MASAIC
Sign In

Data sets

Named collections of conversations you evaluate against in batch — ground-truth transcripts for testing at scale.

A data set is a named collection of conversations — the batch inputs and ground-truth transcripts you evaluate an agent against, so you can measure quality across many cases at once instead of one chat at a time.

Create a data set

  1. In the sidebar, open Testing → Test Sets.
  2. Click Create Data Set.
  3. In the dialog, enter a Name and an optional Description.
  4. Click Create Data Set to save it.

Review the conversations in a data set

  1. From the list, click a data set to open it.
  2. The detail page lists the conversations it contains, with a search box to find one by ID.
  3. Click any conversation to open its transcript in a side panel and read the user and assistant turns.

Use a data set in a test

  1. Open Testing → Tests and create or edit a test case.
  2. Attach a data set so evaluation runs across every conversation it holds.
  3. Run the suite in staging to score the batch before you promote.

Tip: Personas and mock users shape a single run; a data set is how you evaluate many recorded conversations at once. Build data sets from real captured threads so scores reflect production behavior.