{
  "community": [],
  "cross_family_note": "Each board is one result-sealed bundle, shown with its interval and its provenance. Benchmark families share no denominators.",
  "gateway": {
    "bundle_count": 1,
    "bundles": [
      {
        "arms": [
          {
            "arm_id": "direct-openai",
            "baseline": true,
            "conditions": {
              "cold": {
                "availability": {
                  "attempted": 30,
                  "rate": 1.0,
                  "successes": 30,
                  "wilson95": {
                    "confidence": 0.95,
                    "high": 1.0,
                    "low": 0.886482908609522
                  }
                },
                "denominators": {
                  "attempted": 30,
                  "request_failed": 0,
                  "route_failed": 0,
                  "route_unverifiable": 0,
                  "route_verified": 30,
                  "scheduled": 30,
                  "success": 30
                },
                "metrics": {
                  "cache_write_input_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.0,
                    "p95": 0.0
                  },
                  "cached_input_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.0,
                    "p95": 0.0
                  },
                  "charged_cost_usd": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 1.395e-05,
                    "p95": 1.4399999999999998e-05
                  },
                  "cold_end_to_end_first_body_byte_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.2618867295095697,
                    "p95": 0.9161389394081192
                  },
                  "cold_end_to_end_response_headers_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.26175627097836696,
                    "p95": 0.9146265079878497
                  },
                  "cold_end_to_end_semantic_ttft_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.5556630629871506,
                    "p95": 1.384148770500905
                  },
                  "cold_end_to_end_stream_total_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.7653896670381073,
                    "p95": 1.7911753040534668
                  },
                  "input_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 37.0,
                    "p95": 39.55
                  },
                  "measured_cost_usd": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 1.395e-05,
                    "p95": 1.4399999999999998e-05
                  },
                  "output_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 14.0,
                    "p95": 14.0
                  },
                  "request_stream_total_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.7349729375273455,
                    "p95": 1.7633883562404655
                  },
                  "request_to_first_body_byte_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.2320579580264166,
                    "p95": 0.8883519915951177
                  },
                  "request_to_response_headers_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.2319540415192023,
                    "p95": 0.8868395601748482
                  },
                  "request_to_semantic_ttft_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.5268897500063758,
                    "p95": 1.3563618226879037
                  },
                  "setup_dns_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.0016002080228645355,
                    "p95": 0.002891178955906066
                  },
                  "setup_tcp_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.011436458502430469,
                    "p95": 0.012133112322771922
                  },
                  "setup_tls_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.01643262500874698,
                    "p95": 0.022621695898124017
                  },
                  "throughput_tokens_per_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 92.29271620360342,
                    "p95": 123.19624845830998
                  },
                  "total_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 51.0,
                    "p95": 54.0
                  }
                }
              },
              "warm": {
                "availability": {
                  "attempted": 30,
                  "rate": 1.0,
                  "successes": 30,
                  "wilson95": {
                    "confidence": 0.95,
                    "high": 1.0,
                    "low": 0.886482908609522
                  }
                },
                "denominators": {
                  "attempted": 30,
                  "request_failed": 0,
                  "route_failed": 0,
                  "route_unverifiable": 0,
                  "route_verified": 30,
                  "scheduled": 30,
                  "success": 30
                },
                "metrics": {
                  "cache_write_input_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.0,
                    "p95": 0.0
                  },
                  "cached_input_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.0,
                    "p95": 0.0
                  },
                  "charged_cost_usd": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 2.835e-05,
                    "p95": 3.0300000000000005e-05
                  },
                  "input_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 38.0,
                    "p95": 40.0
                  },
                  "measured_cost_usd": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 1.41e-05,
                    "p95": 1.5457499999999996e-05
                  },
                  "output_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 14.0,
                    "p95": 16.19999999999999
                  },
                  "request_stream_total_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.6562053334782831,
                    "p95": 1.2209905081253964
                  },
                  "request_to_first_body_byte_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.18428431247593835,
                    "p95": 0.43743395827186743
                  },
                  "request_to_response_headers_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.1841714994807262,
                    "p95": 0.4361857331299687
                  },
                  "request_to_semantic_ttft_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.5037482295010705,
                    "p95": 0.9772484541055746
                  },
                  "throughput_tokens_per_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 95.10098558742803,
                    "p95": 127.21719451628391
                  },
                  "total_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 52.0,
                    "p95": 54.55
                  }
                }
              }
            },
            "role": "direct"
          },
          {
            "arm_id": "cloudflare-openai",
            "baseline": false,
            "conditions": {
              "cold": {
                "availability": {
                  "attempted": 30,
                  "rate": 1.0,
                  "successes": 30,
                  "wilson95": {
                    "confidence": 0.95,
                    "high": 1.0,
                    "low": 0.886482908609522
                  }
                },
                "denominators": {
                  "attempted": 30,
                  "request_failed": 0,
                  "route_failed": 0,
                  "route_unverifiable": 0,
                  "route_verified": 30,
                  "scheduled": 30,
                  "success": 30
                },
                "metrics": {
                  "cache_write_input_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.0,
                    "p95": 0.0
                  },
                  "cached_input_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.0,
                    "p95": 0.0
                  },
                  "charged_cost_usd": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 1.395e-05,
                    "p95": 1.4332499999999999e-05
                  },
                  "cold_end_to_end_first_body_byte_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.3778885630017612,
                    "p95": 0.8001176937425039
                  },
                  "cold_end_to_end_response_headers_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.37778033351060003,
                    "p95": 0.7999479419959239
                  },
                  "cold_end_to_end_semantic_ttft_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.684436937503051,
                    "p95": 1.0906790498644112
                  },
                  "cold_end_to_end_stream_total_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.8488696874992456,
                    "p95": 1.247032909898553
                  },
                  "input_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 37.0,
                    "p95": 39.55
                  },
                  "measured_cost_usd": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 1.395e-05,
                    "p95": 1.4332499999999999e-05
                  },
                  "output_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 14.0,
                    "p95": 14.0
                  },
                  "request_stream_total_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.8121094164962415,
                    "p95": 1.2154791437787935
                  },
                  "request_to_first_body_byte_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.34092250050161965,
                    "p95": 0.7672035979398056
                  },
                  "request_to_response_headers_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.34081427101045847,
                    "p95": 0.7670338461932256
                  },
                  "request_to_semantic_ttft_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.6555284794885665,
                    "p95": 1.054727060740697
                  },
                  "setup_dns_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.0013577920035459101,
                    "p95": 0.0017405648017302154
                  },
                  "setup_tcp_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.01132097901427187,
                    "p95": 0.0123495559731964
                  },
                  "setup_tls_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.016649396013235673,
                    "p95": 0.0229986027610721
                  },
                  "throughput_tokens_per_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 84.58266220944367,
                    "p95": 113.73809479723356
                  },
                  "total_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 51.0,
                    "p95": 53.55
                  }
                }
              },
              "warm": {
                "availability": {
                  "attempted": 30,
                  "rate": 1.0,
                  "successes": 30,
                  "wilson95": {
                    "confidence": 0.95,
                    "high": 1.0,
                    "low": 0.886482908609522
                  }
                },
                "denominators": {
                  "attempted": 30,
                  "request_failed": 0,
                  "route_failed": 0,
                  "route_unverifiable": 0,
                  "route_verified": 30,
                  "scheduled": 30,
                  "success": 30
                },
                "metrics": {
                  "cache_write_input_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.0,
                    "p95": 0.0
                  },
                  "cached_input_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.0,
                    "p95": 0.0
                  },
                  "charged_cost_usd": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 2.835e-05,
                    "p95": 3.0300000000000005e-05
                  },
                  "input_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 38.0,
                    "p95": 40.0
                  },
                  "measured_cost_usd": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 1.41e-05,
                    "p95": 1.5457499999999996e-05
                  },
                  "output_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 14.0,
                    "p95": 16.19999999999999
                  },
                  "request_stream_total_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.7536033329670317,
                    "p95": 0.8933024374855449
                  },
                  "request_to_first_body_byte_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.26018185398424976,
                    "p95": 0.4104486996948252
                  },
                  "request_to_response_headers_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.2600222500041127,
                    "p95": 0.41031452936585977
                  },
                  "request_to_semantic_ttft_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.5606690415006597,
                    "p95": 0.739080439400277
                  },
                  "throughput_tokens_per_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 96.57889187473157,
                    "p95": 126.43455910330448
