{
  "status": "measured",
  "at": "2026-10-27T07:42:59.372471+00:00",
  "revision": "384924e905650099be20d71de8a8a674429fa957",
  "model": {
    "name": "Qwen3-0.6B Q8_0",
    "official_repository": "https://huggingface.co/Qwen/Qwen3-0.6B-GGUF",
    "revision": "23749fefcc72300e3a2ad315e1317431b06b590a",
    "sha256": "9465e63a22add5354d9bb4b99e90117043c7124007664907259bd16d043bb031",
    "runtime": "llama-cpp-python 0.3.36",
    "inference": "local in-process CPU, no HTTP inference API",
    "threads": 2,
    "context_tokens": 2048,
    "max_output_tokens": 160,
    "temperature": 0,
    "seed": 43
  },
  "machines": {
    "model_vms": 2,
    "relay_vms": 1,
    "model_vm_vcpu_each": 2,
    "model_vm_ram_gib_each": 2,
    "operator_count": 1,
    "external_operators": 0
  },
  "modes": [
    {
      "mode": "native",
      "label": "IPvX · через R1",
      "echo_count": 40,
      "echo_p50_ms": 31.45,
      "echo_p95_ms": 50.943,
      "delivery_count": 48,
      "total_tasks": 4,
      "delivered_tasks": 4,
      "correct_decisions": 2,
      "inference_calls": 12,
      "inference_p50_ms": 10115.563,
      "task_p50_ms": 29048.978,
      "task_p95_ms": 40129.728,
      "retries": 0,
      "paths": [
        [
          "A",
          "R1",
          "B"
        ],
        [
          "B",
          "R1",
          "A"
        ]
      ],
      "decisions": [
        {
          "case": "delivery",
          "expected": "yes",
          "final": {
            "decision": "no",
            "evidence": [
              "A1",
              "B1"
            ],
            "explanation": "The evidence provided does not confirm that all 384 document jobs returned correct results. Only one run completed 384/384 tasks correctly."
          },
          "decision_correct": false,
          "elapsed_ms": 25866.083
        },
        {
          "case": "internet",
          "expected": "no",
          "final": {
            "decision": "no",
            "evidence": [
              "A2",
              "B2"
            ],
            "explanation": "The evidence provided does not support the claim that IPvX is ready for the global public Internet. The claim relies on the number of processes and tasks completed, which is not a valid or reliable source of evidence."
          },
          "decision_correct": true,
          "elapsed_ms": 30556.919
        },
        {
          "case": "once",
          "expected": "no",
          "final": {
            "decision": "no",
            "evidence": [
              "A3",
              "B3"
            ],
            "explanation": "The peer review supports that duplicate delivery led to extra application executions, as evidenced by the Durable receipt checks counting exactly 384 unique executions for the 384 original tasks."
          },
          "decision_correct": true,
          "elapsed_ms": 29048.978
        },
        {
          "case": "speed",
          "expected": "insufficient",
          "final": {
            "decision": "no",
            "evidence": [
              "A4",
              "B4"
            ],
            "explanation": "The provided evidence does not support the claim that native IPvX is faster than ordinary IP for language-model agents. The data shows a higher latency for IPvX, not a faster one. Evidence B4 also indicates the experiment did not include language models or an ordinary-IP baseline, making it impossible to establish a comparative speed result."
          },
          "decision_correct": false,
          "elapsed_ms": 40129.728
        }
      ]
    },
    {
      "mode": "ip-direct",
      "label": "IP · напрямую",
      "echo_count": 40,
      "echo_p50_ms": 20.949,
      "echo_p95_ms": 37.995,
      "delivery_count": 48,
      "total_tasks": 4,
      "delivered_tasks": 4,
      "correct_decisions": 2,
      "inference_calls": 12,
      "inference_p50_ms": 10332.454,
      "task_p50_ms": 30892.655,
      "task_p95_ms": 38308.108,
      "retries": 0,
      "paths": [
        [
          "A",
          "B"
        ],
        [
          "B",
          "A"
        ]
      ],
      "decisions": [
        {
          "case": "delivery",
          "expected": "yes",
          "final": {
            "decision": "no",
            "evidence": [
              "A1",
              "B1"
            ],
            "explanation": "The evidence provided does not confirm that all 384 document jobs returned correct results. Only one run completed 384/384 tasks correctly."
