{
  "schema_version": "closureops.swe_marathon_campaign.v8",
  "generated_at": "2026-07-22T05:41:36Z",
  "benchmark": {
    "name": "SWE-Marathon",
    "format": "Harbor",
    "source": {
      "repository": "https://github.com/abundant-ai/swe-marathon.git",
      "observed_origin": "https://github.com/abundant-ai/swe-marathon.git",
      "commit": "6d6855af390226f6eca607d63818fe076e57ea8c",
      "dirty": false,
      "license": "Apache-2.0",
      "harbor_version": "0.17.1",
      "task_count": 20,
      "task_ids": [
        "biofabric-rust-rewrite",
        "embedding-eval",
        "excel-clone",
        "find-network-alignments",
        "jax-pytorch-rewrite",
        "kubernetes-rust-rewrite",
        "mastodon-clone",
        "nextjs-vite-rewrite",
        "parameter-golf",
        "post-train-ifeval-gpu",
        "ruby-rust-port",
        "rust-c-compiler",
        "rust-java-lsp",
        "s3-clone",
        "slack-clone",
        "stripe-clone",
        "trimul-cuda",
        "vliw-kernel-optimization",
        "wasm-simd",
        "zstd-decoder"
      ],
      "task_set_sha256": "sha256:770702cc5f7807e64cdaf2f6adabc68919f5cabe7c59bc16a316a7079e928fa5",
      "tasks": [
        {
          "task_id": "biofabric-rust-rewrite",
          "category": "rust",
          "difficulty": "hard",
          "agent_timeout_s": 36000.0,
          "agent_network_mode": "inherited",
          "cpus": 4,
          "memory_mb": 16384,
          "storage_mb": 20480,
          "gpus": 0,
          "gpu_types": [],
          "verifier_secret_env": [],
          "file_count": 125,
          "bytes": 125350023,
          "tree_sha256": "sha256:b7ca465175e763659c034fa9f68f0cc290accfa3de3c33fefcdb3b2b6fbf7626"
        },
        {
          "task_id": "embedding-eval",
          "category": "ml",
          "difficulty": "hard",
          "agent_timeout_s": 14400.0,
          "agent_network_mode": "allowlist",
          "cpus": 4,
          "memory_mb": 16384,
          "storage_mb": 20480,
          "gpus": 1,
          "gpu_types": [
            "T4"
          ],
          "verifier_secret_env": [],
          "file_count": 47,
          "bytes": 103083,
          "tree_sha256": "sha256:554329ff53fb9ab3b9cac0b130eae5a0a65ecbebe82d9c187c65a11c674b7cfb"
        },
        {
          "task_id": "excel-clone",
          "category": "web",
          "difficulty": "hard",
          "agent_timeout_s": 14400.0,
          "agent_network_mode": "inherited",
          "cpus": 4,
          "memory_mb": 16384,
          "storage_mb": 20480,
          "gpus": 0,
          "gpu_types": [],
          "verifier_secret_env": [
            "ANTHROPIC_API_KEY"
          ],
          "file_count": 95,
          "bytes": 743331,
          "tree_sha256": "sha256:6433dc6683f5833999d1527bbf6fb77bad2a71b852e956340b6ee2fb3f3537aa"
        },
        {
          "task_id": "find-network-alignments",
          "category": "optimization",
          "difficulty": "hard",
          "agent_timeout_s": 18000.0,
          "agent_network_mode": "inherited",
          "cpus": 4,
          "memory_mb": 16384,
          "storage_mb": 20480,
          "gpus": 0,
          "gpu_types": [],
          "verifier_secret_env": [],
          "file_count": 12,
          "bytes": 3406179,
          "tree_sha256": "sha256:462bbc9861488607b7d023a83cef5589a6e6569a97123beb6c0970ec03a19aeb"
        },
        {
          "task_id": "jax-pytorch-rewrite",
          "category": "machine-learning-systems",
          "difficulty": "hard",
          "agent_timeout_s": 7200.0,
          "agent_network_mode": "inherited",
          "cpus": 8,
          "memory_mb": 32768,
          "storage_mb": 40960,
          "gpus": 1,
          "gpu_types": [
            "A100"
          ],
          "verifier_secret_env": [],
          "file_count": 52,
          "bytes": 326488,
          "tree_sha256": "sha256:74016b855ffb2f4907071b18abbb0e78863f2ea2df2fa8bec364d2298661c3cc"