                  },
                  "total_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 52.0,
                    "p95": 54.55
                  }
                }
              }
            },
            "role": "gateway"
          },
          {
            "arm_id": "concentrate-openai",
            "baseline": false,
            "conditions": {
              "cold": {
                "availability": {
                  "attempted": 30,
                  "rate": 1.0,
                  "successes": 30,
                  "wilson95": {
                    "confidence": 0.95,
                    "high": 1.0,
                    "low": 0.886482908609522
                  }
                },
                "denominators": {
                  "attempted": 30,
                  "request_failed": 0,
                  "route_failed": 0,
                  "route_unverifiable": 0,
                  "route_verified": 30,
                  "scheduled": 30,
                  "success": 30
                },
                "metrics": {
                  "cache_write_input_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.0,
                    "p95": 0.0
                  },
                  "cached_input_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.0,
                    "p95": 0.0
                  },
                  "charged_cost_usd": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 1.395e-05,
                    "p95": 1.4399999999999998e-05
                  },
                  "cold_end_to_end_first_body_byte_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.4355404380185064,
                    "p95": 0.5690667329850838
                  },
                  "cold_end_to_end_response_headers_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.4354242920235265,
                    "p95": 0.5689299000136088
                  },
                  "cold_end_to_end_semantic_ttft_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.793651020998368,
                    "p95": 0.9779021225287579
                  },
                  "cold_end_to_end_stream_total_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.9667132294853218,
                    "p95": 1.8331581669684933
                  },
                  "input_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 37.0,
                    "p95": 39.55
                  },
                  "measured_cost_usd": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 1.395e-05,
                    "p95": 1.4399999999999998e-05
                  },
                  "output_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 14.0,
                    "p95": 14.0
                  },
                  "request_stream_total_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.9297308755049016,
                    "p95": 1.8025811024388503
                  },
                  "request_to_first_body_byte_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.40371929202228785,
                    "p95": 0.5377796893997584
                  },
                  "request_to_response_headers_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.4036018130136654,
                    "p95": 0.5376428564282832
                  },
                  "request_to_semantic_ttft_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.7626268334861379,
                    "p95": 0.9463682267582043
                  },
                  "setup_dns_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.0012215210008434951,
                    "p95": 0.0016650413163006301
                  },
                  "setup_tcp_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.01148339599603787,
                    "p95": 0.014478962498833422
                  },
                  "setup_tls_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.017997916496824473,
                    "p95": 0.02259496236511039
                  },
                  "throughput_tokens_per_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 84.86000426935587,
                    "p95": 122.97578848982381
                  },
                  "total_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 51.0,
                    "p95": 54.0
                  }
                }
              },
              "warm": {
                "availability": {
                  "attempted": 30,
                  "rate": 1.0,
                  "successes": 30,
                  "wilson95": {
                    "confidence": 0.95,
                    "high": 1.0,
                    "low": 0.886482908609522
                  }
                },
                "denominators": {
                  "attempted": 30,
                  "request_failed": 0,
                  "route_failed": 0,
                  "route_unverifiable": 0,
                  "route_verified": 30,
                  "scheduled": 30,
                  "success": 30
                },
                "metrics": {
                  "cache_write_input_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.0,
                    "p95": 0.0
                  },
                  "cached_input_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.0,
                    "p95": 0.0
                  },
                  "charged_cost_usd": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 2.835e-05,
                    "p95": 3.0300000000000005e-05
                  },
                  "input_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 38.0,
                    "p95": 40.0
                  },
                  "measured_cost_usd": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 1.41e-05,
                    "p95": 1.5457499999999996e-05
                  },
                  "output_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 14.0,
                    "p95": 16.19999999999999
                  },
                  "request_stream_total_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.8962873334821779,
                    "p95": 1.0960268963506674
                  },
                  "request_to_first_body_byte_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.36147922949749045,
                    "p95": 0.43618071058008334
                  },
                  "request_to_response_headers_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.3613480414787773,
                    "p95": 0.43604239198320993
                  },
                  "request_to_semantic_ttft_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.7036326875095256,
                    "p95": 0.9330836377979717
                  },
                  "throughput_tokens_per_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 85.26395326652326,
                    "p95": 131.6394817915579
                  },
                  "total_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 52.0,
                    "p95": 54.55
                  }
                }
              }
            },
            "role": "gateway"
          },
          {
            "arm_id": "openrouter-openai",
            "baseline": false,
            "conditions": {
              "cold": {
                "availability": {
                  "attempted": 30,
                  "rate": 1.0,
                  "successes": 30,
                  "wilson95": {
                    "confidence": 0.95,
                    "high": 1.0,
                    "low": 0.886482908609522
                  }
                },
                "denominators": {
                  "attempted": 30,
                  "request_failed": 0,
                  "route_failed": 0,
                  "route_unverifiable": 0,
                  "route_verified": 30,
                  "scheduled": 30,
                  "success": 30
                },
                "metrics": {
                  "cache_write_input_tokens": {
                    "coverage": {
                      "covered": 0,
                      "ratio": 0.0,
                      "total": 30
                    },
                    "p50": null,
                    "p95": null
                  },
                  "cached_input_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.0,
                    "p95": 0.0
                  },
                  "charged_cost_usd": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 1.575e-05,
                    "p95": 1.61325e-05
                  },
                  "cold_end_to_end_first_body_byte_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.499478020996321,
                    "p95": 0.6956120435992488
                  },
                  "cold_end_to_end_response_headers_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.48873749951599166,
                    "p95": 0.6954786559566855
                  },
                  "cold_end_to_end_semantic_ttft_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.5005642494943459,
                    "p95": 0.7029175142524763
                  },
                  "cold_end_to_end_stream_total_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.6901099374808837,
                    "p95": 0.8744508600619156
                  },
                  "input_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 37.0,
                    "p95": 39.55
                  },
                  "measured_cost_usd": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 1.575e-05,
                    "p95": 1.61325e-05
                  },
                  "output_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 17.0,
                    "p95": 17.0
                  },
                  "request_stream_total_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.6617980415176135,
                    "p95": 0.8448033626715187
                  },
                  "request_to_first_body_byte_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.46975895849755034,
                    "p95": 0.667759970485349
                  },
                  "request_to_response_headers_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.459018437017221,
                    "p95": 0.6676265828427856
                  },
                  "request_to_semantic_ttft_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.4708451869955752,
                    "p95": 0.6750654411385766
                  },
                  "setup_dns_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.0011508960160426795,
                    "p95": 0.0015771440521348263
                  },
                  "setup_tcp_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.0111028749961406,
                    "p95": 0.011712333545438013
                  },
                  "setup_tls_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.015560562023892999,
                    "p95": 0.020372152770869435
                  },
                  "throughput_tokens_per_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 100.57065537140309,
                    "p95": 156.85007680269499
                  },
                  "total_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 54.0,
                    "p95": 56.55
                  }
                }
              },
              "warm": {
                "availability": {
                  "attempted": 30,
                  "rate": 1.0,
                  "successes": 30,
                  "wilson95": {
                    "confidence": 0.95,
                    "high": 1.0,
                    "low": 0.886482908609522
                  }
                },
                "denominators": {
                  "attempted": 30,
                  "request_failed": 0,
                  "route_failed": 0,
                  "route_unverifiable": 0,
                  "route_verified": 30,
                  "scheduled": 30,
                  "success": 30
                },
                "metrics": {
                  "cache_write_input_tokens": {
                    "coverage": {
                      "covered": 0,
                      "ratio": 0.0,
                      "total": 30
                    },
                    "p50": null,
                    "p95": null
                  },
                  "cached_input_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.0,
                    "p95": 0.0
                  },
                  "charged_cost_usd": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 3.135e-05,
                    "p95": 3.24e-05
                  },
                  "input_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 38.0,
                    "p95": 40.0
                  },
                  "measured_cost_usd": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 1.56e-05,
                    "p95": 1.62e-05
                  },
                  "output_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 17.0,
                    "p95": 17.0
                  },
                  "request_stream_total_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.5871571250027046,
                    "p95": 0.7637681001157034
                  },
                  "request_to_first_body_byte_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.43211764548323117,
                    "p95": 0.6315843268821475
                  },
                  "request_to_response_headers_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.4233607085188851,
                    "p95": 0.4849301542155444
                  },
                  "request_to_semantic_ttft_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.4328554795065429,
