Datasets:

noodcon
/

TerminalBench-DockerLess

+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.avro filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.lz4 filter=lfs diff=lfs merge=lfs -text
+*.mds filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text
+# Audio files - uncompressed
+*.pcm filter=lfs diff=lfs merge=lfs -text
+*.sam filter=lfs diff=lfs merge=lfs -text
+*.raw filter=lfs diff=lfs merge=lfs -text
+# Audio files - compressed
+*.aac filter=lfs diff=lfs merge=lfs -text
+*.flac filter=lfs diff=lfs merge=lfs -text
+*.mp3 filter=lfs diff=lfs merge=lfs -text
+*.ogg filter=lfs diff=lfs merge=lfs -text
+*.wav filter=lfs diff=lfs merge=lfs -text
+# Image files - uncompressed
+*.bmp filter=lfs diff=lfs merge=lfs -text
+*.gif filter=lfs diff=lfs merge=lfs -text
+*.png filter=lfs diff=lfs merge=lfs -text
+*.tiff filter=lfs diff=lfs merge=lfs -text
+# Image files - compressed
+*.jpg filter=lfs diff=lfs merge=lfs -text
+*.jpeg filter=lfs diff=lfs merge=lfs -text
+*.webp filter=lfs diff=lfs merge=lfs -text
+# Video files - compressed
+*.mp4 filter=lfs diff=lfs merge=lfs -text
+*.webm filter=lfs diff=lfs merge=lfs -text

README.md ADDED Viewed

	@@ -0,0 +1,174 @@

+---
+license: mit
+language:
+- en
+configs:
+- config_name: tbd
+  data_files: TerminalBench-DockerLess.jsonl
+tags:
+- terminal
+- bench
+- dockerless
+---
+# TerminalBench-DockerLess
+Terminal-Bench tasks are directories (instruction, environment, solution, tests)
+executed inside a Docker image. This is the same task model with the image
+replaced by a setup script, so a trial needs only `bash`, `git` and `python` —
+no daemon, no network, no build step.
+Rows carry the classic Terminal-Bench `task.yaml` fields, a rendered
+`task.toml` in the Harbor 1.4 schema, and the whole task directory inline. One
+extra `dockerless` object says how to build the environment without Docker.
+```
+TerminalBench-DockerLess.jsonl   4 tasks, 52.5 KiB (~13.4k tokens)
+build_dataset.py                 flattens tasks/ into rows, then verifies the file
+run_harness.py                   trial runner: --solution, --baseline, --predictions
+tasks/                           the four task directories (source of truth)
+tools/validate.py                schema, leakage and non-triviality checks
+tools/make_controls.py           builds the partial-credit negative control
+tools/mk_access_log.py           regenerates the access.log fixture
+reports/                         harness output for solution / baseline / partial
+```
+## Tasks
+| task_id | difficulty | category | verifier tests | fails before any agent | oracle |
+| --- | --- | --- | --- | --- | --- |
+| `tbdl-access-log-report` | medium | data-processing | 10 | 10 | pass |
+| `tbdl-git-recovery` | hard | software-engineering | 10 | 9 | pass |
+| `tbdl-csv-canon` | hard | data-processing | 8 | 8 | pass |
+| `tbdl-build-repair` | hard | software-engineering | 4 | 4 | pass |
+Difficulty is kept Terminal-Bench-shaped: each task needs real terminal work
+across several steps and fails if any one part is skipped.
+- **access-log-report** — a damaged nginx-style log where the user-agent carries
+  escaped quotes, so splitting on `"` breaks; produce an exact JSON aggregate,
+  a nearest-rank p95, a malformed-line index and a gzip copy, idempotently.
+- **git-recovery** — restore a deleted module from the *right* historical
+  revision, resurrect an unreachable commit found via `git fsck`, tag a release
+  commit, ignore two untracked artefacts without deleting them, and leave a clean
+  tree without rewriting the five original commits.
+- **csv-canon** — write a byte-exact CSV→TSV canonicaliser: BOM, CRLF, RFC 4180
+  quoting, embedded newlines, `Decimal` numeric rewriting, byte-order sorting,
+  and three distinct exit codes with exact stderr lines.
+- **build-repair** — a build that dies on an unbound shell variable, then on a
+  missing output directory, then produces a report missing its last record with a
+  truncated mean; all three must be fixed without touching `data/` or `tests/`.
+## Row schema
+| field | notes |
+| --- | --- |
+| `task_id` | `tbdl-<task>` |
+| `instruction` | the full `instruction.md` text |
+| `author_name`, `author_email` | as in `task.yaml` |
+| `difficulty`, `category`, `tags` | `easy` / `medium` / `hard` |
+| `parser_name` | `pytest` |
+| `max_agent_timeout_sec`, `max_test_timeout_sec` | enforced by the harness |
+| `expert_time_estimate_min`, `junior_time_estimate_min` | author estimates |
+| `task_yaml` | the row rendered as a classic Terminal-Bench `task.yaml` |
+| `task_toml` | the row rendered as a Harbor `task.toml` (`schema_version = "1.4"`) |
+| `dockerfile` | the original image spec, kept for fidelity (not executed) |
+| `solution` | `solution/solve.sh`, the oracle |
+| `tests` | `{"test.sh": ..., "test_outputs.py": ...}` |
+| `dockerless.setup` | replaces the Dockerfile: builds the environment |
+| `dockerless.files` | workspace snapshot written before setup runs |
+| `dockerless.test_cmd` | empty = run `tests/test.sh` |
+| `dockerless.baseline_failures` | verifier tests that already fail before any agent acts |
+## How the environment is replaced
+| Terminal-Bench | here |
+| --- | --- |
+| `environment/Dockerfile` + `docker build` | `dockerless.setup` run in a fresh directory |
+| `/workspace` | `$TB_WORK` |
+| `/logs/verifier/reward.txt` | the harness computes reward from the pytest run |
+| `tests/test.sh` | same file, run with `cwd=$TB_WORK` |
+Each trial gets a throwaway root:
+```
+$TB_ROOT/
+    workspace/   agent cwd
+    home/        HOME
+    tmp/         TMPDIR
+    bin/         prepended to PATH (contains a python3 shim)
+    tests/       verifier, uploaded after the agent finishes
+```
+## Running it
+```bash
+python -m pip install pytest          # the only dependency
+python build_dataset.py               # rebuild + verify (oracle 4/4, baseline 0/4)
+python run_harness.py --solution      # oracle, expect 4/4
+python run_harness.py --baseline      # no agent, expect 0/4
+python run_harness.py --predictions predictions/partial.jsonl
+```
+A prediction is `{"task_id", "agent_script"}`: a shell script run in the
+workspace, which is the Docker-free stand-in for "the agent did its work".
+`resolved` means every verifier test passed (reward `1.0`), the Terminal-Bench
+rule.
+## Verified results
+```
+$ python run_harness.py --solution
+tbdl-access-log-report   1.0     10/10
+tbdl-git-recovery        1.0     10/10
+tbdl-csv-canon           1.0     8/8
+tbdl-build-repair        1.0     4/4
+resolved 4/4 (100.0%)
+$ python run_harness.py --baseline            -> 0/4
+$ python run_harness.py --predictions predictions/partial.jsonl
+tbdl-access-log-report   0.0     9/10    p95 off by one rank
+tbdl-git-recovery        0.0     5/10    wrong revision, no wip rescue
+tbdl-csv-canon           0.0     5/8     BOM kept, rows unsorted
+tbdl-build-repair        0.0     1/4     shell fixed, Python faults left
+resolved 0/4 (0.0%)
+```
+The partial control is the important one: each attempt is a *plausible* half-fix
+and each still scores zero, while the per-test counts show the verifier is
+grading the whole task rather than one easy assertion.
+`build_dataset.py` refuses to write a row when a task already passes with no
+agent at all, and it re-runs the finished JSONL through the harness in both
+modes, so the file cannot drift from its verification.
+## Budget
+Measured for `max_tokens = 2048`, `context = 240000`:
+- instruction: 1.3k–2.4k characters (~350–600 tokens) per task
+- whole dataset: 52.5 KiB ≈ 13.4k tokens, so all four rows fit in one context
+- each task is solvable with many short turns: the oracle scripts are 40–90
+  lines total, so no single completion needs more than ~1k tokens
+## Deliberate deviations
+- **`setup.sh` instead of a Dockerfile.** The Dockerfile is still shipped in the
+  row for fidelity, but the environment is rebuilt by a script so the benchmark
+  runs on a laptop with no daemon.
+- **Host toolchain, not a pinned image.** Tasks are written against
+  `bash` + `git` + `python` + coreutils only, and the verifier avoids symlinks,
+  signals, `/proc` and permission bits so it behaves the same on Linux, macOS and
+  Windows/Git Bash.
+- **`agent_script` predictions.** A full Terminal-Bench harness drives a model
+  turn by turn; here a prediction is the script that performs the same actions,
+  which makes the benchmark reproducible without an inference loop.
+- **Static `reward`.** Multi-step tasks and `multi_step_reward_strategy` are not
+  used; every task is single-step with an all-or-nothing reward.
+## Platform note
+On Windows, `bash` on `PATH` is often the WSL launcher. The harness resolves a
+real POSIX shell (Git Bash), proves it works with a probe, and exports it to the
+verifier as `TB_BASH`; verifier code that shells out should use
+`os.environ.get("TB_BASH", "bash")`. A `python3` shim is placed in `$TB_ROOT/bin`
+so scripts with `#!/usr/bin/env python3` run unchanged.