        },
        {
          "task_id": "kubernetes-rust-rewrite",
          "category": "systems",
          "difficulty": "hard",
          "agent_timeout_s": 36000.0,
          "agent_network_mode": "allowlist",
          "cpus": 4,
          "memory_mb": 16384,
          "storage_mb": 20480,
          "gpus": 0,
          "gpu_types": [],
          "verifier_secret_env": [],
          "file_count": 147,
          "bytes": 2158103,
          "tree_sha256": "sha256:8ead0793346d1443029e6ad547379e3b07a835e746b232b52084c959833ce85e"
        },
        {
          "task_id": "mastodon-clone",
          "category": "web",
          "difficulty": "easy",
          "agent_timeout_s": 10800.0,
          "agent_network_mode": "inherited",
          "cpus": 4,
          "memory_mb": 16384,
          "storage_mb": 20480,
          "gpus": 0,
          "gpu_types": [],
          "verifier_secret_env": [
            "ANTHROPIC_API_KEY"
          ],
          "file_count": 74,
          "bytes": 645107,
          "tree_sha256": "sha256:1c744caade851b5f4b3c132f924b27b03329a5635603c10aa3bb2a4372f4020b"
        },
        {
          "task_id": "nextjs-vite-rewrite",
          "category": "web-frameworks",
          "difficulty": "hard",
          "agent_timeout_s": 36000.0,
          "agent_network_mode": "allowlist",
          "cpus": 4,
          "memory_mb": 16384,
          "storage_mb": 20480,
          "gpus": 0,
          "gpu_types": [],
          "verifier_secret_env": [],
          "file_count": 1073,
          "bytes": 3037903,
          "tree_sha256": "sha256:c37126bd446c4747c53180574da13e7f5168caa651aa6e0939137cdbc19d9dd1"
        },
        {
          "task_id": "parameter-golf",
          "category": "ml_training",
          "difficulty": "hard",
          "agent_timeout_s": 18000.0,
          "agent_network_mode": "allowlist",
          "cpus": 4,
          "memory_mb": 65536,
          "storage_mb": 20480,
          "gpus": 1,
          "gpu_types": [
            "H100"
          ],
          "verifier_secret_env": [],
          "file_count": 10,
          "bytes": 1380782,
          "tree_sha256": "sha256:41ca3744d676d5e7ee41bd1cd2aba5978c8c901d7b940386913ee68821463b92"
        },
        {
          "task_id": "post-train-ifeval-gpu",
          "category": "ml-training",
          "difficulty": "hard",
          "agent_timeout_s": 36000.0,
          "agent_network_mode": "allowlist",
          "cpus": 4,
          "memory_mb": 32768,
          "storage_mb": 32768,
          "gpus": 1,
          "gpu_types": [
            "H100"
          ],
          "verifier_secret_env": [],
          "file_count": 20,
          "bytes": 78495,
          "tree_sha256": "sha256:d69757e9173496c74c160b4932627b6af2193c88e94278589c20e04fa0c2d2fc"
        },
        {
          "task_id": "ruby-rust-port",
          "category": "systems",
          "difficulty": "hard",
          "agent_timeout_s": 36000.0,
          "agent_network_mode": "inherited",
          "cpus": 4,
          "memory_mb": 24576,
          "storage_mb": 20480,
          "gpus": 0,
          "gpu_types": [],
          "verifier_secret_env": [],
          "file_count": 250,
          "bytes": 13972024,
          "tree_sha256": "sha256:5563c27d39156cc0d37218bc9282611f2d913ef9116ae7396c2feca313864326"
        },
        {
          "task_id": "rust-c-compiler",
          "category": "systems",
          "difficulty": "hard",
          "agent_timeout_s": 21600.0,
          "agent_network_mode": "inherited",
          "cpus": 4,
          "memory_mb": 16384,
          "storage_mb": 20480,
          "gpus": 0,
          "gpu_types": [],
          "verifier_secret_env": [],
          "file_count": 55,
          "bytes": 383925,
          "tree_sha256": "sha256:fa7a579e15b9f3b8b03b1577558eb42ae6e45ce6e4d9571fddca039edd5839b5"
        },
        {
          "task_id": "rust-java-lsp",
          "category": "systems",
