Skip to content

Generate Test Cases from Atlas

2 min read

The AI test case generator writes from requirements, context and Canon documents. Atlas adds the one thing none of those supply: the product’s shape around the subject.


  1. Select a node on the map and open its inspector.

  2. Select the generate action — Generate test cases for this node, with the map as context.

  3. The generator opens with the node’s requirements already selected and a preamble describing the node’s position in the map.

  4. Continue as normal — pick your focus lenses, map coverage, review and add. See The Coverage Map.


Atlas does not generate anything itself. It hands the existing generation pipeline two things:

  • The requirements the node covers, so the run is grounded in what was asked for and knows what is already covered.
  • A preamble describing the product’s shape — where the node sits in the containment tree, what routes reach it, and what depends on it.

That second part is the difference. A generator told “test the payment form” writes generic payment tests. A generator told “test the payment form, which sits under Checkout, is reached from the cart, and is depended on by the receipt page” writes tests that cross those seams.


A journey supplies an ordered path rather than a single subject: the steps in sequence, the persona walking them, and the nodes each step lands on.

This is the right way to produce end-to-end cases. A journey-seeded run knows the whole walk, so it can write a case that crosses four screens instead of four cases that each test one.


An Atlas-seeded run is grounded in Canon like any other, and the Ground in Canon toggle behaves the same way. Atlas supplies the structure; Canon supplies the stated behaviour.

The two together are the most complete context the generator can get: what the product is made of and what it promises.


Generated cases go through the usual review flow and land in the folder you choose, exactly as any other run. They are ordinary test cases afterwards.

To have them count toward the node’s coverage, link them — or link the folder they landed in — from the node’s inspector. See Nodes and routes.