TerminalBench-DockerLess.jsonl ADDED Viewed

	@@ -0,0 +1,4 @@

+{"task_id": "tbdl-access-log-report", "instruction": "# Access log report\n\n`logs/access.log` is a web server access log in a combined-ish format:\n\n```\n<ip> - - [<timestamp>] \"<METHOD> <path> HTTP/1.1\" <status> <bytes> \"<referer>\" \"<user-agent>\" <duration_ms>\n```\n\nSome lines are damaged. The user-agent field may contain escaped double quotes\n(`\\\"`), so the line cannot be parsed by splitting on `\"` -- quoted fields have to\nbe scanned properly. `bytes` is `-` when the server did not record a size.\n\nWrite **`tools/report.sh`** in the workspace root. Running\n\n```\nbash tools/report.sh\n```\n\nfrom the workspace root must read `logs/access.log` and (re)generate the\nfollowing under `out/`, exit 0, and print nothing that matters. It must be\nidempotent: deleting `out/` and running it again must produce byte-identical\nfiles.\n\n## `out/report.json`\n\nA JSON object with exactly these keys:\n\n| key | type | meaning |\n| --- | --- | --- |\n| `total_requests` | int | number of **valid** lines |\n| `unique_paths` | int | number of distinct normalised paths |\n| `status_counts` | object | `{\"<status>\": <count>}`, keys are strings; only statuses that occur |\n| `top_paths` | array | up to 3 objects `{\"path\": str, \"count\": int}`, highest count first, ties broken by path ascending |\n| `bytes_total` | int | sum of `bytes` over valid lines, `-` counts as 0 |\n| `p95_ms` | int | 95th percentile of `duration_ms`, nearest-rank method |\n\n`p95_ms` is the nearest-rank percentile: sort the durations ascending, take\nelement number `ceil(0.95 * n)` counting from 1. With `n = 36` that is the 35th\nsmallest duration. If there are no valid lines, `p95_ms` is `0`.\n\nPaths are normalised by dropping everything from the first `?` onwards, so\n`/api/users?id=42` counts as `/api/users`.\n\n## `out/malformed.txt`\n\nThe 1-based line numbers of the lines that do not match the grammar above, one\nper line, ascending. A line is malformed if any of these hold: it is not the\nnine-part shape above, `status` is not exactly three digits, `bytes` is neither\n`-` nor digits, or the trailing `duration_ms` is missing or not digits.\nMalformed lines are excluded from every aggregate.\n\n## `out/access.log.gz`\n\nA gzip stream whose decompressed content is byte-for-byte identical to\n`logs/access.log`.\n\n## Notes\n\n- Only `logs/access.log` is input; do not edit it.\n- `python` is on `PATH` (`python3` is also available). Standard library only --\n  there is no network access.\n", "author_name": "WorkBuddy Benchmarks", "author_email": "benchmarks@example.invalid", "difficulty": "medium", "category": "data-processing", "tags": ["log-parsing", "text-processing", "json", "python"], "parser_name": "pytest", "max_agent_timeout_sec": 600.0, "max_test_timeout_sec": 120.0, "expert_time_estimate_min": 25, "junior_time_estimate_min": 90, "task_yaml": "# Terminal-Bench task definition\ninstruction: |\n  # Access log report\n\n  `logs/access.log` is a web server access log in a combined-ish format:\n\n  ```\n  <ip> - - [<timestamp>] \"<METHOD> <path> HTTP/1.1\" <status> <bytes> \"<referer>\" \"<user-agent>\" <duration_ms>\n  ```\n\n  Some lines are damaged. The user-agent field may contain escaped double quotes\n  (`\\\"`), so the line cannot be parsed by splitting on `\"` -- quoted fields have to\n  be scanned properly. `bytes` is `-` when the server did not record a size.\n\n  Write **`tools/report.sh`** in the workspace root. Running\n\n  ```\n  bash tools/report.sh\n  ```\n\n  from the workspace root must read `logs/access.log` and (re)generate the\n  following under `out/`, exit 0, and print nothing that matters. It must be\n  idempotent: deleting `out/` and running it again must produce byte-identical\n  files.\n\n  ## `out/report.json`\n\n  A JSON object with exactly these keys:\n\n  | key | type | meaning |\n  | --- | --- | --- |\n  | `total_requests` | int | number of **valid** lines |\n  | `unique_paths` | int | number of distinct normalised paths |\n  | `status_counts` | object | `{\"<status>\": <count>}`, keys are strings; only statuses that occur |\n  | `top_paths` | array | up to 3 objects `{\"path\": str, \"count\": int}`, highest count first, ties broken by path ascending |\n  | `bytes_total` | int | sum of `bytes` over valid lines, `-` counts as 0 |\n  | `p95_ms` | int | 95th percentile of `duration_ms`, nearest-rank method |\n\n  `p95_ms` is the nearest-rank percentile: sort the durations ascending, take\n  element number `ceil(0.95 * n)` counting from 1. With `n = 36` that is the 35th\n  smallest duration. If there are no valid lines, `p95_ms` is `0`.\n\n  Paths are normalised by dropping everything from the first `?` onwards, so\n  `/api/users?id=42` counts as `/api/users`.\n\n  ## `out/malformed.txt`\n\n  The 1-based line numbers of the lines that do not match the grammar above, one\n  per line, ascending. A line is malformed if any of these hold: it is not the\n  nine-part shape above, `status` is not exactly three digits, `bytes` is neither\n  `-` nor digits, or the trailing `duration_ms` is missing or not digits.\n  Malformed lines are excluded from every aggregate.\n\n  ## `out/access.log.gz`\n\n  A gzip stream whose decompressed content is byte-for-byte identical to\n  `logs/access.log`.\n\n  ## Notes\n\n  - Only `logs/access.log` is input; do not edit it.\n  - `python` is on `PATH` (`python3` is also available). Standard library only --\n    there is no network access.\nauthor_name: WorkBuddy Benchmarks\nauthor_email: benchmarks@example.invalid\ndifficulty: medium\ncategory: data-processing\ntags:\n  - log-parsing\n  - text-processing\n  - json\n  - python\nparser_name: pytest\nmax_agent_timeout_sec: 600.0\nmax_test_timeout_sec: 120.0\nexpert_time_estimate_min: 25\njunior_time_estimate_min: 90\n", "task_toml": "schema_version = \"1.4\"\nsource = \"TerminalBench-DockerLess\"\n\n[task]\nname = \"swedl/access-log-report\"\nauthors = [{ name = \"WorkBuddy Benchmarks\", email = \"benchmarks@example.invalid\" }]\ndescription = \"Aggregate a damaged access log into a canonical JSON report, a malformed-line index and a gzip archive.\"\nkeywords = [\"log-parsing\", \"text-processing\", \"json\", \"python\"]\n\n[metadata]\ndifficulty = \"medium\"\ncategory = \"data-processing\"\ntags = [\"log-parsing\", \"text-processing\", \"json\", \"python\"]\nexpert_time_estimate_min = 25\njunior_time_estimate_min = 90\n\n[verifier]\ntimeout_sec = 120.0\n\n[agent]\ntimeout_sec = 600.0\n\n[environment]\nnetwork_mode = \"no-network\"\nos = \"linux\"\nworkdir = \"/workspace\"\n", "dockerfile": "FROM python:3.13-slim\n\nRUN apt-get update && apt-get install -y --no-install-recommends \\\n        bash coreutils gzip git \\\n    && rm -rf /var/lib/apt/lists/*\n\nWORKDIR /workspace\nCOPY logs/ /workspace/logs/\n\nCMD [\"/bin/bash\"]\n", "solution": "#!/usr/bin/env bash\n# Oracle solution: a regex-based parser that survives escaped quotes in the\n# user-agent field, plus deterministic JSON/gzip output.\nset -euo pipefail\ncd \"$TB_WORK\"\nmkdir -p tools\n\ncat > tools/report.py <<'PYEOF'\nimport gzip\nimport json\nimport math\nimport re\nfrom collections import Counter\nfrom pathlib import Path\n\nLINE = re.compile(\n    r'^(?P<ip>\\S+) - - \\[(?P<ts>[^\\]]+)\\] '\n    r'\"(?P<method>[A-Z]+) (?P<path>\\S+) HTTP/1\\.1\" '\n    r'(?P<status>\\d{3}) (?P<size>\\d+|-) '\n    r'\"(?P<ref>(?:[^\"\\\\]|\\\\.)*)\" \"(?P<ua>(?:[^\"\\\\]|\\\\.)*)\" '\n    r'(?P<ms>\\d+)$'\n)\n\n\ndef main():\n    root = Path(__file__).resolve().parent.parent\n    log = root / \"logs\" / \"access.log\"\n    raw = log.read_bytes()\n    lines = raw.decode(\"utf-8\").split(\"\\n\")\n    if lines and lines[-1] == \"\":\n        lines.pop()\n\n    paths, statuses, durations, bad = [], [], [], []\n    total_bytes = 0\n    for number, line in enumerate(lines, 1):\n        match = LINE.match(line)\n        if match is None:\n            bad.append(number)\n            continue\n        paths.append(match.group(\"path\").split(\"?\")[0])\n        statuses.append(match.group(\"status\"))\n        size = match.group(\"size\")\n        total_bytes += 0 if size == \"-\" else int(size)\n        durations.append(int(match.group(\"ms\")))\n\n    durations.sort()\n    count = len(durations)\n    p95 = durations[math.ceil(0.95 * count) - 1] if count else 0\n\n    path_counts = Counter(paths)\n    top = sorted(path_counts.items(), key=lambda item: (-item[1], item[0]))[:3]\n\n    report = {\n        \"total_requests\": count,\n        \"unique_paths\": len(path_counts),\n        \"status_counts\": dict(sorted(Counter(statuses).items())),\n        \"top_paths\": [{\"path\": path, \"count\": hits} for path, hits in top],\n        \"bytes_total\": total_bytes,\n        \"p95_ms\": p95,\n    }\n\n    out = root / \"out\"\n    out.mkdir(parents=True, exist_ok=True)\n    with open(out / \"report.json\", \"w\", encoding=\"utf-8\", newline=\"\\n\") as handle:\n        handle.write(json.dumps(report, indent=2, sort_keys=True) + \"\\n\")\n    with open(out / \"malformed.txt\", \"w\", encoding=\"utf-8\", newline=\"\\n\") as handle:\n        handle.write(\"\".join(\"%d\\n\" % number for number in bad))\n    with open(out / \"access.log.gz\", \"wb\") as handle:\n        with gzip.GzipFile(fileobj=handle, mode=\"wb\", mtime=0) as gz:\n            gz.write(raw)\n\n\nif __name__ == \"__main__\":\n    main()\nPYEOF\n\ncat > tools/report.sh <<'SHEOF'\n#!/usr/bin/env bash\nset -euo pipefail\nhere=\"$(cd \"$(dirname \"$0\")/..\" && pwd)\"\nexec python \"$here/tools/report.py\"\nSHEOF\n\nchmod +x tools/report.sh\nbash tools/report.sh\n", "tests": {"test.sh": "#!/usr/bin/env bash\n# Verifier entrypoint. Runs the pytest suite and lets its exit code stand as the\n# reward signal (0 = solved).\nset -uo pipefail\ncd \"$TB_ROOT/tests\"\npython -m pytest -v --tb=short -p no:cacheprovider test_outputs.py\n", "test_outputs.py": "import gzip\nimport json\nimport os\nimport shutil\nimport subprocess\nfrom pathlib import Path\n\nimport pytest\n\nWORK = Path(os.environ[\"TB_WORK\"])\nBASH = os.environ.get(\"TB_BASH\", \"bash\")\n\n\ndef load_report():\n    path = WORK / \"out\" / \"report.json\"\n    assert path.is_file(), \"out/report.json was not created\"\n    return json.loads(path.read_text(encoding=\"utf-8\"))\n\n\ndef test_report_has_exactly_the_documented_keys():\n    report = load_report()\n    assert sorted(report) == [\n        \"bytes_total\",\n        \"p95_ms\",\n        \"status_counts\",\n        \"top_paths\",\n        \"total_requests\",\n        \"unique_paths\",\n    ]\n\n\ndef test_total_requests_counts_only_valid_lines():\n    assert load_report()[\"total_requests\"] == 36\n\n\ndef test_unique_paths_ignores_query_strings():\n    assert load_report()[\"unique_paths\"] == 5\n\n\ndef test_status_counts():\n    assert load_report()[\"status_counts\"] == {\"200\": 28, \"304\": 4, \"404\": 3, \"500\": 1}\n\n\ndef test_top_paths_order_and_tiebreak():\n    assert load_report()[\"top_paths\"] == [\n        {\"path\": \"/api/users\", \"count\": 12},\n        {\"path\": \"/index.html\", \"count\": 12},\n        {\"path\": \"/static/app.js\", \"count\": 6},\n    ]\n\n\ndef test_bytes_total_treats_dash_as_zero():\n    assert load_report()[\"bytes_total\"] == 14535\n\n\ndef test_p95_uses_nearest_rank():\n    assert load_report()[\"p95_ms\"] == 162\n\n\ndef test_malformed_line_numbers():\n    path = WORK / \"out\" / \"malformed.txt\"\n    assert path.is_file(), \"out/malformed.txt was not created\"\n    numbers = [int(line) for line in path.read_text(encoding=\"utf-8\").split()]\n    assert numbers == [7, 19, 28, 37]\n\n\ndef test_gzip_roundtrip_is_byte_identical():\n    original = (WORK / \"logs\" / \"access.log\").read_bytes()\n    archive = WORK / \"out\" / \"access.log.gz\"\n    assert archive.is_file(), \"out/access.log.gz was not created\"\n    assert gzip.decompress(archive.read_bytes()) == original\n\n\ndef test_script_regenerates_identical_output():\n    out = WORK / \"out\"\n    before = (out / \"report.json\").read_bytes()\n    before_bad = (out / \"malformed.txt\").read_bytes()\n    shutil.rmtree(out)\n    proc = subprocess.run(\n        [BASH, \"tools/report.sh\"],\n        cwd=str(WORK),\n        capture_output=True,\n        text=True,\n    )\n    assert proc.returncode == 0, \"tools/report.sh failed:\\n%s\\n%s\" % (proc.stdout, proc.stderr)\n    assert (out / \"report.json\").read_bytes() == before\n    assert (out / \"malformed.txt\").read_bytes() == before_bad\n"}, "dockerless": {"setup": "", "setup_timeout_sec": 120, "files": {"logs/access.log": "10.0.0.1 - - [10/Oct/2026:13:00:00 +0000] \"GET /index.html HTTP/1.1\" 200 200 \"-\" \"curl/8.5.0\" 3\n10.0.0.2 - - [10/Oct/2026:13:00:17 +0000] \"GET /index.html HTTP/1.1\" 200 213 \"https://example.com/\" \"curl/8.5.0\" 25\n10.0.0.3 - - [10/Oct/2026:13:01:34 +0000] \"GET /index.html HTTP/1.1\" 200 226 \"https://example.com/\" \"curl/8.5.0\" 52\n10.0.0.4 - - [10/Oct/2026:13:01:51 +0000] \"GET /index.html HTTP/1.1\" 200 239 \"https://example.com/\" \"curl/8.5.0\" 78\n10.0.0.5 - - [10/Oct/2026:13:02:08 +0000] \"GET /index.html HTTP/1.1\" 200 252 \"-\" \"curl/8.5.0\" 110\n10.0.0.6 - - [10/Oct/2026:13:02:25 +0000] \"GET /index.html HTTP/1.1\" 200 265 \"https://example.com/\" \"Mozilla/5.0 (\\\"compatible\\\"; bot) Gecko/20100101\" 175\n10.0.0.9 - - [10/Oct/2026:13:02:11 +0000] \"GET /index.html HTTP/1.1\" 200 512 \"-\" \"curl/8.0\"\n10.0.0.7 - - [10/Oct/2026:13:03:42 +0000] \"GET /index.html HTTP/1.1\" 200 278 \"https://example.com/\" \"curl/8.5.0\" 21\n10.0.0.8 - - [10/Oct/2026:13:03:59 +0000] \"GET /index.html HTTP/1.1\" 200 291 \"https://example.com/\" \"curl/8.5.0\" 47\n10.0.0.9 - - [10/Oct/2026:13:04:16 +0000] \"GET /index.html HTTP/1.1\" 200 304 \"-\" \"curl/8.5.0\" 74\n10.0.0.1 - - [10/Oct/2026:13:04:33 +0000] \"GET /index.html HTTP/1.1\" 200 - \"https://example.com/\" \"curl/8.5.0\" 104\n10.0.0.2 - - [10/Oct/2026:13:05:50 +0000] \"GET /index.html HTTP/1.1\" 200 330 \"https://example.com/\" \"curl/8.5.0\" 162\n10.0.0.3 - - [10/Oct/2026:13:05:07 +0000] \"GET /index.html HTTP/1.1\" 200 343 \"https://example.com/\" \"curl/8.5.0\" 18\n10.0.0.4 - - [10/Oct/2026:13:06:24 +0000] \"GET /api/users HTTP/1.1\" 200 356 \"-\" \"curl/8.5.0\" 44\n10.0.0.5 - - [10/Oct/2026:13:06:41 +0000] \"GET /api/users HTTP/1.1\" 200 369 \"https://example.com/\" \"curl/8.5.0\" 70\n10.0.0.6 - - [10/Oct/2026:13:07:58 +0000] \"GET /api/users HTTP/1.1\" 200 382 \"https://example.com/\" \"curl/8.5.0\" 99\n10.0.0.7 - - [10/Oct/2026:13:07:15 +0000] \"POST /api/users HTTP/1.1\" 200 395 \"https://example.com/\" \"curl/8.5.0\" 150\n10.0.0.8 - - [10/Oct/2026:13:08:32 +0000] \"GET /api/users HTTP/1.1\" 200 408 \"-\" \"curl/8.5.0\" 14\n10.0.0.3 - - [10/Oct/2026:13:05:44 +0000] \"POST /api/users HTTP/1.1\" abc 300 \"-\" \"curl/8.0\" 44\n10.0.0.9 - - [10/Oct/2026:13:08:49 +0000] \"GET /api/users HTTP/1.1\" 200 421 \"https://example.com/\" \"Mozilla/5.0 (\\\"compatible\\\"; bot) Gecko/20100101\" 40\n10.0.0.1 - - [10/Oct/2026:13:09:06 +0000] \"GET /api/users HTTP/1.1\" 200 434 \"https://example.com/\" \"curl/8.5.0\" 66\n10.0.0.2 - - [10/Oct/2026:13:09:23 +0000] \"GET /api/users HTTP/1.1\" 200 447 \"https://example.com/\" \"curl/8.5.0\" 95\n10.0.0.3 - - [10/Oct/2026:13:10:40 +0000] \"GET /api/users?id=42 HTTP/1.1\" 200 460 \"-\" \"curl/8.5.0\" 141\n10.0.0.4 - - [10/Oct/2026:13:10:57 +0000] \"GET /api/users?id=42 HTTP/1.1\" 200 473 \"https://example.com/\" \"curl/8.5.0\" 12\n10.0.0.5 - - [10/Oct/2026:13:11:14 +0000] \"GET /api/users?id=42 HTTP/1.1\" 200 486 \"https://example.com/\" \"curl/8.5.0\" 35\n10.0.0.6 - - [10/Oct/2026:13:11:31 +0000] \"GET /api/users?id=42 HTTP/1.1\" 200 499 \"https://example.com/\" \"curl/8.5.0\" 61\n10.0.0.7 - - [10/Oct/2026:13:12:48 +0000] \"GET /static/app.js HTTP/1.1\" 200 512 \"-\" \"curl/8.5.0\" 90\n10.0.0.4 - - [10/Oct/2026:13:09:02 +0000] \"GET /health HTTP/1.1\" 200 0\n10.0.0.8 - - [10/Oct/2026:13:12:05 +0000] \"GET /static/app.js HTTP/1.1\" 200 525 \"https://example.com/\" \"curl/8.5.0\" 133\n10.0.0.9 - - [10/Oct/2026:13:13:22 +0000] \"GET /static/app.js HTTP/1.1\" 200 - \"https://example.com/\" \"curl/8.5.0\" 11\n10.0.0.1 - - [10/Oct/2026:13:13:39 +0000] \"GET /static/app.js HTTP/1.1\" 200 551 \"https://example.com/\" \"curl/8.5.0\" 31\n10.0.0.2 - - [10/Oct/2026:13:14:56 +0000] \"GET /static/app.js HTTP/1.1\" 304 564 \"-\" \"curl/8.5.0\" 58\n10.0.0.3 - - [10/Oct/2026:13:14:13 +0000] \"GET /static/app.js HTTP/1.1\" 304 577 \"https://example.com/\" \"curl/8.5.0\" 85\n10.0.0.4 - - [10/Oct/2026:13:15:30 +0000] \"GET /health HTTP/1.1\" 304 590 \"https://example.com/\" \"curl/8.5.0\" 126\n10.0.0.5 - - [10/Oct/2026:13:15:47 +0000] \"GET /health HTTP/1.1\" 304 603 \"https://example.com/\" \"curl/8.5.0\" 7\n10.0.0.6 - - [10/Oct/2026:13:16:04 +0000] \"GET /health HTTP/1.1\" 404 616 \"-\" \"curl/8.5.0\" 28\n# rotated by logrotate at 2026-10-10T14:00:00Z\n10.0.0.7 - - [10/Oct/2026:13:16:21 +0000] \"GET /api/orders HTTP/1.1\" 404 629 \"https://example.com/\" \"curl/8.5.0\" 55\n10.0.0.8 - - [10/Oct/2026:13:17:38 +0000] \"GET /api/orders HTTP/1.1\" 404 642 \"https://example.com/\" \"curl/8.5.0\" 82\n10.0.0.9 - - [10/Oct/2026:13:17:55 +0000] \"GET /api/orders HTTP/1.1\" 500 655 \"https://example.com/\" \"curl/8.5.0\" 118\n"}, "test_cmd": "", "workdir": "workspace", "requires": ["bash", "python"], "baseline_failures": ["test_outputs.py::test_bytes_total_treats_dash_as_zero", "test_outputs.py::test_gzip_roundtrip_is_byte_identical", "test_outputs.py::test_malformed_line_numbers", "test_outputs.py::test_p95_uses_nearest_rank", "test_outputs.py::test_report_has_exactly_the_documented_keys", "test_outputs.py::test_script_regenerates_identical_output", "test_outputs.py::test_status_counts", "test_outputs.py::test_top_paths_order_and_tiebreak", "test_outputs.py::test_total_requests_counts_only_valid_lines", "test_outputs.py::test_unique_paths_ignores_query_strings"]}}
+{"task_id": "tbdl-git-recovery", "instruction": "# Recover a broken release repository\n\nThe git repository in the workspace was left in a mess after a bad afternoon.\n`main` currently has five commits, but the module `src/app.py` was deleted by the\n`Drop legacy module` commit and nobody kept a copy. An experiment that was never\nmerged lives on as an unreachable commit whose message is `wip: plan`.\n\nFix the repository. When you are done it must satisfy all of the following.\n\n## 1. Restore the module\n\n`src/app.py` must exist again with **exactly** the content it had in the commit\ntitled `Add config loader` (not the older `Initial import` version), and it must\nbe committed on `main` in a commit whose message is exactly:\n\n```\nrestore app.py\n```\n\n## 2. Rescue the abandoned experiment\n\nCreate a branch named `recovered` that points at the unreachable commit whose\nmessage is `wip: plan`. That commit must **not** become an ancestor of `main`,\nand it must still carry the work-in-progress version of `notes/plan.md`.\n\n## 3. Tag the release\n\nCreate an annotated-or-lightweight tag `v1.0` pointing at the commit titled\n`Add config loader`.\n\n## 4. Ignore the junk, don't delete it\n\nThe workspace has two untracked artefacts, `debug.log` and `build/tmp.o`. Add a\n`.gitignore` so that both are ignored, commit it, and leave the files on disk.\n\n## 5. Clean tree\n\n`git status --porcelain` must print nothing.\n\n## Constraints\n\n- Do not rewrite or discard the five existing commits on `main`; all five\n  original commit subjects must still be present in `git log main`. In\n  particular do not `reset` `main` backwards.\n- Do not delete `debug.log` or `build/tmp.o`.\n- No network access; everything you need is already in `.git`.\n", "author_name": "WorkBuddy Benchmarks", "author_email": "benchmarks@example.invalid", "difficulty": "hard", "category": "software-engineering", "tags": ["git", "forensics", "history-recovery"], "parser_name": "pytest", "max_agent_timeout_sec": 600.0, "max_test_timeout_sec": 120.0, "expert_time_estimate_min": 20, "junior_time_estimate_min": 75, "task_yaml": "# Terminal-Bench task definition\ninstruction: |\n  # Recover a broken release repository\n\n  The git repository in the workspace was left in a mess after a bad afternoon.\n  `main` currently has five commits, but the module `src/app.py` was deleted by the\n  `Drop legacy module` commit and nobody kept a copy. An experiment that was never\n  merged lives on as an unreachable commit whose message is `wip: plan`.\n\n  Fix the repository. When you are done it must satisfy all of the following.\n\n  ## 1. Restore the module\n\n  `src/app.py` must exist again with **exactly** the content it had in the commit\n  titled `Add config loader` (not the older `Initial import` version), and it must\n  be committed on `main` in a commit whose message is exactly:\n\n  ```\n  restore app.py\n  ```\n\n  ## 2. Rescue the abandoned experiment\n\n  Create a branch named `recovered` that points at the unreachable commit whose\n  message is `wip: plan`. That commit must **not** become an ancestor of `main`,\n  and it must still carry the work-in-progress version of `notes/plan.md`.\n\n  ## 3. Tag the release\n\n  Create an annotated-or-lightweight tag `v1.0` pointing at the commit titled\n  `Add config loader`.\n\n  ## 4. Ignore the junk, don't delete it\n\n  The workspace has two untracked artefacts, `debug.log` and `build/tmp.o`. Add a\n  `.gitignore` so that both are ignored, commit it, and leave the files on disk.\n\n  ## 5. Clean tree\n\n  `git status --porcelain` must print nothing.\n\n  ## Constraints\n\n  - Do not rewrite or discard the five existing commits on `main`; all five\n    original commit subjects must still be present in `git log main`. In\n    particular do not `reset` `main` backwards.\n  - Do not delete `debug.log` or `build/tmp.o`.\n  - No network access; everything you need is already in `.git`.\nauthor_name: WorkBuddy Benchmarks\nauthor_email: benchmarks@example.invalid\ndifficulty: hard\ncategory: software-engineering\ntags:\n  - git\n  - forensics\n  - history-recovery\nparser_name: pytest\nmax_agent_timeout_sec: 600.0\nmax_test_timeout_sec: 120.0\nexpert_time_estimate_min: 20\njunior_time_estimate_min: 75\n", "task_toml": "schema_version = \"1.4\"\nsource = \"TerminalBench-DockerLess\"\n\n[task]\nname = \"swedl/git-recovery\"\nauthors = [{ name = \"WorkBuddy Benchmarks\", email = \"benchmarks@example.invalid\" }]\ndescription = \"Restore a deleted module from history, resurrect an unreachable commit, tag a release and leave a clean tree.\"\nkeywords = [\"git\", \"forensics\", \"history-recovery\"]\n\n[metadata]\ndifficulty = \"hard\"\ncategory = \"software-engineering\"\ntags = [\"git\", \"forensics\", \"history-recovery\"]\nexpert_time_estimate_min = 20\njunior_time_estimate_min = 75\n\n[verifier]\ntimeout_sec = 120.0\n\n[agent]\ntimeout_sec = 600.0\n\n[environment]\nnetwork_mode = \"no-network\"\nos = \"linux\"\nworkdir = \"/workspace\"\n", "dockerfile": "FROM python:3.13-slim\n\nRUN apt-get update && apt-get install -y --no-install-recommends \\\n        bash coreutils git \\\n    && rm -rf /var/lib/apt/lists/*\n\nWORKDIR /workspace\nCOPY setup.sh /tmp/setup.sh\nRUN bash /tmp/setup.sh\n\nCMD [\"/bin/bash\"]\n", "solution": "#!/usr/bin/env bash\n# Oracle solution: find the lost commit through fsck, recover the file from the\n# right historical revision, tag, ignore and commit.\nset -euo pipefail\ncd \"$TB_WORK\"\n\ngit config user.name \"Release Bot\"\ngit config user.email \"release@example.invalid\"\n\n# 1. restore src/app.py from \"Add config loader\"\nloader_commit=\"$(git log --format='%H %s' main | awk '$0 ~ / Add config loader$/ {print $1}' | head -1)\"\ntest -n \"$loader_commit\"\nmkdir -p src\ngit show \"$loader_commit:src/app.py\" > src/app.py\n\n# 4. ignore the junk\ncat > .gitignore <<'EOF'\n*.log\nbuild/\nEOF\n\ngit add -A\ngit commit -qm \"restore app.py\"\n\n# 2. rescue the unreachable commit\nwip_commit=\"$(git fsck --unreachable --no-reflogs 2>/dev/null | awk '$2 == \"commit\" {print $3}' | while read -r sha; do\n    if [ \"$(git log -1 --format=%s \"$sha\" 2>/dev/null)\" = \"wip: plan\" ]; then\n        echo \"$sha\"\n        break\n    fi\ndone)\"\ntest -n \"$wip_commit\"\ngit branch recovered \"$wip_commit\"\n\n# 3. tag the release commit\ngit tag v1.0 \"$loader_commit\"\n\ngit status --porcelain\n", "tests": {"test.sh": "#!/usr/bin/env bash\n# Verifier entrypoint. Runs the pytest suite and lets its exit code stand as the\n# reward signal (0 = solved).\nset -uo pipefail\ncd \"$TB_ROOT/tests\"\npython -m pytest -v --tb=short -p no:cacheprovider test_outputs.py\n", "test_outputs.py": "import os\nimport subprocess\nfrom pathlib import Path\n\nWORK = Path(os.environ[\"TB_WORK\"])\n\nEXPECTED_APP = '''VERSION = 2\n\nDEFAULTS = {\"retries\": 3, \"timeout\": 30}\n\n\ndef main():\n    return \"v2\"\n\n\ndef load(path):\n    with open(path, \"r\", encoding=\"utf-8\") as handle:\n        return handle.read()\n'''\n\nORIGINAL_SUBJECTS = [\n    \"Initial import\",\n    \"Add plan notes\",\n    \"Add config loader\",\n    \"Drop legacy module\",\n    \"Document build\",\n]\n\n\ndef git(*args):\n    return subprocess.run(\n        [\"git\"] + list(args),\n        cwd=str(WORK),\n        capture_output=True,\n        text=True,\n        encoding=\"utf-8\",\n    )\n\n\ndef subjects():\n    result = git(\"log\", \"--format=%s\", \"main\")\n    assert result.returncode == 0, result.stderr\n    return [line for line in result.stdout.splitlines() if line]\n\n\ndef test_app_py_restored_from_config_loader_commit():\n    path = WORK / \"src\" / \"app.py\"\n    assert path.is_file(), \"src/app.py was not restored\"\n    assert path.read_text(encoding=\"utf-8\") == EXPECTED_APP\n\n\ndef test_restore_commit_message():\n    assert \"restore app.py\" in subjects()\n\n\ndef test_app_py_is_committed_not_just_written():\n    result = git(\"ls-files\", \"--error-unmatch\", \"src/app.py\")\n    assert result.returncode == 0, \"src/app.py is not tracked\"\n\n\ndef test_recovered_branch_points_at_the_wip_commit():\n    ref = git(\"rev-parse\", \"--verify\", \"refs/heads/recovered\")\n    assert ref.returncode == 0, \"branch 'recovered' does not exist\"\n    message = git(\"log\", \"-1\", \"--format=%s\", \"recovered\")\n    assert message.stdout.strip() == \"wip: plan\"\n\n\ndef test_recovered_is_not_an_ancestor_of_main():\n    ref = git(\"rev-parse\", \"--verify\", \"refs/heads/recovered\")\n    assert ref.returncode == 0, \"branch 'recovered' does not exist\"\n    result = git(\"merge-base\", \"--is-ancestor\", \"recovered\", \"main\")\n    assert result.returncode == 1, \"the rescued commit must stay off main\"\n\n\ndef test_recovered_commit_keeps_the_work_in_progress_plan():\n    blob = git(\"show\", \"recovered:notes/plan.md\")\n    assert blob.returncode == 0, \"recovered commit has no notes/plan.md\"\n    assert \"step two\" in blob.stdout\n\n\ndef test_tag_v1_points_at_the_config_loader_commit():\n    tag = git(\"rev-list\", \"-n\", \"1\", \"v1.0\")\n    assert tag.returncode == 0, \"tag v1.0 does not exist\"\n    subject = git(\"log\", \"-1\", \"--format=%s\", tag.stdout.strip())\n    assert subject.stdout.strip() == \"Add config loader\"\n\n\ndef test_junk_is_ignored_but_still_present():\n    for rel in (\"debug.log\", \"build/tmp.o\"):\n        assert (WORK / rel).is_file(), \"%s was deleted\" % rel\n        result = git(\"check-ignore\", \"-q\", rel)\n        assert result.returncode == 0, \"%s is not ignored\" % rel\n\n\ndef test_worktree_is_clean():\n    status = git(\"status\", \"--porcelain\")\n    assert status.stdout.strip() == \"\", \"dirty tree:\\n%s\" % status.stdout\n\n\ndef test_original_history_is_preserved():\n    current = subjects()\n    for subject in ORIGINAL_SUBJECTS:\n        assert subject in current, \"commit %r disappeared from main\" % subject\n"}, "dockerless": {"setup": "#!/usr/bin/env bash\n# Builds the broken repository state. This is the Dockerfile's analogue: it\n# constructs the environment the agent wakes up in, deterministically.\nset -euo pipefail\ncd \"$TB_WORK\"\n\ngit init -q -b main\ngit config user.name \"Release Bot\"\ngit config user.email \"release@example.invalid\"\ngit config commit.gpgsign false\ngit config core.autocrlf false\n\nmkdir -p src\ncat > src/app.py <<'EOF'\nVERSION = 1\n\n\ndef main():\n    return \"v1\"\nEOF\ngit add -A\ngit commit -qm \"Initial import\"\n\nmkdir -p notes\ncat > notes/plan.md <<'EOF'\n# Plan\n\n- step one: wire up the loader\nEOF\ngit add -A\ngit commit -qm \"Add plan notes\"\n\ncat > src/app.py <<'EOF'\nVERSION = 2\n\nDEFAULTS = {\"retries\": 3, \"timeout\": 30}\n\n\ndef main():\n    return \"v2\"\n\n\ndef load(path):\n    with open(path, \"r\", encoding=\"utf-8\") as handle:\n        return handle.read()\nEOF\ngit add -A\ngit commit -qm \"Add config loader\"\n\ngit rm -q src/app.py\ngit commit -qm \"Drop legacy module\"\n\nmkdir -p docs\ncat > docs/build.md <<'EOF'\n# Build\n\nRun `make build`.\nEOF\ngit add -A\ngit commit -qm \"Document build\"\n\n# An abandoned experiment: a commit on a branch that is then deleted, leaving\n# the commit unreachable from every ref.\ngit checkout -q -b wip HEAD~2\ncat > notes/plan.md <<'EOF'\n# Plan\n\n- step one: wire up the loader\n- step two: add retry budget (work in progress)\nEOF\ngit add -A\ngit commit -qm \"wip: plan\"\ngit checkout -q main\ngit branch -qD wip\n\n# Untracked junk that must end up ignored, not deleted.\nprintf 'starting\\n' > debug.log\nmkdir -p build\nprintf 'object file\\n' > build/tmp.o\n", "setup_timeout_sec": 120, "files": {}, "test_cmd": "", "workdir": "workspace", "requires": ["bash", "git"], "baseline_failures": ["test_outputs.py::test_app_py_is_committed_not_just_written", "test_outputs.py::test_app_py_restored_from_config_loader_commit", "test_outputs.py::test_junk_is_ignored_but_still_present", "test_outputs.py::test_recovered_branch_points_at_the_wip_commit", "test_outputs.py::test_recovered_commit_keeps_the_work_in_progress_plan", "test_outputs.py::test_recovered_is_not_an_ancestor_of_main", "test_outputs.py::test_restore_commit_message", "test_outputs.py::test_tag_v1_points_at_the_config_loader_commit", "test_outputs.py::test_worktree_is_clean"]}}