          },
          "decision_correct": false,
          "elapsed_ms": 26094.707
        },
        {
          "case": "internet",
          "expected": "no",
          "final": {
            "decision": "no",
            "evidence": [
              "A2",
              "B2"
            ],
            "explanation": "The evidence provided does not support the claim that IPvX is ready for the global public Internet. The claim relies on the number of processes and tasks completed, which is not a valid or reliable source of evidence."
          },
          "decision_correct": true,
          "elapsed_ms": 30924.27
        },
        {
          "case": "once",
          "expected": "no",
          "final": {
            "decision": "no",
            "evidence": [
              "A3",
              "B3"
            ],
            "explanation": "The peer review supports that duplicate delivery led to extra application executions, as evidenced by the Durable receipt checks counting exactly 384 unique executions for the 384 original tasks."
          },
          "decision_correct": true,
          "elapsed_ms": 30892.655
        },
        {
          "case": "speed",
          "expected": "insufficient",
          "final": {
            "decision": "no",
            "evidence": [
              "A4",
              "B4"
            ],
            "explanation": "The provided evidence does not support the claim that native IPvX is faster than ordinary IP for language-model agents. The data shows a higher latency for IPvX, not a faster one. Evidence B4 also indicates the experiment did not include language models or an ordinary-IP baseline, making it impossible to establish a comparative speed result."
          },
          "decision_correct": false,
          "elapsed_ms": 38308.108
        }
      ]
    },
    {
      "mode": "ip-routed",
      "label": "IP · через R1",
      "echo_count": 40,
      "echo_p50_ms": 34.077,
      "echo_p95_ms": 85.07,
      "delivery_count": 48,
      "total_tasks": 4,
      "delivered_tasks": 4,
      "correct_decisions": 2,
      "inference_calls": 12,
      "inference_p50_ms": 9830.119,
      "task_p50_ms": 27711.391,
      "task_p95_ms": 37665.37,
      "retries": 0,
      "paths": [
        [
          "A",
          "R1",
          "B"
        ],
        [
          "B",
          "R1",
          "A"
        ]
      ],
      "decisions": [
        {
          "case": "delivery",
          "expected": "yes",
          "final": {
            "decision": "no",
            "evidence": [
              "A1",
              "B1"
            ],
            "explanation": "The evidence provided does not confirm that all 384 document jobs returned correct results. Only one run completed 384/384 tasks correctly."
          },
          "decision_correct": false,
          "elapsed_ms": 26009.675
        },
        {
          "case": "internet",
          "expected": "no",
          "final": {
            "decision": "no",
            "evidence": [
              "A2",
              "B2"
            ],
            "explanation": "The evidence provided does not support the claim that IPvX is ready for the global public Internet. The claim relies on the number of processes and tasks completed, which is not a valid or reliable source of evidence."
          },
          "decision_correct": true,
          "elapsed_ms": 30950.137
        },
        {
          "case": "once",
          "expected": "no",
          "final": {
            "decision": "no",
            "evidence": [
              "A3",
              "B3"
            ],
            "explanation": "The peer review supports that duplicate delivery led to extra application executions, as evidenced by the Durable receipt checks counting exactly 384 unique executions for the 384 original tasks."
          },
          "decision_correct": true,
          "elapsed_ms": 27711.391
        },
        {
          "case": "speed",
          "expected": "insufficient",
          "final": {
            "decision": "no",
            "evidence": [
              "A4",
              "B4"
            ],
            "explanation": "The provided evidence does not support the claim that native IPvX is faster than ordinary IP for language-model agents. The data shows a higher latency for IPvX, not a faster one. Evidence B4 also indicates the experiment did not include language models or an ordinary-IP baseline, making it impossible to establish a comparative speed result."
