Data and connections

Build a clear input basis before asking a model for results.

Move from received data to an input basis you can use in a test. Review labels, choose how predictions are supplied, and keep the versions connected to the result.

Read the guide

Check the input basis

Import samples with an explicit task and version. Inspect the reference labels and publish reviewed corrections before relying on them for scoring.

Prepare a Dataset for testing

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

Prepare data

Inspect and correct Dataset labels

Turn quality findings into reviewed annotations and a new version.

Prepare data

Choose the prediction source

Verify the request and response mapping for a model behind an API, or preserve the producer and sample mapping when predictions already exist elsewhere.

Test a model behind an API

Verify the request and prediction mapping before running the Dataset.

Prepare data

Score predictions produced elsewhere

Evaluate existing outputs against an explicit reference and saved plan.

Test models

Put the prepared basis to work

Compare candidates using the same question and Dataset. Share the measurements with their source context and interpretation limits.

Compare model candidates on the same question

Use a common Dataset and test basis to examine candidate differences.

Test models

Share findings with their evidence

Give the recipient a result they can understand in context.

Share evidence