                    "p95": 0.6547882412996835
                  },
                  "throughput_tokens_per_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 105.94427864870052,
                    "p95": 415.5487273251547
                  },
                  "total_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 54.5,
                    "p95": 57.0
                  }
                }
              }
            },
            "role": "gateway"
          },
          {
            "arm_id": "vercel-openai",
            "baseline": false,
            "conditions": {
              "cold": {
                "availability": {
                  "attempted": 30,
                  "rate": 1.0,
                  "successes": 30,
                  "wilson95": {
                    "confidence": 0.95,
                    "high": 1.0,
                    "low": 0.886482908609522
                  }
                },
                "denominators": {
                  "attempted": 30,
                  "request_failed": 0,
                  "route_failed": 0,
                  "route_unverifiable": 0,
                  "route_verified": 30,
                  "scheduled": 30,
                  "success": 30
                },
                "metrics": {
                  "cache_write_input_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.0,
                    "p95": 0.0
                  },
                  "cached_input_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.0,
                    "p95": 0.0
                  },
                  "charged_cost_usd": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 1.395e-05,
                    "p95": 1.4399999999999998e-05
                  },
                  "cold_end_to_end_first_body_byte_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.8351534999965224,
                    "p95": 2.5231243521120583
                  },
                  "cold_end_to_end_response_headers_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.8292382500076201,
                    "p95": 2.5188931835087667
                  },
                  "cold_end_to_end_semantic_ttft_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.8415064374858048,
                    "p95": 2.5277202040539075
                  },
                  "cold_end_to_end_stream_total_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 1.004722916521132,
                    "p95": 2.752883272795584
                  },
                  "input_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 37.0,
                    "p95": 39.55
                  },
                  "measured_cost_usd": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 1.395e-05,
                    "p95": 1.4399999999999998e-05
                  },
                  "output_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 14.0,
                    "p95": 14.0
                  },
                  "request_stream_total_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.9506968329951633,
                    "p95": 2.719361543457485
                  },
                  "request_to_first_body_byte_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.7828884375048801,
                    "p95": 2.489602622773959
                  },
                  "request_to_response_headers_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.7733296249934938,
                    "p95": 2.4853714541706675
                  },
                  "request_to_semantic_ttft_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.7888890624744818,
                    "p95": 2.4941984747158084
                  },
                  "setup_dns_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.001817666518036276,
                    "p95": 0.14762862310744762
                  },
                  "setup_tcp_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.005584062513662502,
                    "p95": 0.006914308253908529
                  },
                  "setup_tls_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.024053103988990188,
                    "p95": 0.028209633677033705
                  },
                  "throughput_tokens_per_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 87.90931630091568,
                    "p95": 345.2667300252649
                  },
                  "total_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 51.0,
                    "p95": 54.0
                  }
                }
              },
              "warm": {
                "availability": {
                  "attempted": 30,
                  "rate": 1.0,
                  "successes": 30,
                  "wilson95": {
                    "confidence": 0.95,
                    "high": 1.0,
                    "low": 0.886482908609522
                  }
                },
                "denominators": {
                  "attempted": 30,
                  "request_failed": 0,
                  "route_failed": 0,
                  "route_unverifiable": 0,
                  "route_verified": 30,
                  "scheduled": 30,
                  "success": 30
                },
                "metrics": {
                  "cache_write_input_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.0,
                    "p95": 0.0
                  },
                  "cached_input_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.0,
                    "p95": 0.0
                  },
                  "charged_cost_usd": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 2.835e-05,
                    "p95": 3.0382500000000003e-05
                  },
                  "input_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 38.0,
                    "p95": 40.0
                  },
                  "measured_cost_usd": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 1.4175e-05,
                    "p95": 1.62825e-05
                  },
                  "output_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 14.0,
                    "p95": 18.0
                  },
                  "request_stream_total_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.7264135834993795,
                    "p95": 2.4583423890871896
                  },
                  "request_to_first_body_byte_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.5721195829974022,
                    "p95": 1.6005666395387363
                  },
                  "request_to_response_headers_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.5588160619954579,
                    "p95": 1.5710355376475476
                  },
                  "request_to_semantic_ttft_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 0.5721195829974022,
                    "p95": 1.6050971270393333
                  },
                  "throughput_tokens_per_s": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 94.37047761218236,
                    "p95": 449.19378363279077
                  },
                  "total_tokens": {
                    "coverage": {
                      "covered": 30,
                      "ratio": 1.0,
                      "total": 30
                    },
                    "p50": 52.5,
                    "p95": 55.0
                  }
                }
              }
            },
            "role": "gateway"
          }
        ],
        "baseline_arm_id": "direct-openai",
        "complete_blocks": {
          "cold": 30,
          "warm": 30
        },
        "contrasts": [
          {
            "arm_id": "cloudflare-openai",
            "conditions": {
              "cold": {
                "cache_write_input_tokens": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                },
                "cached_input_tokens": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                },
                "charged_cost_usd": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                },
                "cold_end_to_end_first_body_byte_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.12170043648802675,
                    "low": 0.06726702000014484,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.09042116702767089
                },
                "cold_end_to_end_response_headers_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.1216524159826804,
                    "low": 0.06718133254471469,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.09037456254009157
                },
                "cold_end_to_end_semantic_ttft_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.1595789585262537,
                    "low": 0.007124416501028463,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.09380425000563264
                },
                "cold_end_to_end_stream_total_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.18731377049698494,
                    "low": 0.011488624528283253,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.11504197953036055
                },
                "input_tokens": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                },
                "measured_cost_usd": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                },
                "output_tokens": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                },
                "request_stream_total_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.1890914993367912,
                    "low": 0.015864896005950868,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.11766906300908886
                },
                "request_to_first_body_byte_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.11944389549898915,
                    "low": 0.07481614596326835,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0899678754794877
                },
                "request_to_response_headers_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.12039591697975993,
                    "low": 0.07240795850520954,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.08992870850488544
                },
                "request_to_semantic_ttft_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.1626327504927758,
                    "low": 0.011061979486839846,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.09514824996585958
                },
                "setup_dns_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": -0.0001253119553439319,
                    "low": -0.0003124999930150807,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": -0.00027477103867568076
                },
                "setup_tcp_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.00021356251090765,
                    "low": -0.0003019794530700892,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": -1.8708029529079795e-05
                },
                "setup_tls_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0019422090263105929,
                    "low": -0.0012832079664804041,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.00019310397328808904
                },
                "throughput_tokens_per_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 20.860273527171852,
                    "low": -20.957597029407538,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 1.9003459749691913
                },
                "total_tokens": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                }
              },
              "warm": {
                "cache_write_input_tokens": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                },
                "cached_input_tokens": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                },
                "charged_cost_usd": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                },
                "input_tokens": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                },
                "measured_cost_usd": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                },
                "output_tokens": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                },
                "request_stream_total_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.12259283295134082,
                    "low": -0.012794332986231893,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.05114497951581143
                },
                "request_to_first_body_byte_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.11200035398360342,
                    "low": 0.051706875558011234,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.07910862501012161
                },
                "request_to_response_headers_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.11193162450217642,
                    "low": 0.04981233406579122,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.08056397899053991
                },
                "request_to_semantic_ttft_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.11582944041074303,
                    "low": 0.0025938330218195915,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.037403958005597815
                },
                "throughput_tokens_per_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 24.219404585049492,
                    "low": -20.987276097366294,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": -0.2077389019195195
                },
                "total_tokens": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                }
              }
            },