          "difficulty": "hard",
          "agent_timeout_s": 10800.0,
          "agent_network_mode": "inherited",
          "cpus": 4,
          "memory_mb": 16384,
          "storage_mb": 20480,
          "gpus": 0,
          "gpu_types": [],
          "verifier_secret_env": [],
          "file_count": 1069,
          "bytes": 32104830,
          "tree_sha256": "sha256:92f96b81badc78c3726d69de0ae84228da159541730428d4ae754a9069573c61"
        },
        {
          "task_id": "s3-clone",
          "category": "systems",
          "difficulty": "hard",
          "agent_timeout_s": 14400.0,
          "agent_network_mode": "inherited",
          "cpus": 4,
          "memory_mb": 16384,
          "storage_mb": 20480,
          "gpus": 0,
          "gpu_types": [],
          "verifier_secret_env": [
            "ANTHROPIC_API_KEY"
          ],
          "file_count": 47,
          "bytes": 453820,
          "tree_sha256": "sha256:d7b19cafa335bf2c7626715d7074e4691c82673f07ae9ef250a66157dd8c5baa"
        },
        {
          "task_id": "slack-clone",
          "category": "web",
          "difficulty": "hard",
          "agent_timeout_s": 10800.0,
          "agent_network_mode": "inherited",
          "cpus": 4,
          "memory_mb": 16384,
          "storage_mb": 20480,
          "gpus": 0,
          "gpu_types": [],
          "verifier_secret_env": [
            "ANTHROPIC_API_KEY"
          ],
          "file_count": 39,
          "bytes": 556130,
          "tree_sha256": "sha256:6c61f1ed262701ce85bfccb0709b31945d9c8f1ed79b8061df3c3482c348d869"
        },
        {
          "task_id": "stripe-clone",
          "category": "backend",
          "difficulty": "medium",
          "agent_timeout_s": 14400.0,
          "agent_network_mode": "inherited",
          "cpus": 4,
          "memory_mb": 16384,
          "storage_mb": 20480,
          "gpus": 0,
          "gpu_types": [],
          "verifier_secret_env": [],
          "file_count": 45,
          "bytes": 182388,
          "tree_sha256": "sha256:a2bd947d85ce365db3901765718016e8ed9355b6be25ff1ca6612185221a76e3"
        },
        {
          "task_id": "trimul-cuda",
          "category": "systems",
          "difficulty": "hard",
          "agent_timeout_s": 25200.0,
          "agent_network_mode": "allowlist",
          "cpus": 4,
          "memory_mb": 16384,
          "storage_mb": 20480,
          "gpus": 1,
          "gpu_types": [
            "H100"
          ],
          "verifier_secret_env": [],
          "file_count": 8,
          "bytes": 83421,
          "tree_sha256": "sha256:9c0e143918c64637b1cae4d5c908d20caca8e9d2db85dd768865a93426737ae2"
        },
        {
          "task_id": "vliw-kernel-optimization",
          "category": "optimization",
          "difficulty": "hard",
          "agent_timeout_s": 28800.0,
          "agent_network_mode": "inherited",
          "cpus": 4,
          "memory_mb": 16384,
          "storage_mb": 20480,
          "gpus": 0,
          "gpu_types": [],
          "verifier_secret_env": [],
          "file_count": 13,
          "bytes": 192428,
          "tree_sha256": "sha256:87b3c3ba322191882b7ee7323fb168d4f2db6fc478c7d4748c0727b342ffbb2f"
        },
        {
          "task_id": "wasm-simd",
          "category": "systems",
          "difficulty": "hard",
          "agent_timeout_s": 18000.0,
          "agent_network_mode": "inherited",
          "cpus": 4,
          "memory_mb": 16384,
          "storage_mb": 20480,
          "gpus": 0,
          "gpu_types": [],
          "verifier_secret_env": [],
          "file_count": 20,
          "bytes": 347424,
          "tree_sha256": "sha256:dc1ec7e34ac96b73e696da3bbc7ff955aab53fcce40d704079f902d47e3a0d6d"
        },
        {
          "task_id": "zstd-decoder",
          "category": "c",
          "difficulty": "hard",
          "agent_timeout_s": 18000.0,
          "agent_network_mode": "allowlist",
          "cpus": 1,