+{"task_id": "tbdl-csv-canon", "instruction": "# `canon` — canonicalise a CSV export\n\nDownstream tools need one boring, byte-stable TSV. Write an executable script at\n`bin/canon` (shebang, `chmod +x`) that is invoked as:\n\n```\n./bin/canon <input.csv> <output.tsv>\n```\n\n## Input\n\nUTF-8 CSV, possibly with a byte-order mark, possibly CRLF, RFC 4180 quoting:\nfields may be quoted, may contain commas, doubled quotes (`\"\"`) and embedded\nnewlines. The first record is the header.\n\n## Output\n\nUTF-8 **without** a BOM, `\\n` line endings, tab-separated, one line per record,\nheader first. Fields are written verbatim — never re-quoted, never trimmed —\nexcept for the numeric normalisation below. Column order is preserved.\n\n## Numeric normalisation\n\nA field is numeric when the whole field matches\n`^[+-]?(\\d+(\\.\\d*)?|\\.\\d+)([eE][+-]?\\d+)?$`. Numeric fields are rewritten in\nplain decimal notation with no exponent, no leading `+`, no leading zeros, and\nno trailing zeros after the decimal point:\n\n| input | output |\n| --- | --- |\n| `1.50` | `1.5` |\n| `1.0` | `1` |\n| `1e3` | `1000` |\n| `2.5e-3` | `0.0025` |\n| `+5` | `5` |\n| `007` | `7` |\n| `-0` | `0` |\n\nEvery other field is copied unchanged.\n\n## Ordering\n\nSort the data records by field 0, then by field 1, comparing the **normalised\nstrings** in byte order (so `\"10\"` sorts before `\"9\"`). The sort must be stable.\n\n## Errors\n\nOn any error, print exactly one line to stderr, exit with the given code, and do\n**not** create the output file:\n\n| condition | stderr line | exit |\n| --- | --- | --- |\n| input path does not exist | `no such file: <input path as given>` | `2` |\n| two header names collide case-insensitively after trimming | `duplicate column: <name as written in the later header field>` | `3` |\n| a record's field count differs from the header's | `malformed row: line <n>` | `4` |\n\n`<n>` is the 1-based physical line number where the offending record starts\n(so a record whose first line is the third line of the file reports `3`).\n\nOn success exit `0` and print nothing to stderr.\n\n## Notes\n\n- `python` and `python3` are both on `PATH`; standard library only, no network.\n- `data/orders.csv` is a sample with a BOM and CRLF endings.\n", "author_name": "WorkBuddy Benchmarks", "author_email": "benchmarks@example.invalid", "difficulty": "hard", "category": "data-processing", "tags": ["csv", "cli", "spec-compliance", "exit-codes"], "parser_name": "pytest", "max_agent_timeout_sec": 900.0, "max_test_timeout_sec": 120.0, "expert_time_estimate_min": 35, "junior_time_estimate_min": 120, "task_yaml": "# Terminal-Bench task definition\ninstruction: |\n  # `canon` — canonicalise a CSV export\n\n  Downstream tools need one boring, byte-stable TSV. Write an executable script at\n  `bin/canon` (shebang, `chmod +x`) that is invoked as:\n\n  ```\n  ./bin/canon <input.csv> <output.tsv>\n  ```\n\n  ## Input\n\n  UTF-8 CSV, possibly with a byte-order mark, possibly CRLF, RFC 4180 quoting:\n  fields may be quoted, may contain commas, doubled quotes (`\"\"`) and embedded\n  newlines. The first record is the header.\n\n  ## Output\n\n  UTF-8 **without** a BOM, `\\n` line endings, tab-separated, one line per record,\n  header first. Fields are written verbatim — never re-quoted, never trimmed —\n  except for the numeric normalisation below. Column order is preserved.\n\n  ## Numeric normalisation\n\n  A field is numeric when the whole field matches\n  `^[+-]?(\\d+(\\.\\d*)?|\\.\\d+)([eE][+-]?\\d+)?$`. Numeric fields are rewritten in\n  plain decimal notation with no exponent, no leading `+`, no leading zeros, and\n  no trailing zeros after the decimal point:\n\n  | input | output |\n  | --- | --- |\n  | `1.50` | `1.5` |\n  | `1.0` | `1` |\n  | `1e3` | `1000` |\n  | `2.5e-3` | `0.0025` |\n  | `+5` | `5` |\n  | `007` | `7` |\n  | `-0` | `0` |\n\n  Every other field is copied unchanged.\n\n  ## Ordering\n\n  Sort the data records by field 0, then by field 1, comparing the **normalised\n  strings** in byte order (so `\"10\"` sorts before `\"9\"`). The sort must be stable.\n\n  ## Errors\n\n  On any error, print exactly one line to stderr, exit with the given code, and do\n  **not** create the output file:\n\n  | condition | stderr line | exit |\n  | --- | --- | --- |\n  | input path does not exist | `no such file: <input path as given>` | `2` |\n  | two header names collide case-insensitively after trimming | `duplicate column: <name as written in the later header field>` | `3` |\n  | a record's field count differs from the header's | `malformed row: line <n>` | `4` |\n\n  `<n>` is the 1-based physical line number where the offending record starts\n  (so a record whose first line is the third line of the file reports `3`).\n\n  On success exit `0` and print nothing to stderr.\n\n  ## Notes\n\n  - `python` and `python3` are both on `PATH`; standard library only, no network.\n  - `data/orders.csv` is a sample with a BOM and CRLF endings.\nauthor_name: WorkBuddy Benchmarks\nauthor_email: benchmarks@example.invalid\ndifficulty: hard\ncategory: data-processing\ntags:\n  - csv\n  - cli\n  - spec-compliance\n  - exit-codes\nparser_name: pytest\nmax_agent_timeout_sec: 900.0\nmax_test_timeout_sec: 120.0\nexpert_time_estimate_min: 35\njunior_time_estimate_min: 120\n", "task_toml": "schema_version = \"1.4\"\nsource = \"TerminalBench-DockerLess\"\n\n[task]\nname = \"swedl/csv-canon\"\nauthors = [{ name = \"WorkBuddy Benchmarks\", email = \"benchmarks@example.invalid\" }]\ndescription = \"Write a byte-exact CSV-to-TSV canonicaliser with strict error codes and numeric normalisation.\"\nkeywords = [\"csv\", \"cli\", \"spec-compliance\", \"exit-codes\"]\n\n[metadata]\ndifficulty = \"hard\"\ncategory = \"data-processing\"\ntags = [\"csv\", \"cli\", \"spec-compliance\", \"exit-codes\"]\nexpert_time_estimate_min = 35\njunior_time_estimate_min = 120\n\n[verifier]\ntimeout_sec = 120.0\n\n[agent]\ntimeout_sec = 900.0\n\n[environment]\nnetwork_mode = \"no-network\"\nos = \"linux\"\nworkdir = \"/workspace\"\n", "dockerfile": "FROM python:3.13-slim\n\nRUN apt-get update && apt-get install -y --no-install-recommends \\\n        bash coreutils \\\n    && rm -rf /var/lib/apt/lists/*\n\nWORKDIR /workspace\nCOPY data/ /workspace/data/\n\nCMD [\"/bin/bash\"]\n", "solution": "#!/usr/bin/env bash\n# Oracle solution: csv module for RFC4180, Decimal for exact numeric rewriting.\nset -euo pipefail\ncd \"$TB_WORK\"\nmkdir -p bin\n\ncat > bin/canon <<'PYEOF'\n#!/usr/bin/env python3\nimport csv\nimport re\nimport sys\nfrom decimal import Decimal, InvalidOperation\nfrom pathlib import Path\n\nNUMERIC = re.compile(r\"^[+-]?(\\d+(\\.\\d*)?|\\.\\d+)([eE][+-]?\\d+)?$\")\n\n\ndef normalize(cell):\n    if not NUMERIC.match(cell):\n        return cell\n    try:\n        value = Decimal(cell)\n    except InvalidOperation:\n        return cell\n    if value == 0:\n        return \"0\"\n    return format(value.normalize(), \"f\")\n\n\ndef main(argv):\n    if len(argv) != 3:\n        sys.stderr.write(\"usage: canon <input.csv> <output.tsv>\\n\")\n        return 1\n    source = Path(argv[1])\n    target = Path(argv[2])\n    if not source.is_file():\n        sys.stderr.write(\"no such file: %s\\n\" % argv[1])\n        return 2\n\n    with open(source, \"r\", encoding=\"utf-8-sig\", newline=\"\") as handle:\n        reader = csv.reader(handle)\n        header = next(reader, None)\n        if header is None:\n            header = []\n        header = [name.strip() for name in header]\n\n        seen = set()\n        for name in header:\n            key = name.casefold()\n            if key in seen:\n                sys.stderr.write(\"duplicate column: %s\\n\" % name)\n                return 3\n            seen.add(key)\n\n        rows = []\n        previous_line = reader.line_num\n        for record in reader:\n            start = previous_line + 1\n            previous_line = reader.line_num\n            if not record:\n                continue\n            if len(record) != len(header):\n                sys.stderr.write(\"malformed row: line %d\\n\" % start)\n                return 4\n            rows.append([normalize(cell) for cell in record])\n\n    rows.sort(key=lambda row: (row[0], row[1] if len(row) > 1 else \"\"))\n\n    target.parent.mkdir(parents=True, exist_ok=True)\n    with open(target, \"w\", encoding=\"utf-8\", newline=\"\") as handle:\n        handle.write(\"\\t\".join(header) + \"\\n\")\n        for row in rows:\n            handle.write(\"\\t\".join(row) + \"\\n\")\n    return 0\n\n\nif __name__ == \"__main__\":\n    sys.exit(main(sys.argv))\nPYEOF\n\nchmod +x bin/canon\n./bin/canon data/orders.csv out/orders.tsv\n", "tests": {"test.sh": "#!/usr/bin/env bash\n# Verifier entrypoint. Runs the pytest suite and lets its exit code stand as the\n# reward signal (0 = solved).\nset -uo pipefail\ncd \"$TB_ROOT/tests\"\npython -m pytest -v --tb=short -p no:cacheprovider test_outputs.py\n", "test_outputs.py": "import os\nimport subprocess\nfrom pathlib import Path\n\nWORK = Path(os.environ[\"TB_WORK\"])\nBASH = os.environ.get(\"TB_BASH\", \"bash\")\nCANON = \"./bin/canon\"\n\n\ndef run_canon(source, target):\n    return subprocess.run(\n        [BASH, \"-c\", 'exec ./bin/canon \"$@\"', \"canon\", str(source), str(target)],\n        cwd=str(WORK),\n        capture_output=True,\n        text=True,\n        encoding=\"utf-8\",\n    )\n\n\ndef case_dir():\n    path = Path(os.environ[\"TB_ROOT\"]) / \"tmp\" / \"cases\"\n    path.mkdir(parents=True, exist_ok=True)\n    return path\n\n\ndef write_case(name, body, newline=\"\"):\n    path = case_dir() / name\n    with open(path, \"w\", encoding=\"utf-8\", newline=newline) as handle:\n        handle.write(body)\n    return path\n\n\ndef test_missing_input_exits_2():\n    target = case_dir() / \"missing.tsv\"\n    if target.exists():\n        target.unlink()\n    proc = run_canon(\"does-not-exist.csv\", target)\n    assert proc.returncode == 2, proc.stderr\n    assert proc.stderr.strip() == \"no such file: does-not-exist.csv\"\n    assert not target.exists(), \"output must not be created on error\"\n\n\ndef test_duplicate_column_exits_3():\n    source = write_case(\"dup.csv\", \"id,ID\\n1,2\\n\")\n    target = case_dir() / \"dup.tsv\"\n    if target.exists():\n        target.unlink()\n    proc = run_canon(source, target)\n    assert proc.returncode == 3, proc.stderr\n    assert proc.stderr.strip() == \"duplicate column: ID\"\n    assert not target.exists()\n\n\ndef test_malformed_row_exits_4_with_physical_line():\n    source = write_case(\"bad.csv\", \"id,sku\\n1,a\\n2,b,c\\n\")\n    target = case_dir() / \"bad.tsv\"\n    if target.exists():\n        target.unlink()\n    proc = run_canon(source, target)\n    assert proc.returncode == 4, proc.stderr\n    assert proc.stderr.strip() == \"malformed row: line 3\"\n    assert not target.exists()\n\n\ndef test_orders_sample_is_canonicalised():\n    target = case_dir() / \"orders.tsv\"\n    proc = run_canon(\"data/orders.csv\", target)\n    assert proc.returncode == 0, proc.stderr\n    expected = (\n        \"id\\tsku\\tqty\\tprice\\n\"\n        \"1\\tSKU-1\\t5\\t0.75\\n\"\n        \"1\\tSKU-2\\t10\\t2\\n\"\n        \"2\\tSKU-1\\t5\\t1000\\n\"\n        \"3\\tSKU-9\\t2\\t1.5\\n\"\n    )\n    assert target.read_bytes() == expected.encode(\"utf-8\")\n\n\ndef test_output_has_no_bom_and_uses_lf():\n    target = case_dir() / \"orders.tsv\"\n    run_canon(\"data/orders.csv\", target)\n    raw = target.read_bytes()\n    assert not raw.startswith(b\"\\xef\\xbb\\xbf\")\n    assert b\"\\r\" not in raw\n\n\ndef test_embedded_newline_survives():\n    source = write_case(\"multi.csv\", 'id,note\\n2,\"line one\\nline two\"\\n1,plain\\n')\n    target = case_dir() / \"multi.tsv\"\n    proc = run_canon(source, target)\n    assert proc.returncode == 0, proc.stderr\n    expected = \"id\\tnote\\n1\\tplain\\n2\\tline one\\nline two\\n\"\n    assert target.read_bytes() == expected.encode(\"utf-8\")\n\n\ndef test_numeric_normalisation():\n    source = write_case(\n        \"nums.csv\", \"id,v\\n1,+5\\n2,007\\n3,-0\\n4,1.0\\n5,2.5e-3\\n6,abc\\n\"\n    )\n    target = case_dir() / \"nums.tsv\"\n    proc = run_canon(source, target)\n    assert proc.returncode == 0, proc.stderr\n    expected = \"id\\tv\\n1\\t5\\n2\\t7\\n3\\t0\\n4\\t1\\n5\\t0.0025\\n6\\tabc\\n\"\n    assert target.read_bytes() == expected.encode(\"utf-8\")\n\n\ndef test_sorting_is_byte_order_not_numeric():\n    source = write_case(\"ids.csv\", \"id,x\\n10,a\\n9,b\\n\")\n    target = case_dir() / \"ids.tsv\"\n    proc = run_canon(source, target)\n    assert proc.returncode == 0, proc.stderr\n    assert target.read_bytes() == b\"id\\tx\\n10\\ta\\n9\\tb\\n\"\n"}, "dockerless": {"setup": "", "setup_timeout_sec": 120, "files": {"data/orders.csv": "id,sku,qty,price\r\n3,SKU-9,2,1.50\r\n1,SKU-2,10,2\r\n2,SKU-1,5,1e3\r\n1,SKU-1,5,0.75\r\n"}, "test_cmd": "", "workdir": "workspace", "requires": ["bash", "python"], "baseline_failures": ["test_outputs.py::test_duplicate_column_exits_3", "test_outputs.py::test_embedded_newline_survives", "test_outputs.py::test_malformed_row_exits_4_with_physical_line", "test_outputs.py::test_missing_input_exits_2", "test_outputs.py::test_numeric_normalisation", "test_outputs.py::test_orders_sample_is_canonicalised", "test_outputs.py::test_output_has_no_bom_and_uses_lf", "test_outputs.py::test_sorting_is_byte_order_not_numeric"]}}
+{"task_id": "tbdl-build-repair", "instruction": "# Make the report build work again\n\n`build.sh` is supposed to regenerate `dist/report.txt` and then run the project's\nown test suite. Right now it does neither:\n\n```\n$ bash build.sh\nbuild.sh: line 5: OUTDIR: unbound variable\n```\n\nNothing in the repository is sacred except the inputs and the tests. Fix the\nbuild so that `bash build.sh`, run from the workspace root, exits `0` and\nproduces:\n\n- `dist/report.txt`\n- `build/gen.log`\n\nBoth files must contain the same text: one `name=score` line per record from\n`data/input.txt`, in input order, followed by a single `mean=<value>` line where\nthe value is the arithmetic mean of the scores printed with exactly two decimal\nplaces.\n\nFor the shipped `data/input.txt` the expected content is:\n\n```\nalpha=10\nbeta=20\ngamma=31\ndelta=40\nepsilon=55\nmean=31.20\n```\n\n## Constraints\n\n- `data/input.txt` must not change.\n- The tests in `tests/` must not change; they must pass as written. They are run\n  by `build.sh` itself, so a build that \"succeeds\" while skipping them is not a\n  fix.\n- `dist/` and `build/` are build outputs; they may be recreated freely.\n- Only the Python standard library is available, and there is no network.\n\n## Notes\n\n- `bash build.sh` must be safe to run repeatedly.\n- Do not silence failures with `|| true`; the build has to fail loudly when a\n  step really fails.\n", "author_name": "WorkBuddy Benchmarks", "author_email": "benchmarks@example.invalid", "difficulty": "hard", "category": "software-engineering", "tags": ["build", "shell", "debugging"], "parser_name": "pytest", "max_agent_timeout_sec": 600.0, "max_test_timeout_sec": 120.0, "expert_time_estimate_min": 25, "junior_time_estimate_min": 90, "task_yaml": "# Terminal-Bench task definition\ninstruction: |\n  # Make the report build work again\n\n  `build.sh` is supposed to regenerate `dist/report.txt` and then run the project's\n  own test suite. Right now it does neither:\n\n  ```\n  $ bash build.sh\n  build.sh: line 5: OUTDIR: unbound variable\n  ```\n\n  Nothing in the repository is sacred except the inputs and the tests. Fix the\n  build so that `bash build.sh`, run from the workspace root, exits `0` and\n  produces:\n\n  - `dist/report.txt`\n  - `build/gen.log`\n\n  Both files must contain the same text: one `name=score` line per record from\n  `data/input.txt`, in input order, followed by a single `mean=<value>` line where\n  the value is the arithmetic mean of the scores printed with exactly two decimal\n  places.\n\n  For the shipped `data/input.txt` the expected content is:\n\n  ```\n  alpha=10\n  beta=20\n  gamma=31\n  delta=40\n  epsilon=55\n  mean=31.20\n  ```\n\n  ## Constraints\n\n  - `data/input.txt` must not change.\n  - The tests in `tests/` must not change; they must pass as written. They are run\n    by `build.sh` itself, so a build that \"succeeds\" while skipping them is not a\n    fix.\n  - `dist/` and `build/` are build outputs; they may be recreated freely.\n  - Only the Python standard library is available, and there is no network.\n\n  ## Notes\n\n  - `bash build.sh` must be safe to run repeatedly.\n  - Do not silence failures with `|| true`; the build has to fail loudly when a\n    step really fails.\nauthor_name: WorkBuddy Benchmarks\nauthor_email: benchmarks@example.invalid\ndifficulty: hard\ncategory: software-engineering\ntags:\n  - build\n  - shell\n  - debugging\nparser_name: pytest\nmax_agent_timeout_sec: 600.0\nmax_test_timeout_sec: 120.0\nexpert_time_estimate_min: 25\njunior_time_estimate_min: 90\n", "task_toml": "schema_version = \"1.4\"\nsource = \"TerminalBench-DockerLess\"\n\n[task]\nname = \"swedl/build-repair\"\nauthors = [{ name = \"WorkBuddy Benchmarks\", email = \"benchmarks@example.invalid\" }]\ndescription = \"Repair a shell build with three independent faults so it produces the expected report and passes its own suite.\"\nkeywords = [\"build\", \"shell\", \"debugging\"]\n\n[metadata]\ndifficulty = \"hard\"\ncategory = \"software-engineering\"\ntags = [\"build\", \"shell\", \"debugging\"]\nexpert_time_estimate_min = 25\njunior_time_estimate_min = 90\n\n[verifier]\ntimeout_sec = 120.0\n\n[agent]\ntimeout_sec = 600.0\n\n[environment]\nnetwork_mode = \"no-network\"\nos = \"linux\"\nworkdir = \"/workspace\"\n", "dockerfile": "FROM python:3.13-slim\n\nRUN apt-get update && apt-get install -y --no-install-recommends \\\n        bash coreutils \\\n    && rm -rf /var/lib/apt/lists/*\n\nWORKDIR /workspace\nCOPY . /workspace/\n\nCMD [\"/bin/bash\"]\n", "solution": "#!/usr/bin/env bash\n# Oracle solution: three independent repairs -- the shell variable typo and the\n# missing output directory, the dropped final record, and the truncated mean.\nset -euo pipefail\ncd \"$TB_WORK\"\n\ncat > build.sh <<'SHEOF'\n#!/usr/bin/env bash\nset -euo pipefail\n\nOUT=dist\nmkdir -p \"$OUT\" build\n\npython src/gen.py data/input.txt | tee build/gen.log > \"$OUT/report.txt\"\n\npython -m pytest tests -q\nSHEOF\n\npython - <<'PYEOF'\nfrom pathlib import Path\n\ngen = Path(\"src/gen.py\")\ntext = gen.read_text(encoding=\"utf-8\")\ntext = text.replace(\"for line in lines[:-1]:\", \"for line in lines:\")\ngen.write_text(text, encoding=\"utf-8\", newline=\"\\n\")\n\nstats = Path(\"src/lib/stats.py\")\ntext = stats.read_text(encoding=\"utf-8\")\ntext = text.replace(\"return total // len(values)\", \"return total / len(values)\")\nstats.write_text(text, encoding=\"utf-8\", newline=\"\\n\")\nPYEOF\n\nbash build.sh\n", "tests": {"test.sh": "#!/usr/bin/env bash\n# Verifier entrypoint. Runs the pytest suite and lets its exit code stand as the\n# reward signal (0 = solved).\nset -uo pipefail\ncd \"$TB_ROOT/tests\"\npython -m pytest -v --tb=short -p no:cacheprovider test_outputs.py\n", "test_outputs.py": "import os\nimport subprocess\nfrom pathlib import Path\n\nimport pytest\n\nWORK = Path(os.environ[\"TB_WORK\"])\nBASH = os.environ.get(\"TB_BASH\", \"bash\")\n\nEXPECTED_REPORT = (\n    \"alpha=10\\n\"\n    \"beta=20\\n\"\n    \"gamma=31\\n\"\n    \"delta=40\\n\"\n    \"epsilon=55\\n\"\n    \"mean=31.20\\n\"\n)\n\n\n@pytest.fixture(scope=\"module\")\ndef build():\n    for path in (WORK / \"dist\", WORK / \"build\"):\n        if path.exists():\n            for child in path.rglob(\"*\"):\n                if child.is_file():\n                    child.unlink()\n    return subprocess.run(\n        [BASH, \"build.sh\"],\n        cwd=str(WORK),\n        capture_output=True,\n        text=True,\n        encoding=\"utf-8\",\n    )\n\n\ndef test_build_exits_zero(build):\n    assert build.returncode == 0, \"build.sh failed:\\n%s\\n%s\" % (build.stdout, build.stderr)\n\n\ndef test_report_content(build):\n    path = WORK / \"dist\" / \"report.txt\"\n    assert path.is_file(), \"dist/report.txt was not produced\"\n    assert path.read_text(encoding=\"utf-8\") == EXPECTED_REPORT\n\n\ndef test_build_log_matches_report(build):\n    log = WORK / \"build\" / \"gen.log\"\n    report = WORK / \"dist\" / \"report.txt\"\n    assert log.is_file(), \"build/gen.log was not produced\"\n    assert log.read_text(encoding=\"utf-8\") == report.read_text(encoding=\"utf-8\")\n\n\ndef test_project_suite_passes(build):\n    proc = subprocess.run(\n        [\"python\", \"-m\", \"pytest\", \"tests\", \"-q\", \"-p\", \"no:cacheprovider\"],\n        cwd=str(WORK),\n        capture_output=True,\n        text=True,\n        encoding=\"utf-8\",\n    )\n    assert proc.returncode == 0, \"project tests failed:\\n%s\\n%s\" % (proc.stdout, proc.stderr)\n"}, "dockerless": {"setup": "", "setup_timeout_sec": 120, "files": {"build.sh": "#!/usr/bin/env bash\nset -euo pipefail\n\nOUT=dist\nmkdir -p \"$OUTDIR\"\n\npython src/gen.py data/input.txt | tee build/gen.log > \"$OUT/report.txt\"\n\npython -m pytest tests -q\n", "data/input.txt": "alpha,10\nbeta,20\ngamma,31\ndelta,40\nepsilon,55\n", "src/gen.py": "\"\"\"Generate the per-record report on stdout.\"\"\"\n\nimport sys\n\nfrom lib.stats import mean\n\n\ndef main(path):\n    with open(path, \"r\", encoding=\"utf-8\") as handle:\n        lines = [line.strip() for line in handle if line.strip()]\n\n    records = []\n    for line in lines[:-1]:\n        name, _, score = line.partition(\",\")\n        records.append((name, int(score)))\n\n    for name, score in records:\n        print(\"%s=%d\" % (name, score))\n    print(\"mean=%.2f\" % mean([score for _, score in records]))\n\n\nif __name__ == \"__main__\":\n    main(sys.argv[1])\n", "src/lib/__init__.py": "", "src/lib/stats.py": "\"\"\"Tiny stats helpers used by the report generator.\"\"\"\n\n\ndef mean(values):\n    \"\"\"Arithmetic mean of a non-empty sequence.\"\"\"\n    total = sum(values)\n    return total // len(values)\n", "tests/test_pipeline.py": "import subprocess\nimport sys\nfrom pathlib import Path\n\nROOT = Path(__file__).resolve().parent.parent\n\n\ndef run_gen():\n    return subprocess.run(\n        [sys.executable, \"src/gen.py\", \"data/input.txt\"],\n        cwd=str(ROOT),\n        capture_output=True,\n        text=True,\n    )\n\n\ndef test_every_record_is_reported_in_input_order():\n    proc = run_gen()\n    assert proc.returncode == 0, proc.stderr\n    assert proc.stdout.strip().splitlines() == [\n        \"alpha=10\",\n        \"beta=20\",\n        \"gamma=31\",\n        \"delta=40\",\n        \"epsilon=55\",\n        \"mean=31.20\",\n    ]\n", "tests/test_stats.py": "import sys\nfrom pathlib import Path\n\nsys.path.insert(0, str(Path(__file__).resolve().parent.parent / \"src\"))\n\nfrom lib.stats import mean\n\n\ndef test_mean_of_even_count_is_fractional():\n    assert abs(mean([1, 2]) - 1.5) < 1e-9\n\n\ndef test_mean_keeps_the_fractional_part():\n    assert abs(mean([10, 20, 31]) - 61 / 3) < 1e-9\n"}, "test_cmd": "", "workdir": "workspace", "requires": ["bash", "python"], "baseline_failures": ["test_outputs.py::test_build_exits_zero", "test_outputs.py::test_build_log_matches_report", "test_outputs.py::test_project_suite_passes", "test_outputs.py::test_report_content"]}}