            "direct_arm": "direct-openai"
          },
          {
            "arm_id": "concentrate-openai",
            "conditions": {
              "cold": {
                "cache_write_input_tokens": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                },
                "cached_input_tokens": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                },
                "charged_cost_usd": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                },
                "cold_end_to_end_first_body_byte_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.19392779149347916,
                    "low": 0.14470543750212528,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.16657393801142462
                },
                "cold_end_to_end_response_headers_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.18906864503514953,
                    "low": 0.1446818329859525,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.1665636670368258
                },
                "cold_end_to_end_semantic_ttft_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.23637485451763496,
                    "low": 0.08468770905165002,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.19116072947508655
                },
                "cold_end_to_end_stream_total_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.2920525830122642,
                    "low": 0.05204991699429229,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.19803156197303906
                },
                "input_tokens": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                },
                "measured_cost_usd": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                },
                "output_tokens": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                },
                "request_stream_total_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.27374437503749505,
                    "low": 0.06171727098990232,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.19892531199730001
                },
                "request_to_first_body_byte_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.1927400419372134,
                    "low": 0.14380220850580372,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.16858725002384745
                },
                "request_to_response_headers_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.1900320767555964,
                    "low": 0.1426413330482319,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.16857697904924862
                },
                "request_to_semantic_ttft_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.23524766700575128,
                    "low": 0.0965253540489357,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.19317883349140175
                },
                "setup_dns_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": -0.00024479100829921663,
                    "low": -0.000430583517299965,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": -0.00032902153907343745
                },
                "setup_tcp_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0005222299951128662,
                    "low": -0.00036754150642082095,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.000377853459212929
                },
                "setup_tls_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0027130624803248793,
                    "low": -0.00022147901472635567,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0011074999929405749
                },
                "throughput_tokens_per_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 27.941385781837674,
                    "low": -32.401504564727,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 3.817695001056485
                },
                "total_tokens": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                }
              },
              "warm": {
                "cache_write_input_tokens": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                },
                "cached_input_tokens": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                },
                "charged_cost_usd": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                },
                "input_tokens": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                },
                "measured_cost_usd": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                },
                "output_tokens": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                },
                "request_stream_total_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.2706499999621883,
                    "low": 0.156668666517362,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.20070966650382616
                },
                "request_to_first_body_byte_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.20377760447445326,
                    "low": 0.14063983474479755,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.16801072948146611
                },
                "request_to_response_headers_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.20367116702254862,
                    "low": 0.14306247950298712,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.16795933348475955
                },
                "request_to_semantic_ttft_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.23001316701993346,
                    "low": 0.1422742500435561,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.20233466601348482
                },
                "throughput_tokens_per_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 7.422840232342814,
                    "low": -15.673043016897921,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": -5.29159021579887
                },
                "total_tokens": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                }
              }
            },
            "direct_arm": "direct-openai"
          },
          {
            "arm_id": "openrouter-openai",
            "conditions": {
              "cold": {
                "cache_write_input_tokens": {
                  "coverage": {
                    "covered": 0,
                    "ratio": 0.0,
                    "total": 30
                  },
                  "interval": null,
                  "median_gateway_minus_direct": null
                },
                "cached_input_tokens": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                },
                "charged_cost_usd": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 1.7999999999999997e-06,
                    "low": 1.7999999999999997e-06,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 1.7999999999999997e-06
                },
                "cold_end_to_end_first_body_byte_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.2527823124837596,
                    "low": 0.2226975834928453,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.2416072295163758
                },
                "cold_end_to_end_response_headers_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.24588058301014826,
                    "low": 0.20653462497284636,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.23038235399872065
                },
                "cold_end_to_end_semantic_ttft_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": -0.009755562525242567,
                    "low": -0.17344358400441706,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": -0.08814597950549796
                },
                "cold_end_to_end_stream_total_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.002021333493757993,
                    "low": -0.2061335839680396,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": -0.09709120853221975
                },
                "input_tokens": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                },
                "measured_cost_usd": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 1.7999999999999997e-06,
                    "low": 1.7999999999999997e-06,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 1.7999999999999997e-06
                },
                "output_tokens": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 3.0,
                    "low": 3.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 3.0
                },
                "request_stream_total_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.014045981059461138,
                    "low": -0.17590204198495485,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": -0.0945279790321365
                },
                "request_to_first_body_byte_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.2520663960312959,
                    "low": 0.2309046603484603,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.24342847950174473
                },
                "request_to_response_headers_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.24409531254786998,
                    "low": 0.2121493960148655,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.23311016647494398
                },
                "request_to_semantic_ttft_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": -0.00858166697435081,
                    "low": -0.17558802096755244,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": -0.08753445904585533
                },
                "setup_dns_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": -0.0002869585296139121,
                    "low": -0.0005898955278098583,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": -0.00044333349796943367
                },
                "setup_tcp_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": -8.949998300522566e-05,
                    "low": -0.0006410834612324834,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": -0.0003782699932344258
                },
                "setup_tls_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.00126479149912484,
                    "low": -0.0020909784943796694,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": -0.0005678745219483972
                },
                "throughput_tokens_per_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 47.94634096848077,
                    "low": -2.0912354822923547,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 27.41498365396285
                },
                "total_tokens": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 3.0,
                    "low": 3.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 3.0
                }
              },
              "warm": {
                "cache_write_input_tokens": {
                  "coverage": {
                    "covered": 0,
                    "ratio": 0.0,
                    "total": 30
                  },
                  "interval": null,
                  "median_gateway_minus_direct": null
                },
                "cached_input_tokens": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                },
                "charged_cost_usd": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 3.5999999999999994e-06,
                    "low": 2.6999999999999996e-06,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 3.0000000000000018e-06
                },
                "input_tokens": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                },
                "measured_cost_usd": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 1.7999999999999997e-06,
                    "low": 1.2000000000000004e-06,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 1.7999999999999997e-06
                },
                "output_tokens": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 3.0,
                    "low": 2.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 3.0
                },
                "request_stream_total_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.005193853547098115,
                    "low": -0.12351249950006604,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": -0.055522812996059656
                },
                "request_to_first_body_byte_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.2694512909511104,
                    "low": 0.2287311260588467,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.2501706040056888
                },
                "request_to_response_headers_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.25157781250891276,
                    "low": 0.2024219169979915,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.2296486044942867
                },
                "request_to_semantic_ttft_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": -0.018306938029127195,
                    "low": -0.1089712499815505,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": -0.054442000488052145
                },
                "throughput_tokens_per_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 27.1157458830417,
                    "low": 4.206847860683574,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 16.020852197066944
                },
                "total_tokens": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 3.0,
                    "low": 2.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 3.0
                }
              }
            },
            "direct_arm": "direct-openai"
          },
          {
            "arm_id": "vercel-openai",
            "conditions": {
              "cold": {