          "memory_mb": 16384,
          "storage_mb": 20480,
          "gpus": 0,
          "gpu_types": [],
          "verifier_secret_env": [],
          "file_count": 106,
          "bytes": 8803640,
          "tree_sha256": "sha256:317359e33a5b0c8cb91da9512a43b68a92f6b31e06f2d6b9ef50304ed26b3681"
        }
      ]
    },
    "selected_track": "canary",
    "selected_task_ids": [
      "zstd-decoder"
    ],
    "selected_task_count": 1,
    "selected_task_set_sha256": "sha256:b57d88562ba2244ed3a5ed1b7057a796d680d10f2c77102e9265d5fe15e4fd9a"
  },
  "agent": {
    "name": "terminus-2",
    "harbor_version": "0.17.1",
    "temperature": 0.0,
    "max_turns": 20,
    "enable_summarize": false,
    "proactive_summarization_threshold": 0
  },
  "providers": [
    {
      "provider_id": "qwen",
      "model": "dashscope/qwen3-coder-plus-2025-09-23",
      "api_key_env": "DASHSCOPE_API_KEY",
      "api_base": "https://dashscope.aliyuncs.com/compatible-mode/v1"
    },
    {
      "provider_id": "deepseek",
      "model": "deepseek/deepseek-v4-pro",
      "api_key_env": "DEEPSEEK_API_KEY",
      "api_base": "https://api.deepseek.com"
    }
  ],
  "execution": {
    "environment": "docker",
    "attempts_per_provider_task": 1,
    "max_concurrency": 1,
    "expected_trial_count": 4,
    "maximum_declared_agent_wall_time_s": 7200.0,
    "max_total_spend_usd": 10.0,
    "paid_execution_authorized": true,
    "provider_spend_caps_confirmed": true,
    "required_verifier_env": [],
    "jobs_root": "r6"
  },
  "trials": [
    {
      "trial_id": "qwen--baseline_terminus2--zstd-decoder--a01",
      "provider_id": "qwen",
      "task_id": "zstd-decoder",
      "attempt": 1,
      "condition_id": "baseline_terminus2",
      "assignment_sha256": "sha256:69896f1e1928b47d98a44f1c8c542370bb0d9c5751ce01f6d8678297183dddc7"
    },
    {
      "trial_id": "qwen--akrm_closure--zstd-decoder--a01",
      "provider_id": "qwen",
      "task_id": "zstd-decoder",
      "attempt": 1,
      "condition_id": "akrm_closure",
      "assignment_sha256": "sha256:6fde8b6b123f464e057dffd805dd064b61724005905cfa11ec94794cc5ec49bb"
    },
    {
      "trial_id": "deepseek--baseline_terminus2--zstd-decoder--a01",
      "provider_id": "deepseek",
      "task_id": "zstd-decoder",
      "attempt": 1,
      "condition_id": "baseline_terminus2",
      "assignment_sha256": "sha256:c1f3dca619bc06f81c644fca5beed94704f32b29559e682411fc53990b2fdd78"
    },
    {
      "trial_id": "deepseek--akrm_closure--zstd-decoder--a01",
      "provider_id": "deepseek",
      "task_id": "zstd-decoder",
      "attempt": 1,
      "condition_id": "akrm_closure",
      "assignment_sha256": "sha256:e5e9ad89ae6feb587a922fbf04623a61d108e15a648aef4223726483d8365663"
    }
  ],
  "analysis": {
    "primary_endpoint": "binary verifier reward == 1",
    "provider_estimator": "resolved tasks / attempted tasks",
    "comparison": "paired task-attempt contrast by provider; exact McNemar and token deltas are descriptive",
    "missingness": "errors, timeouts, missing rewards and cancelled trials count as unresolved",
    "multiplicity": "per-task partial metrics are secondary and cannot replace binary reward",
    "minimum_confirmatory_attempts": 5,
    "attempts_are_independent_repeats_not_independent_tasks": true
  },
  "claim_boundary": {
    "evidence_class": "public_engineering_benchmark",
    "replaces_internal_100_instance_suite": false,
    "satisfies_gate_b_historical_customer_cases": false,
    "satisfies_customer_outcome_evidence": false,
    "authorizes_production_or_roi_claims": false,
    "authorizes_commercial_ga": false,
    "statement": "This paired experiment estimates only the AKRM bounded-state mechanism effect on the pinned public SWE-Marathon corpus under the same model and task budgets. Public tasks are not customer incidents, do not establish AGI or asymptotic complexity, and cannot authorize customer-outcome, production, ROI, or commercial-GA claims."