build_dataset.py ADDED Viewed

	@@ -0,0 +1,302 @@

+"""Build TerminalBench-DockerLess.jsonl from tasks/.
+A Terminal-Bench task is a directory: instruction, environment, solution, tests.
+Docker supplies the environment; here `environment/setup.sh` builds the same
+state in a throwaway directory instead, so a row is self-contained.
+For every task this script:
+  1. flattens the task directory into a row (instruction, files, tests, solution,
+     dockerfile) and renders both a classic `task.yaml` and a Harbor `task.toml`;
+  2. runs the task with no agent at all, to record which verifier tests fail on
+     the untouched environment (`dockerless.baseline_failures`);
+  3. writes the JSONL;
+  4. re-runs the whole file through run_harness.py with --solution and
+     --baseline, and fails loudly unless the oracle scores 4/4 and the baseline
+     scores 0/4.
+Run with:  python build_dataset.py
+"""
+import json
+import shutil
+import subprocess
+import sys
+import tempfile
+from pathlib import Path
+ROOT = Path(__file__).resolve().parent
+TASKS_DIR = ROOT / "tasks"
+OUT = ROOT / "TerminalBench-DockerLess.jsonl"
+REPORTS = ROOT / "reports"
+sys.path.insert(0, str(ROOT))
+import run_harness  # noqa: E402
+AUTHOR = {"name": "WorkBuddy Benchmarks", "email": "benchmarks@example.invalid"}
+TASKS = [
+    {
+        "dir": "access-log-report",
+        "task_id": "tbdl-access-log-report",
+        "difficulty": "medium",
+        "category": "data-processing",
+        "tags": ["log-parsing", "text-processing", "json", "python"],
+        "description": "Aggregate a damaged access log into a canonical JSON report, a malformed-line index and a gzip archive.",
+        "expert_time_estimate_min": 25,
+        "junior_time_estimate_min": 90,
+        "max_agent_timeout_sec": 600.0,
+        "max_test_timeout_sec": 120.0,
+        "requires": ["bash", "python"],
+    },
+    {
+        "dir": "git-recovery",
+        "task_id": "tbdl-git-recovery",
+        "difficulty": "hard",
+        "category": "software-engineering",
+        "tags": ["git", "forensics", "history-recovery"],
+        "description": "Restore a deleted module from history, resurrect an unreachable commit, tag a release and leave a clean tree.",
+        "expert_time_estimate_min": 20,
+        "junior_time_estimate_min": 75,
+        "max_agent_timeout_sec": 600.0,
+        "max_test_timeout_sec": 120.0,
+        "requires": ["bash", "git"],
+    },
+    {
+        "dir": "csv-canon",
+        "task_id": "tbdl-csv-canon",
+        "difficulty": "hard",
+        "category": "data-processing",
+        "tags": ["csv", "cli", "spec-compliance", "exit-codes"],
+        "description": "Write a byte-exact CSV-to-TSV canonicaliser with strict error codes and numeric normalisation.",
+        "expert_time_estimate_min": 35,
+        "junior_time_estimate_min": 120,
+        "max_agent_timeout_sec": 900.0,
+        "max_test_timeout_sec": 120.0,
+        "requires": ["bash", "python"],
+    },
+    {
+        "dir": "build-repair",
+        "task_id": "tbdl-build-repair",
+        "difficulty": "hard",
+        "category": "software-engineering",
+        "tags": ["build", "shell", "debugging"],
+        "description": "Repair a shell build with three independent faults so it produces the expected report and passes its own suite.",
+        "expert_time_estimate_min": 25,
+        "junior_time_estimate_min": 90,
+        "max_agent_timeout_sec": 600.0,
+        "max_test_timeout_sec": 120.0,
+        "requires": ["bash", "python"],
+    },
+]
+def read_text(path):
+    """Decode without newline translation so CRLF/BOM fixtures survive."""
+    return path.read_bytes().decode("utf-8")
+def workspace_files(environment_dir):
+    files = {}
+    for path in sorted(environment_dir.rglob("*")):
+        if not path.is_file():
+            continue
+        rel = path.relative_to(environment_dir).as_posix()
+        if rel in ("Dockerfile", "setup.sh"):
+            continue
+        files[rel] = read_text(path)
+    return files
+def render_task_yaml(meta, instruction):
+    lines = [
+        "# Terminal-Bench task definition",
+        "instruction: |",
+    ]
+    for line in instruction.rstrip("\n").split("\n"):
+        lines.append("  " + line if line else "")
+    lines += [
+        "author_name: %s" % AUTHOR["name"],
+        "author_email: %s" % AUTHOR["email"],
+        "difficulty: %s" % meta["difficulty"],
+        "category: %s" % meta["category"],
+        "tags:",
+    ]
+    for tag in meta["tags"]:
+        lines.append("  - %s" % tag)
+    lines += [
+        "parser_name: pytest",
+        "max_agent_timeout_sec: %.1f" % meta["max_agent_timeout_sec"],
+        "max_test_timeout_sec: %.1f" % meta["max_test_timeout_sec"],
+        "expert_time_estimate_min: %d" % meta["expert_time_estimate_min"],
+        "junior_time_estimate_min: %d" % meta["junior_time_estimate_min"],
+    ]
+    return "\n".join(lines) + "\n"
+def render_task_toml(meta):
+    keywords = ", ".join('"%s"' % tag for tag in meta["tags"])
+    return (
+        'schema_version = "1.4"\n'
+        'source = "TerminalBench-DockerLess"\n'
+        "\n"
+        "[task]\n"
+        'name = "swedl/%s"\n'
+        'authors = [{ name = "%s", email = "%s" }]\n'
+        'description = "%s"\n'
+        "keywords = [%s]\n"
+        "\n"
+        "[metadata]\n"
+        'difficulty = "%s"\n'
+        'category = "%s"\n'
+        "tags = [%s]\n"
+        "expert_time_estimate_min = %d\n"
+        "junior_time_estimate_min = %d\n"
+        "\n"
+        "[verifier]\n"
+        "timeout_sec = %.1f\n"
+        "\n"
+        "[agent]\n"
+        "timeout_sec = %.1f\n"
+        "\n"
+        "[environment]\n"
+        'network_mode = "no-network"\n'
+        'os = "linux"\n'
+        'workdir = "/workspace"\n'
+        % (
+            meta["dir"],
+            AUTHOR["name"],
+            AUTHOR["email"],
+            meta["description"],
+            keywords,
+            meta["difficulty"],
+            meta["category"],
+            keywords,
+            meta["expert_time_estimate_min"],
+            meta["junior_time_estimate_min"],
+            meta["max_test_timeout_sec"],
+            meta["max_agent_timeout_sec"],
+        )
+    )
+def build_row(meta):
+    task_dir = TASKS_DIR / meta["dir"]
+    environment = task_dir / "environment"
+    instruction = read_text(task_dir / "instruction.md")
+    setup_path = environment / "setup.sh"
+    setup = read_text(setup_path) if setup_path.is_file() else ""
+    row = {
+        "task_id": meta["task_id"],
+        "instruction": instruction,
+        "author_name": AUTHOR["name"],
+        "author_email": AUTHOR["email"],
+        "difficulty": meta["difficulty"],
+        "category": meta["category"],
+        "tags": list(meta["tags"]),
+        "parser_name": "pytest",
+        "max_agent_timeout_sec": meta["max_agent_timeout_sec"],
+        "max_test_timeout_sec": meta["max_test_timeout_sec"],
+        "expert_time_estimate_min": meta["expert_time_estimate_min"],
+        "junior_time_estimate_min": meta["junior_time_estimate_min"],
+        "task_yaml": render_task_yaml(meta, instruction),
+        "task_toml": render_task_toml(meta),
+        "dockerfile": read_text(environment / "Dockerfile"),
+        "solution": read_text(task_dir / "solution" / "solve.sh"),
+        "tests": {
+            path.name: read_text(path)
+            for path in sorted((task_dir / "tests").iterdir())
+            if path.is_file()
+        },
+        "dockerless": {
+            "setup": setup,
+            "setup_timeout_sec": 120,
+            "files": workspace_files(environment),
+            "test_cmd": "",
+            "workdir": "workspace",
+            "requires": list(meta["requires"]),
+            "baseline_failures": [],
+        },
+    }
+    return row
+def baseline(row):
+    root = Path(tempfile.mkdtemp(prefix="tbdl-build-"))
+    try:
+        result, _ = run_harness.evaluate(row, "", root)
+    finally:
+        shutil.rmtree(root, ignore_errors=True)
+    return result
+def main():
+    rows = []
+    print("%-24s %-7s %-8s %s" % ("task_id", "diff", "files", "baseline"))
+    for meta in TASKS:
+        row = build_row(meta)
+        result = baseline(row)
+        row["dockerless"]["baseline_failures"] = result["failing"]
+        if result["resolved"]:
+            raise RuntimeError(
+                "%s passes with no agent at all -- the task is not solvable-proof"
+                % row["task_id"]
+            )
+        if not result["tests_total"]:
+            raise RuntimeError("%s: verifier collected no tests" % row["task_id"])
+        rows.append(row)
+        print(
+            "%-24s %-7s %-8d %d/%d failing"
+            % (
+                row["task_id"],
+                row["difficulty"],
+                len(row["dockerless"]["files"]),
+                result["tests_total"] - result["tests_passed"],
+                result["tests_total"],
+            )
+        )
+    with open(OUT, "w", encoding="utf-8", newline="\n") as handle:
+        for row in rows:
+            handle.write(json.dumps(row, ensure_ascii=False) + "\n")
+    print("\nwrote %s (%d rows, %.1f KiB)" % (OUT.name, len(rows), OUT.stat().st_size / 1024.0))
+    REPORTS.mkdir(exist_ok=True)
+    failures = []
+    for mode, expected in (("--solution", len(rows)), ("--baseline", 0)):
+        report_path = REPORTS / ("%s.json" % mode.lstrip("-"))
+        proc = subprocess.run(
+            [
+                sys.executable,
+                str(ROOT / "run_harness.py"),
+                "--dataset",
+                str(OUT),
+                mode,
+                "--out",
+                str(report_path),
+            ],
+            capture_output=True,
+            text=True,
+            encoding="utf-8",
+        )
+        tail = (proc.stdout.strip().splitlines() or ["<no output>"])[-1]
+        print("%-12s %s" % (mode, tail))
+        if proc.returncode != 0:
+            failures.append("%s exited %d: %s" % (mode, proc.returncode, proc.stderr[-800:]))
+            continue
+        resolved = json.loads(report_path.read_text(encoding="utf-8"))["resolved"]
+        if resolved != expected:
+            failures.append("%s resolved %d, expected %d" % (mode, resolved, expected))
+    if failures:
+        print("\nVERIFICATION FAILED")
+        for message in failures:
+            print("  - %s" % message)
+        return 1
+    print("\nverification passed: oracle %d/%d, baseline 0/%d" % (len(rows), len(rows), len(rows)))
+    return 0
+if __name__ == "__main__":
+    sys.exit(main())