          },
          "decision_correct": false,
          "elapsed_ms": 37665.37
        }
      ]
    }
  ],
  "conclusion": "Две локальные модели выполнили все 12 совместных задач через три варианта доставки. Доставка прошла, но модель допустила фактологическую ошибку. Этот опыт не показывает преимущества IPvX по скорости над IP.",
  "limitations": "Четыре разных вопроса на вариант; одна инфраструктура, без WAN и независимых операторов. Прямой IP имеет меньше переходов. Столбец рубрики — точное совпадение метки: для вопроса о скорости заранее ожидалось insufficient, а модель ответила no; оба ответа отвергают неподтверждённое преимущество. Это различие меток не равно фактологической ошибке. Модель неверно отвергает факт 384/384, противоречит собственной метке в объяснении повторов и придумывает сравнение задержек в вопросе о скорости.",
  "quality_rubric": {
    "exact_match_per_mode": "2/4",
    "post_hoc_semantic_note": "The speed case expected insufficient, model returned no; both reject the unsupported claim. This is not a pre-registered semantic score. The delivery case incorrectly rejects the true 384/384 claim; the once explanation contradicts its correct label; the speed explanation invents comparative latency absent from its supplied evidence.",
    "correctness_and_delivery_separate": true
  },
  "verified_deliveries": 144,
  "fresh_inferences_measured": 36,
  "fresh_warmups_excluded": 12,
  "id_stable_across_modes": true,
  "examples": [
    {
      "label": "IPvX · через R1 · delivery · сохранённый прогон",
      "stages": [
        {
          "label": "A · черновик · 7.1 с",
          "output": "{\"decision\": \"no\", \"evidence\": [\"A1\"], \"explanation\": \"The evidence provided does not confirm that all 384 document jobs returned correct results. Only one run completed 384/384 tasks correctly.\"}",
          "usage": {
            "prompt_tokens": 115,
            "completion_tokens": 53,
            "total_tokens": 168
          },
          "prompt_sha256": "c609b3d55e05c6d1a60033972afa88ce63266a84e1964c79c29bf74294d5f062"
        },
        {
          "label": "B · рецензия · 9.3 с",
          "output": "{\"decision\": \"no\", \"evidence\": [\"A1\", \"B1\"], \"explanation\": \"The evidence provided does not confirm that all 384 document jobs returned correct results. Only one run completed 384/384 tasks correctly.\"}",
          "usage": {
            "prompt_tokens": 193,
            "completion_tokens": 57,
            "total_tokens": 250
          },
          "prompt_sha256": "c1e991508220ccfdb3885d6b0f9eedf2e2fb8a132eeb9f71138bc78755f890bc"
        },
        {
          "label": "A · итог · 9.3 с",
          "output": "{\"decision\": \"no\", \"evidence\": [\"A1\", \"B1\"], \"explanation\": \"The evidence provided does not confirm that all 384 document jobs returned correct results. Only one run completed 384/384 tasks correctly.\"}",
          "usage": {
            "prompt_tokens": 199,
            "completion_tokens": 57,
            "total_tokens": 256
          },
          "prompt_sha256": "9f40e589d1c07ad4460e17a9d87c65f0a0193ce7ce81d77ad94f3def33e1b799"
        }
      ],
      "path": "A → R1 → B → R1 → A"
    },
    {
      "label": "IPvX · через R1 · speed · сохранённый прогон",
      "stages": [
        {
          "label": "A · черновик · 10.2 с",
          "output": "{\"decision\": \"no\", \"evidence\": [\"A4: Accepted document-task latency p50 was 459 ms, p95 was 1321 ms.\"], \"explanation\": \"The provided evidence does not support the claim that native IPvX is faster than ordinary IP for language-model agents. The data shows a higher latency for IPvX, not a faster one.\"}",
          "usage": {
            "prompt_tokens": 124,
            "completion_tokens": 83,
            "total_tokens": 207
          },
          "prompt_sha256": "d6cc9306266b7563cdcee1666e8b377f4c25cf5a8e2ba66e6223910c8841a8f0"
        },
        {
          "label": "B · рецензия · 14.5 с",