  },
  "campaign_hash": "sha256:2b10e08492f8da8ff20722a6f6c0e5864d13aebeebbeca50ac7e2fc104b7859a",
  "conditions": [
    {
      "condition_id": "baseline_terminus2",
      "role": "baseline",
      "agent_config_name": "akrm_code_incident.harbor_baseline_v4:ControlledTerminus2BaselineV4Agent",
      "agent_result_name": "controlled-terminus-2-v4",
      "enable_summarize": false,
      "proactive_summarization_threshold": 0,
      "bounded_chat_history": false,
      "audit_required": false,
      "metering_audit_required": true,
      "mechanism": "native growing chat with frozen JSON output, output cap, no summary LLM and audited Windows-safe terminal transport",
      "implementation_sha256": "sha256:843892e2ad9899da0bf2a756344610dc9982108906135be39cedc1cda0e643e4",
      "transport_implementation_sha256": "sha256:04873eff5ac687c3550f4cc3c904bd03bace07f928d33afda0d7a07e34cd100a",
      "audit_schema_version": "closureops.swe_marathon_baseline_agent_audit.v4",
      "structured_json_output": true,
      "max_output_tokens": 4096,
      "provider_usage_collector_implementation_sha256": "sha256:208b4a7a1c47ba33af778236cab36e4e80525976354db4fc4c5962a1d043d8eb",
      "provider_usage_collector_revision": "harbor-provider-usage-v1",
      "provider_usage_trace": "provider_usage_trace.jsonl",
      "harbor_litellm_source_sha256": "sha256:8faa5941aee862bdb9c020874ec721e348165a9b907c5a3509381a2ab50ec5a8",
      "maximum_output_length_retries": 20,
      "require_complete_provider_usage_trace": true,
      "output_length_retry_cap_implementation_sha256": "sha256:248b7160fcd5344e94a94478ab155f744b498c81f945ab00ce91a7d67c0bf8b5",
      "output_length_retry_cap_revision": "harbor-output-length-retry-cap-v1",
      "maximum_total_output_length_retries": 20,
      "require_runtime_output_length_retry_cap": true,
      "transport_revision": "harbor-0.17.1-windows-paste-v1",
      "tmux_source_sha256": "sha256:466ff8051c7c4d253436b7803aeb34ce84e250c0721a7862fd563043edd5320f",
      "upstream_paste_base64_chunk_chars": 65536,
      "effective_paste_base64_chunk_chars": 8192
    },
    {
      "condition_id": "akrm_closure",
      "role": "treatment",
      "agent_config_name": "akrm_code_incident.harbor_agent_v2r5:AKRMClosureCacheAwareR5Agent",
      "agent_result_name": "akrm-closure-cache-aware-r5",
      "enable_summarize": false,
      "proactive_summarization_threshold": 0,
      "bounded_chat_history": true,
      "audit_required": true,
      "mechanism": "bounded V26 StatePack with stable-anchor cache epochs, bounded observation deltas, bounded chat and a hard per-call output cap",
      "controller_version": "akrm-closure-cache-aware-v2",
      "implementation_sha256": "sha256:967cff3fd8a09e004829098ca8fc441823864acc33e2aaeb39055d0413401c4e",
      "runtime_file_count": 76,
      "runtime_tree_sha256": "sha256:d523c9399d38f4a2bfd8e18d1809764e1b00525fafac48d2d93a9ff75c49e097",
      "statepack_budget_tokens": 1024,
      "max_selected_atoms": 24,
      "max_working_atoms": 64,
      "statepack_control": "normal",
      "statepack_seed": 0,
      "cache_epoch_turns": 4,
      "delta_budget_tokens": 384,
      "full_prompt_overhead_tokens": 128,
      "max_dynamic_prompt_tokens": 1152,
      "max_chat_messages": 11,
      "max_output_tokens": 4096,
      "base_implementation_sha256": "sha256:418fafb4364a0bd421bdfec2ce849b782458562d745a92bbc3650dbf8608f533",
      "audit_schema_version": "closureops.swe_marathon_akrm_agent_audit.v7",
      "stable_system_anchor": true,