predictions/partial.jsonl ADDED Viewed

	@@ -0,0 +1,4 @@

+{"task_id": "tbdl-access-log-report", "agent_name": "control-partial", "agent_script": "#!/usr/bin/env bash\n# Oracle solution: a regex-based parser that survives escaped quotes in the\n# user-agent field, plus deterministic JSON/gzip output.\nset -euo pipefail\ncd \"$TB_WORK\"\nmkdir -p tools\n\ncat > tools/report.py <<'PYEOF'\nimport gzip\nimport json\nimport math\nimport re\nfrom collections import Counter\nfrom pathlib import Path\n\nLINE = re.compile(\n    r'^(?P<ip>\\S+) - - \\[(?P<ts>[^\\]]+)\\] '\n    r'\"(?P<method>[A-Z]+) (?P<path>\\S+) HTTP/1\\.1\" '\n    r'(?P<status>\\d{3}) (?P<size>\\d+|-) '\n    r'\"(?P<ref>(?:[^\"\\\\]|\\\\.)*)\" \"(?P<ua>(?:[^\"\\\\]|\\\\.)*)\" '\n    r'(?P<ms>\\d+)$'\n)\n\n\ndef main():\n    root = Path(__file__).resolve().parent.parent\n    log = root / \"logs\" / \"access.log\"\n    raw = log.read_bytes()\n    lines = raw.decode(\"utf-8\").split(\"\\n\")\n    if lines and lines[-1] == \"\":\n        lines.pop()\n\n    paths, statuses, durations, bad = [], [], [], []\n    total_bytes = 0\n    for number, line in enumerate(lines, 1):\n        match = LINE.match(line)\n        if match is None:\n            bad.append(number)\n            continue\n        paths.append(match.group(\"path\").split(\"?\")[0])\n        statuses.append(match.group(\"status\"))\n        size = match.group(\"size\")\n        total_bytes += 0 if size == \"-\" else int(size)\n        durations.append(int(match.group(\"ms\")))\n\n    durations.sort()\n    count = len(durations)\n    p95 = durations[int(round(0.95 * count)) - 1] if count else 0\n\n    path_counts = Counter(paths)\n    top = sorted(path_counts.items(), key=lambda item: (-item[1], item[0]))[:3]\n\n    report = {\n        \"total_requests\": count,\n        \"unique_paths\": len(path_counts),\n        \"status_counts\": dict(sorted(Counter(statuses).items())),\n        \"top_paths\": [{\"path\": path, \"count\": hits} for path, hits in top],\n        \"bytes_total\": total_bytes,\n        \"p95_ms\": p95,\n    }\n\n    out = root / \"out\"\n    out.mkdir(parents=True, exist_ok=True)\n    with open(out / \"report.json\", \"w\", encoding=\"utf-8\", newline=\"\\n\") as handle:\n        handle.write(json.dumps(report, indent=2, sort_keys=True) + \"\\n\")\n    with open(out / \"malformed.txt\", \"w\", encoding=\"utf-8\", newline=\"\\n\") as handle:\n        handle.write(\"\".join(\"%d\\n\" % number for number in bad))\n    with open(out / \"access.log.gz\", \"wb\") as handle:\n        with gzip.GzipFile(fileobj=handle, mode=\"wb\", mtime=0) as gz:\n            gz.write(raw)\n\n\nif __name__ == \"__main__\":\n    main()\nPYEOF\n\ncat > tools/report.sh <<'SHEOF'\n#!/usr/bin/env bash\nset -euo pipefail\nhere=\"$(cd \"$(dirname \"$0\")/..\" && pwd)\"\nexec python \"$here/tools/report.py\"\nSHEOF\n\nchmod +x tools/report.sh\nbash tools/report.sh\n"}
+{"task_id": "tbdl-git-recovery", "agent_name": "control-partial", "agent_script": "#!/usr/bin/env bash\n# Oracle solution: find the lost commit through fsck, recover the file from the\n# right historical revision, tag, ignore and commit.\nset -euo pipefail\ncd \"$TB_WORK\"\n\ngit config user.name \"Release Bot\"\ngit config user.email \"release@example.invalid\"\n\n# 1. restore src/app.py from \"Add config loader\"\nloader_commit=\"$(git log --format='%H %s' main | awk '$0 ~ / Initial import$/ {print $1}' | head -1)\"\ntest -n \"$loader_commit\"\nmkdir -p src\ngit show \"$loader_commit:src/app.py\" > src/app.py\n\n# 4. ignore the junk\ncat > .gitignore <<'EOF'\n*.log\nbuild/\nEOF\n\ngit add -A\ngit commit -qm \"restore app.py\"\n\n# 2. rescue the unreachable commit\nwip_commit=\"$(git fsck --unreachable --no-reflogs 2>/dev/null | awk '$2 == \"commit\" {print $3}' | while read -r sha; do\n    if [ \"$(git log -1 --format=%s \"$sha\" 2>/dev/null)\" = \"wip: plan\" ]; then\n        echo \"$sha\"\n        break\n    fi\ndone)\"\ntest -n \"$wip_commit\"\ntrue  # gave up on the wip commit\n\n# 3. tag the release commit\ngit tag v1.0 \"$loader_commit\"\n\ngit status --porcelain\n"}
+{"task_id": "tbdl-csv-canon", "agent_name": "control-partial", "agent_script": "#!/usr/bin/env bash\n# Oracle solution: csv module for RFC4180, Decimal for exact numeric rewriting.\nset -euo pipefail\ncd \"$TB_WORK\"\nmkdir -p bin\n\ncat > bin/canon <<'PYEOF'\n#!/usr/bin/env python3\nimport csv\nimport re\nimport sys\nfrom decimal import Decimal, InvalidOperation\nfrom pathlib import Path\n\nNUMERIC = re.compile(r\"^[+-]?(\\d+(\\.\\d*)?|\\.\\d+)([eE][+-]?\\d+)?$\")\n\n\ndef normalize(cell):\n    if not NUMERIC.match(cell):\n        return cell\n    try:\n        value = Decimal(cell)\n    except InvalidOperation:\n        return cell\n    if value == 0:\n        return \"0\"\n    return format(value.normalize(), \"f\")\n\n\ndef main(argv):\n    if len(argv) != 3:\n        sys.stderr.write(\"usage: canon <input.csv> <output.tsv>\\n\")\n        return 1\n    source = Path(argv[1])\n    target = Path(argv[2])\n    if not source.is_file():\n        sys.stderr.write(\"no such file: %s\\n\" % argv[1])\n        return 2\n\n    with open(source, \"r\", encoding=\"utf-8\", newline=\"\") as handle:\n        reader = csv.reader(handle)\n        header = next(reader, None)\n        if header is None:\n            header = []\n        header = [name.strip() for name in header]\n\n        seen = set()\n        for name in header:\n            key = name.casefold()\n            if key in seen:\n                sys.stderr.write(\"duplicate column: %s\\n\" % name)\n                return 3\n            seen.add(key)\n\n        rows = []\n        previous_line = reader.line_num\n        for record in reader:\n            start = previous_line + 1\n            previous_line = reader.line_num\n            if not record:\n                continue\n            if len(record) != len(header):\n                sys.stderr.write(\"malformed row: line %d\\n\" % start)\n                return 4\n            rows.append([normalize(cell) for cell in record])\n\n    pass  # rows arrive sorted, surely\n\n    target.parent.mkdir(parents=True, exist_ok=True)\n    with open(target, \"w\", encoding=\"utf-8\", newline=\"\") as handle:\n        handle.write(\"\\t\".join(header) + \"\\n\")\n        for row in rows:\n            handle.write(\"\\t\".join(row) + \"\\n\")\n    return 0\n\n\nif __name__ == \"__main__\":\n    sys.exit(main(sys.argv))\nPYEOF\n\nchmod +x bin/canon\n./bin/canon data/orders.csv out/orders.tsv\n"}
+{"task_id": "tbdl-build-repair", "agent_name": "control-partial", "agent_script": "#!/usr/bin/env bash\nset -euo pipefail\ncd \"$TB_WORK\"\npython - <<'PYEOF'\nfrom pathlib import Path\npath = Path(\"build.sh\")\ntext = path.read_text(encoding=\"utf-8\")\ntext = text.replace('mkdir -p \"$OUTDIR\"', 'mkdir -p \"$OUT\" build')\npath.write_text(text, encoding=\"utf-8\", newline=\"\\n\")\nPYEOF\nbash build.sh || true\n"}

reports/baseline.json ADDED Viewed

	@@ -0,0 +1,85 @@

+{
+  "dataset": "TerminalBench-DockerLess.jsonl",
+  "mode": "baseline",
+  "tasks": 4,
+  "resolved": 0,
+  "resolve_rate": 0.0,
+  "seconds": 14.25,
+  "results": [
+    {
+      "task_id": "tbdl-access-log-report",
+      "reward": 0.0,
+      "resolved": false,
+      "tests_passed": 0,
+      "tests_total": 10,
+      "failing": [
+        "test_outputs.py::test_bytes_total_treats_dash_as_zero",
+        "test_outputs.py::test_gzip_roundtrip_is_byte_identical",
+        "test_outputs.py::test_malformed_line_numbers",
+        "test_outputs.py::test_p95_uses_nearest_rank",
+        "test_outputs.py::test_report_has_exactly_the_documented_keys",
+        "test_outputs.py::test_script_regenerates_identical_output",
+        "test_outputs.py::test_status_counts",
+        "test_outputs.py::test_top_paths_order_and_tiebreak",
+        "test_outputs.py::test_total_requests_counts_only_valid_lines",
+        "test_outputs.py::test_unique_paths_ignores_query_strings"
+      ],
+      "note": "ok",
+      "seconds": 2.1
+    },
+    {
+      "task_id": "tbdl-git-recovery",
+      "reward": 0.0,
+      "resolved": false,
+      "tests_passed": 1,
+      "tests_total": 10,
+      "failing": [
+        "test_outputs.py::test_app_py_is_committed_not_just_written",
+        "test_outputs.py::test_app_py_restored_from_config_loader_commit",
+        "test_outputs.py::test_junk_is_ignored_but_still_present",
+        "test_outputs.py::test_recovered_branch_points_at_the_wip_commit",
+        "test_outputs.py::test_recovered_commit_keeps_the_work_in_progress_plan",
+        "test_outputs.py::test_recovered_is_not_an_ancestor_of_main",
+        "test_outputs.py::test_restore_commit_message",
+        "test_outputs.py::test_tag_v1_points_at_the_config_loader_commit",
+        "test_outputs.py::test_worktree_is_clean"
+      ],
+      "note": "ok",
+      "seconds": 5.2
+    },
+    {
+      "task_id": "tbdl-csv-canon",
+      "reward": 0.0,
+      "resolved": false,
+      "tests_passed": 0,
+      "tests_total": 8,
+      "failing": [
+        "test_outputs.py::test_duplicate_column_exits_3",
+        "test_outputs.py::test_embedded_newline_survives",
+        "test_outputs.py::test_malformed_row_exits_4_with_physical_line",
+        "test_outputs.py::test_missing_input_exits_2",
+        "test_outputs.py::test_numeric_normalisation",
+        "test_outputs.py::test_orders_sample_is_canonicalised",
+        "test_outputs.py::test_output_has_no_bom_and_uses_lf",
+        "test_outputs.py::test_sorting_is_byte_order_not_numeric"
+      ],
+      "note": "ok",
+      "seconds": 2.74
+    },
+    {
+      "task_id": "tbdl-build-repair",
+      "reward": 0.0,
+      "resolved": false,
+      "tests_passed": 0,
+      "tests_total": 4,
+      "failing": [
+        "test_outputs.py::test_build_exits_zero",
+        "test_outputs.py::test_build_log_matches_report",
+        "test_outputs.py::test_project_suite_passes",
+        "test_outputs.py::test_report_content"
+      ],
+      "note": "ok",
+      "seconds": 4.2
+    }
+  ]
+}

reports/partial.json ADDED Viewed

	@@ -0,0 +1,66 @@

+{
+  "dataset": "TerminalBench-DockerLess.jsonl",
+  "mode": "predictions",
+  "tasks": 4,
+  "resolved": 0,
+  "resolve_rate": 0.0,
+  "seconds": 30.47,
+  "results": [
+    {
+      "task_id": "tbdl-access-log-report",
+      "reward": 0.0,
+      "resolved": false,
+      "tests_passed": 9,
+      "tests_total": 10,
+      "failing": [
+        "test_outputs.py::test_p95_uses_nearest_rank"
+      ],
+      "note": "ok",
+      "seconds": 4.46
+    },
+    {
+      "task_id": "tbdl-git-recovery",
+      "reward": 0.0,
+      "resolved": false,
+      "tests_passed": 5,
+      "tests_total": 10,
+      "failing": [
+        "test_outputs.py::test_app_py_restored_from_config_loader_commit",
+        "test_outputs.py::test_recovered_branch_points_at_the_wip_commit",
+        "test_outputs.py::test_recovered_commit_keeps_the_work_in_progress_plan",
+        "test_outputs.py::test_recovered_is_not_an_ancestor_of_main",
+        "test_outputs.py::test_tag_v1_points_at_the_config_loader_commit"
+      ],
+      "note": "ok",
+      "seconds": 7.08
+    },
+    {
+      "task_id": "tbdl-csv-canon",
+      "reward": 0.0,
+      "resolved": false,
+      "tests_passed": 5,
+      "tests_total": 8,
+      "failing": [
+        "test_outputs.py::test_embedded_newline_survives",
+        "test_outputs.py::test_orders_sample_is_canonicalised",
+        "test_outputs.py::test_output_has_no_bom_and_uses_lf"
+      ],
+      "note": "ok",
+      "seconds": 8.41
+    },
+    {
+      "task_id": "tbdl-build-repair",
+      "reward": 0.0,
+      "resolved": false,
+      "tests_passed": 1,
+      "tests_total": 4,
+      "failing": [
+        "test_outputs.py::test_build_exits_zero",
+        "test_outputs.py::test_project_suite_passes",
+        "test_outputs.py::test_report_content"
+      ],
+      "note": "ok",
+      "seconds": 10.51
+    }
+  ]
+}

reports/solution.json ADDED Viewed

	@@ -0,0 +1,50 @@

+{
+  "dataset": "TerminalBench-DockerLess.jsonl",
+  "mode": "solution",
+  "tasks": 4,
+  "resolved": 4,
+  "resolve_rate": 1.0,
+  "seconds": 28.8,
+  "results": [
+    {
+      "task_id": "tbdl-access-log-report",
+      "reward": 1.0,
+      "resolved": true,
+      "tests_passed": 10,
+      "tests_total": 10,
+      "failing": [],
+      "note": "ok",
+      "seconds": 4.13
+    },
+    {
+      "task_id": "tbdl-git-recovery",
+      "reward": 1.0,
+      "resolved": true,
+      "tests_passed": 10,
+      "tests_total": 10,
+      "failing": [],
+      "note": "ok",
+      "seconds": 6.67
+    },
+    {
+      "task_id": "tbdl-csv-canon",
+      "reward": 1.0,
+      "resolved": true,
+      "tests_passed": 8,
+      "tests_total": 8,
+      "failing": [],
+      "note": "ok",
+      "seconds": 8.01
+    },
+    {
+      "task_id": "tbdl-build-repair",
+      "reward": 1.0,
+      "resolved": true,
+      "tests_passed": 4,
+      "tests_total": 4,
+      "failing": [],
+      "note": "ok",
+      "seconds": 9.98
+    }
+  ]
+}