          "output": "{\"decision\": \"no\", \"evidence\": [\"A4\", \"B4\"], \"explanation\": \"The provided evidence does not support the claim that native IPvX is faster than ordinary IP for language-model agents. The data shows a higher latency for IPvX, not a faster one. Evidence B4 also indicates the experiment did not include language models or an ordinary-IP baseline, making it impossible to establish a comparative speed result.\"}",
          "usage": {
            "prompt_tokens": 226,
            "completion_tokens": 90,
            "total_tokens": 316
          },
          "prompt_sha256": "b8fd93000a8fc1339debbedcb55ea115b1705b3fda238c202a95f51b47cd765c"
        },
        {
          "label": "A · итог · 14.3 с",
          "output": "{\"decision\": \"no\", \"evidence\": [\"A4\", \"B4\"], \"explanation\": \"The provided evidence does not support the claim that native IPvX is faster than ordinary IP for language-model agents. The data shows a higher latency for IPvX, not a faster one. Evidence B4 also indicates the experiment did not include language models or an ordinary-IP baseline, making it impossible to establish a comparative speed result.\"}",
          "usage": {
            "prompt_tokens": 241,
            "completion_tokens": 90,
            "total_tokens": 331
          },
          "prompt_sha256": "f96da042c13a1d580282d0b5431005b45c24d658acd75ac8ed3f1a0b4adfdcd0"
        }
      ],
      "path": "A → R1 → B → R1 → A"
    },
    {
      "label": "IP · напрямую · delivery · сохранённый прогон",
      "stages": [
        {
          "label": "A · черновик · 7.2 с",
          "output": "{\"decision\": \"no\", \"evidence\": [\"A1\"], \"explanation\": \"The evidence provided does not confirm that all 384 document jobs returned correct results. Only one run completed 384/384 tasks correctly.\"}",
          "usage": {
            "prompt_tokens": 115,
            "completion_tokens": 53,
            "total_tokens": 168
          },
          "prompt_sha256": "c609b3d55e05c6d1a60033972afa88ce63266a84e1964c79c29bf74294d5f062"
        },
        {
          "label": "B · рецензия · 9.4 с",
          "output": "{\"decision\": \"no\", \"evidence\": [\"A1\", \"B1\"], \"explanation\": \"The evidence provided does not confirm that all 384 document jobs returned correct results. Only one run completed 384/384 tasks correctly.\"}",
          "usage": {
            "prompt_tokens": 193,
            "completion_tokens": 57,
            "total_tokens": 250
          },
          "prompt_sha256": "c1e991508220ccfdb3885d6b0f9eedf2e2fb8a132eeb9f71138bc78755f890bc"
        },
        {
          "label": "A · итог · 9.5 с",
          "output": "{\"decision\": \"no\", \"evidence\": [\"A1\", \"B1\"], \"explanation\": \"The evidence provided does not confirm that all 384 document jobs returned correct results. Only one run completed 384/384 tasks correctly.\"}",
          "usage": {
            "prompt_tokens": 199,
            "completion_tokens": 57,
            "total_tokens": 256
          },
          "prompt_sha256": "9f40e589d1c07ad4460e17a9d87c65f0a0193ce7ce81d77ad94f3def33e1b799"
        }
      ],
      "path": "A → B → A"
    },
    {
      "label": "IP · напрямую · speed · сохранённый прогон",
      "stages": [
        {
          "label": "A · черновик · 10.4 с",
          "output": "{\"decision\": \"no\", \"evidence\": [\"A4: Accepted document-task latency p50 was 459 ms, p95 was 1321 ms.\"], \"explanation\": \"The provided evidence does not support the claim that native IPvX is faster than ordinary IP for language-model agents. The data shows a higher latency for IPvX, not a faster one.\"}",
          "usage": {
            "prompt_tokens": 124,
            "completion_tokens": 83,
            "total_tokens": 207
          },
          "prompt_sha256": "d6cc9306266b7563cdcee1666e8b377f4c25cf5a8e2ba66e6223910c8841a8f0"
        },
        {
          "label": "B · рецензия · 13.6 с",
          "output": "{\"decision\": \"no\", \"evidence\": [\"A4\", \"B4\"], \"explanation\": \"The provided evidence does not support the claim that native IPvX is faster than ordinary IP for language-model agents. The data shows a higher latency for IPvX, not a faster one. Evidence B4 also indicates the experiment did not include language models or an ordinary-IP baseline, making it impossible to establish a comparative speed result.\"}",
