Prepare a Harness for Trialground
Prepare an exact runnable candidate before evaluating it against a Benchmark Version.
Prerequisites
A Harness Draft, the intended Case input contract, and access to the required execution settings.
Steps
- Inspect the Harness files, entry point, declared Python dependencies, and how it consumes Case input and materials.
- Use the editor’s validation and debug workflow with supplied inputs to identify packaging or execution problems.
- Publish the intended Draft as a saved Harness Version. Retain the exact Version identity for the candidate.
- Open Benchmark Evaluations, choose the intended Benchmark Version, and configure managed execution for that saved candidate.
Object and state changes
Publishing creates a saved candidate boundary. Debugging supplied inputs and running the benchmark are separate operations.
Success criteria
The evaluation identifies an exact Harness Version with compatible dependencies and the intended input architecture.
Common failure modes
Trying to evaluate an editable Draft; assuming a successful debug input establishes benchmark performance.