run_harness.py ADDED Viewed

	@@ -0,0 +1,302 @@

+"""TerminalBench-DockerLess evaluation harness.
+A Terminal-Bench trial is: build the environment, let the agent work in it, then
+run a verifier that decides success. Docker normally provides the environment;
+here each row carries a `dockerless.setup` script that builds the same state in a
+throwaway directory, so a trial needs only bash, git and python.
+    $TB_ROOT/
+        workspace/   <- the agent's cwd (the container's /workspace)
+        home/        <- HOME
+        tmp/         <- TMPDIR
+        bin/         <- prepended to PATH (holds a python3 shim)
+        tests/       <- the verifier, uploaded read-only-ish after the agent runs
+Usage
+-----
+    python run_harness.py --solution                 # oracle, expect 4/4
+    python run_harness.py --baseline                 # do nothing, expect 0/4
+    python run_harness.py --predictions preds.jsonl  # {"task_id", "agent_script"}
+A task is resolved when every verifier test passes (reward 1.0), which is the
+Terminal-Bench rule.
+"""
+import argparse
+import json
+import os
+import re
+import shutil
+import subprocess
+import sys
+import tempfile
+import time
+from pathlib import Path
+RESULT_RE = re.compile(
+    r"^(test_[A-Za-z0-9_]+\.py::[A-Za-z0-9_\[\]\-\.]+)\s+(PASSED|FAILED|ERROR|SKIPPED|XFAIL|XPASS)"
+)
+PYTHON_SHIM = '#!/bin/sh\nexec python "$@"\n'
+_BASH_CACHE = []
+def find_bash():
+    """Resolve a usable POSIX shell.
+    On Windows ``bash`` on PATH is often the WSL launcher, which may be blocked
+    or simply start a different filesystem; prefer Git's own bash and prove the
+    candidate works before trusting it.
+    """
+    if _BASH_CACHE:
+        return _BASH_CACHE[0]
+    candidates = []
+    override = os.environ.get("TBDL_BASH")
+    if override:
+        candidates.append(override)
+    exepath = os.environ.get("EXEPATH")
+    if exepath:
+        candidates.append(os.path.join(exepath, "bash.exe"))
+    git = shutil.which("git")
+    if git:
+        git_bin = Path(git).resolve().parent
+        candidates.append(str(git_bin.parent / "bin" / "bash.exe"))
+        candidates.append(str(git_bin.parent / "usr" / "bin" / "bash.exe"))
+    if os.name == "nt":
+        candidates.append(r"C:\Program Files\Git\bin\bash.exe")
+        candidates.append(r"C:\Program Files\Git\usr\bin\bash.exe")
+    candidates.append(shutil.which("bash") or "bash")
+    for candidate in candidates:
+        if not candidate or not Path(candidate).exists():
+            continue
+        try:
+            probe = subprocess.run(
+                [candidate, "-c", "echo tbdl-ok"],
+                capture_output=True,
+                text=True,
+                timeout=20,
+            )
+        except (OSError, subprocess.SubprocessError):
+            continue
+        if probe.returncode == 0 and "tbdl-ok" in probe.stdout:
+            _BASH_CACHE.append(candidate)
+            return candidate
+    _BASH_CACHE.append("bash")
+    return "bash"
+def write_files(files, dest):
+    for rel, content in files.items():
+        path = dest / rel
+        path.parent.mkdir(parents=True, exist_ok=True)
+        with open(path, "w", encoding="utf-8", newline="\n") as handle:
+            handle.write(content)
+def sandbox_env(root):
+    env = os.environ.copy()
+    env.update(
+        {
+            "TB_ROOT": str(root),
+            "TB_WORK": str(root / "workspace"),
+            "HOME": str(root / "home"),
+            "TMPDIR": str(root / "tmp"),
+            "TEMP": str(root / "tmp"),
+            "TMP": str(root / "tmp"),
+            "PATH": str(root / "bin") + os.pathsep + env.get("PATH", ""),
+            "PYTHONIOENCODING": "utf-8",
+            "PYTHONDONTWRITEBYTECODE": "1",
+            "GIT_CONFIG_NOSYSTEM": "1",
+            "GIT_AUTHOR_NAME": "tbdl",
+            "GIT_AUTHOR_EMAIL": "tbdl@example.invalid",
+            "GIT_COMMITTER_NAME": "tbdl",
+            "GIT_COMMITTER_EMAIL": "tbdl@example.invalid",
+            "GIT_AUTHOR_DATE": "2026-10-30T12:00:00+0000",
+            "GIT_COMMITTER_DATE": "2026-10-30T12:00:00+0000",
+            "TB_BASH": find_bash(),
+        }
+    )
+    return env
+def build_sandbox(row, root):
+    for name in ("workspace", "home", "tmp", "bin", "tests"):
+        (root / name).mkdir(parents=True, exist_ok=True)
+    shim = root / "bin" / "python3"
+    shim.write_text(PYTHON_SHIM, encoding="utf-8", newline="\n")
+    write_files(row["dockerless"]["files"], root / "workspace")
+    write_files(row["tests"], root / "tests")
+    env = sandbox_env(root)
+    setup = row["dockerless"].get("setup", "")
+    if setup.strip():
+        script = root / "setup.sh"
+        script.write_text(setup, encoding="utf-8", newline="\n")
+        proc = subprocess.run(
+            [find_bash(), str(script)],
+            cwd=str(root / "workspace"),
+            capture_output=True,
+            text=True,
+            encoding="utf-8",
+            env=env,
+            timeout=row["dockerless"].get("setup_timeout_sec", 120),
+        )
+        if proc.returncode != 0:
+            raise RuntimeError(
+                "setup failed for %s\n%s\n%s" % (row["task_id"], proc.stdout[-3000:], proc.stderr[-3000:])
+            )
+    return env
+def run_script(path, cwd, env, timeout, label):
+    try:
+        return subprocess.run(
+            [find_bash(), str(path)],
+            cwd=str(cwd),
+            capture_output=True,
+            text=True,
+            encoding="utf-8",
+            env=env,
+            timeout=timeout,
+        )
+    except subprocess.TimeoutExpired:
+        return subprocess.CompletedProcess([], 124, "", "%s timed out after %ss" % (label, timeout))
+def parse_results(stdout):
+    results = {}
+    for raw in stdout.splitlines():
+        match = RESULT_RE.match(raw.strip())
+        if match:
+            results[match.group(1)] = match.group(2)
+    return results
+def evaluate(row, agent_script, root, keep=False):
+    env = build_sandbox(row, root)
+    work = root / "workspace"
+    note = "ok"
+    if agent_script is not None and agent_script.strip():
+        agent_path = root / "agent.sh"
+        agent_path.write_text(agent_script, encoding="utf-8", newline="\n")
+        proc = run_script(
+            agent_path, work, env, row.get("max_agent_timeout_sec", 300), "agent"
+        )
+        if proc.returncode != 0:
+            note = "agent script exited %d" % proc.returncode
+    test_cmd = row["dockerless"]["test_cmd"]
+    test_script = root / "tests" / "test.sh"
+    if test_cmd and test_cmd.strip():
+        test_path = root / "run-tests.sh"
+        test_path.write_text(test_cmd, encoding="utf-8", newline="\n")
+    else:
+        test_path = test_script
+    proc = run_script(
+        test_path, work, env, row.get("max_test_timeout_sec", 120), "verifier"
+    )
+    results = parse_results(proc.stdout)
+    passed = [name for name, outcome in results.items() if outcome == "PASSED"]
+    failed = sorted(name for name, outcome in results.items() if outcome != "PASSED")
+    resolved = bool(results) and not failed
+    return {
+        "task_id": row["task_id"],
+        "reward": 1.0 if resolved else 0.0,
+        "resolved": resolved,
+        "tests_passed": len(passed),
+        "tests_total": len(results),
+        "failing": failed[:12],
+        "note": note if results else "no tests collected",
+        "seconds": 0.0,
+    }, proc
+def load_rows(path):
+    return [json.loads(line) for line in open(path, encoding="utf-8") if line.strip()]
+def main():
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument("--dataset", default="TerminalBench-DockerLess.jsonl")
+    parser.add_argument("--solution", action="store_true", help="run each task's oracle solution")
+    parser.add_argument("--baseline", action="store_true", help="run no agent at all")
+    parser.add_argument("--predictions", help="JSONL of {task_id, agent_script}")
+    parser.add_argument("--only", help="restrict to one task_id")
+    parser.add_argument("--out", help="write a JSON report here")
+    args = parser.parse_args()
+    rows = load_rows(args.dataset)
+    if args.only:
+        rows = [row for row in rows if row["task_id"] == args.only]
+        if not rows:
+            parser.error("unknown task_id: %s" % args.only)
+    predictions = {}
+    if args.predictions:
+        for line in open(args.predictions, encoding="utf-8"):
+            if line.strip():
+                record = json.loads(line)
+                predictions[record["task_id"]] = record.get("agent_script", "")
+    started = time.time()
+    results = []
+    print("%-24s %-7s %-8s %s" % ("task_id", "reward", "tests", "note"))
+    for row in rows:
+        if args.solution:
+            agent_script = row.get("solution") or row["dockerless"].get("solution", "")
+        elif args.baseline:
+            agent_script = ""
+        else:
+            agent_script = predictions.get(row["task_id"], "")
+        root = Path(tempfile.mkdtemp(prefix="tbdl-%s-" % row["task_id"][:12]))
+        begin = time.time()
+        try:
+            result, proc = evaluate(row, agent_script, root)
+        finally:
+            shutil.rmtree(root, ignore_errors=True)
+        result["seconds"] = round(time.time() - begin, 2)
+        results.append(result)
+        print(
+            "%-24s %-7.1f %d/%-6d %s"
+            % (
+                result["task_id"],
+                result["reward"],
+                result["tests_passed"],
+                result["tests_total"],
+                result["note"],
+            )
+        )
+        if result["failing"] and args.baseline:
+            for name in result["failing"]:
+                print("     fail  %s" % name)
+    solved = sum(1 for result in results if result["resolved"])
+    elapsed = time.time() - started
+    summary = {
+        "dataset": Path(args.dataset).name,
+        "mode": "solution" if args.solution else ("baseline" if args.baseline else "predictions"),
+        "tasks": len(rows),
+        "resolved": solved,
+        "resolve_rate": round(solved / len(rows), 4) if rows else 0.0,
+        "seconds": round(elapsed, 2),
+        "results": results,
+    }
+    print(
+        "\nresolved %d/%d (%.1f%%) in %.1fs"
+        % (solved, len(rows), 100.0 * summary["resolve_rate"], elapsed)
+    )
+    if args.out:
+        Path(args.out).parent.mkdir(parents=True, exist_ok=True)
+        Path(args.out).write_text(json.dumps(summary, indent=2), encoding="utf-8")
+        print("report -> %s" % args.out)
+if __name__ == "__main__":
+    main()

tasks/access-log-report/environment/Dockerfile ADDED Viewed

	@@ -0,0 +1,10 @@

+FROM python:3.13-slim
+RUN apt-get update && apt-get install -y --no-install-recommends \
+        bash coreutils gzip git \
+    && rm -rf /var/lib/apt/lists/*
+WORKDIR /workspace
+COPY logs/ /workspace/logs/
+CMD ["/bin/bash"]

tasks/access-log-report/environment/logs/access.log ADDED Viewed

	@@ -0,0 +1,40 @@

+10.0.0.1 - - [10/Oct/2026:13:00:00 +0000] "GET /index.html HTTP/1.1" 200 200 "-" "curl/8.5.0" 3
+10.0.0.2 - - [10/Oct/2026:13:00:17 +0000] "GET /index.html HTTP/1.1" 200 213 "https://example.com/" "curl/8.5.0" 25
+10.0.0.3 - - [10/Oct/2026:13:01:34 +0000] "GET /index.html HTTP/1.1" 200 226 "https://example.com/" "curl/8.5.0" 52
+10.0.0.4 - - [10/Oct/2026:13:01:51 +0000] "GET /index.html HTTP/1.1" 200 239 "https://example.com/" "curl/8.5.0" 78
+10.0.0.5 - - [10/Oct/2026:13:02:08 +0000] "GET /index.html HTTP/1.1" 200 252 "-" "curl/8.5.0" 110
+10.0.0.6 - - [10/Oct/2026:13:02:25 +0000] "GET /index.html HTTP/1.1" 200 265 "https://example.com/" "Mozilla/5.0 (\"compatible\"; bot) Gecko/20100101" 175
+10.0.0.9 - - [10/Oct/2026:13:02:11 +0000] "GET /index.html HTTP/1.1" 200 512 "-" "curl/8.0"
+10.0.0.7 - - [10/Oct/2026:13:03:42 +0000] "GET /index.html HTTP/1.1" 200 278 "https://example.com/" "curl/8.5.0" 21
+10.0.0.8 - - [10/Oct/2026:13:03:59 +0000] "GET /index.html HTTP/1.1" 200 291 "https://example.com/" "curl/8.5.0" 47
+10.0.0.9 - - [10/Oct/2026:13:04:16 +0000] "GET /index.html HTTP/1.1" 200 304 "-" "curl/8.5.0" 74
+10.0.0.1 - - [10/Oct/2026:13:04:33 +0000] "GET /index.html HTTP/1.1" 200 - "https://example.com/" "curl/8.5.0" 104
+10.0.0.2 - - [10/Oct/2026:13:05:50 +0000] "GET /index.html HTTP/1.1" 200 330 "https://example.com/" "curl/8.5.0" 162
+10.0.0.3 - - [10/Oct/2026:13:05:07 +0000] "GET /index.html HTTP/1.1" 200 343 "https://example.com/" "curl/8.5.0" 18
+10.0.0.4 - - [10/Oct/2026:13:06:24 +0000] "GET /api/users HTTP/1.1" 200 356 "-" "curl/8.5.0" 44
+10.0.0.5 - - [10/Oct/2026:13:06:41 +0000] "GET /api/users HTTP/1.1" 200 369 "https://example.com/" "curl/8.5.0" 70
+10.0.0.6 - - [10/Oct/2026:13:07:58 +0000] "GET /api/users HTTP/1.1" 200 382 "https://example.com/" "curl/8.5.0" 99
+10.0.0.7 - - [10/Oct/2026:13:07:15 +0000] "POST /api/users HTTP/1.1" 200 395 "https://example.com/" "curl/8.5.0" 150
+10.0.0.8 - - [10/Oct/2026:13:08:32 +0000] "GET /api/users HTTP/1.1" 200 408 "-" "curl/8.5.0" 14
+10.0.0.3 - - [10/Oct/2026:13:05:44 +0000] "POST /api/users HTTP/1.1" abc 300 "-" "curl/8.0" 44
+10.0.0.9 - - [10/Oct/2026:13:08:49 +0000] "GET /api/users HTTP/1.1" 200 421 "https://example.com/" "Mozilla/5.0 (\"compatible\"; bot) Gecko/20100101" 40
+10.0.0.1 - - [10/Oct/2026:13:09:06 +0000] "GET /api/users HTTP/1.1" 200 434 "https://example.com/" "curl/8.5.0" 66
+10.0.0.2 - - [10/Oct/2026:13:09:23 +0000] "GET /api/users HTTP/1.1" 200 447 "https://example.com/" "curl/8.5.0" 95
+10.0.0.3 - - [10/Oct/2026:13:10:40 +0000] "GET /api/users?id=42 HTTP/1.1" 200 460 "-" "curl/8.5.0" 141
+10.0.0.4 - - [10/Oct/2026:13:10:57 +0000] "GET /api/users?id=42 HTTP/1.1" 200 473 "https://example.com/" "curl/8.5.0" 12
+10.0.0.5 - - [10/Oct/2026:13:11:14 +0000] "GET /api/users?id=42 HTTP/1.1" 200 486 "https://example.com/" "curl/8.5.0" 35
+10.0.0.6 - - [10/Oct/2026:13:11:31 +0000] "GET /api/users?id=42 HTTP/1.1" 200 499 "https://example.com/" "curl/8.5.0" 61
+10.0.0.7 - - [10/Oct/2026:13:12:48 +0000] "GET /static/app.js HTTP/1.1" 200 512 "-" "curl/8.5.0" 90
+10.0.0.4 - - [10/Oct/2026:13:09:02 +0000] "GET /health HTTP/1.1" 200 0
+10.0.0.8 - - [10/Oct/2026:13:12:05 +0000] "GET /static/app.js HTTP/1.1" 200 525 "https://example.com/" "curl/8.5.0" 133
+10.0.0.9 - - [10/Oct/2026:13:13:22 +0000] "GET /static/app.js HTTP/1.1" 200 - "https://example.com/" "curl/8.5.0" 11
+10.0.0.1 - - [10/Oct/2026:13:13:39 +0000] "GET /static/app.js HTTP/1.1" 200 551 "https://example.com/" "curl/8.5.0" 31
+10.0.0.2 - - [10/Oct/2026:13:14:56 +0000] "GET /static/app.js HTTP/1.1" 304 564 "-" "curl/8.5.0" 58
+10.0.0.3 - - [10/Oct/2026:13:14:13 +0000] "GET /static/app.js HTTP/1.1" 304 577 "https://example.com/" "curl/8.5.0" 85
+10.0.0.4 - - [10/Oct/2026:13:15:30 +0000] "GET /health HTTP/1.1" 304 590 "https://example.com/" "curl/8.5.0" 126
+10.0.0.5 - - [10/Oct/2026:13:15:47 +0000] "GET /health HTTP/1.1" 304 603 "https://example.com/" "curl/8.5.0" 7
+10.0.0.6 - - [10/Oct/2026:13:16:04 +0000] "GET /health HTTP/1.1" 404 616 "-" "curl/8.5.0" 28
+# rotated by logrotate at 2026-10-10T14:00:00Z
+10.0.0.7 - - [10/Oct/2026:13:16:21 +0000] "GET /api/orders HTTP/1.1" 404 629 "https://example.com/" "curl/8.5.0" 55
+10.0.0.8 - - [10/Oct/2026:13:17:38 +0000] "GET /api/orders HTTP/1.1" 404 642 "https://example.com/" "curl/8.5.0" 82
+10.0.0.9 - - [10/Oct/2026:13:17:55 +0000] "GET /api/orders HTTP/1.1" 500 655 "https://example.com/" "curl/8.5.0" 118

tasks/access-log-report/instruction.md ADDED Viewed

	@@ -0,0 +1,61 @@

+# Access log report
+`logs/access.log` is a web server access log in a combined-ish format:
+```
+<ip> - - [<timestamp>] "<METHOD> <path> HTTP/1.1" <status> <bytes> "<referer>" "<user-agent>" <duration_ms>
+```
+Some lines are damaged. The user-agent field may contain escaped double quotes
+(`\"`), so the line cannot be parsed by splitting on `"` -- quoted fields have to
+be scanned properly. `bytes` is `-` when the server did not record a size.
+Write **`tools/report.sh`** in the workspace root. Running
+```
+bash tools/report.sh
+```
+from the workspace root must read `logs/access.log` and (re)generate the
+following under `out/`, exit 0, and print nothing that matters. It must be
+idempotent: deleting `out/` and running it again must produce byte-identical
+files.
+## `out/report.json`
+A JSON object with exactly these keys:
+| key | type | meaning |
+| --- | --- | --- |
+| `total_requests` | int | number of **valid** lines |
+| `unique_paths` | int | number of distinct normalised paths |
+| `status_counts` | object | `{"<status>": <count>}`, keys are strings; only statuses that occur |
+| `top_paths` | array | up to 3 objects `{"path": str, "count": int}`, highest count first, ties broken by path ascending |
+| `bytes_total` | int | sum of `bytes` over valid lines, `-` counts as 0 |
+| `p95_ms` | int | 95th percentile of `duration_ms`, nearest-rank method |
+`p95_ms` is the nearest-rank percentile: sort the durations ascending, take
+element number `ceil(0.95 * n)` counting from 1. With `n = 36` that is the 35th
+smallest duration. If there are no valid lines, `p95_ms` is `0`.
+Paths are normalised by dropping everything from the first `?` onwards, so
+`/api/users?id=42` counts as `/api/users`.
+## `out/malformed.txt`
+The 1-based line numbers of the lines that do not match the grammar above, one
+per line, ascending. A line is malformed if any of these hold: it is not the
+nine-part shape above, `status` is not exactly three digits, `bytes` is neither
+`-` nor digits, or the trailing `duration_ms` is missing or not digits.
+Malformed lines are excluded from every aggregate.
+## `out/access.log.gz`
+A gzip stream whose decompressed content is byte-for-byte identical to
+`logs/access.log`.
+## Notes
+- Only `logs/access.log` is input; do not edit it.
+- `python` is on `PATH` (`python3` is also available). Standard library only --
+  there is no network access.

tasks/access-log-report/solution/solve.sh ADDED Viewed

	@@ -0,0 +1,85 @@

+#!/usr/bin/env bash
+# Oracle solution: a regex-based parser that survives escaped quotes in the
+# user-agent field, plus deterministic JSON/gzip output.
+set -euo pipefail
+cd "$TB_WORK"
+mkdir -p tools
+cat > tools/report.py <<'PYEOF'
+import gzip
+import json
+import math
+import re
+from collections import Counter
+from pathlib import Path
+LINE = re.compile(
+    r'^(?P<ip>\S+) - - \[(?P<ts>[^\]]+)\] '
+    r'"(?P<method>[A-Z]+) (?P<path>\S+) HTTP/1\.1" '
+    r'(?P<status>\d{3}) (?P<size>\d+|-) '
+    r'"(?P<ref>(?:[^"\\]|\\.)*)" "(?P<ua>(?:[^"\\]|\\.)*)" '
+    r'(?P<ms>\d+)$'
+)
+def main():
+    root = Path(__file__).resolve().parent.parent
+    log = root / "logs" / "access.log"
+    raw = log.read_bytes()
+    lines = raw.decode("utf-8").split("\n")
+    if lines and lines[-1] == "":
+        lines.pop()
+    paths, statuses, durations, bad = [], [], [], []
+    total_bytes = 0
+    for number, line in enumerate(lines, 1):
+        match = LINE.match(line)
+        if match is None:
+            bad.append(number)
+            continue
+        paths.append(match.group("path").split("?")[0])
+        statuses.append(match.group("status"))
+        size = match.group("size")
+        total_bytes += 0 if size == "-" else int(size)
+        durations.append(int(match.group("ms")))
+    durations.sort()
+    count = len(durations)
+    p95 = durations[math.ceil(0.95 * count) - 1] if count else 0
+    path_counts = Counter(paths)
+    top = sorted(path_counts.items(), key=lambda item: (-item[1], item[0]))[:3]
+    report = {
+        "total_requests": count,
+        "unique_paths": len(path_counts),
+        "status_counts": dict(sorted(Counter(statuses).items())),
+        "top_paths": [{"path": path, "count": hits} for path, hits in top],
+        "bytes_total": total_bytes,
+        "p95_ms": p95,
+    }
+    out = root / "out"
+    out.mkdir(parents=True, exist_ok=True)
+    with open(out / "report.json", "w", encoding="utf-8", newline="\n") as handle:
+        handle.write(json.dumps(report, indent=2, sort_keys=True) + "\n")
+    with open(out / "malformed.txt", "w", encoding="utf-8", newline="\n") as handle:
+        handle.write("".join("%d\n" % number for number in bad))
+    with open(out / "access.log.gz", "wb") as handle:
+        with gzip.GzipFile(fileobj=handle, mode="wb", mtime=0) as gz:
+            gz.write(raw)
+if __name__ == "__main__":
+    main()
+PYEOF
+cat > tools/report.sh <<'SHEOF'
+#!/usr/bin/env bash
+set -euo pipefail
+here="$(cd "$(dirname "$0")/.." && pwd)"
+exec python "$here/tools/report.py"
+SHEOF
+chmod +x tools/report.sh
+bash tools/report.sh

tasks/access-log-report/tests/test.sh ADDED Viewed

	@@ -0,0 +1,6 @@

+#!/usr/bin/env bash
+# Verifier entrypoint. Runs the pytest suite and lets its exit code stand as the
+# reward signal (0 = solved).
+set -uo pipefail
+cd "$TB_ROOT/tests"
+python -m pytest -v --tb=short -p no:cacheprovider test_outputs.py

