vibrantlabsai/pool-desk

Pool Desk

An OpenEnv environment for the OpenEnv Arena: 50 terminal tasks built on real public data, covering all eight Arena domains. Each episode is one task in a fresh container with its own image. The model works in a shell (exec, write_file), then calls evaluate, and a pytest verifier checks the files it leaves behind.

What it teaches

The tasks are multi-step professional work on real, messy data, where getting the answer exactly right needs domain conventions. Examples:

• footing a bank Call Report, or turning year-to-date income into quarterly figures;
• reconciling agencies' hurricane wind estimates;
• computing survey standard errors with replicate weights;
• triaging vulnerabilities against CISA KEV and EPSS;
• QA of a localization file against its source;
• reconciling an aggregate report against its detail records;
• auditing a bridge inventory;
• resolving the versions affected by a security advisory.

Each task is meant to be finished in 3–5 actions: look at the data, write one script, run it, evaluate. The difficulty sits in the domain logic, not in exploring the files.

Domains

Arena domain	Tasks
finance-economics	6
natural-science	6
mathematics-or-formal-reasoning	7
cybersecurity	7
media-content-production	6
office-white-collar	6
industrial-physical-systems	6
software-engineering	6

They were chosen from a reviewed pool of 80, ten per domain. The selection rules and reasons are in selection.json.

How every task was built and checked

1. Raw data. Each task uses a pinned slice of a public dataset with its licence recorded (provenance.json). Where the clean public source could be fetched from inside the sandbox to read off the answer, the slice is disguised: task-local ids, no names or geography, and seeded per-entity scaling.
2. Measurement. The base model, Qwen3.8-27B with thinking off, made 8 or more attempts. Sampling was temperature 1.0 and top_p 1.0, calibrated against the Arena trainer's own per-task rewards. The reference model, Qwen3.8-2.4T-A95B with low reasoning, made 4 attempts.
   • A task is kept when the base model solves 25–75% of its pooled attempts and the reference model at least 50%.
   • Every batch for unchanged content is pooled into one verdict.
   • Each task's measurement.json has the numbers.
3. Task review. Separate reviewer agents applied a fixed protocol:
   • static checks;
   • three clean builds;
   • the reference solution scoring 1, three times out of three;
   • doing nothing scoring 0;
   • a Learnable measurement;
   • a rubric (instruction matches tests, the tests can't be gamed, "do not modify" rules are enforced, no answers in the image), adapted from Terminal-Bench's Apache-2.0 task rubric;
   • a reward-hacking scan of the transcripts.
4. Red-team review.
   • Defensive verifier checks: integrity manifest, hidden tests and solution unreachable, empty outputs scoring 0.
   • A shortcut audit: the reference model is asked to pass without doing the work, and every passing transcript is judged.
   • A check for routes to the answer through public sources.

The verifier is hardened against an agent running as root. It stops leftover processes and hashes the interpreter, the standard library and a snapshot of the packages the tests import. It runs pytest isolated from anything the agent installed. Reward is binary: 1.0 if every test passes, otherwise 0.0.

Experiments, by tag

Each task's tb2/task.toml carries two tags:

• budget_tier. extended (8,192 completion and 16,384 context tokens) is used on some finance and science tasks that need room for larger inputs. Every other task is standard (4,096 / 8,192).
• format_profile. format-stress is used on some cybersecurity and media tasks. These have an eval-style markdown instruction and content that is awkward to quote inline (spaces in paths, quotes, regexes, embedded JSON), so writing a script with write_file is the reliable way through. Every other task is standard.

Mathematics is a control domain, with no experiment tasks.

Files

• tasks.jsonl: one line per task (task_id, arena_domain, budget_tier, format_profile, limits).
• tasks/<domain>/<task_id>/:
  • tb2/: the Terminal-Bench task (instruction, tests, task.toml);
  • data/: the input files;
  • solution/solve.sh: the reference solution;
  • notes/: the generator and the build notes;
  • Dockerfile;
  • provenance.json;
  • measurement.json: a summary.
• selection.json: how the 50 were picked from the pool.
• LICENSES.md: every task's source and licence.

Licences

The tasks draw on many sources under different permissive terms. Most are US federal government works (public domain). The others include:

• CC0 (MusicBrainz);
• CC-BY-4.0 (e.g. GitHub Advisory Database, PUDL, Red Hat VEX, Loghub);
• ODC-BY;
• Artistic-2.0 (Cremona's ecdata);
• the UK Open Government Licence v3.0;
• Apache-2.0, MIT, BSD and PSF;
• source-specific open terms.

No task uses share-alike or non-commercial data; the one such task in the pool was left out. Attribution for each task is in LICENSES.md and its provenance.json. Reuse must keep that attribution.

No SkillsBench task, data slice or verifier was copied. Any personal names in the sources are pseudonymized or removed.