{
  "schema": "anvil-serving.publication-run-plan/v1",
  "kind": "Description of retained completed cells, not a reconstructed claim of predeclaration",
  "functional": {
    "preflight_repetitions": 3,
    "agentic_repetitions": 3,
    "context_cases": [
      "native-needle",
      "native-order",
      "native-distractor"
    ],
    "context_positions": [
      0.1,
      0.5,
      0.9
    ],
    "context_output_reserve": 8192,
    "image_attempts": 12
  },
  "capacity": {
    "context_tokens": 8192,
    "concurrency": 1,
    "requests": 12,
    "max_tokens": 8192,
    "response_words": 128,
    "prompt_cache_mode": "unique",
    "request_canaries": true,
    "controlled_output_policy": "strict",
    "context_seed": 17,
    "sampling": {
      "temperature": {
        "effective_request": 0.9,
        "requested": 0.9,
        "sent": true
      },
      "top_p": {
        "effective_request": 0.95,
        "requested": 0.95,
        "sent": true
      }
    },
    "serve_flags": {
      "no_thinking": false,
      "reasoning_effort": null,
      "shared_prefix_burst": false,
      "thinking_mode": "default"
    }
  },
  "swe": {
    "selection": {
      "instance_ids": [
        "astropy__astropy-12907",
        "django__django-10097",
        "matplotlib__matplotlib-13989",
        "scikit-learn__scikit-learn-10297",
        "sympy__sympy-11618"
      ],
      "kind": "explicit_instance_ids",
      "sha256": "675265ec1242101c95fdf9e44d4c2e9c6159b9879cbae3f61642740265a66970"
    },
    "dataset": "princeton-nlp/SWE-bench_Verified",
    "split": "test",
    "harnesses": {
      "agent": {
        "name": "mini-swe-agent",
        "revision": "a83fcae82d2a08f0ee0c688f9d137b3566c097f8"
      },
      "grader": {
        "name": "swe-bench",
        "revision": "f7bbbb2ccdf479001d6467c9e34af59e44a840f9"
      },
      "grader_containment": {
        "adapter_sha256": "92ede9ca7793cfed0f89f00c16cbfc5a7275f609a9bdcf115ef673709ebe90f1",
        "receipt": "[operator identity redacted]",
        "source_hashes": {
          "swebench/harness/docker_build.py": "5278842b60a7d38256f95f93c915dc84de2b8b4f286e9baae1b19280f768e484",
          "swebench/harness/run_evaluation.py": "6959f0b4e4eaf979771f529b88e3e9df1daa7fe86bc4291feec2e7d320bf7f2e"
        }
      },
      "python_environment": {
        "architecture": "x86_64",
        "platform": "linux",
        "python": {
          "implementation": "CPython",
          "version": "3.14.4"
        },
        "resolved_packages_sha256": "558d4728e007ae1dfa707114f8c089b6590acd259db7a8ae5cabc06dc45b0a34",
        "schema": "anvil-serving.swe-python-environment/v1"
      }
    },
    "container_limits": {
      "mem_limit": 8589934592,
      "memswap_limit": 8589934592,
      "nano_cpus": 4000000000,
      "network_mode": "none",
      "pids_limit": 512
    }
  },
  "prospective_gates": [
    "Same model/image host cache trial must isolate host KV and state pool changes.",
    "32K/C1 Strata trial is feasibility only; any final contract requires measured quality, context, concurrency and vision parity or explicitly approved changes.",
    "No comparative speed without eligible matched populations and an otherwise matched no-speculation control for speculation gain."
  ]
}