                "cache_write_input_tokens": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                },
                "cached_input_tokens": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                },
                "charged_cost_usd": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                },
                "cold_end_to_end_first_body_byte_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.6373836660059169,
                    "low": 0.4620176244643517,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.5194777495053131
                },
                "cold_end_to_end_response_headers_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.6328783954959363,
                    "low": 0.4505084067335702,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.5110465420002583
                },
                "cold_end_to_end_semantic_ttft_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.29971533350180835,
                    "low": 0.17978658198262565,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.2288713965099305
                },
                "cold_end_to_end_stream_total_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.3379475625115447,
                    "low": 0.16034895749180578,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.238228269998217
                },
                "input_tokens": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                },
                "measured_cost_usd": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                },
                "output_tokens": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                },
                "request_stream_total_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.29213631301536225,
                    "low": 0.14674216747516766,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.23466897898470052
                },
                "request_to_first_body_byte_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.6006696879921947,
                    "low": 0.45213562549906783,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.5163948324916419
                },
                "request_to_response_headers_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.5961508960172068,
                    "low": 0.43974770803470165,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.5079636249865871
                },
                "request_to_semantic_ttft_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.2967062824573074,
                    "low": 0.17082722901250236,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.22735554198152386
                },
                "setup_dns_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.00042562547605484724,
                    "low": -5.208348738960922e-05,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.00022552150767296553
                },
                "setup_tcp_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": -0.005180478998227045,
                    "low": -0.006166687497170642,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": -0.005376624991185963
                },
                "setup_tls_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.009415479988092557,
                    "low": 0.005901666474528611,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.008299104461912066
                },
                "throughput_tokens_per_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 32.281512534957905,
                    "low": -8.347501492040571,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 9.449165425328447
                },
                "total_tokens": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                }
              },
              "warm": {
                "cache_write_input_tokens": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                },
                "cached_input_tokens": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                },
                "charged_cost_usd": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                },
                "input_tokens": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                },
                "measured_cost_usd": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                },
                "output_tokens": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                },
                "request_stream_total_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.22041020850883797,
                    "low": -0.04030391649575904,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.07145018698065542
                },
                "request_to_first_body_byte_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.4653479169937782,
                    "low": 0.33173170901136473,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.36314587452216074
                },
                "request_to_response_headers_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.46518966602161527,
                    "low": 0.32805150002241135,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.3480733534961473
                },
                "request_to_semantic_ttft_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.21406552052940242,
                    "low": 0.013774916500551626,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.05843802049639635
                },
                "throughput_tokens_per_s": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 31.988273684160433,
                    "low": -12.96412002218355,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 10.75924396155878
                },
                "total_tokens": {
                  "coverage": {
                    "covered": 30,
                    "ratio": 1.0,
                    "total": 30
                  },
                  "interval": {
                    "confidence": 0.95,
                    "high": 0.0,
                    "low": 0.0,
                    "replicates": 2000
                  },
                  "median_gateway_minus_direct": 0.0
                }
              }
            },
            "direct_arm": "direct-openai"
          }
        ],
        "date": "2026-07-27",
        "experiment_digest": "e3649dbc6dc423c5209cdc09c73f0d1275bfa6c231a5fa8300e444dc89e6f943",
        "experiment_id": "gpt-4o-mini-five-way-managed-probe-r30-2026-07-27",
        "family": "gateway_probe",
        "id": "2026-07-27-gpt4o-mini-managed-30",
        "kind": "release",
        "link": null,
        "model_match": "rolling_alias",
        "path": null,
        "price_digest": "56ef4049554cc2dc3cb79debc2f5c49d503cfa97c5744d35ffe912815179aea5",
        "result_count": 300,
        "results_path": "gateway-probe/2026-07-27-gpt4o-mini-managed-30/results.jsonl",
        "schedule_digest": "e6b8e2ac79abf1e3b6c6b4ce6b691747a1f9f6bf76c3c78bdb00051e99c097be",
        "scheduled_blocks_per_condition": 30,
        "submitter": "OpenBench",
        "title": "GPT-4o mini managed gateway request benchmark",
        "track": "request_probe",
        "verified_with_commit": "2cd722da4d657d1a7d7aeeee7e92de1f7e0261ff"
      }
    ],
    "note": "Gateway Bench measures request-level transport and serving telemetry. Cold and warm denominators are separate and are never merged.",
    "skipped": []
  },
  "generated_by": "obench site",
  "harness": {
    "bundle_count": 5,
    "bundles": [
      {
        "also_seen_as": [],
        "arms": [
          {
            "cost_per_solve_usd": null,
            "fresh_tokens_per_solve": 5518.583333333333,
            "harness": "cursor",
            "median_wall_s": 42.908500000000004,
            "model": "gpt-5.6-sol",
            "n": 42,
            "solve_rate": 0.8571428571428571,
            "solved": 36,
            "token_telemetry_bases": [
              "harness_reported"
            ],
            "token_telemetry_coverage": {
              "covered_rows": 42,
              "native_covered_rows": 42,
              "proxy_covered_rows": 0,
              "ratio": 1.0,
              "total_rows": 42
            },
            "token_telemetry_source": "native",
            "tokens_cache_read_per_solve": 407067.22222222225,
            "tokens_cache_write_per_solve": 45776.333333333336,
            "tokens_input_uncached_per_solve": 72.13888888888889,
            "tokens_output_per_solve": 5446.444444444444,
            "wilson95": [
              0.7215861764782122,
              0.9328412431581007
            ]
          },
          {
            "cost_per_solve_usd": null,
            "fresh_tokens_per_solve": null,
            "harness": "devin",
            "median_wall_s": 68.4285,
            "model": "gpt-5.6-sol",
            "n": 42,
            "solve_rate": 0.8095238095238095,
            "solved": 34,
            "token_telemetry_bases": [],
            "token_telemetry_coverage": {
              "covered_rows": 0,
              "native_covered_rows": 0,
              "proxy_covered_rows": 0,
              "ratio": 0.0,
              "total_rows": 42
            },
            "token_telemetry_source": null,
            "tokens_cache_read_per_solve": null,
            "tokens_cache_write_per_solve": null,
            "tokens_input_uncached_per_solve": null,
            "tokens_output_per_solve": null,
            "wilson95": [
              0.6669893264500356,
              0.9001811039014356
            ]
          },
          {
            "cost_per_solve_usd": null,
            "fresh_tokens_per_solve": 46215.58823529412,
            "harness": "grokbuild",
            "median_wall_s": 86.2995,
            "model": "gpt-5.6-sol",
            "n": 42,
            "solve_rate": 0.8095238095238095,
            "solved": 34,
            "token_telemetry_bases": [
              "proxy_measured"
            ],
            "token_telemetry_coverage": {
              "covered_rows": 42,
              "native_covered_rows": 0,
              "proxy_covered_rows": 42,
              "ratio": 1.0,
              "total_rows": 42
            },
            "token_telemetry_source": "proxy",
            "tokens_cache_read_per_solve": 368079.0588235294,
            "tokens_cache_write_per_solve": 0.0,
            "tokens_input_uncached_per_solve": 38726.82352941176,
            "tokens_output_per_solve": 7488.764705882353,
            "wilson95": [
              0.6669893264500356,
              0.9001811039014356
            ]
          },
          {
            "cost_per_solve_usd": null,
            "fresh_tokens_per_solve": 42634.5,
            "harness": "opencode",
            "median_wall_s": 60.524,
            "model": "gpt-5.6-sol",
            "n": 42,
            "solve_rate": 0.8095238095238095,
            "solved": 34,
            "token_telemetry_bases": [
              "vendor_split"
            ],
            "token_telemetry_coverage": {
              "covered_rows": 42,
              "native_covered_rows": 42,
              "proxy_covered_rows": 0,
              "ratio": 1.0,
              "total_rows": 42
            },
            "token_telemetry_source": "native",
            "tokens_cache_read_per_solve": 313916.23529411765,
            "tokens_cache_write_per_solve": 0.0,
            "tokens_input_uncached_per_solve": 36707.55882352941,
            "tokens_output_per_solve": 5926.941176470588,
            "wilson95": [
              0.6669893264500356,
              0.9001811039014356
            ]
          },
          {
            "cost_per_solve_usd": null,
            "fresh_tokens_per_solve": 40255.0,
            "harness": "claude",
            "median_wall_s": 58.793,
            "model": "gpt-5.6-sol",
            "n": 42,
            "solve_rate": 0.7619047619047619,
            "solved": 32,
            "token_telemetry_bases": [
              "proxy_measured"
            ],
            "token_telemetry_coverage": {
              "covered_rows": 42,
              "native_covered_rows": 42,
              "proxy_covered_rows": 42,
              "ratio": 1.0,
              "total_rows": 42
            },
            "token_telemetry_source": "proxy",
            "tokens_cache_read_per_solve": 153488.0,
            "tokens_cache_write_per_solve": 0.0,
            "tokens_input_uncached_per_solve": 34455.125,
            "tokens_output_per_solve": 5799.875,
            "wilson95": [
              0.6147216491768112,
              0.8651917918898182
            ]
          },
          {
            "cost_per_solve_usd": null,
            "fresh_tokens_per_solve": 37602.0625,
            "harness": "pi",
            "median_wall_s": 39.670500000000004,
            "model": "gpt-5.6-sol",
            "n": 42,
            "solve_rate": 0.7619047619047619,
            "solved": 32,
            "token_telemetry_bases": [
              "proxy_measured"
            ],
            "token_telemetry_coverage": {
              "covered_rows": 42,
              "native_covered_rows": 42,
              "proxy_covered_rows": 42,
              "ratio": 1.0,
              "total_rows": 42
            },
            "token_telemetry_source": "proxy",
            "tokens_cache_read_per_solve": 111408.0,
            "tokens_cache_write_per_solve": 0.0,
            "tokens_input_uncached_per_solve": 32915.71875,
            "tokens_output_per_solve": 4686.34375,
            "wilson95": [
              0.6147216491768112,
              0.8651917918898182
            ]
          },
          {
            "cost_per_solve_usd": null,
