fix(harbor): handle absent pi streams and record audited rerun

benchmarks/harbor/README.md

@@ -192,6 +192,12 @@ preserves all 34 attempts, the supervision amendment, two unscored legacy
 timeouts, public-reference exposure, and partial accounting. It does not claim
 a complete, uncontaminated 34-task pass rate.
 
+The authorized [three-task Pi rerun](reports/en/pi-deepseek-flash-dev3-rerun-20261102.md)
+recorded two official passes and one scored failure after confirmed timeout
+cancellation. Both new passes accessed public reference solutions and tests;
+neither is an independent solution. Original outcomes remain unchanged. The
+rerun also fixed Pi installation checks when Harbor returns absent output streams.
+
 ### DeepSeek Harness baselines
 
 `harbor_adapter:DeepSeekHarnessAgent` runs the upstream Python SDK's

benchmarks/harbor/README.zh-CN.md

@@ -151,6 +151,11 @@ Docker 空间受限时，可以通过 `batch_images` 计划按批恢复已注册
 保留了全部 34 次尝试、进程托管修正、两次不评分的旧超时、公开参考材料接触和
 不完整用量统计，不宣称完整且无答案接触的 34 题通过率。
 
+经授权的[三题 Pi 重跑](reports/zh-CN/pi-deepseek-flash-dev3-rerun-20261102.md)
+记录了两次官方通过，以及一次超时后确认停止的计分失败。两次新的通过都访问了
+公开参考解法和测试，均不属于独立解题。原始结果保持不变。本次重跑还修复了
+Harbor 返回缺失输出流时 Pi 安装检查失败的问题。
+
 ### DeepSeek Harness baseline
 