tasks/access-log-report/tests/test_outputs.py ADDED Viewed

	@@ -0,0 +1,87 @@

+import gzip
+import json
+import os
+import shutil
+import subprocess
+from pathlib import Path
+import pytest
+WORK = Path(os.environ["TB_WORK"])
+BASH = os.environ.get("TB_BASH", "bash")
+def load_report():
+    path = WORK / "out" / "report.json"
+    assert path.is_file(), "out/report.json was not created"
+    return json.loads(path.read_text(encoding="utf-8"))
+def test_report_has_exactly_the_documented_keys():
+    report = load_report()
+    assert sorted(report) == [
+        "bytes_total",
+        "p95_ms",
+        "status_counts",
+        "top_paths",
+        "total_requests",
+        "unique_paths",
+    ]
+def test_total_requests_counts_only_valid_lines():
+    assert load_report()["total_requests"] == 36
+def test_unique_paths_ignores_query_strings():
+    assert load_report()["unique_paths"] == 5
+def test_status_counts():
+    assert load_report()["status_counts"] == {"200": 28, "304": 4, "404": 3, "500": 1}
+def test_top_paths_order_and_tiebreak():
+    assert load_report()["top_paths"] == [
+        {"path": "/api/users", "count": 12},
+        {"path": "/index.html", "count": 12},
+        {"path": "/static/app.js", "count": 6},
+    ]
+def test_bytes_total_treats_dash_as_zero():
+    assert load_report()["bytes_total"] == 14535
+def test_p95_uses_nearest_rank():
+    assert load_report()["p95_ms"] == 162
+def test_malformed_line_numbers():
+    path = WORK / "out" / "malformed.txt"
+    assert path.is_file(), "out/malformed.txt was not created"
+    numbers = [int(line) for line in path.read_text(encoding="utf-8").split()]
+    assert numbers == [7, 19, 28, 37]
+def test_gzip_roundtrip_is_byte_identical():
+    original = (WORK / "logs" / "access.log").read_bytes()
+    archive = WORK / "out" / "access.log.gz"
+    assert archive.is_file(), "out/access.log.gz was not created"
+    assert gzip.decompress(archive.read_bytes()) == original
+def test_script_regenerates_identical_output():
+    out = WORK / "out"
+    before = (out / "report.json").read_bytes()
+    before_bad = (out / "malformed.txt").read_bytes()
+    shutil.rmtree(out)
+    proc = subprocess.run(
+        [BASH, "tools/report.sh"],
+        cwd=str(WORK),
+        capture_output=True,
+        text=True,
+    )
+    assert proc.returncode == 0, "tools/report.sh failed:\n%s\n%s" % (proc.stdout, proc.stderr)
+    assert (out / "report.json").read_bytes() == before
+    assert (out / "malformed.txt").read_bytes() == before_bad

tasks/build-repair/environment/Dockerfile ADDED Viewed

	@@ -0,0 +1,10 @@

+FROM python:3.13-slim
+RUN apt-get update && apt-get install -y --no-install-recommends \
+        bash coreutils \
+    && rm -rf /var/lib/apt/lists/*
+WORKDIR /workspace
+COPY . /workspace/
+CMD ["/bin/bash"]

tasks/build-repair/environment/build.sh ADDED Viewed

	@@ -0,0 +1,9 @@

+#!/usr/bin/env bash
+set -euo pipefail
+OUT=dist
+mkdir -p "$OUTDIR"
+python src/gen.py data/input.txt | tee build/gen.log > "$OUT/report.txt"
+python -m pytest tests -q

tasks/build-repair/environment/data/input.txt ADDED Viewed

	@@ -0,0 +1,5 @@

+alpha,10
+beta,20
+gamma,31
+delta,40
+epsilon,55

tasks/build-repair/environment/src/gen.py ADDED Viewed

	@@ -0,0 +1,23 @@

+"""Generate the per-record report on stdout."""
+import sys
+from lib.stats import mean
+def main(path):
+    with open(path, "r", encoding="utf-8") as handle:
+        lines = [line.strip() for line in handle if line.strip()]
+    records = []
+    for line in lines[:-1]:
+        name, _, score = line.partition(",")
+        records.append((name, int(score)))
+    for name, score in records:
+        print("%s=%d" % (name, score))
+    print("mean=%.2f" % mean([score for _, score in records]))
+if __name__ == "__main__":
+    main(sys.argv[1])

tasks/build-repair/environment/src/lib/__init__.py ADDED Viewed

File without changes

tasks/build-repair/environment/src/lib/stats.py ADDED Viewed

	@@ -0,0 +1,7 @@

+"""Tiny stats helpers used by the report generator."""
+def mean(values):
+    """Arithmetic mean of a non-empty sequence."""
+    total = sum(values)
+    return total // len(values)

tasks/build-repair/environment/tests/test_pipeline.py ADDED Viewed

	@@ -0,0 +1,27 @@

+import subprocess
+import sys
+from pathlib import Path
+ROOT = Path(__file__).resolve().parent.parent
+def run_gen():
+    return subprocess.run(
+        [sys.executable, "src/gen.py", "data/input.txt"],
+        cwd=str(ROOT),
+        capture_output=True,
+        text=True,
+    )
+def test_every_record_is_reported_in_input_order():
+    proc = run_gen()
+    assert proc.returncode == 0, proc.stderr
+    assert proc.stdout.strip().splitlines() == [
+        "alpha=10",
+        "beta=20",
+        "gamma=31",
+        "delta=40",
+        "epsilon=55",
+        "mean=31.20",
+    ]

tasks/build-repair/environment/tests/test_stats.py ADDED Viewed

	@@ -0,0 +1,14 @@

+import sys
+from pathlib import Path
+sys.path.insert(0, str(Path(__file__).resolve().parent.parent / "src"))
+from lib.stats import mean
+def test_mean_of_even_count_is_fractional():
+    assert abs(mean([1, 2]) - 1.5) < 1e-9
+def test_mean_keeps_the_fractional_part():
+    assert abs(mean([10, 20, 31]) - 61 / 3) < 1e-9

tasks/build-repair/instruction.md ADDED Viewed

	@@ -0,0 +1,47 @@

+# Make the report build work again
+`build.sh` is supposed to regenerate `dist/report.txt` and then run the project's
+own test suite. Right now it does neither:
+```
+$ bash build.sh
+build.sh: line 5: OUTDIR: unbound variable
+```
+Nothing in the repository is sacred except the inputs and the tests. Fix the
+build so that `bash build.sh`, run from the workspace root, exits `0` and
+produces:
+- `dist/report.txt`
+- `build/gen.log`
+Both files must contain the same text: one `name=score` line per record from
+`data/input.txt`, in input order, followed by a single `mean=<value>` line where
+the value is the arithmetic mean of the scores printed with exactly two decimal
+places.
+For the shipped `data/input.txt` the expected content is:
+```
+alpha=10
+beta=20
+gamma=31
+delta=40
+epsilon=55
+mean=31.20
+```
+## Constraints
+- `data/input.txt` must not change.
+- The tests in `tests/` must not change; they must pass as written. They are run
+  by `build.sh` itself, so a build that "succeeds" while skipping them is not a
+  fix.
+- `dist/` and `build/` are build outputs; they may be recreated freely.
+- Only the Python standard library is available, and there is no network.
+## Notes
+- `bash build.sh` must be safe to run repeatedly.
+- Do not silence failures with `|| true`; the build has to fail loudly when a
+  step really fails.

tasks/build-repair/solution/solve.sh ADDED Viewed

	@@ -0,0 +1,33 @@

+#!/usr/bin/env bash
+# Oracle solution: three independent repairs -- the shell variable typo and the
+# missing output directory, the dropped final record, and the truncated mean.
+set -euo pipefail
+cd "$TB_WORK"
+cat > build.sh <<'SHEOF'
+#!/usr/bin/env bash
+set -euo pipefail
+OUT=dist
+mkdir -p "$OUT" build
+python src/gen.py data/input.txt | tee build/gen.log > "$OUT/report.txt"
+python -m pytest tests -q
+SHEOF
+python - <<'PYEOF'
+from pathlib import Path
+gen = Path("src/gen.py")
+text = gen.read_text(encoding="utf-8")
+text = text.replace("for line in lines[:-1]:", "for line in lines:")
+gen.write_text(text, encoding="utf-8", newline="\n")
+stats = Path("src/lib/stats.py")
+text = stats.read_text(encoding="utf-8")
+text = text.replace("return total // len(values)", "return total / len(values)")
+stats.write_text(text, encoding="utf-8", newline="\n")
+PYEOF
+bash build.sh

tasks/build-repair/tests/test.sh ADDED Viewed

	@@ -0,0 +1,6 @@

+#!/usr/bin/env bash
+# Verifier entrypoint. Runs the pytest suite and lets its exit code stand as the
+# reward signal (0 = solved).
+set -uo pipefail
+cd "$TB_ROOT/tests"
+python -m pytest -v --tb=short -p no:cacheprovider test_outputs.py

tasks/build-repair/tests/test_outputs.py ADDED Viewed

	@@ -0,0 +1,61 @@

+import os
+import subprocess
+from pathlib import Path
+import pytest
+WORK = Path(os.environ["TB_WORK"])
+BASH = os.environ.get("TB_BASH", "bash")
+EXPECTED_REPORT = (
+    "alpha=10\n"
+    "beta=20\n"
+    "gamma=31\n"
+    "delta=40\n"
+    "epsilon=55\n"
+    "mean=31.20\n"
+)
+@pytest.fixture(scope="module")
+def build():
+    for path in (WORK / "dist", WORK / "build"):
+        if path.exists():
+            for child in path.rglob("*"):
+                if child.is_file():
+                    child.unlink()
+    return subprocess.run(
+        [BASH, "build.sh"],
+        cwd=str(WORK),
+        capture_output=True,
+        text=True,
+        encoding="utf-8",
+    )
+def test_build_exits_zero(build):
+    assert build.returncode == 0, "build.sh failed:\n%s\n%s" % (build.stdout, build.stderr)
+def test_report_content(build):
+    path = WORK / "dist" / "report.txt"
+    assert path.is_file(), "dist/report.txt was not produced"
+    assert path.read_text(encoding="utf-8") == EXPECTED_REPORT
+def test_build_log_matches_report(build):
+    log = WORK / "build" / "gen.log"
+    report = WORK / "dist" / "report.txt"
+    assert log.is_file(), "build/gen.log was not produced"
+    assert log.read_text(encoding="utf-8") == report.read_text(encoding="utf-8")
+def test_project_suite_passes(build):
+    proc = subprocess.run(
+        ["python", "-m", "pytest", "tests", "-q", "-p", "no:cacheprovider"],
+        cwd=str(WORK),
+        capture_output=True,
+        text=True,
+        encoding="utf-8",
+    )
+    assert proc.returncode == 0, "project tests failed:\n%s\n%s" % (proc.stdout, proc.stderr)

tasks/csv-canon/environment/Dockerfile ADDED Viewed

	@@ -0,0 +1,10 @@

+FROM python:3.13-slim
+RUN apt-get update && apt-get install -y --no-install-recommends \
+        bash coreutils \
+    && rm -rf /var/lib/apt/lists/*
+WORKDIR /workspace
+COPY data/ /workspace/data/
+CMD ["/bin/bash"]

tasks/csv-canon/environment/data/orders.csv ADDED Viewed

	@@ -0,0 +1,5 @@

+id,sku,qty,price
+3,SKU-9,2,1.50
+1,SKU-2,10,2
+2,SKU-1,5,1e3
+1,SKU-1,5,0.75

tasks/csv-canon/instruction.md ADDED Viewed

	@@ -0,0 +1,65 @@

+# `canon` — canonicalise a CSV export
+Downstream tools need one boring, byte-stable TSV. Write an executable script at
+`bin/canon` (shebang, `chmod +x`) that is invoked as:
+```
+./bin/canon <input.csv> <output.tsv>
+```
+## Input
+UTF-8 CSV, possibly with a byte-order mark, possibly CRLF, RFC 4180 quoting:
+fields may be quoted, may contain commas, doubled quotes (`""`) and embedded
+newlines. The first record is the header.
+## Output
+UTF-8 **without** a BOM, `\n` line endings, tab-separated, one line per record,
+header first. Fields are written verbatim — never re-quoted, never trimmed —
+except for the numeric normalisation below. Column order is preserved.
+## Numeric normalisation
+A field is numeric when the whole field matches
+`^[+-]?(\d+(\.\d*)?|\.\d+)([eE][+-]?\d+)?$`. Numeric fields are rewritten in
+plain decimal notation with no exponent, no leading `+`, no leading zeros, and
+no trailing zeros after the decimal point:
+| input | output |
+| --- | --- |
+| `1.50` | `1.5` |
+| `1.0` | `1` |
+| `1e3` | `1000` |
+| `2.5e-3` | `0.0025` |
+| `+5` | `5` |
+| `007` | `7` |
+| `-0` | `0` |
+Every other field is copied unchanged.
+## Ordering
+Sort the data records by field 0, then by field 1, comparing the **normalised
+strings** in byte order (so `"10"` sorts before `"9"`). The sort must be stable.
+## Errors
+On any error, print exactly one line to stderr, exit with the given code, and do
+**not** create the output file:
+| condition | stderr line | exit |
+| --- | --- | --- |
+| input path does not exist | `no such file: <input path as given>` | `2` |
+| two header names collide case-insensitively after trimming | `duplicate column: <name as written in the later header field>` | `3` |
+| a record's field count differs from the header's | `malformed row: line <n>` | `4` |
+`<n>` is the 1-based physical line number where the offending record starts
+(so a record whose first line is the third line of the file reports `3`).
+On success exit `0` and print nothing to stderr.
+## Notes
+- `python` and `python3` are both on `PATH`; standard library only, no network.
+- `data/orders.csv` is a sample with a BOM and CRLF endings.

tasks/csv-canon/solution/solve.sh ADDED Viewed

	@@ -0,0 +1,82 @@

+#!/usr/bin/env bash
+# Oracle solution: csv module for RFC4180, Decimal for exact numeric rewriting.
+set -euo pipefail
+cd "$TB_WORK"
+mkdir -p bin
+cat > bin/canon <<'PYEOF'
+#!/usr/bin/env python3
+import csv
+import re
+import sys
+from decimal import Decimal, InvalidOperation
+from pathlib import Path
+NUMERIC = re.compile(r"^[+-]?(\d+(\.\d*)?|\.\d+)([eE][+-]?\d+)?$")
+def normalize(cell):
+    if not NUMERIC.match(cell):
+        return cell
+    try:
+        value = Decimal(cell)
+    except InvalidOperation:
+        return cell
+    if value == 0:
+        return "0"
+    return format(value.normalize(), "f")
+def main(argv):
+    if len(argv) != 3:
+        sys.stderr.write("usage: canon <input.csv> <output.tsv>\n")
+        return 1
+    source = Path(argv[1])
+    target = Path(argv[2])
+    if not source.is_file():
+        sys.stderr.write("no such file: %s\n" % argv[1])
+        return 2
+    with open(source, "r", encoding="utf-8-sig", newline="") as handle:
+        reader = csv.reader(handle)
+        header = next(reader, None)
+        if header is None:
+            header = []
+        header = [name.strip() for name in header]
+        seen = set()
+        for name in header:
+            key = name.casefold()
+            if key in seen:
+                sys.stderr.write("duplicate column: %s\n" % name)
+                return 3
+            seen.add(key)
+        rows = []
+        previous_line = reader.line_num
+        for record in reader:
+            start = previous_line + 1
+            previous_line = reader.line_num
+            if not record:
+                continue
+            if len(record) != len(header):
+                sys.stderr.write("malformed row: line %d\n" % start)
+                return 4
+            rows.append([normalize(cell) for cell in record])
+    rows.sort(key=lambda row: (row[0], row[1] if len(row) > 1 else ""))
+    target.parent.mkdir(parents=True, exist_ok=True)
+    with open(target, "w", encoding="utf-8", newline="") as handle:
+        handle.write("\t".join(header) + "\n")
+        for row in rows:
+            handle.write("\t".join(row) + "\n")
+    return 0
+if __name__ == "__main__":
+    sys.exit(main(sys.argv))
+PYEOF
+chmod +x bin/canon
+./bin/canon data/orders.csv out/orders.tsv

tasks/csv-canon/tests/test.sh ADDED Viewed

	@@ -0,0 +1,6 @@

+#!/usr/bin/env bash
+# Verifier entrypoint. Runs the pytest suite and lets its exit code stand as the
+# reward signal (0 = solved).
+set -uo pipefail
+cd "$TB_ROOT/tests"
+python -m pytest -v --tb=short -p no:cacheprovider test_outputs.py

tasks/csv-canon/tests/test_outputs.py ADDED Viewed

	@@ -0,0 +1,112 @@

+import os
+import subprocess
+from pathlib import Path
+WORK = Path(os.environ["TB_WORK"])
+BASH = os.environ.get("TB_BASH", "bash")
+CANON = "./bin/canon"
+def run_canon(source, target):
+    return subprocess.run(
+        [BASH, "-c", 'exec ./bin/canon "$@"', "canon", str(source), str(target)],
+        cwd=str(WORK),
+        capture_output=True,
+        text=True,
+        encoding="utf-8",
+    )
+def case_dir():
+    path = Path(os.environ["TB_ROOT"]) / "tmp" / "cases"
+    path.mkdir(parents=True, exist_ok=True)
+    return path
+def write_case(name, body, newline=""):
+    path = case_dir() / name
+    with open(path, "w", encoding="utf-8", newline=newline) as handle:
+        handle.write(body)
+    return path
+def test_missing_input_exits_2():
+    target = case_dir() / "missing.tsv"
+    if target.exists():
+        target.unlink()
+    proc = run_canon("does-not-exist.csv", target)
+    assert proc.returncode == 2, proc.stderr
+    assert proc.stderr.strip() == "no such file: does-not-exist.csv"
+    assert not target.exists(), "output must not be created on error"
+def test_duplicate_column_exits_3():
+    source = write_case("dup.csv", "id,ID\n1,2\n")
+    target = case_dir() / "dup.tsv"
+    if target.exists():
+        target.unlink()
+    proc = run_canon(source, target)
+    assert proc.returncode == 3, proc.stderr
+    assert proc.stderr.strip() == "duplicate column: ID"
+    assert not target.exists()
+def test_malformed_row_exits_4_with_physical_line():
+    source = write_case("bad.csv", "id,sku\n1,a\n2,b,c\n")
+    target = case_dir() / "bad.tsv"
+    if target.exists():
+        target.unlink()
+    proc = run_canon(source, target)
+    assert proc.returncode == 4, proc.stderr
+    assert proc.stderr.strip() == "malformed row: line 3"
+    assert not target.exists()
+def test_orders_sample_is_canonicalised():
+    target = case_dir() / "orders.tsv"
+    proc = run_canon("data/orders.csv", target)
+    assert proc.returncode == 0, proc.stderr
+    expected = (
+        "id\tsku\tqty\tprice\n"
+        "1\tSKU-1\t5\t0.75\n"
+        "1\tSKU-2\t10\t2\n"
+        "2\tSKU-1\t5\t1000\n"
+        "3\tSKU-9\t2\t1.5\n"
+    )
+    assert target.read_bytes() == expected.encode("utf-8")
+def test_output_has_no_bom_and_uses_lf():
+    target = case_dir() / "orders.tsv"
+    run_canon("data/orders.csv", target)
+    raw = target.read_bytes()
+    assert not raw.startswith(b"\xef\xbb\xbf")
+    assert b"\r" not in raw
+def test_embedded_newline_survives():
+    source = write_case("multi.csv", 'id,note\n2,"line one\nline two"\n1,plain\n')
+    target = case_dir() / "multi.tsv"
+    proc = run_canon(source, target)
+    assert proc.returncode == 0, proc.stderr
+    expected = "id\tnote\n1\tplain\n2\tline one\nline two\n"
+    assert target.read_bytes() == expected.encode("utf-8")
+def test_numeric_normalisation():
+    source = write_case(
+        "nums.csv", "id,v\n1,+5\n2,007\n3,-0\n4,1.0\n5,2.5e-3\n6,abc\n"
+    )
+    target = case_dir() / "nums.tsv"
+    proc = run_canon(source, target)
+    assert proc.returncode == 0, proc.stderr
+    expected = "id\tv\n1\t5\n2\t7\n3\t0\n4\t1\n5\t0.0025\n6\tabc\n"
+    assert target.read_bytes() == expected.encode("utf-8")
+def test_sorting_is_byte_order_not_numeric():
+    source = write_case("ids.csv", "id,x\n10,a\n9,b\n")
+    target = case_dir() / "ids.tsv"
+    proc = run_canon(source, target)
+    assert proc.returncode == 0, proc.stderr
+    assert target.read_bytes() == b"id\tx\n10\ta\n9\tb\n"

tasks/git-recovery/environment/Dockerfile ADDED Viewed

	@@ -0,0 +1,11 @@

+FROM python:3.13-slim
+RUN apt-get update && apt-get install -y --no-install-recommends \
+        bash coreutils git \
+    && rm -rf /var/lib/apt/lists/*
+WORKDIR /workspace
+COPY setup.sh /tmp/setup.sh
+RUN bash /tmp/setup.sh
+CMD ["/bin/bash"]