            "fresh_tokens_per_solve": 117107.35483870968,
            "harness": "codex",
            "median_wall_s": 94.603,
            "model": "gpt-5.6-sol",
            "n": 42,
            "solve_rate": 0.7380952380952381,
            "solved": 31,
            "token_telemetry_bases": [
              "proxy_measured"
            ],
            "token_telemetry_coverage": {
              "covered_rows": 42,
              "native_covered_rows": 0,
              "proxy_covered_rows": 42,
              "ratio": 1.0,
              "total_rows": 42
            },
            "token_telemetry_source": "proxy",
            "tokens_cache_read_per_solve": 1164510.9677419355,
            "tokens_cache_write_per_solve": 0.0,
            "tokens_input_uncached_per_solve": 102174.32258064517,
            "tokens_output_per_solve": 14933.032258064517,
            "wilson95": [
              0.5893098001502446,
              0.8469751462739642
            ]
          }
        ],
        "caveats": [],
        "claim": null,
        "countable_rows": 312,
        "date": "2026-07-21",
        "excluded_counts": {
          "infra": 2,
          "rate_limited": 1
        },
        "family": "harness",
        "has_caveats": false,
        "id": "2026-07-21-gpt56",
        "kind": "release",
        "link": null,
        "matched": {
          "duplicate_cells_excluded": 0,
          "duplicate_rows_excluded": 0,
          "groups_compared": 7,
          "matched_cells_per_group": 42,
          "matched_rows": 294,
          "unmatched_countable_rows": 18
        },
        "models": [
          "gpt-5.6-sol"
        ],
        "path": "releases/2026-07-21-gpt56/index.html",
        "raw_rows": 315,
        "results_path": "releases/2026-07-21-gpt56/results.jsonl",
        "results_sha256": "546f167576aac07d8d4c889d559ac86d9f9ad427c5fa53e5a421059f9db498b6",
        "submitter": null,
        "table": "matched",
        "task_set_digest": "d949d46e8aae81489a91d1925ec14f91e4667f5c69ccb314e8fdc611d966c059",
        "title": "OpenBench: gpt-5.6 across 7 harnesses"
      },
      {
        "also_seen_as": [],
        "arms": [
          {
            "cost_per_solve_usd": null,
            "fresh_tokens_per_solve": 83322.96666666666,
            "harness": "claude",
            "median_wall_s": 47.6005,
            "model": "deepseek-v4-flash",
            "n": 38,
            "solve_rate": 0.7894736842105263,
            "solved": 30,
            "token_telemetry_bases": [
              "proxy_measured"
            ],
            "token_telemetry_coverage": {
              "covered_rows": 38,
              "native_covered_rows": 37,
              "proxy_covered_rows": 38,
              "ratio": 1.0,
              "total_rows": 38
            },
            "token_telemetry_source": "proxy",
            "tokens_cache_read_per_solve": 2053623.4666666666,
            "tokens_cache_write_per_solve": 0.0,
            "tokens_input_uncached_per_solve": 35233.7,
            "tokens_output_per_solve": 48089.26666666667,
            "wilson95": [
              0.6365391162355784,
              0.8892533964790406
            ]
          },
          {
            "cost_per_solve_usd": null,
            "fresh_tokens_per_solve": 69624.10714285714,
            "harness": "pi",
            "median_wall_s": 24.4575,
            "model": "deepseek-v4-flash",
            "n": 38,
            "solve_rate": 0.7368421052631579,
            "solved": 28,
            "token_telemetry_bases": [
              "proxy_measured"
            ],
            "token_telemetry_coverage": {
              "covered_rows": 38,
              "native_covered_rows": 35,
              "proxy_covered_rows": 38,
              "ratio": 1.0,
              "total_rows": 38
            },
            "token_telemetry_source": "proxy",
            "tokens_cache_read_per_solve": 1729696.0,
            "tokens_cache_write_per_solve": 0.0,
            "tokens_input_uncached_per_solve": 37986.607142857145,
            "tokens_output_per_solve": 31637.5,
            "wilson95": [
              0.5799084979362467,
              0.8502853761029869
            ]
          },
          {
            "cost_per_solve_usd": null,
            "fresh_tokens_per_solve": 71318.76923076923,
            "harness": "opencode",
            "median_wall_s": 40.271,
            "model": "deepseek-v4-flash",
            "n": 38,
            "solve_rate": 0.6842105263157895,
            "solved": 26,
            "token_telemetry_bases": [
              "proxy_measured"
            ],
            "token_telemetry_coverage": {
              "covered_rows": 38,
              "native_covered_rows": 37,
              "proxy_covered_rows": 38,
              "ratio": 1.0,
              "total_rows": 38
            },
            "token_telemetry_source": "proxy",
            "tokens_cache_read_per_solve": 1584019.6923076923,
            "tokens_cache_write_per_solve": 0.0,
            "tokens_input_uncached_per_solve": 37710.61538461538,
            "tokens_output_per_solve": 33608.153846153844,
            "wilson95": [
              0.5254394356005149,
              0.8091557997633336
            ]
          },
          {
            "cost_per_solve_usd": null,
            "fresh_tokens_per_solve": 131326.68,
            "harness": "grokbuild",
            "median_wall_s": 27.67,
            "model": "deepseek-v4-flash",
            "n": 38,
            "solve_rate": 0.6578947368421053,
            "solved": 25,
            "token_telemetry_bases": [
              "proxy_measured"
            ],
            "token_telemetry_coverage": {
              "covered_rows": 38,
              "native_covered_rows": 0,
              "proxy_covered_rows": 38,
              "ratio": 1.0,
              "total_rows": 38
            },
            "token_telemetry_source": "proxy",
            "tokens_cache_read_per_solve": 1745745.92,
            "tokens_cache_write_per_solve": 0.0,
            "tokens_input_uncached_per_solve": 81795.08,
            "tokens_output_per_solve": 49531.6,
            "wilson95": [
              0.4989181162907454,
              0.7878777997354104
            ]
          },
          {
            "cost_per_solve_usd": null,
            "fresh_tokens_per_solve": 92723.33333333333,
            "harness": "codex",
            "median_wall_s": 27.182,
            "model": "deepseek-v4-flash",
            "n": 38,
            "solve_rate": 0.631578947368421,
            "solved": 24,
            "token_telemetry_bases": [
              "proxy_measured"
            ],
            "token_telemetry_coverage": {
              "covered_rows": 38,
              "native_covered_rows": 33,
              "proxy_covered_rows": 38,
              "ratio": 1.0,
              "total_rows": 38
            },
            "token_telemetry_source": "proxy",
            "tokens_cache_read_per_solve": 1212576.0,
            "tokens_cache_write_per_solve": 0.0,
            "tokens_input_uncached_per_solve": 40351.541666666664,
            "tokens_output_per_solve": 52371.791666666664,
            "wilson95": [
              0.4728367282426011,
              0.766159868445862
            ]
          }
        ],
        "caveats": [
          "Archived first-party matrix published without rerunning models. The source contains 225 rows across 5 harnesses, 15 tasks, and 3 trials.",
          "The bundled results SHA is sealed. Current-tree task checks report expected post-run definition drift; recorded task digests are retained as run-time fingerprints.",
          "The unified board ranks only task/trial cells shared by every harness. The historical release page retains its older per-arm countable view, so its denominators differ.",
          "Harness and host/container version stamps vary in the archive. Token columns use the complete counting-proxy lane for every displayed harness."
        ],
        "claim": null,
        "countable_rows": 217,
        "date": "2026-07-20",
        "excluded_counts": {
          "infra": 8
        },
        "family": "harness",
        "has_caveats": true,
        "id": "2026-07-20-deepseek",
        "kind": "release",
        "link": null,
        "matched": {
          "duplicate_cells_excluded": 0,
          "duplicate_rows_excluded": 0,
          "groups_compared": 5,
          "matched_cells_per_group": 38,
          "matched_rows": 190,
          "unmatched_countable_rows": 27
        },
        "models": [
          "deepseek-v4-flash"
        ],
        "path": "releases/2026-07-20-deepseek/index.html",
        "raw_rows": 225,
        "results_path": "releases/2026-07-20-deepseek/results.jsonl",
        "results_sha256": "ec6b9d58fd3aa6d001db67dc8333e1590c7b1acaa65595f05347413ad23ef414",
        "submitter": null,
        "table": "matched",
        "task_set_digest": "1c4805e7fcd0fd8ff18912c3f745ad243edd2dfd2448f933efe5d0a79966f5bc",
        "title": "OpenBench: deepseek across 5 harnesses"
      },
      {
        "also_seen_as": [],
        "arms": [
          {
            "cost_per_solve_usd": null,
            "fresh_tokens_per_solve": 36706.57142857143,
            "harness": "cursor",
            "median_wall_s": 25.938,
            "model": "grok-4.5",
            "n": 39,
            "solve_rate": 0.8974358974358975,
            "solved": 35,
            "token_telemetry_bases": [
              "harness_reported"
            ],
            "token_telemetry_coverage": {
              "covered_rows": 39,
              "native_covered_rows": 39,
              "proxy_covered_rows": 0,
              "ratio": 1.0,
              "total_rows": 39
            },
            "token_telemetry_source": "native",
            "tokens_cache_read_per_solve": 247504.45714285714,
            "tokens_cache_write_per_solve": 0.0,
            "tokens_input_uncached_per_solve": 28600.485714285714,
            "tokens_output_per_solve": 8106.085714285714,
            "wilson95": [
              0.7642084129775517,
              0.9593873444171301
            ]
          },
          {
            "cost_per_solve_usd": null,
            "fresh_tokens_per_solve": null,
            "harness": "grokbuild",
            "median_wall_s": 42.2565,
            "model": "grok-4.5",
            "n": 39,
            "solve_rate": 0.8717948717948718,
            "solved": 34,
            "token_telemetry_bases": [],
            "token_telemetry_coverage": {
              "covered_rows": 0,
              "native_covered_rows": 0,
              "proxy_covered_rows": 0,
              "ratio": 0.0,
              "total_rows": 39
            },
            "token_telemetry_source": null,
            "tokens_cache_read_per_solve": null,
            "tokens_cache_write_per_solve": null,
            "tokens_input_uncached_per_solve": null,
            "tokens_output_per_solve": null,
            "wilson95": [
              0.7329395096236953,
              0.9439726505197168
            ]
          },
          {
            "cost_per_solve_usd": null,
            "fresh_tokens_per_solve": null,
            "harness": "pi",
            "median_wall_s": 25.284,
            "model": "grok-4.5",
            "n": 39,
            "solve_rate": 0.7948717948717948,
            "solved": 31,
            "token_telemetry_bases": [],
            "token_telemetry_coverage": {
              "covered_rows": 0,
              "native_covered_rows": 38,
              "proxy_covered_rows": 0,
              "ratio": 0.0,
              "total_rows": 39
            },
            "token_telemetry_source": null,
            "tokens_cache_read_per_solve": null,
            "tokens_cache_write_per_solve": null,
            "tokens_input_uncached_per_solve": null,
            "tokens_output_per_solve": null,
            "wilson95": [
              0.6446572703687137,
              0.8922040980208888
            ]
          },
          {
            "cost_per_solve_usd": null,
            "fresh_tokens_per_solve": 55008.13333333333,
            "harness": "opencode",
            "median_wall_s": 21.2275,
            "model": "grok-4.5",
            "n": 39,
            "solve_rate": 0.7692307692307693,
            "solved": 30,
            "token_telemetry_bases": [
              "vendor_split"
            ],
            "token_telemetry_coverage": {
              "covered_rows": 39,
              "native_covered_rows": 39,
              "proxy_covered_rows": 0,
              "ratio": 1.0,
              "total_rows": 39
            },
            "token_telemetry_source": "native",
            "tokens_cache_read_per_solve": 250734.93333333332,
            "tokens_cache_write_per_solve": 0.0,
            "tokens_input_uncached_per_solve": 45493.666666666664,
            "tokens_output_per_solve": 9514.466666666667,
            "wilson95": [
              0.6166343954261705,
              0.8735433757121625
            ]
          }
        ],
        "caveats": [
          "Archived first-party matrix published without rerunning models. The source contains 180 rows across 4 harnesses, 15 tasks, and 3 trials.",
          "The bundled results SHA is sealed. Current-tree task checks report expected post-run definition drift; recorded task digests are retained as run-time fingerprints.",
          "The unified board ranks only task/trial cells shared by every harness. The historical release page retains its older per-arm countable view, so its denominators differ.",
          "No counting-proxy telemetry was retained for this run. Complete native split telemetry is available only where the board reports 100% native coverage."