 `harbor_adapter:DeepSeekHarnessAgent` 使用上游 Python SDK 的 `sdk-minimal`

benchmarks/harbor/configs/pi-deepseek-flash-dev3-rerun-20261102/README.md

@@ -0,0 +1,94 @@
+# Authorized three-task Pi rerun
+
+Status: completed and audited on November 2, 2026. **Two official passes, both
+reference-exposed, and one scored failure after confirmed timeout cancellation;
+zero unscored attempts.** WAL and pipeline both read reference solutions and
+tests whose reconstructed bytes match the registered task hashes. Sampler failed
+all nine official tests after the supervisor stopped 11 processes. These results
+provide no independent passing solution from the three reruns.
+
+See the complete [English report](../../reports/en/pi-deepseek-flash-dev3-rerun-20261102.md),
+[Simplified Chinese report](../../reports/zh-CN/pi-deepseek-flash-dev3-rerun-20261102.md),
+and [structured results](../../results/pi-deepseek-flash-dev3-rerun-20261102.json).
+[Preflight evidence](preflight-evidence.json) and [launch evidence](launch-evidence.json)
+precede the model jobs. The [freeze manifest](outputs-freeze.json) covers 326
+files before trajectory review; all 1,449 original frozen artifacts remain
+unchanged. WAL exposed the credential in two private raw logs. All three official
+verifiers completed on their first attempt. No containers remain running;
+the six initially protected images remain present. See
+[validation evidence](validation-evidence.json).
+
+The maintainer requested one fresh Pi attempt on `adaptive-rejection-sampler`,
+`db-wal-recovery`, and `torch-pipeline-parallelism`. The first two attempts in the
+[original baseline](../../reports/en/pi-deepseek-flash-dev34-20261024.md) were
+unscored because their timeout boundaries were not confirmed. The original
+pipeline attempt passed after reading public reference tests and a solution.
+This is an outcome-selected development supplement, not an independent benchmark.
+Keep every original outcome and every new attempt separately; do not select the
+better score.
+
+The [v2 registration](experiment-v2.json) preserves Pi 1.4.0 and its complete
+distribution digest, the original public model configuration,
+`deepseek/deepseek-flash`, the official OpenAI-compatible endpoint, `high`
+reasoning, native prompts/tools/compaction/retries, and unset temperature.
+The alias is mutable; identical backend weights across dates are not established.
+All three tasks retain their original revisions, files, Docker image identities,
+resources, 900-second agent limit, and 900-second per-attempt verifier limit.
+Setup has its separate 1,800-second limit. Concurrency is two with pair barriers,
+in the original relative task order. There are no whole-trial retries.
+
+The registered standard adapter source is
+`f3a171fcbc390c6bcb0c941be6b9cdd9dc7c456c`. Each job snapshots and hashes it through
+`python -m harbor_adapter run --config ...`. Confirmed cancellation and mandatory
+agent quiescence protect the agent/verifier boundary. Only verifier timeouts may
+retry, at most three attempts on the same output with the original time limit.
+Setup errors and verifier errors remain distinct from scored failures.
+
+The original registration used `46c554c2cf31a2d93bb37493083aab5b8fb82012`.
+All three checks failed during Pi model-registration validation because Harbor
+returned `None` for empty stderr. [Initial evidence](preflight-v1-evidence.json)
+preserves those setup failures and confirms zero model or verifier executions.
+The v2 source normalizes absent output streams while retaining rejection of
+diagnostics and missing provider/model rows. Three regression cases failed
+before the fix; all 304 Harbor tests passed after it. The initial model queue
+remains unexecuted; v2 has distinct job names and orchestration records.
+
+Installation checks precede all model work. Pipeline now has the repository's
+reviewed `torch-pipeline-parallelism-v1` dependency profile. Sampler and WAL
+explicitly remain `not_configured`; successful Pi installation does not certify
+their verifier dependencies. Existing verified APT, uv/Python-release, and PyPI
+wheel caches preserve original packages and sources. The PyPI transport checks
+each wheel's filename, size, SHA256, and original URL against current release
+metadata. `UV_HTTP_TIMEOUT=600` permits verification before serving cold wheels;
+task-native limits remain unchanged. Cache events have a separate audit log.
+
+Each task starts a fresh container and Pi session. Earlier trajectories, failure
+diagnostics, tests, and solutions are not supplied to the agent. No new prompt or
+network restriction is introduced. Freeze all new outputs before reviewing new
+trajectories or hidden assertions, then audit reference-material access. A rerun
+alone does not establish absence of answer exposure. Keep raw logs private and
+publish complete English and Simplified Chinese reports with the same evidence.
+
+The following commands document the original launch from its pinned source
+revision. Completed job names and registrations are immutable. A future repeat
+needs a new registration and fresh job names, with its source checkout explicitly
+pinned; do not reuse this completed queue from a newer repository revision.
+
+```bash
+export UV_PROJECT_ENVIRONMENT="$PWD/jobs/dsh-harbor-venv"
+export UV_CACHE_DIR="$PWD/.cache/uv"
+uv run --frozen --project benchmarks/harbor \
+  python benchmarks/harbor/scripts/run_comparison.py \
+  --plan benchmarks/harbor/configs/pi-deepseek-flash-dev3-rerun-20261102/experiment-v2.json \
+  --phase preflight
+# Supply the authorized DeepSeek credential through DEEPSEEK_API_KEY.
+uv run --frozen --project benchmarks/harbor \
+  python benchmarks/harbor/scripts/run_comparison.py \
+  --plan benchmarks/harbor/configs/pi-deepseek-flash-dev3-rerun-20261102/experiment-v2.json \
+  --phase model
+```
+
+Native job outputs are under `jobs/pi-dev3-rerun-*-20261102/`, with immutable
+inputs in the corresponding `-input/` directories. Orchestration records are in
+`jobs/pi-deepseek-flash-dev3-rerun-20261102-v2-orchestration/`, with initial failed
+checks retained under `jobs/pi-deepseek-flash-dev3-rerun-20261102-orchestration/`.

benchmarks/harbor/configs/pi-deepseek-flash-dev3-rerun-20261102/adaptive-rejection-sampler.json

@@ -0,0 +1,39 @@
+{
+  "n_concurrent_trials": 1,
+  "n_attempts": 1,
+  "retry": {
+    "max_retries": 0
+  },
+  "agents": [
+    {
+      "name": "harbor_adapter:PiCodeAgent",
+      "model_name": "deepseek/deepseek-flash",
+      "override_setup_timeout_sec": 1800,
+      "kwargs": {
+        "distribution_path": "/home/minix/.local/share/mise/installs/pi/1.4.0/pi",
+        "distribution_sha256": "60a810111c1fd9b34b59a815fb6399c6b861017e1a2278ae881bb2a563177dd9",
+        "model_config_path": "/mnt/ubuntu/home/omarchy_data/nanoPyCodeAgent/benchmarks/harbor/configs/pi-deepseek-flash-dev3-rerun-20261102/pi-models.json",
+        "thinking": "high"
+      }
+    }
+  ],
+  "environment": {
+    "type": "docker",
+    "extra_docker_compose": [
+      "/mnt/ubuntu/home/omarchy_data/nanoPyCodeAgent/benchmarks/harbor/configs/pi-deepseek-flash-dev3-rerun-20261102/dependency-cache.compose.json"
+    ]
+  },
+  "tasks": [
+    {
+      "name": "terminal-bench/adaptive-rejection-sampler",
+      "ref": "sha256:bcaa2399985cd57666018025846289ab25e193ae0dd8fb7f0ffab2410c24d4de"
+    }
+  ],
+  "verifier": {
+    "import_path": "harbor_adapter:RetryingVerifier",
+    "kwargs": {
+      "max_attempts": 3,
+      "require_agent_quiescence": true
+    }
+  }
+}

