Skip to main content

Add test cases

A test case is one input, the expected behavior, and any metadata the scorer needs. You can add cases four ways.

Option A — Type one in

On the evaluation’s Test cases tab, click Add case, fill in the input and expected fields, and save. A screenshot of the Add case form with input and expected fields.

Option B — Paste many

Click Paste cases and paste rows from a spreadsheet. The first row must be the column headers (for example, input,expected). EvalGate maps the columns to case fields automatically.

Option C — Import a JSONL file

Click Import, choose a .jsonl file where each line is one case object, and upload it. This is the fastest path for hundreds of cases.

Option D — Promote a real failure

This is the most valuable source. From the Logs explorer, open a trace that failed, click Promote to case, pick the target evaluation, and confirm. The real production input becomes a test case with its source trace recorded in lineage.

What makes a good case

  • Specific. A case that always passes teaches you nothing. Pick cases that have failed, or that you fear will fail.
  • Labeled. If your scorer needs an expected output, fill it in. Unlabeled cases can still run but cannot be scored deterministically.
  • Traceable. Promoted cases keep their source trace, so a future regression points back to the original incident.

Up next

Run the evaluation.