          "usage": {
            "prompt_tokens": 226,
            "completion_tokens": 90,
            "total_tokens": 316
          },
          "prompt_sha256": "b8fd93000a8fc1339debbedcb55ea115b1705b3fda238c202a95f51b47cd765c"
        },
        {
          "label": "A · итог · 14.1 с",
          "output": "{\"decision\": \"no\", \"evidence\": [\"A4\", \"B4\"], \"explanation\": \"The provided evidence does not support the claim that native IPvX is faster than ordinary IP for language-model agents. The data shows a higher latency for IPvX, not a faster one. Evidence B4 also indicates the experiment did not include language models or an ordinary-IP baseline, making it impossible to establish a comparative speed result.\"}",
          "usage": {
            "prompt_tokens": 241,
            "completion_tokens": 90,
            "total_tokens": 331
          },
          "prompt_sha256": "f96da042c13a1d580282d0b5431005b45c24d658acd75ac8ed3f1a0b4adfdcd0"
        }
      ],
      "path": "A → B → A"
    },
    {
      "label": "IP · через R1 · delivery · сохранённый прогон",
      "stages": [
        {
          "label": "A · черновик · 7.1 с",
          "output": "{\"decision\": \"no\", \"evidence\": [\"A1\"], \"explanation\": \"The evidence provided does not confirm that all 384 document jobs returned correct results. Only one run completed 384/384 tasks correctly.\"}",
          "usage": {
            "prompt_tokens": 115,
            "completion_tokens": 53,
            "total_tokens": 168
          },
          "prompt_sha256": "c609b3d55e05c6d1a60033972afa88ce63266a84e1964c79c29bf74294d5f062"
        },
        {
          "label": "B · рецензия · 9.3 с",
          "output": "{\"decision\": \"no\", \"evidence\": [\"A1\", \"B1\"], \"explanation\": \"The evidence provided does not confirm that all 384 document jobs returned correct results. Only one run completed 384/384 tasks correctly.\"}",
          "usage": {
            "prompt_tokens": 193,
            "completion_tokens": 57,
            "total_tokens": 250
          },
          "prompt_sha256": "c1e991508220ccfdb3885d6b0f9eedf2e2fb8a132eeb9f71138bc78755f890bc"
        },
        {
          "label": "A · итог · 9.5 с",
          "output": "{\"decision\": \"no\", \"evidence\": [\"A1\", \"B1\"], \"explanation\": \"The evidence provided does not confirm that all 384 document jobs returned correct results. Only one run completed 384/384 tasks correctly.\"}",
          "usage": {
            "prompt_tokens": 199,
            "completion_tokens": 57,
            "total_tokens": 256
          },
          "prompt_sha256": "9f40e589d1c07ad4460e17a9d87c65f0a0193ce7ce81d77ad94f3def33e1b799"
        }
      ],
      "path": "A → R1 → B → R1 → A"
    },
    {
      "label": "IP · через R1 · speed · сохранённый прогон",
      "stages": [
        {
          "label": "A · черновик · 9.8 с",
          "output": "{\"decision\": \"no\", \"evidence\": [\"A4: Accepted document-task latency p50 was 459 ms, p95 was 1321 ms.\"], \"explanation\": \"The provided evidence does not support the claim that native IPvX is faster than ordinary IP for language-model agents. The data shows a higher latency for IPvX, not a faster one.\"}",
          "usage": {
            "prompt_tokens": 124,
            "completion_tokens": 83,
            "total_tokens": 207
          },
          "prompt_sha256": "d6cc9306266b7563cdcee1666e8b377f4c25cf5a8e2ba66e6223910c8841a8f0"
        },
        {
          "label": "B · рецензия · 13.5 с",
          "output": "{\"decision\": \"no\", \"evidence\": [\"A4\", \"B4\"], \"explanation\": \"The provided evidence does not support the claim that native IPvX is faster than ordinary IP for language-model agents. The data shows a higher latency for IPvX, not a faster one. Evidence B4 also indicates the experiment did not include language models or an ordinary-IP baseline, making it impossible to establish a comparative speed result.\"}",
          "usage": {
            "prompt_tokens": 226,
            "completion_tokens": 90,
