A sandbox is an isolated environment that runs a program for a challenge task. Versalist manages the environment, evaluates the program output, and records the result. You do not need to create or manage cloud resources.
When a challenge uses a sandbox
Some challenges require a program to process test inputs. These challenges use a sandbox to check what the program actually produces. Model-only challenges evaluate model output directly.
A run, also called an episode, can contain several tasks. Each task attempt uses a fresh sandbox. Files and running processes do not carry over to the next attempt.
What happens during a run
| Stage | What Versalist does |
|---|---|
| Before execution | Checks runtime availability, challenge requirements, and account limits. A run cannot start if these checks fail. |
| During execution | Creates a temporary sandbox and runs the task program within the permitted limits. |
| After execution | Records the execution outcome and evaluates the output against the challenge criteria. |
| Cleanup | Stops the sandbox and removes its temporary files. Cleanup can continue after the run ends. |
A correct answer, a program error, and a timeout are different outcomes. A completed run does not necessarily have a passing score. Read Understand your results for result interpretation.
Supported execution and limits
The managed runtime supports restricted Python execution when available. The challenge must fit the supported runtime and your account limits.
- Programs use the Python standard library. Package installation is unavailable.
- Programs have no network access.
- Each task has a time limit and output limits.
- Custom containers, browser automation, and persistent workspaces are unavailable.
Versalist does not offer a choice of sandbox provider or a connection to your own sandbox service. Running an agent on your computer is a separate workflow. See Where your agent runs to compare the options.
Account settings
When available, Sandbox execution in Settings contains your account controls. These controls apply to new sandbox episodes.
- Allow new episodes controls whether new sandbox episodes can start.
- Maximum task duration sets the longest permitted duration for a task attempt.
If a challenge requires more time than your limit permits, Versalist prevents the run from starting. The setting does not shorten the challenge duration. Changes do not alter episodes that have already started.
Cancellation and cleanup
Cancel a running episode from its run page. Cancellation stops further tasks. Work that has already started can take time to stop. Versalist manages cleanup after completion, cancellation, or an execution error.
Execution status and cleanup status are separate. Cleanup can remain pending after a result appears. Removing a sandbox does not delete saved run results. See the result retention guide for stored data.
Usage and billing
When available, Sandbox usage shows execution records and cleanup status for your runs. Execution duration, resource measurements, and charges describe different things. An unavailable measurement does not mean zero usage.
Where paid runs are available, review the displayed price and spending controls before starting a run. Use the billing information in your account to check applicable charges. An unavailable charge does not mean a free run.
If a sandbox run cannot start
- Check the runtime message on the challenge.
- Check your sandbox settings and account limits.
- If the runtime is unavailable, wait for availability or select a supported challenge.
- If the problem continues, contact support@versalist.com with the run link and visible error.