        ],
        "claim": null,
        "countable_rows": 174,
        "date": "2026-07-20",
        "excluded_counts": {
          "infra": 6
        },
        "family": "harness",
        "has_caveats": true,
        "id": "2026-07-20-grok45",
        "kind": "release",
        "link": null,
        "matched": {
          "duplicate_cells_excluded": 0,
          "duplicate_rows_excluded": 0,
          "groups_compared": 4,
          "matched_cells_per_group": 39,
          "matched_rows": 156,
          "unmatched_countable_rows": 18
        },
        "models": [
          "grok-4.5"
        ],
        "path": "releases/2026-07-20-grok45/index.html",
        "raw_rows": 180,
        "results_path": "releases/2026-07-20-grok45/results.jsonl",
        "results_sha256": "3905ac339eea53b2e66fd2eb6b3a670174d364a40111777d4428619aba3f0081",
        "submitter": null,
        "table": "matched",
        "task_set_digest": "1c4805e7fcd0fd8ff18912c3f745ad243edd2dfd2448f933efe5d0a79966f5bc",
        "title": "OpenBench: grok-4.5 across 4 harnesses"
      },
      {
        "also_seen_as": [],
        "arms": [
          {
            "cost_per_solve_usd": null,
            "fresh_tokens_per_solve": 5856.0,
            "harness": "aider",
            "median_wall_s": 39.0355,
            "model": "deepseek-v4-flash",
            "n": 8,
            "solve_rate": 1.0,
            "solved": 8,
            "token_telemetry_bases": [
              "proxy_measured"
            ],
            "token_telemetry_coverage": {
              "covered_rows": 8,
              "native_covered_rows": 0,
              "proxy_covered_rows": 8,
              "ratio": 1.0,
              "total_rows": 8
            },
            "token_telemetry_source": "proxy",
            "tokens_cache_read_per_solve": 336.0,
            "tokens_cache_write_per_solve": 0.0,
            "tokens_input_uncached_per_solve": 2307.625,
            "tokens_output_per_solve": 3548.375,
            "wilson95": [
              0.6755843804891231,
              1.0
            ]
          },
          {
            "cost_per_solve_usd": null,
            "fresh_tokens_per_solve": 13669.375,
            "harness": "opencode",
            "median_wall_s": 31.720999999999997,
            "model": "deepseek-v4-flash",
            "n": 8,
            "solve_rate": 1.0,
            "solved": 8,
            "token_telemetry_bases": [
              "proxy_measured"
            ],
            "token_telemetry_coverage": {
              "covered_rows": 8,
              "native_covered_rows": 8,
              "proxy_covered_rows": 8,
              "ratio": 1.0,
              "total_rows": 8
            },
            "token_telemetry_source": "proxy",
            "tokens_cache_read_per_solve": 91600.0,
            "tokens_cache_write_per_solve": 0.0,
            "tokens_input_uncached_per_solve": 10327.875,
            "tokens_output_per_solve": 3341.5,
            "wilson95": [
              0.6755843804891231,
              1.0
            ]
          },
          {
            "cost_per_solve_usd": null,
            "fresh_tokens_per_solve": 7742.25,
            "harness": "pi",
            "median_wall_s": 26.444,
            "model": "deepseek-v4-flash",
            "n": 8,
            "solve_rate": 1.0,
            "solved": 8,
            "token_telemetry_bases": [
              "proxy_measured"
            ],
            "token_telemetry_coverage": {
              "covered_rows": 8,
              "native_covered_rows": 8,
              "proxy_covered_rows": 8,
              "ratio": 1.0,
              "total_rows": 8
            },
            "token_telemetry_source": "proxy",
            "tokens_cache_read_per_solve": 35456.0,
            "tokens_cache_write_per_solve": 0.0,
            "tokens_input_uncached_per_solve": 4927.75,
            "tokens_output_per_solve": 2814.5,
            "wilson95": [
              0.6755843804891231,
              1.0
            ]
          }
        ],
        "caveats": [
          "Harness mode is not apples-to-apples: aider ran as a one-shot --message invocation, while pi and opencode ran full agentic loops. Token comparison reflects that difference in harness mode, not a pure like-for-like agent loop contest.",
          "One aider cell excluded: 1 of 9 aider cells was infrastructure-classified and is excluded from solve-rate denominators (aider reported as 8/8). pi and opencode are 9/9.",
          "Uniform total-token basis: totals include uncached input, output, and cache reads from split fields; the vendor aggregate is never used. Values are 6,192 vs 48,262 vs 111,380 tokens/solve (aider vs pi vs opencode).",
          "Verifiable bundle: this release includes results.jsonl and provenance.json. Re-check digests with obench verify docs/releases/2026-07-20-aider-showcase."
