Environments and data
Build and test an environment
Build a small environment around an inspectable task, explicit expected outcomes and a rubric that covers failure cases. This SAVRN Cloud walkthrough connects local configuration to a sample evaluation and identifies the versioned components a future executable package must preserve.
SAVRN Cloud · Coming soon · Public preview · Browser-local simulation · No live compute or payments
Assemble a local task
Open Datasets and choose a synthetic collection. In Environments, use Configure to review the rubric, threshold of 50–100 and timeout of 10–120 seconds. Changes increment its local version. Use Run to connect the environment to a demonstration evaluation; Export configuration retains the sample setup.
The following manifest is an illustrative documentation example. It is not an import format or a released API schema:
{
"simulation": true,
"schema_status": "illustrative",
"name": "Fictional note extraction",
"task_revision": "demo-1",
"dataset": "synthetic-notes",
"scoring": "exact fields with documented exceptions",
"network": "none required"
}
Review the outcome
Inspect the local evaluation result and ask whether each case has a clear expected outcome. Include at least one missing-field case and one ambiguous case in the planned fixture. Record how the rubric handles each instead of relying on an unexplained aggregate score.
The application demonstrates record relationships and sample results. It does not execute this manifest or validate a package format.
Before connected services launch
Publish a supported environment contract, versioned example package and real validation procedure. Pin datasets, harness code, dependencies and verifier versions. Test repeatability, resource bounds, malformed inputs and cleanup in the qualified runtime. Provide commands only for tooling that actually exists and has been released. Keep the baseline result and known scoring limitations with the package.
Build toward the work that matters.
Tell SAVRN what your institution needs to run, who reviews the results and where its data must stay. That workload defines the next service to qualify.
Discuss an AI project