Datasets:
task_id stringlengths 6 11 | domain stringclasses 7
values | prompt stringlengths 927 3.71k |
|---|---|---|
finance-18 | finance | Operate from the current directory by executing shell commands; the environment provides python3 with access to numpy, pandas and Pillow. Once completed, respond with {"answer": "done"} to trigger scoring of your files.
Set up a discount curve and value a set of instruments. The file quotes.csv contains the columns ty... |
math-1 | math | Operate from the current directory by executing shell commands; the environment provides python3 with access to numpy, pandas and Pillow. Once completed, respond with {"answer": "done"} to trigger scoring of your files.
Answer every exact-integer problem listed below. Because the values are too large for a naive brute... |
media-12 | media | Operate from the current directory by executing shell commands; the environment provides python3 with access to numpy, pandas and Pillow. Once completed, respond with {"answer": "done"} to trigger scoring of your files.
stream.mcf is a corrupted media container. Its structure consists of the magic bytes 'MCF1' followe... |
office-3 | office | Operate from the current directory by executing shell commands; the environment provides python3 with access to numpy, pandas and Pillow. Once completed, respond with {"answer": "done"} to trigger scoring of your files.
The team roster is located in roster.csv (columns: full_name, aliases separated by pipes), and meet... |
science-12 | science | Operate from the current directory by executing shell commands; the environment provides python3 with access to numpy, pandas and Pillow. Once completed, respond with {"answer": "done"} to trigger scoring of your files.
Combine the chemical kinetics model found in params.json (containing rate constants k1,k2,k3) with ... |
security-21 | security | Operate from the current directory by executing shell commands; the environment provides python3 with access to numpy, pandas and Pillow. Once completed, respond with {"answer": "done"} to trigger scoring of your files.
Audit a synthetic firewall ruleset. The file rules.csv has columns id,action,proto,src,dst,dport,en... |
software-17 | software | Operate from the current directory by executing shell commands; the environment provides python3 with access to numpy, pandas and Pillow. Once completed, respond with {"answer": "done"} to trigger scoring of your files.
Create `kv.py` (stub) containing `TxError(Exception)` and a `Store` class implementing an in-memory... |
finance-21 | finance | Operate from the current directory by executing shell commands; the environment provides python3 with access to numpy, pandas and Pillow. Once completed, respond with {"answer": "done"} to trigger scoring of your files.
Determine the contents of credit-card statements. The file transactions.csv contains the columns da... |
math-3 | math | Operate from the current directory by executing shell commands; the environment provides python3 with access to numpy, pandas and Pillow. Once completed, respond with {"answer": "done"} to trigger scoring of your files.
Resolve the following computational problems precisely (python3 and numpy are available; Python int... |
media-13 | media | Operate from the current directory by executing shell commands; the environment provides python3 with access to numpy, pandas and Pillow. Once completed, respond with {"answer": "done"} to trigger scoring of your files.
The file stream.mcf is a corrupted media container. Its structure begins with the magic bytes 'MCF1... |
office-14 | office | Operate from the current directory by executing shell commands; the environment provides python3 with access to numpy, pandas and Pillow. Once completed, respond with {"answer": "done"} to trigger scoring of your files.
Perform intercompany ledger reconciliation for the corporate group with a cutoff of 2026-09-30. ent... |
science-13 | science | Operate from the current directory by executing shell commands; the environment provides python3 with access to numpy, pandas and Pillow. Once completed, respond with {"answer": "done"} to trigger scoring of your files.
Using the 4-species kinetics defined in params.json (k1,k2,k3,kf,kr, dose, t_dose), begin integrati... |
software-23 | software | Operate from the current directory by executing shell commands; the environment provides python3 with access to numpy, pandas and Pillow. Once completed, respond with {"answer": "done"} to trigger scoring of your files.
Create `cronlib.py` (stub): define `next_times(expr, after, count) -> list[str]`, which acts as a c... |
math-7 | math | Operate from the current directory by executing shell commands; the environment provides python3 with access to numpy, pandas and Pillow. Once completed, respond with {"answer": "done"} to trigger scoring of your files.
The file roads.csv contains a road network where lines beginning with # are comments, and if a pair... |
media-19 | media | Operate from the current directory by executing shell commands; the environment provides python3 with access to numpy, pandas and Pillow. Once completed, respond with {"answer": "done"} to trigger scoring of your files.
playlist.m3u8 is a synthetic HLS media playlist. Determine its timeline structure. The rules are as... |
office-15 | office | Operate from the current directory by executing shell commands; the environment provides python3 with access to numpy, pandas and Pillow. Once completed, respond with {"answer": "done"} to trigger scoring of your files.
Perform reconciliation of the intercompany ledgers for a corporate group with a cutoff date of 2026... |
science-16 | science | Operate from the current directory by executing shell commands; the environment provides python3 with access to numpy, pandas and Pillow. Once completed, respond with {"answer": "done"} to trigger scoring of your files.
The file dist.csv contains a symmetric 13x13 matrix of estimated pairwise evolutionary distances, w... |
software-27 | software | Operate from the current directory by executing shell commands; the environment provides python3 with access to numpy, pandas and Pillow. Once completed, respond with {"answer": "done"} to trigger scoring of your files.
Create `interp.py` (stub) defining `InterpError(Exception)` and `interpolate(text, env) -> str`, wh... |
math-14 | math | Operate from the current directory by executing shell commands; the environment provides python3 with access to numpy, pandas and Pillow. Once completed, respond with {"answer": "done"} to trigger scoring of your files.
