "session_recall_passed": true, "thinking_mode": "enabled", "tool_call_passed": true, "ttft_p50_ms": 24868.237495422363, "usable_context_tokens": 240000, "voice_latency_ms": null }, "selection": { "context_targets": [ 32768, 131072, 240000 ], "endpoint_already_loaded": true, "requests_per_context": 1, "suites": [ "chat", "context", "tool", "session", "intelligence" ] }, "session": { "checks": [ { "attempt_count": 3, "attempts": [ { "attempt": 1, "budget": { "legacy_max_tokens_as_visible": false, "max_completion_tokens": 2560, "reasoning_headroom_tokens": 2048, "visible_answer_tokens": 512 }, "content": "RIVER-918", "content_excerpt": "RIVER-918", "expected": "RIVER-918", "failure_class": null, "finish_reason": "stop", "latency_ms": 793.0104732513428, "reasoning_chars": 365, "reasoning_excerpt": "The user is asking for the session code that was previously provided.\nThe previous message was: \"Remember this session code: RIVER-918. Reply with ok.\"\nThe user's current instruction is: \"What session", "reasoning_field": "reasoning", "reasoning_tokens": null, "status": "passed", "usage": { "completion_tokens": 107, "prompt_tokens": 55, "prompt_tokens_details": null, "total_tokens": 162 } }, { "attempt": 2, "budget": { "legacy_max_tokens_as_visible": false, "max_completion_tokens": 2560, "reasoning_headroom_tokens": 2048, "visible_answer_tokens": 512 }, "content": "RIVER-918", "content_excerpt": "RIVER-918", "expected": "RIVER-918", "failure_class": null, "finish_reason": "stop", "latency_ms": 793.6651706695557, "reasoning_chars": 365, "reasoning_excerpt": "The user is asking for the session code that was previously provided.\nThe previous message was: \"Remember this session code: RIVER-918. Reply with ok.\"\nThe user's current instruction is: \"What session", "reasoning_field": "reasoning", "reasoning_tokens": null, "status": "passed", "usage": { "completion_tokens": 107, "prompt_tokens": 55, "prompt_tokens_details": null, "total_tokens": 162 } }, { "attempt": 3, "budget": { "legacy_max_tokens_as_visible": false, "max_completion_tokens": 2560, "reasoning_headroom_tokens": 2048, "visible_answer_tokens": 512 }, "content": "RIVER-918", "content_excerpt": "RIVER-918", "expected": "RIVER-918", "failure_class": null, "finish_reason": "stop", "latency_ms": 790.8487319946289, "reasoning_chars": 365, "reasoning_excerpt": "The user is asking for the session code that was previously provided.\nThe previous message was: \"Remember this session code: RIVER-918. Reply with ok.\"\nThe user's current instruction is: \"What session", "reasoning_field": "reasoning", "reasoning_tokens": null, "status": "passed", "usage": { "completion_tokens": 107, "prompt_tokens": 55, "prompt_tokens_details": null, "total_tokens": 162 } } ], "name": "single_request_multiturn_recall", "pass_count": 3, "pass_rate": 1.0, "required_pass_rate": 1.0, "status": "passed" } ], "status": "passed" }, "source_recipe": { "ref": null, "serve_command": "vllm serve google/gemma-4-26B-A4B-it --revision 01e5b3ee840d3a9e0b0b493c593e85398a30ef75 --max-model-len 262144 --kv-cache-dtype fp8" }, "suites": {}, "thinking": { "chat_template_kwargs": { "enable_thinking": true }, "control_evidence": "C:\\Users\\operator\\ai-code\\anvil-serving\\docs\\findings\\2026-07-16-gemma4-chat-template-bakeoff-evidence\\heavy-26b-thinking-control.json", "control_evidence_sha256": "7206426785261e7f2da653d02afe3ae76fd8b0ce4aa0f413ea859e1e4feab643", "control_mechanism": "chat_template_kwargs", "control_requested": { "enable_thinking": true }, "control_status": "verified", "mode": "enabled", "reasoning_effort": null, "unsupported": false }, "timing": { "chat": { "e2e_p50_ms": 24868.23797225952, "e2e_p95_ms": 55764.29343223572, "output_tokens": 0, "ttft_p50_ms": 24868.237495422363, "ttft_p95_ms": 55764.29319381714 }, "wall_ms": 106373.70586395264 }, "tool": { "checks": [ { "attempt_count": 3, "attempts": [ { "arguments": { "zip": "98101" }, "attempt": 1, "budget": { "legacy_max_tokens_as_visible": false, "max_completion_tokens": 2560, "reasoning_headroom_tokens": 2048, "visible_answer_tokens": 512 }, "content": "", "content_excerpt": "", "failure_class": null, "finish_reason": "tool_calls", "latency_ms": 226.668119430542, "reasoning_chars": 0, "reasoning_excerpt": "", "reasoning_field": null, "reasoning_tokens": null, "status": "passed", "tool_call_count": 1, "usage": { "completion_tokens": 25, "prompt_tokens": 81, "prompt_tokens_details": null, "total_tokens": 106 }, "valid_tool_call_count": 1, "validation_errors": [] }, { "arguments": { "zip": "98101" }, "attempt": 2, "budget": { "legacy_max_tokens_as_visible": false, "max_completion_tokens": 2560, "reasoning_headroom_tokens": 2048, "visible_answer_tokens": 512 }, "content": "", "content_excerpt": "", "failure_class": null, "finish_reason": "tool_calls", "latency_ms": 208.74738693237305, "reasoning_chars": 0, "reasoning_excerpt": "", "reasoning_field": null, "reasoning_tokens": null, "status": "passed", "tool_call_count": 1, "usage": { "completion_tokens": 25, "prompt_tokens": 81, "prompt_tokens_details": null, "total_tokens": 106 }, "valid_tool_call_count": 1, "validation_errors": [] }, { "arguments": { "zip": "98101" }, "attempt": 3, "budget": { "legacy_max_tokens_as_visible": false, "max_completion_tokens": 2560, "reasoning_headroom_tokens": 2048, "visible_answer_tokens": 512 }, "content": "", "content_excerpt": "", "failure_class": null, "finish_reason": "tool_calls", "latency_ms": 207.31759071350098, "reasoning_chars": 0, "reasoning_excerpt": "", "reasoning_field": null, "reasoning_tokens": null, "status": "passed", "tool_call_count": 1, "usage": { "completion_tokens": 25, "prompt_tokens": 81, "prompt_tokens_details": null, "total_tokens": 106 }, "valid_tool_call_count": 1, "validation_errors": [] } ], "expected_arguments": { "zip": "98101" }, "expected_function": "record_weather_zip", "name": "openai_tool_call_smoke", "pass_count": 3, "pass_rate": 1.0, "required_pass_rate": 1.0, "status": "passed" } ], "status": "passed" }, "voice": { "llm_latency_ms": null, "status": "not_run", "stt_latency_ms": null, "total_turn_latency_ms": null, "tts_latency_ms": null } }