Prepare a Dataset for testing

Bring samples and labels into Vault with a clear task and version.

Use this workflow when a Dataset arrives from a labeling tool, public source, or another team. Inspect the source before treating its labels as the reference for a test.

Best for

Practitioners bringing their own data
Teams preparing repeatable comparisons

What you’ll need

The retained sample files and annotations
A known evaluation task

What you’ll keep

A checked Dataset version
A compatible Model or verified API profile

Try this with the assistant

Add the relevant saved records to your chat, then adapt this request to your work.

Help me inspect this Dataset for testing. Identify the task, annotation formats, classes, splits, missing labels, and import warnings. Explain any proposed conversion before preparing an import. Do not infer missing reference labels.

Identify the task and source layout

Choose classification, detection, segmentation, oriented boxes, or the supported tabular task that matches the question. Compare the source with the format guide.

Review the import interpretation

Inspect counts, warnings, and annotation options. When several formats are usable, select the intended annotations. Review conversions before accepting a different Ground Truth meaning.

Inspect representative samples

Check class names, annotation boundaries, and split membership. Keep known-negative samples distinct from unfinished labels. Use Quality and the labelling studio when correction is needed.

Pin the version for the Run

Select the reviewed Dataset version with a compatible inference source in a saved Test plan’s Run. Later data changes use Update data so the earlier test remains identifiable.