The alphabet consists of {a,b,c}. The file params.json specifies the queries (lengths n) and the fo... |
office-17 | office | Operate from the current directory by executing shell commands; the environment provides python3 with access to numpy, pandas and Pillow. Once completed, respond with {"answer": "done"} to trigger scoring of your files.
Create a 14-day schedule commencing on Monday 2026-11-02 (weeks run Mon-Sun, meaning days 1-7 and 8... |
science-20 | science | Operate from the current directory by executing shell commands; the environment provides python3 with access to numpy, pandas and Pillow. Once completed, respond with {"answer": "done"} to trigger scoring of your files.
Identify the earthquake source using the provided phase picks. The file stations.csv provides colum... |
math-17 | math | Operate from the current directory by executing shell commands; the environment provides python3 with access to numpy, pandas and Pillow. Once completed, respond with {"answer": "done"} to trigger scoring of your files.
Facility location problem. The file depots.csv has columns (id,fixed,cap); activating a depot incur... |
office-22 | office | Operate from the current directory by executing shell commands; the environment provides python3 with access to numpy, pandas and Pillow. Once completed, respond with {"answer": "done"} to trigger scoring of your files.
sheet.csv is a compact spreadsheet consisting of 5 columns (A to E) and 9 rows without a header: li... |
office-23 | office | Operate from the current directory by executing shell commands; the environment provides python3 with access to numpy, pandas and Pillow. Once completed, respond with {"answer": "done"} to trigger scoring of your files.
The file sheet.csv is a compact spreadsheet consisting of 12 rows and 5 columns (A through E) witho... |
Mixed Skills v6: eight-domain verified workspace tasks
One OpenEnv server, 190 deterministic procedurally generated tasks across the eight evaluation domains: software engineering, industrial and physical systems, natural science, office work, finance and economics, math and formal reasoning, cybersecurity, and media production. Each task gives the agent a scratch workspace with input files and a written specification. The agent works with shell commands and is scored by a hidden verifier on the files it leaves behind.
tasks.jsonl lists every task id with its domain and the exact first-turn prompt. The image is
docker.io/chrismun/mixed-env:v6 (linux/amd64); it contains 190 task ids and the submission names the 24 used for training.
Interaction
Each turn the model replies with exactly one JSON object:
{"command": "<shell command>"}runs in the workspace (python3 with numpy, pandas and Pillow). At most 12 commands per episode; each command's output is truncated to its last 1500 characters, so strong solutions inspect withheadand do the work in a script.{"answer": "done"}ends the episode and scores the workspace.
Observations carry the command output, the commands left and. There is no reminder text. Replies that are not a single JSON object end the episode, which is then scored on the workspace as it stands.
Reward
Terminal and in [0, 1]. Every verifier is a weighted set of independent sub-checks recomputed from the task inputs, so partial progress earns partial reward and only a fully correct deliverable scores 1.0. References live only inside the image, agent commands run as an unprivileged user with memory and file-size limits, and verifiers that execute agent code do so in a subprocess with a timeout. An untouched workspace scores at most 0.25 on every task.
Calibration
Tasks were measured with Qwen3.8-27B (thinking off) inside a tool-calling GRPO training harness, because pass rates there differ a lot from a strict JSON-reply driver (the two correlate only 0.25 per task). Each task was sampled 8 to 56 times across profiling and training passes. The 24 training tasks are those whose pooled pass rate (shrunk toward 0.5) falls between 0.08 and 0.45.
- Pooled mean pass@1 of the 24 tasks: about 0.33, with the hardest near 0.1 (science and finance) and the easiest near 0.45. Selecting on low observed rates regresses upward, so the true start is probably a little higher.
- Every task runs bare: no reply-format reminder and no per-step hint.
- Prompt text: the task statements and the three global strings were rewritten by Qwen3.8-27B from the original statements under strict gates (every number, file name, identifier, quoted string and JSON example must survive verbatim). Difficulty of the rewrites was not re-matched task by task.
- No transcript in the profiling runs touched verifier or task source.
Domains
| Domain | Skills exercised |
|---|---|
| software | multi-file bug fixing against hidden tests, implementing modules from a spec, CLI conversion, migration scripts, simulator debugging, performance and dependency problems |
| industrial | sensor log analysis and fault reports, control tuning by simulation, pipe and pump sizing, tariff billing, inventory policy, flow-line simulation |
| science | curve and rate fitting, sequence analysis and protein mass, ODE integration, hypothesis tests, titration equilibria, spectral analysis |
| office | payables reconciliation, meeting-minutes extraction, room scheduling, expense audits, payroll, contact deduplication |
| finance | amortization, market-data analytics, ledger audit, bond pricing, tax lots, FX arbitrage |
| math | exact number theory, proofs by computation, assignment optimization, shortest routes, constructive codes, absorbing Markov chains, impartial games |
| security | log incident analysis, hardening vulnerable code, layered-encoding and crypto forensics, detection rules, advisory matching (all synthetic, defensive) |
| media | subtitle retiming, image pipelines, audio synthesis, SVG charts, timecode conformance, markdown to RSS |
Files
tasks.jsonl:task_id,domain,promptfor each of the 24 training tasks, in training order (domains interleaved).
- Downloads last month
- 77