Skip to main content

​
Create an evaluation

An evaluation is a durable place to define the behavior you want to measure, the cases that exercise it, and the checks that decide what counts as good.

​
Step 1 — Open the builder

Open Evaluations, then choose New evaluation. The creation page is a builder, not a one-field dialog. At the top of the working surface, enter:
  • Evaluation Name — required.
  • Description — optional.

​
Step 2 — Add at least one check

Use Add a check on the left. You can click or drag a template onto the working surface. Templates are a starting point, not a requirement to accept their defaults. Choose a check that matches the behavior you actually care about. The builder will not create the evaluation until it has a name and at least one check.

​
Step 3 — Configure the checks

Open a selected check’s settings to review its Overview, Criteria, and Settings. Keep the requirement specific enough that a future regression has an interpretable reason for failing. Then choose Create evaluation. EvalGate creates the evaluation and opens its detail page. Model-backed checks still require a connected AI provider.

​
Step 4 — Finish the runnable setup

On the evaluation detail page, Overview contains the current test cases and the run setup. Add at least one case and, when the page says configuration is required, bind the approved executor/target/evaluator before running. See Add test cases for the supported case paths.

​
Up next

Run the evaluation.