{
  "status": "passed",
  "created_at": "2026-09-27T10:34:26.139443+00:00",
  "runtime": {
    "python": "3.12.3",
    "platform": "Linux-4.19.0-gvisor-x86_64-with-glibc2.39",
    "gemmadecision": null,
    "granian": "2.8.3",
    "torch": "2.13.0+cu130",
    "transformers": "5.17.0",
    "vllm": "0.30.0",
    "pydantic-ai-slim": "2.51.0",
    "httpx": "0.28.1",
    "gpu_inventory": [
      "NVIDIA H100 80GB HBM3, 580.95.05, 81559 MiB"
    ]
  },
  "backend": "vllm",
  "command": [
    "python",
    "-m",
    "gemmadecision.cli",
    "serve",
    "--model-path",
    "MODEL_DIR",
    "--device",
    "cuda",
    "--port",
    "18700",
    "--backend",
    "vllm",
    "--cache-size",
    "0",
    "--batch-wait-ms",
    "1",
    "--max-batch-size",
    "32"
  ],
  "model_source": "MODEL_DIR",
  "cache_size": 0,
  "batch_wait_ms": 1,
  "max_batch_size": 32,
  "max_batch_tokens": 8192,
  "accuracy_evaluated": false,
  "authentication": "random ephemeral bearer key, never written to report",
  "measurements": [
    {
      "concurrency": 1,
      "requests": 32,
      "candidate_pairs": 128,
      "input_tokens_including_repeated_state": 5416,
      "cached_pairs": 0,
      "elapsed_s": 1.5150376290000054,
      "requests_per_s": 21.121587601174944,
      "candidate_pairs_per_s": 84.48635040469978,
      "request_latency_ms_p50": 45.64897000000201,
      "request_latency_ms_p95": 56.10913290000639,
      "request_latency_ms_min": 43.44772999999691,
      "request_latency_ms_max": 57.278682999992725,
      "model_batcher_calls": 32,
      "requests_per_batcher_call": 1.0,
      "completed_requests_delta": 32,
      "latency_scope": "HTTP SDK call, excluding client semaphore wait; includes server queue and inference"
    },
    {
      "concurrency": 8,
      "requests": 32,
      "candidate_pairs": 128,
      "input_tokens_including_repeated_state": 5416,
      "cached_pairs": 0,
      "elapsed_s": 0.45641346200000044,
      "requests_per_s": 70.11186712104467,
      "candidate_pairs_per_s": 280.44746848417867,
      "request_latency_ms_p50": 113.98175450000281,
      "request_latency_ms_p95": 124.41482229999252,
      "request_latency_ms_min": 48.02211300000181,
      "request_latency_ms_max": 126.43978300002345,
      "model_batcher_calls": 8,
      "requests_per_batcher_call": 4.0,
      "completed_requests_delta": 32,
      "latency_scope": "HTTP SDK call, excluding client semaphore wait; includes server queue and inference"
    },
    {
      "concurrency": 32,
      "requests": 32,
      "candidate_pairs": 128,
      "input_tokens_including_repeated_state": 5544,
      "cached_pairs": 0,
      "elapsed_s": 0.28632362600001215,
      "requests_per_s": 111.7616469414181,
      "candidate_pairs_per_s": 447.0465877656724,
      "request_latency_ms_p50": 276.59070299998234,
      "request_latency_ms_p95": 282.13342995000517,
      "request_latency_ms_min": 89.02169999998932,
      "request_latency_ms_max": 282.29464800000414,
      "model_batcher_calls": 2,
      "requests_per_batcher_call": 16.0,
      "completed_requests_delta": 32,
      "latency_scope": "HTTP SDK call, excluding client semaphore wait; includes server queue and inference"
    }
  ],
  "server_log": "server-results.server.log",
  "batching_note": "model_batches counts scheduler calls, not individual encoder forward chunks",
  "health_after_startup": {
    "status": "ready",
    "serving_runtime": "granian-rust",
    "model": "rajan2k/GemmaDecision-270M@785d530221c990671f29976902540101bb9c7647",
    "backend": "vllm",
    "cache_size": 0,
    "temperature": 4.136820402388508,
    "max_state_tokens": 2048,
    "max_action_tokens": 768,
    "vllm_version": "0.30.0",
    "device": "cuda",
    "dtype": "bfloat16",
    "head_device": "cpu",
    "head_dtype": "float32",
    "pooling_type": "LAST",
    "normalization": "float32_after_pooling",
    "context_limit": 3072,
    "prefix_caching": false,
    "model_revision": "785d530221c990671f29976902540101bb9c7647",
    "verified": true,
    "queued_requests": 0,
    "completed_requests": 0,
    "model_batches": 0
  },
  "startup_to_ready_s": 62.420137823000005,
  "checks": {
    "all_three_question_types": {
      "model": "rajan2k/GemmaDecision-270M@785d530221c990671f29976902540101bb9c7647",
      "answers": {
        "route": {
          "type": "choice",
          "choice": "cards",
          "probabilities": {
            "cards": 0.8216023096492325,
            "transfers": 0.02476313430266723,
            "cash": 0.026322632087462206,
            "other": 0.12731192396063812
          },
          "confidence": 0.8216023096492325,