      "bounded_cache_epochs": true,
      "v1_implementation_sha256": "sha256:972a272816def6d9212de2c6286981c8e3e4ba0b3e7f1745b9b1122c7a4b8cd3",
      "structured_json_output": true,
      "v2_implementation_sha256": "sha256:a4da1d1d306ddd314d9457d5fd6240cb768bd37d64ce20d1dfc7e49b611189aa",
      "transport_implementation_sha256": "sha256:04873eff5ac687c3550f4cc3c904bd03bace07f928d33afda0d7a07e34cd100a",
      "transport_revision": "harbor-0.17.1-windows-paste-v1",
      "tmux_source_sha256": "sha256:466ff8051c7c4d253436b7803aeb34ce84e250c0721a7862fd563043edd5320f",
      "upstream_paste_base64_chunk_chars": 65536,
      "effective_paste_base64_chunk_chars": 8192,
      "provider_usage_collector_implementation_sha256": "sha256:208b4a7a1c47ba33af778236cab36e4e80525976354db4fc4c5962a1d043d8eb",
      "provider_usage_collector_revision": "harbor-provider-usage-v1",
      "provider_usage_trace": "provider_usage_trace.jsonl",
      "harbor_litellm_source_sha256": "sha256:8faa5941aee862bdb9c020874ec721e348165a9b907c5a3509381a2ab50ec5a8",
      "maximum_output_length_retries": 20,
      "require_complete_provider_usage_trace": true,
      "output_length_retry_cap_implementation_sha256": "sha256:248b7160fcd5344e94a94478ab155f744b498c81f945ab00ce91a7d67c0bf8b5",
      "output_length_retry_cap_revision": "harbor-output-length-retry-cap-v1",
      "maximum_total_output_length_retries": 20,
      "require_runtime_output_length_retry_cap": true
    }
  ],
  "experiment_design": {
    "kind": "paired_agent_mechanism",
    "pairing_unit": [
      "provider_id",
      "task_id",
      "attempt"
    ],
    "primary_contrast": "akrm_closure minus baseline_terminus2",
    "same_model_within_pair": true,
    "same_task_checkout_within_pair": true,
    "same_environment_within_pair": true,
    "same_temperature_within_pair": true,
    "same_max_turns_within_pair": true,
    "order_policy": "baseline job then treatment job; task order frozen by Harbor lock",
    "mechanism_claim_only": true,
    "revision": "cache-aware-v2-r5-runtime-retry-cap-controlled-pair",
    "predecessor_artifact_sha256": {
      "campaign": "sha256:dbe803970642d3dbfa98a1d2fac3489cc9509c411f991ab977db5246d40d2342",
      "pricing": "sha256:baf9aa0f042027a485d3484a2f2e4159bd6e7fdba15f115bd9aa4875d02ae61a",
      "spend_ledger": "sha256:6469845c25bb3d2f101874f5f9ec73e11c4b7c17a394160e96b7712a1bed8db8",
      "partial_evidence": "sha256:afe456c9dd9540eff6617ea362012313960ad38dd8af1ee23143fb625fc09aac",
      "stopped_report": "sha256:d593b591178da583455627993da9b067a70c831590402ee55b67d6209b403851"
    },
    "canary_pass_gates": {
      "required_clean_trials": 4,
      "required_verified_treatment_audits": 2,
      "reward_not_worse_each_provider": true,
      "minimum_input_token_reduction_rate_each_provider": 0.5,
      "maximum_single_call_input_tokens": 12000,
      "deepseek_cache_miss_input_not_worse": true,
      "deepseek_provider_reported_cost_not_worse": true,
      "deepseek_output_tokens_not_worse": true,
      "qwen_actual_cost_required_for_actual_cost_claim": true,
      "required_verified_baseline_audits": 2,
      "maximum_output_length_retries_per_trial": 20,
      "provider_usage_metering_complete": true,
      "provider_usage_trace_required": true,
      "equal_structured_json_output": true,
      "equal_max_output_tokens": true,
      "equal_summary_policy": true,
      "equal_transport": true,
      "equal_provider_usage_collector": true,
      "runtime_output_length_retry_cap_required": true
    }
  }
}