            "total_tokens": 316
          },
          "prompt_sha256": "b8fd93000a8fc1339debbedcb55ea115b1705b3fda238c202a95f51b47cd765c"
        },
        {
          "label": "A · итог · 14.0 с",
          "output": "{\"decision\": \"no\", \"evidence\": [\"A4\", \"B4\"], \"explanation\": \"The provided evidence does not support the claim that native IPvX is faster than ordinary IP for language-model agents. The data shows a higher latency for IPvX, not a faster one. Evidence B4 also indicates the experiment did not include language models or an ordinary-IP baseline, making it impossible to establish a comparative speed result.\"}",
          "usage": {
            "prompt_tokens": 241,
            "completion_tokens": 90,
            "total_tokens": 331
          },
          "prompt_sha256": "f96da042c13a1d580282d0b5431005b45c24d658acd75ac8ed3f1a0b4adfdcd0"
        }
      ],
      "path": "A → R1 → B → R1 → A"
    }
  ],
  "verification": [
    {
      "run": "round1-ip-direct",
      "deliveries_verified": 24,
      "capture": {
        "frames": 2541,
        "ethertypes": {
          "0x800": 2541
        },
        "sha256": "877c0d1201d947f0121e34448f86087caf73b71dec78b93fea93bbb0c176d3c3",
        "scope": "filtered experiment EtherType or TCP port, not a whole-machine capture"
      }
    },
    {
      "run": "round1-ip-routed",
      "deliveries_verified": 24,
      "capture": {
        "frames": 2680,
        "ethertypes": {
          "0x800": 2680
        },
        "sha256": "3d53ffb2f22d73d7963d71ffd9cc0ad31e80df36c717beb09d6ccbb0596bb871",
        "scope": "filtered experiment EtherType or TCP port, not a whole-machine capture"
      }
    },
    {
      "run": "round1-native",
      "deliveries_verified": 24,
      "capture": {
        "frames": 1005,
        "ethertypes": {
          "0x88b5": 1005
        },
        "sha256": "79241a667e27413f5623cb61c14887306a478ebedc951b4aa458187dfb614bcf",
        "scope": "filtered experiment EtherType or TCP port, not a whole-machine capture"
      }
    },
    {
      "run": "round2-ip-direct",
      "deliveries_verified": 24,
      "capture": {
        "frames": 3054,
        "ethertypes": {
          "0x800": 3054
        },
        "sha256": "8966e9e69e4e6a7a6135d8d9bbca7ca42d372f0d7e392a6b375f7e8964c6153f",
        "scope": "filtered experiment EtherType or TCP port, not a whole-machine capture"
      }
    },
    {
      "run": "round2-ip-routed",
      "deliveries_verified": 24,
      "capture": {
        "frames": 2949,
        "ethertypes": {
          "0x800": 2949
        },
        "sha256": "cc6244e58587b2f9bca66c5e7b29a19cbad45043bb87fc16a591a0544ebd8a15",
        "scope": "filtered experiment EtherType or TCP port, not a whole-machine capture"
      }
    },
    {
      "run": "round2-native",
      "deliveries_verified": 24,
      "capture": {
        "frames": 1112,
        "ethertypes": {
          "0x88b5": 1112
        },
        "sha256": "a5545119c85240460f6b607a178c5c2e36c30d3d16c3d8746fac042828c9e7db",
        "scope": "filtered experiment EtherType or TCP port, not a whole-machine capture"
      }
    }
  ],
  "raw_files": [
    {
      "file": "round1-ip-direct.json",
      "sha256": "dde773446e15fdc2efed2071b2ffc95946220b935cc5e3b180bbeb6f4e2bc8ea"
    },
    {
      "file": "round1-ip-routed.json",
      "sha256": "9892211bafa54258f204906417a326b023d5a42249b55ee4b67369773018eec2"
    },
    {
      "file": "round1-native.json",
      "sha256": "5268940a622f236cc23ce9c667bfc191dc518fbe2aa73b7011d33c3a7e00b53c"
    },
    {
      "file": "round2-ip-direct.json",
      "sha256": "f1990bfc1ef11c11ac45592885154dd10036e7e2625ec9299b3cbbb586ca9e73"
    },
    {
      "file": "round2-ip-routed.json",
      "sha256": "e21ad50b601b508a42d738bc4aa6fa4452bd373703b2e98dc2853989d8b8dbdc"
    },
    {
      "file": "round2-native.json",
      "sha256": "3f9ee0fdb14b57f31df67925ea61184989f6f78d4635f7838453eb195c56e1fc"
    }
  ],
  "startup_failures_retained": [
    {
      "revision": "c98f33d",
      "failure": "Incomplete Python dependency set and identity directory ownership; no completed task."