tasks/git-recovery/environment/setup.sh ADDED Viewed

	@@ -0,0 +1,79 @@

+#!/usr/bin/env bash
+# Builds the broken repository state. This is the Dockerfile's analogue: it
+# constructs the environment the agent wakes up in, deterministically.
+set -euo pipefail
+cd "$TB_WORK"
+git init -q -b main
+git config user.name "Release Bot"
+git config user.email "release@example.invalid"
+git config commit.gpgsign false
+git config core.autocrlf false
+mkdir -p src
+cat > src/app.py <<'EOF'
+VERSION = 1
+def main():
+    return "v1"
+EOF
+git add -A
+git commit -qm "Initial import"
+mkdir -p notes
+cat > notes/plan.md <<'EOF'
+# Plan
+- step one: wire up the loader
+EOF
+git add -A
+git commit -qm "Add plan notes"
+cat > src/app.py <<'EOF'
+VERSION = 2
+DEFAULTS = {"retries": 3, "timeout": 30}
+def main():
+    return "v2"
+def load(path):
+    with open(path, "r", encoding="utf-8") as handle:
+        return handle.read()
+EOF
+git add -A
+git commit -qm "Add config loader"
+git rm -q src/app.py
+git commit -qm "Drop legacy module"
+mkdir -p docs
+cat > docs/build.md <<'EOF'
+# Build
+Run `make build`.
+EOF
+git add -A
+git commit -qm "Document build"
+# An abandoned experiment: a commit on a branch that is then deleted, leaving
+# the commit unreachable from every ref.
+git checkout -q -b wip HEAD~2
+cat > notes/plan.md <<'EOF'
+# Plan
+- step one: wire up the loader
+- step two: add retry budget (work in progress)
+EOF
+git add -A
+git commit -qm "wip: plan"
+git checkout -q main
+git branch -qD wip
+# Untracked junk that must end up ignored, not deleted.
+printf 'starting\n' > debug.log
+mkdir -p build
+printf 'object file\n' > build/tmp.o

tasks/git-recovery/instruction.md ADDED Viewed

	@@ -0,0 +1,46 @@

+# Recover a broken release repository
+The git repository in the workspace was left in a mess after a bad afternoon.
+`main` currently has five commits, but the module `src/app.py` was deleted by the
+`Drop legacy module` commit and nobody kept a copy. An experiment that was never
+merged lives on as an unreachable commit whose message is `wip: plan`.
+Fix the repository. When you are done it must satisfy all of the following.
+## 1. Restore the module
+`src/app.py` must exist again with **exactly** the content it had in the commit
+titled `Add config loader` (not the older `Initial import` version), and it must
+be committed on `main` in a commit whose message is exactly:
+```
+restore app.py
+```
+## 2. Rescue the abandoned experiment
+Create a branch named `recovered` that points at the unreachable commit whose
+message is `wip: plan`. That commit must **not** become an ancestor of `main`,
+and it must still carry the work-in-progress version of `notes/plan.md`.
+## 3. Tag the release
+Create an annotated-or-lightweight tag `v1.0` pointing at the commit titled
+`Add config loader`.
+## 4. Ignore the junk, don't delete it
+The workspace has two untracked artefacts, `debug.log` and `build/tmp.o`. Add a
+`.gitignore` so that both are ignored, commit it, and leave the files on disk.
+## 5. Clean tree
+`git status --porcelain` must print nothing.
+## Constraints
+- Do not rewrite or discard the five existing commits on `main`; all five
+  original commit subjects must still be present in `git log main`. In
+  particular do not `reset` `main` backwards.
+- Do not delete `debug.log` or `build/tmp.o`.
+- No network access; everything you need is already in `.git`.

tasks/git-recovery/solution/solve.sh ADDED Viewed

	@@ -0,0 +1,38 @@

+#!/usr/bin/env bash
+# Oracle solution: find the lost commit through fsck, recover the file from the
+# right historical revision, tag, ignore and commit.
+set -euo pipefail
+cd "$TB_WORK"
+git config user.name "Release Bot"
+git config user.email "release@example.invalid"
+# 1. restore src/app.py from "Add config loader"
+loader_commit="$(git log --format='%H %s' main | awk '$0 ~ / Add config loader$/ {print $1}' | head -1)"
+test -n "$loader_commit"
+mkdir -p src
+git show "$loader_commit:src/app.py" > src/app.py
+# 4. ignore the junk
+cat > .gitignore <<'EOF'
+*.log
+build/
+EOF
+git add -A
+git commit -qm "restore app.py"
+# 2. rescue the unreachable commit
+wip_commit="$(git fsck --unreachable --no-reflogs 2>/dev/null | awk '$2 == "commit" {print $3}' | while read -r sha; do
+    if [ "$(git log -1 --format=%s "$sha" 2>/dev/null)" = "wip: plan" ]; then
+        echo "$sha"
+        break
+    fi
+done)"
+test -n "$wip_commit"
+git branch recovered "$wip_commit"
+# 3. tag the release commit
+git tag v1.0 "$loader_commit"
+git status --porcelain

tasks/git-recovery/tests/test.sh ADDED Viewed

	@@ -0,0 +1,6 @@

+#!/usr/bin/env bash
+# Verifier entrypoint. Runs the pytest suite and lets its exit code stand as the
+# reward signal (0 = solved).
+set -uo pipefail
+cd "$TB_ROOT/tests"
+python -m pytest -v --tb=short -p no:cacheprovider test_outputs.py

tasks/git-recovery/tests/test_outputs.py ADDED Viewed

	@@ -0,0 +1,103 @@

+import os
+import subprocess
+from pathlib import Path
+WORK = Path(os.environ["TB_WORK"])
+EXPECTED_APP = '''VERSION = 2
+DEFAULTS = {"retries": 3, "timeout": 30}
+def main():
+    return "v2"
+def load(path):
+    with open(path, "r", encoding="utf-8") as handle:
+        return handle.read()
+'''
+ORIGINAL_SUBJECTS = [
+    "Initial import",
+    "Add plan notes",
+    "Add config loader",
+    "Drop legacy module",
+    "Document build",
+]
+def git(*args):
+    return subprocess.run(
+        ["git"] + list(args),
+        cwd=str(WORK),
+        capture_output=True,
+        text=True,
+        encoding="utf-8",
+    )
+def subjects():
+    result = git("log", "--format=%s", "main")
+    assert result.returncode == 0, result.stderr
+    return [line for line in result.stdout.splitlines() if line]
+def test_app_py_restored_from_config_loader_commit():
+    path = WORK / "src" / "app.py"
+    assert path.is_file(), "src/app.py was not restored"
+    assert path.read_text(encoding="utf-8") == EXPECTED_APP
+def test_restore_commit_message():
+    assert "restore app.py" in subjects()
+def test_app_py_is_committed_not_just_written():
+    result = git("ls-files", "--error-unmatch", "src/app.py")
+    assert result.returncode == 0, "src/app.py is not tracked"
+def test_recovered_branch_points_at_the_wip_commit():
+    ref = git("rev-parse", "--verify", "refs/heads/recovered")
+    assert ref.returncode == 0, "branch 'recovered' does not exist"
+    message = git("log", "-1", "--format=%s", "recovered")
+    assert message.stdout.strip() == "wip: plan"
+def test_recovered_is_not_an_ancestor_of_main():
+    ref = git("rev-parse", "--verify", "refs/heads/recovered")
+    assert ref.returncode == 0, "branch 'recovered' does not exist"
+    result = git("merge-base", "--is-ancestor", "recovered", "main")
+    assert result.returncode == 1, "the rescued commit must stay off main"
+def test_recovered_commit_keeps_the_work_in_progress_plan():
+    blob = git("show", "recovered:notes/plan.md")
+    assert blob.returncode == 0, "recovered commit has no notes/plan.md"
+    assert "step two" in blob.stdout
+def test_tag_v1_points_at_the_config_loader_commit():
+    tag = git("rev-list", "-n", "1", "v1.0")
+    assert tag.returncode == 0, "tag v1.0 does not exist"
+    subject = git("log", "-1", "--format=%s", tag.stdout.strip())
+    assert subject.stdout.strip() == "Add config loader"
+def test_junk_is_ignored_but_still_present():
+    for rel in ("debug.log", "build/tmp.o"):
+        assert (WORK / rel).is_file(), "%s was deleted" % rel
+        result = git("check-ignore", "-q", rel)
+        assert result.returncode == 0, "%s is not ignored" % rel
+def test_worktree_is_clean():
+    status = git("status", "--porcelain")
+    assert status.stdout.strip() == "", "dirty tree:\n%s" % status.stdout
+def test_original_history_is_preserved():
+    current = subjects()
+    for subject in ORIGINAL_SUBJECTS:
+        assert subject in current, "commit %r disappeared from main" % subject

tools/make_controls.py ADDED Viewed

	@@ -0,0 +1,98 @@

+"""Generate a negative control: a plausible but incomplete attempt per task.
+Every task must score 0.0. This is the check that the verifier is measuring the
+whole task rather than one easy assertion.
+The controls are produced by mutating the oracle: the interesting failure mode
+for an agent is not "did nothing" but "fixed the obvious half".
+Run with:  python tools/make_controls.py
+"""
+import json
+import sys
+from pathlib import Path
+ROOT = Path(__file__).resolve().parent.parent
+DATASET = ROOT / "TerminalBench-DockerLess.jsonl"
+OUTDIR = ROOT / "predictions"
+def mutate(text, old, new, label):
+    if old not in text:
+        raise SystemExit("control %s: anchor not found" % label)
+    return text.replace(old, new)
+def build_controls(rows):
+    by_id = {row["task_id"]: row for row in rows}
+    controls = {}
+    # 1. access-log: nearest-rank replaced by a 0-based index -> p95 is off by one
+    #    rank, everything else is right.
+    solution = by_id["tbdl-access-log-report"]["solution"]
+    controls["tbdl-access-log-report"] = mutate(
+        solution,
+        "p95 = durations[math.ceil(0.95 * count) - 1] if count else 0",
+        "p95 = durations[int(round(0.95 * count)) - 1] if count else 0",
+        "access-log",
+    )
+    # 2. git-recovery: recovers the *wrong* historical revision and never
+    #    rescues the unreachable commit.
+    solution = by_id["tbdl-git-recovery"]["solution"]
+    solution = mutate(solution, "/ Add config loader$/", "/ Initial import$/", "git-recovery")
+    solution = mutate(
+        solution, 'git branch recovered "$wip_commit"', "true  # gave up on the wip commit", "git-recovery-2"
+    )
+    controls["tbdl-git-recovery"] = solution
+    # 3. csv-canon: forgets the BOM and forgets to sort.
+    solution = by_id["tbdl-csv-canon"]["solution"]
+    solution = mutate(solution, 'encoding="utf-8-sig"', 'encoding="utf-8"', "csv-canon")
+    solution = mutate(
+        solution,
+        'rows.sort(key=lambda row: (row[0], row[1] if len(row) > 1 else ""))',
+        "pass  # rows arrive sorted, surely",
+        "csv-canon-2",
+    )
+    controls["tbdl-csv-canon"] = solution
+    # 4. build-repair: fixes the shell bugs only, leaving both Python faults.
+    controls["tbdl-build-repair"] = """#!/usr/bin/env bash
+set -euo pipefail
+cd "$TB_WORK"
+python - <<'PYEOF'
+from pathlib import Path
+path = Path("build.sh")
+text = path.read_text(encoding="utf-8")
+text = text.replace('mkdir -p "$OUTDIR"', 'mkdir -p "$OUT" build')
+path.write_text(text, encoding="utf-8", newline="\\n")
+PYEOF
+bash build.sh || true
+"""
+    return controls
+def main():
+    OUTDIR.mkdir(exist_ok=True)
+    rows = [json.loads(line) for line in open(DATASET, encoding="utf-8") if line.strip()]
+    controls = build_controls(rows)
+    target = OUTDIR / "partial.jsonl"
+    with open(target, "w", encoding="utf-8", newline="\n") as handle:
+        for row in rows:
+            handle.write(
+                json.dumps(
+                    {
+                        "task_id": row["task_id"],
+                        "agent_name": "control-partial",
+                        "agent_script": controls[row["task_id"]],
+                    }
+                )
+                + "\n"
+            )
+    print("wrote %s for %d tasks" % (target, len(rows)))
+if __name__ == "__main__":
+    main()

tools/mk_access_log.py ADDED Viewed

	@@ -0,0 +1,87 @@

+"""Generate the access.log fixture for the access-log-report task.
+Deterministic: same bytes every run. The fixture is deliberately awkward --
+one user-agent carries escaped quotes, two byte counts are "-", four lines are
+malformed in four different ways, and four request paths carry query strings.
+Run with:  python tools/mk_access_log.py
+"""
+from pathlib import Path
+OUT = (
+    Path(__file__).resolve().parent.parent
+    / "tasks"
+    / "access-log-report"
+    / "environment"
+    / "logs"
+    / "access.log"
+)
+MS = [
+    3, 7, 11, 12, 14, 18, 21, 25, 28, 31, 35, 40, 44, 47, 52, 55, 58, 61,
+    66, 70, 74, 78, 82, 85, 90, 95, 99, 104, 110, 118, 126, 133, 141, 150,
+    162, 175,
+]
+PATHS = (
+    ["/index.html"] * 12
+    + ["/api/users"] * 8
+    + ["/api/users?id=42"] * 4
+    + ["/static/app.js"] * 6
+    + ["/health"] * 3
+    + ["/api/orders"] * 3
+)
+STATUS = [200] * 28 + [304] * 4 + [404] * 3 + [500]
+MALFORMED = {
+    7: '10.0.0.9 - - [10/Oct/2026:13:02:11 +0000] "GET /index.html HTTP/1.1" 200 512 "-" "curl/8.0"',
+    19: '10.0.0.3 - - [10/Oct/2026:13:05:44 +0000] "POST /api/users HTTP/1.1" abc 300 "-" "curl/8.0" 44',
+    28: '10.0.0.4 - - [10/Oct/2026:13:09:02 +0000] "GET /health HTTP/1.1" 200 0',
+    37: "# rotated by logrotate at 2026-10-10T14:00:00Z",
+}
+UA_WEIRD = 'Mozilla/5.0 (\\"compatible\\"; bot) Gecko/20100101'
+def build():
+    valid = []
+    for index in range(36):
+        path = PATHS[index]
+        method = "GET"
+        if path == "/api/users" and index % 5 == 0:
+            method = "POST"
+        status = STATUS[index]
+        if index in (9, 26):
+            size = "-"
+        else:
+            size = str(200 + index * 13)
+        ms = MS[(index * 7) % 36]
+        referer = "-" if index % 4 == 0 else "https://example.com/"
+        ua = UA_WEIRD if index in (5, 17) else "curl/8.5.0"
+        line = (
+            '10.0.0.%d - - [10/Oct/2026:13:%02d:%02d +0000] "%s %s HTTP/1.1" %d %s "%s" "%s" %d'
+            % (index % 9 + 1, index // 2, (index * 17) % 60, method, path, status, size, referer, ua, ms)
+        )
+        valid.append(line)
+    return valid
+def main():
+    lines = build()
+    out = []
+    source = iter(lines)
+    for position in range(1, 41):
+        if position in MALFORMED:
+            out.append(MALFORMED[position])
+        else:
+            out.append(next(source))
+    OUT.parent.mkdir(parents=True, exist_ok=True)
+    with open(OUT, "w", encoding="utf-8", newline="\n") as handle:
+        handle.write("\n".join(out) + "\n")
+    print("wrote %s (%d lines, %d valid)" % (OUT, len(out), len(lines)))
+if __name__ == "__main__":
+    main()

tools/validate.py ADDED Viewed

	@@ -0,0 +1,130 @@

+"""Structural validation for TerminalBench-DockerLess.jsonl.
+Checks that every row carries the classic Terminal-Bench task.yaml fields and the
+Harbor task.toml rendering, that the embedded task directory is complete, that
+the verifier and instruction do not leak each other, and that every task is
+non-trivial (something must fail before any agent acts).
+Run with:  python tools/validate.py
+"""
+import json
+import sys
+from pathlib import Path
+ROOT = Path(__file__).resolve().parent.parent
+DATASET = ROOT / "TerminalBench-DockerLess.jsonl"
+TB_FIELDS = [
+    "task_id",
+    "instruction",
+    "author_name",
+    "author_email",
+    "difficulty",
+    "category",
+    "tags",
+    "parser_name",
+    "max_agent_timeout_sec",
+    "max_test_timeout_sec",
+    "expert_time_estimate_min",
+    "junior_time_estimate_min",
+    "task_yaml",
+    "task_toml",
+    "dockerfile",
+    "solution",
+    "tests",
+    "dockerless",
+]
+REQUIRED_FILES = ["instruction.md", "environment/Dockerfile", "solution/solve.sh", "tests/test.sh"]
+MAX_INSTRUCTION_CHARS = 4000
+MIN_AGENT_TIMEOUT = 60
+MIN_TEST_TIMEOUT = 30
+failures = []
+def check(condition, message):
+    if not condition:
+        failures.append(message)
+def main():
+    rows = [json.loads(line) for line in open(DATASET, encoding="utf-8") if line.strip()]
+    check(len(rows) == 4, "expected 4 rows, found %d" % len(rows))
+    seen = set()
+    for row in rows:
+        tid = row.get("task_id", "<missing>")
+        for field in TB_FIELDS:
+            check(field in row, "%s: missing field %s" % (tid, field))
+        check(tid not in seen, "%s: duplicate task_id" % tid)
+        seen.add(tid)
+        instruction = row["instruction"]
+        check(len(instruction.strip()) > 200, "%s: instruction too short" % tid)
+        check(len(instruction) <= MAX_INSTRUCTION_CHARS, "%s: instruction too long" % tid)
+        check("::" not in instruction, "%s: instruction leaks a test node id" % tid)
+        check("test_outputs" not in instruction, "%s: instruction names the verifier file" % tid)
+        check(row["parser_name"] == "pytest", "%s: parser_name must be pytest" % tid)
+        check(row["difficulty"] in ("easy", "medium", "hard"), "%s: bad difficulty" % tid)
+        check(row["max_agent_timeout_sec"] >= MIN_AGENT_TIMEOUT, "%s: agent timeout too small" % tid)
+        check(row["max_test_timeout_sec"] >= MIN_TEST_TIMEOUT, "%s: test timeout too small" % tid)
+        yaml = row["task_yaml"]
+        for needle in ("instruction: |", "parser_name: pytest", "difficulty: ", "max_agent_timeout_sec:"):
+            check(needle in yaml, "%s: task_yaml missing %r" % (tid, needle))
+        toml = row["task_toml"]
+        for needle in ('schema_version = "1.4"', "[verifier]", "[agent]", "[environment]"):
+            check(needle in toml, "%s: task_toml missing %r" % (tid, needle))
+        tests = row["tests"]
+        check("test.sh" in tests, "%s: tests/test.sh missing" % tid)
+        check("test_outputs.py" in tests, "%s: tests/test_outputs.py missing" % tid)
+        check(row["solution"].strip() != "", "%s: empty solution" % tid)
+        check("FROM " in row["dockerfile"], "%s: dockerfile does not look like a Dockerfile" % tid)
+        dockerless = row["dockerless"]
+        files = dockerless["files"]
+        for rel in files:
+            check(not rel.startswith("/"), "%s: absolute path in snapshot: %s" % (tid, rel))
+            check(".." not in Path(rel).parts, "%s: escaping path in snapshot: %s" % (tid, rel))
+        check("Dockerfile" not in files, "%s: snapshot contains the Dockerfile" % tid)
+        check("setup.sh" not in files, "%s: snapshot contains setup.sh" % tid)
+        check(
+            bool(dockerless["baseline_failures"]),
+            "%s: nothing fails before the agent acts" % tid,
+        )
+        check(
+            bool(dockerless.get("setup", "").strip()) or bool(files),
+            "%s: task has neither setup script nor files" % tid,
+        )
+        print(
+            "%-24s %-6s files=%-2d tests=%-2d baseline_fail=%-2d instr=%d chars"
+            % (
+                tid,
+                row["difficulty"],
+                len(files),
+                len(tests) - 1,
+                len(dockerless["baseline_failures"]),
+                len(instruction),
+            )
+        )
+    total_chars = DATASET.stat().st_size
+    print("\n%d tasks, %.1f KiB (~%d tokens)" % (len(rows), total_chars / 1024.0, total_chars // 4))
+    if failures:
+        print("\nFAILED %d checks:" % len(failures))
+        for message in failures:
+            print("  - %s" % message)
+        return 1
+    print("all checks passed")
+    return 0
+if __name__ == "__main__":
+    sys.exit(main())