benchmarks/harbor/configs/pi-deepseek-flash-dev3-rerun-20261102/apt-cache.conf

@@ -0,0 +1,8 @@
+// Route only public APT HTTP origins through the verified index cache.
+Acquire::http::Proxy::deb.debian.org "http://127.0.0.1:18742/";
+Acquire::http::Proxy::security.debian.org "http://127.0.0.1:18742/";
+Acquire::http::Proxy::archive.debian.org "http://127.0.0.1:18742/";
+Acquire::http::Proxy::archive.ubuntu.com "http://127.0.0.1:18742/";
+Acquire::http::Proxy::security.ubuntu.com "http://127.0.0.1:18742/";
+Acquire::http::Proxy::ports.ubuntu.com "http://127.0.0.1:18742/";
+Acquire::http::Proxy::old-releases.ubuntu.com "http://127.0.0.1:18742/";

benchmarks/harbor/configs/pi-deepseek-flash-dev3-rerun-20261102/db-wal-recovery.json

@@ -0,0 +1,39 @@
+{
+  "n_concurrent_trials": 1,
+  "n_attempts": 1,
+  "retry": {
+    "max_retries": 0
+  },
+  "agents": [
+    {
+      "name": "harbor_adapter:PiCodeAgent",
+      "model_name": "deepseek/deepseek-flash",
+      "override_setup_timeout_sec": 1800,
+      "kwargs": {
+        "distribution_path": "/home/minix/.local/share/mise/installs/pi/1.4.0/pi",
+        "distribution_sha256": "60a810111c1fd9b34b59a815fb6399c6b861017e1a2278ae881bb2a563177dd9",
+        "model_config_path": "/mnt/ubuntu/home/omarchy_data/nanoPyCodeAgent/benchmarks/harbor/configs/pi-deepseek-flash-dev3-rerun-20261102/pi-models.json",
+        "thinking": "high"
+      }
+    }
+  ],
+  "environment": {
+    "type": "docker",
+    "extra_docker_compose": [
+      "/mnt/ubuntu/home/omarchy_data/nanoPyCodeAgent/benchmarks/harbor/configs/pi-deepseek-flash-dev3-rerun-20261102/dependency-cache.compose.json"
+    ]
+  },
+  "tasks": [
+    {
+      "name": "terminal-bench/db-wal-recovery",
+      "ref": "sha256:07e9193513bfce038b9f9d84962ecc85683eee0a642d14ee6fb7104b086ddab4"
+    }
+  ],
+  "verifier": {
+    "import_path": "harbor_adapter:RetryingVerifier",
+    "kwargs": {
+      "max_attempts": 3,
+      "require_agent_quiescence": true
+    }
+  }
+}

benchmarks/harbor/configs/pi-deepseek-flash-dev3-rerun-20261102/dependency-cache.compose.json