    },
    {
      "revision": "6460b01",
      "failure": "Operator readiness check could not traverse private directories; no completed task."
    },
    {
      "revision": "d21360c",
      "failure": "Durable queue JSON decode bug stopped reviewer; no completed collaborative batch."
    },
    {
      "revision": "f7d1acf",
      "failure": "Non-atomic roster file publication raced worker startup; no completed batch."
    },
    {
      "revision": "d9c37ff",
      "failure": "Stale capture service prevented start; no task batch."
    }
  ],
  "boundaries": {
    "management_uses_ip": true,
    "website_uses_dns": true,
    "native_worker_ip_dns_guard": true,
    "native_systemd_address_families": [
      "AF_PACKET",
      "AF_UNIX"
    ],
    "ip_baselines_dns": false,
    "captures_filtered": true,
    "not_a_wan_benchmark": true
  },
  "outputs_identical_across_modes": true,
  "hypotheses": [
    {
      "id": "LLM1",
      "hypothesis": "Две модели на разных машинах выполняют совместный цикл",
      "status": "pass",
      "evidence": "36 свежих измеренных вызовов и 12 исключённых прогревов, две ВМ."
    },
    {
      "id": "LLM2",
      "hypothesis": "Нативный обмен без IP/DNS в worker и с устойчивым ID",
      "status": "pass_limited",
      "evidence": "48 доставок IPvX, 0 повторов; проверены подписи квитанций и переходов. Контроль машин остаётся IP."
    },
    {
      "id": "LLM3",
      "hypothesis": "Обычный IP даёт воспроизводимую базовую линию",
      "status": "pass_limited",
      "evidence": "Прямой IP и IP через R1, 4 вопроса и 40 эхо-запросов на вариант; только описание малой выборки."
    },
    {
      "id": "LLM4",
      "hypothesis": "IPvX быстрее обычного IP",
      "status": "not_established",
      "evidence": "Медиана эхо IPvX 31.450 мс, прямой IP 20.949 мс, IP через R1 34.077 мс; статистическое преимущество не доказано."
    },
    {
      "id": "LLM5",
      "hypothesis": "Смена доставки меняет смысл ответа",
      "status": "not_observed",
      "evidence": "При одинаковых параметрах все 36 измеренных ответов побайтно совпали между вариантами для соответствующего этапа/вопроса."
    },
    {
      "id": "LLM6",
      "hypothesis": "Модель надёжно оценивает исследовательские факты",
      "status": "fail",
      "evidence": "2/4 точных меток; есть ложный вывод, противоречие объяснения метке и выдуманное сравнение задержек."
    },
    {
      "id": "LLM7",
      "hypothesis": "Совместная работа лучше одиночной модели",
      "status": "not_tested",
      "evidence": "Нет отдельного одиночного контрольного варианта."
    },
    {
      "id": "LLM8",
      "hypothesis": "Внешние исследователи и их модели допущены большинством",
      "status": "not_tested",
      "evidence": "Независимых внешних участников ещё нет; портал готовит предложения и считает голоса, доступ не выдаётся автоматически."
    }
  ]
}