Test robustness
Examine how performance changes under selected conditions.
Robustness testing examines an input or operating change while retaining the source basis for comparison. Choose changes that are relevant to the product’s intended use.
Define the concern
Describe the condition and why it matters: weather, image quality, spatial changes, adversarial conditions, or supported tabular changes. Available tests depend on the task and your organization’s capabilities.
Prepare the comparison
Save a Test plan with the reason for each test and its settings. Select the Dataset version and inference source, then review the baseline and changed conditions in the Run package. Use confirmed criteria when the work supports acceptance.
Examine changes in context
Inspect the resulting scores and examples. Retain the clean reference, transformation settings, sample coverage, and any omitted tests. A score under one selected condition supports a conclusion about that tested configuration.
Choose a fixed test or a search
Use a Run for the specified tests and conditions. Use a Stress Test when you want to search one numeric condition parameter for a degradation boundary within a budget.
Record the finding and its limits in the report or Assessment review.