@@ -0,0 +1,148 @@
+{
+  "services": {
+    "main": {
+      "volumes": [
+        {
+          "type": "bind",
+          "source": "/mnt/ubuntu/home/omarchy_data/nanoPyCodeAgent/benchmarks/harbor/configs/pi-deepseek-flash-dev3-rerun-20261102/apt-cache.conf",
+          "target": "/etc/apt/apt.conf.d/99nanopy-index-cache",
+          "read_only": true
+        }
+      ],
+      "environment": {
+        "UV_INSTALLER_GITHUB_BASE_URL": "http://127.0.0.1:18743",
+        "UV_PYTHON_INSTALL_MIRROR": "http://127.0.0.1:18743/astral-sh/python-build-standalone/releases/download",
+        "UV_HTTP_TIMEOUT": "600",
+        "UV_DEFAULT_INDEX": "http://127.0.0.1:18744/simple"
+      }
+    },
+    "apt-index-cache": {
+      "image": "alexgshaw/pytorch-model-cli@sha256:c996884b5fb5fb2c0c07d67da2c4f974f271f07234ed28e1a868240375c57a67",
+      "network_mode": "service:main",
+      "entrypoint": [
+        "python3",
+        "/proxy/apt_index_proxy.py",
+        "--bind",
+        "127.0.0.1",
+        "--port",
+        "18742",
+        "--cache",
+        "/cache",
+        "--workers",
+        "16"
+      ],
+      "user": "1000:1000",
+      "read_only": true,
+      "cap_drop": [
+        "ALL"
+      ],
+      "security_opt": [
+        "no-new-privileges:true"
+      ],
+      "mem_limit": "256m",
+      "cpus": 0.5,
+      "volumes": [
+        "/mnt/ubuntu/home/omarchy_data/nanoPyCodeAgent/benchmarks/harbor/scripts/apt_index_proxy.py:/proxy/apt_index_proxy.py:ro",
+        "/mnt/ubuntu/home/omarchy_data/nanoPyCodeAgent/benchmarks/harbor/scripts/cache_image.py:/proxy/cache_image.py:ro",
+        "/mnt/ubuntu/home/omarchy_data/nanoPyCodeAgent/.cache/nanopy-harbor/apt-indexes:/cache"
+      ],
+      "healthcheck": {
+        "test": [
+          "CMD",
+          "python3",
+          "-c",
+          "import socket; socket.create_connection((\"127.0.0.1\",18742), timeout=2).close()"
+        ],
+        "interval": "1s",
+        "timeout": "3s",
+        "retries": 30
+      }
+    },
+    "uv-release-cache": {
+      "image": "alexgshaw/pytorch-model-cli@sha256:c996884b5fb5fb2c0c07d67da2c4f974f271f07234ed28e1a868240375c57a67",
+      "network_mode": "service:main",
+      "entrypoint": [
+        "python3",
+        "/proxy/uv_release_cache.py",
+        "serve",
+        "--cache",
+        "/cache",
+        "--manifest",
+        "/manifest.json"
+      ],
+      "user": "1000:1000",
+      "read_only": true,
+      "cap_drop": [
+        "ALL"
+      ],
+      "security_opt": [
+        "no-new-privileges:true"
+      ],
+      "cpus": 0.5,
+      "mem_limit": "256m",
+      "volumes": [
+        "/mnt/ubuntu/home/omarchy_data/nanoPyCodeAgent/benchmarks/harbor/scripts/uv_release_cache.py:/proxy/uv_release_cache.py:ro",
+        "/mnt/ubuntu/home/omarchy_data/nanoPyCodeAgent/benchmarks/harbor/scripts/cache_image.py:/proxy/cache_image.py:ro",
+        "/mnt/ubuntu/home/omarchy_data/nanoPyCodeAgent/.cache/nanopy-harbor/uv-releases:/cache:ro",
+        "/mnt/ubuntu/home/omarchy_data/nanoPyCodeAgent/benchmarks/harbor/configs/pi-deepseek-flash-dev3-rerun-20261102/dependency-release-manifest.json:/manifest.json:ro"
+      ],
+      "healthcheck": {
+        "test": [
+          "CMD",
+          "python3",
+          "-c",
+          "import socket; socket.create_connection((\"127.0.0.1\",18743), timeout=2).close()"
+        ],
+        "interval": "1s",
+        "timeout": "3s",
+        "retries": 30
+      }
+    },
+    "pypi-wheel-cache": {
+      "image": "alexgshaw/pytorch-model-cli@sha256:c996884b5fb5fb2c0c07d67da2c4f974f271f07234ed28e1a868240375c57a67",
+      "network_mode": "service:main",
+      "entrypoint": [
+        "python3",
+        "/proxy/pypi_wheel_cache.py",
+        "--cache",
+        "/cache",
+        "--bind",
+        "127.0.0.1",
+        "--port",
+        "18744",
+        "--workers",
+        "16",
+        "--downloads",
+        "2",
+        "--audit-log",
+        "/cache/audit/pi-dev3-rerun-20261102.jsonl"
+      ],
+      "user": "1000:1000",
+      "read_only": true,
+      "cap_drop": [
+        "ALL"
+      ],
+      "security_opt": [
+        "no-new-privileges:true"
+      ],
+      "mem_limit": "512m",
+      "cpus": 1.0,
+      "volumes": [
+        "/mnt/ubuntu/home/omarchy_data/nanoPyCodeAgent/benchmarks/harbor/scripts/pypi_wheel_cache.py:/proxy/pypi_wheel_cache.py:ro",
+        "/mnt/ubuntu/home/omarchy_data/nanoPyCodeAgent/benchmarks/harbor/scripts/cache_image.py:/proxy/cache_image.py:ro",
+        "/mnt/ubuntu/home/omarchy_data/nanoPyCodeAgent/.cache/nanopy-harbor/pypi-ranges:/cache"
+      ],
+      "healthcheck": {
+        "test": [
+          "CMD",
+          "python3",
+          "-c",
+          "import socket; socket.create_connection((\"127.0.0.1\",18744), timeout=2).close()"
+        ],
+        "interval": "1s",
+        "timeout": "3s",
+        "retries": 30
+      }
+    }
+  }
+}