          "scores": {
            "cards": -1.1303789615631104,
            "transfers": -15.617112159729004,
            "cash": -15.364463806152344,
            "other": -8.843961715698242
          }
        },
        "urgent": {
          "type": "noul",
          "noul": 0.2576204300029682,
          "probabilities": {
            "false": 0.7423795699970318,
            "true": 0.2576204300029682
          },
          "confidence": 0.7423795699970318,
          "scores": {
            "false": 4.544091701507568,
            "true": 0.16579118371009827
          }
        },
        "priority": {
          "type": "score",
          "score": 1.333615933281818,
          "probabilities": {
            "0": 0.002901199523831955,
            "1": 0.6605816676705178,
            "2": 0.33651713280565015
          },
          "confidence": 0.6605816676705178,
          "scores": {
            "0": -15.325399398803711,
            "1": 7.129247188568115,
            "2": 4.339078903198242
          }
        }
      },
      "usage": {
        "input_tokens": 334,
        "output_tokens": 0,
        "total_tokens": 334,
        "cached_pairs": 0
      },
      "temperature": 4.136820402388508,
      "probabilities_source": "softmax_ranking_scores_external_temperature",
      "latency_ms": 73.21720399997389
    },
    "pydantic_ai_output": {
      "route": "cards",
      "urgent": false
    },
    "pydantic_ai_output_type_valid": true,
    "rank_endpoint_valid": true,
    "missing_key_status": 401,
    "overlong_state_status": 422,
    "accuracy_evaluated": false
  },
  "warmup_requests": 3,
  "health_after_measurement": {
    "status": "ready",
    "serving_runtime": "granian-rust",
    "model": "rajan2k/GemmaDecision-270M@785d530221c990671f29976902540101bb9c7647",
    "backend": "vllm",
    "cache_size": 0,
    "temperature": 4.136820402388508,
    "max_state_tokens": 2048,
    "max_action_tokens": 768,
    "vllm_version": "0.30.0",
    "device": "cuda",
    "dtype": "bfloat16",
    "head_device": "cpu",
    "head_dtype": "float32",
    "pooling_type": "LAST",
    "normalization": "float32_after_pooling",
    "context_limit": 3072,
    "prefix_caching": false,
    "model_revision": "785d530221c990671f29976902540101bb9c7647",
    "verified": true,
    "queued_requests": 0,
    "completed_requests": 102,
    "model_batches": 49
  },
  "simple_api": {
    "passed": true,
    "first_result": "billing",
    "second_result": "technical",
    "first_call_including_load_s": 10.71267967,
    "warm_decide_call_s": 0.017516731000000618,
    "ranking": {
      "model": "rajan2k/GemmaDecision-270M@785d530221c990671f29976902540101bb9c7647",
      "ranked": [
        {
          "rank": 1,
          "candidate": "0",
          "text": "billing",
          "score": -4.178100109100342,
          "probability": 0.9804118154565832
        },
        {
          "rank": 2,
          "candidate": "1",
          "text": "technical",
          "score": -20.36566925048828,
          "probability": 0.019588184543416823
        }
      ],
      "usage": {
        "input_tokens": 32,
        "output_tokens": 0,
        "total_tokens": 32,
        "cached_pairs": 0
      },
      "temperature": 4.136820402388508,
      "probabilities_source": "softmax_ranking_scores_external_temperature",
      "latency_ms": 169.77471199999172
    },
    "engine": {
      "model": "rajan2k/GemmaDecision-270M@785d530221c990671f29976902540101bb9c7647",
      "backend": "torch",
      "cache_size": 0,
      "temperature": 4.136820402388508,
      "max_state_tokens": 2048,
      "max_action_tokens": 768,
      "device": "cuda",
      "dtype": "bfloat16",
      "head_dtype": "float32",
      "strict": false,
      "max_batch_tokens": 8192,
      "max_batch_size": 32,
      "model_revision": "785d530221c990671f29976902540101bb9c7647",
      "verified": true
    },
    "native_pydantic_ai_local": {
      "passed": true,
      "output": "billing",
      "warm_agent_call_s": 0.14082161099997847,
      "provider_url": "local://gemmadecision",
      "provider_details": {
        "confidence": {
          "response": 0.9885687139371604
        },
        "probabilities": {
          "response": {
            "billing": 0.9885687139371604,
            "technical": 0.01143128606283964
          }
        },
        "scores": {},
        "probabilities_source": "softmax_scores",
        "confidence_note": "Ranking probabilities; validate calibration on your own data."
      },
      "shared_engine_identity_unchanged": true
    },
    "accuracy_evaluated": false
  },
  "total_elapsed_s": 81.868308141
}