        ],
        "claim": "Seed community submission: the verified aider showcase bundle (same artifact as docs/releases/2026-07-20-aider-showcase).",
        "countable_rows": 26,
        "date": "2026-07-20",
        "excluded_counts": {
          "infra": 1
        },
        "family": "harness",
        "has_caveats": true,
        "id": "openbench-aider-showcase",
        "kind": "community",
        "link": "https://github.com/minghinmatthewlam/openbench/tree/main/docs/releases/2026-07-20-aider-showcase",
        "matched": {
          "duplicate_cells_excluded": 0,
          "duplicate_rows_excluded": 0,
          "groups_compared": 3,
          "matched_cells_per_group": 8,
          "matched_rows": 24,
          "unmatched_countable_rows": 2
        },
        "models": [
          "deepseek-v4-flash"
        ],
        "path": "../data/community/openbench-aider-showcase/index.html",
        "raw_rows": 27,
        "results_path": "../data/community/openbench-aider-showcase/results.jsonl",
        "results_sha256": "677642cc0649ecfd42231dbdf49fc5aa0b2c2a8161ccb879358bfeec40e45192",
        "submitter": "openbench",
        "table": "matched",
        "task_set_digest": "8b757f8abf3816cac2ddc80e5ceaefa3f4a323c47c2431258a655d4d3b7f2e43",
        "title": "OpenBench showcase: BYO aider vs pi vs opencode (deepseek-v4-flash)"
      },
      {
        "also_seen_as": [],
        "arms": [
          {
            "cost_per_solve_usd": null,
            "fresh_tokens_per_solve": null,
            "harness": "pi",
            "median_wall_s": 693.4925000000001,
            "model": "glm-5.2",
            "n": 8,
            "solve_rate": 1.0,
            "solved": 8,
            "token_telemetry_bases": [],
            "token_telemetry_coverage": {
              "covered_rows": 0,
              "native_covered_rows": 6,
              "proxy_covered_rows": 0,
              "ratio": 0.0,
              "total_rows": 8
            },
            "token_telemetry_source": null,
            "tokens_cache_read_per_solve": null,
            "tokens_cache_write_per_solve": null,
            "tokens_input_uncached_per_solve": null,
            "tokens_output_per_solve": null,
            "wilson95": [
              0.6755843804891231,
              1.0
            ]
          },
          {
            "cost_per_solve_usd": null,
            "fresh_tokens_per_solve": 94197.42857142857,
            "harness": "opencode",
            "median_wall_s": 661.219,
            "model": "glm-5.2",
            "n": 8,
            "solve_rate": 0.875,
            "solved": 7,
            "token_telemetry_bases": [
              "vendor_split"
            ],
            "token_telemetry_coverage": {
              "covered_rows": 8,
              "native_covered_rows": 8,
              "proxy_covered_rows": 0,
              "ratio": 1.0,
              "total_rows": 8
            },
            "token_telemetry_source": "native",
            "tokens_cache_read_per_solve": 973165.7142857143,
            "tokens_cache_write_per_solve": 0.0,
            "tokens_input_uncached_per_solve": 54762.0,
            "tokens_output_per_solve": 39435.42857142857,
            "wilson95": [
              0.5291051942301385,
              0.9775830911367039
            ]
          },
          {
            "cost_per_solve_usd": null,
            "fresh_tokens_per_solve": null,
            "harness": "claude",
            "median_wall_s": 568.97,
            "model": "glm-5.2",
            "n": 8,
            "solve_rate": 0.625,
            "solved": 5,
            "token_telemetry_bases": [],
            "token_telemetry_coverage": {
              "covered_rows": 0,
              "native_covered_rows": 4,
              "proxy_covered_rows": 0,
              "ratio": 0.0,
              "total_rows": 8
            },
            "token_telemetry_source": null,
            "tokens_cache_read_per_solve": null,
            "tokens_cache_write_per_solve": null,
            "tokens_input_uncached_per_solve": null,
            "tokens_output_per_solve": null,
            "wilson95": [
              0.3057378545838018,
              0.863158240538479
            ]
          },
          {
            "cost_per_solve_usd": null,
            "fresh_tokens_per_solve": null,
            "harness": "grokbuild",
            "median_wall_s": 267.572,
            "model": "glm-5.2",
            "n": 8,
            "solve_rate": 0.625,
            "solved": 5,
            "token_telemetry_bases": [],
            "token_telemetry_coverage": {
              "covered_rows": 0,
              "native_covered_rows": 0,
              "proxy_covered_rows": 0,
              "ratio": 0.0,
              "total_rows": 8
            },
            "token_telemetry_source": null,
            "tokens_cache_read_per_solve": null,
            "tokens_cache_write_per_solve": null,
            "tokens_input_uncached_per_solve": null,
            "tokens_output_per_solve": null,
            "wilson95": [
              0.3057378545838018,
              0.863158240538479
            ]
          },
          {
            "cost_per_solve_usd": null,
            "fresh_tokens_per_solve": null,
            "harness": "codex",
            "median_wall_s": 801.077,
            "model": "glm-5.2",
            "n": 8,
            "solve_rate": 0.5,
            "solved": 4,
            "token_telemetry_bases": [],
            "token_telemetry_coverage": {
              "covered_rows": 0,
              "native_covered_rows": 4,
              "proxy_covered_rows": 0,
              "ratio": 0.0,
              "total_rows": 8
            },
            "token_telemetry_source": null,
            "tokens_cache_read_per_solve": null,
            "tokens_cache_write_per_solve": null,
            "tokens_input_uncached_per_solve": null,
            "tokens_output_per_solve": null,
            "wilson95": [
              0.2152125268244419,
              0.7847874731755581
            ]
          }
        ],
        "caveats": [
          "Quarantine-safe derived bundle: all 15 terminal-bench/cancel-async-tasks rows were excluded because its load-sensitive checker is binding-quarantined. No model was rerun.",
          "The bundled results SHA is sealed. Current-tree task checks report expected post-run definition drift; recorded task digests are retained as run-time fingerprints.",
          "The retained archive lacks image digests, version-source labels, and an explicit per-row timeout field; contemporaneous version-stamp drift is documented in the bundle README.",
          "The unified board ranks only task/trial cells shared by every harness. The release page also shows the per-arm countable view, so its denominators differ."
        ],
        "claim": null,
        "countable_rows": 44,
        "date": "2026-07-09",
        "excluded_counts": {
          "infra": 1
        },
        "family": "harness",
        "has_caveats": true,
        "id": "2026-07-09-glm52",
        "kind": "release",
        "link": null,
        "matched": {
          "duplicate_cells_excluded": 0,
          "duplicate_rows_excluded": 0,
          "groups_compared": 5,
          "matched_cells_per_group": 8,
          "matched_rows": 40,
          "unmatched_countable_rows": 4
        },
        "models": [
          "glm-5.2"
        ],
        "path": "releases/2026-07-09-glm52/index.html",
        "raw_rows": 45,
        "results_path": "releases/2026-07-09-glm52/results.jsonl",
        "results_sha256": "5434f3f8aa66a708936cc322e26b23dd5670caa5532163f1c27fe3cb1a57349c",
        "submitter": null,
        "table": "matched",
        "task_set_digest": "46075fa27b81914497dc37b6ff595ef49dd09b04154ac9f9e93c33892aa0168f",
        "title": "OpenBench: GLM 5.2 quarantine-safe archive"
      }
    ],
    "note": "Scores are not comparable across bundles: task sets, trial counts, and timeout caps differ.",
    "skipped": [
      {
        "id": "2026-07-02-m3",
        "kind": "release",
        "path": "releases/2026-07-02-m3/index.html",
        "reason": "no results.jsonl (HTML-only release page)"
      },
      {
        "id": "2026-07-20-kimi-k3",
        "kind": "release",
        "path": "releases/2026-07-20-kimi-k3/index.html",
        "reason": "no results.jsonl (HTML-only release page)"
      }
    ]
  },
  "packs": [
    {
      "content_sha256": "3b1e576398a8c8f6d42a514ce83ec610f81742edd79f733453bbff681dbeca58",
      "description": "Tiny polarity-checked smoke tasks (make-it-run, fix-failing-test)",
      "id": "openbench/core-smoke",
      "kind": "tasks",
      "latest": "1.0.0",
      "license": "Apache-2.0",
      "name": "core-smoke",
      "org": "openbench",
      "source": "data/packs/openbench-core-smoke",
      "tasks": [
        "make-it-run",
        "fix-failing-test"
      ]
    }
  ],
  "releases": [
    {
      "date": "2026-07-02",
      "id": "2026-07-02-m3",
      "models": [
        "gpt-5.5-medium"
      ],
      "path": "releases/2026-07-02-m3/index.html",
      "title": "OpenBench M3 — Harness Comparison"
    },
    {
      "date": "2026-07-09",
      "id": "2026-07-09-glm52",
      "models": [
        "glm-5.2"
      ],
      "path": "releases/2026-07-09-glm52/index.html",
      "title": "OpenBench: GLM 5.2 quarantine-safe archive"
    },
    {
      "date": "2026-07-20",
      "id": "2026-07-20-grok45",
      "models": [
        "grok-4.5"
      ],
      "path": "releases/2026-07-20-grok45/index.html",
      "title": "OpenBench: grok-4.5 across 4 harnesses"
    },
    {
      "date": "2026-07-20",
      "id": "2026-07-20-kimi-k3",
      "models": [
        "kimi-k3"
      ],
      "path": "releases/2026-07-20-kimi-k3/index.html",
      "title": "OpenBench: kimi-k3 across 3 harnesses"
    },
    {
      "date": "2026-07-20",
      "id": "2026-07-20-deepseek",
      "models": [
        "deepseek-v4-flash"
      ],
      "path": "releases/2026-07-20-deepseek/index.html",
      "title": "OpenBench: deepseek across 5 harnesses"
    },
    {
      "date": "2026-07-21",
      "id": "2026-07-21-gpt56",
      "models": [
        "gpt-5.6-sol"
      ],
      "path": "releases/2026-07-21-gpt56/index.html",
      "title": "OpenBench: gpt-5.6 across 7 harnesses"
    }
  ],
  "schema_version": 5,
  "site_metadata": {
    "canonical_url": "https://openbench.run/openbench/",
    "social_image_url": "https://openbench.run/openbench/og.png"
  }
}
