@diegohaz any ideas from the top of your head why tests could be flaky on these rare occasions, and a sleep() before press.Tab() makes them more stable? Is this something that we could have addressed in @ariakit/test instead?

All items within an Ariakit composite widget remain tabbable until they are completely processed internally (for example, to determine their order). After that, only the active item stays tabbable. This happens asynchronously after the initial render.

This is to guarantee that all elements are accessible to keyboard users until we can respond to arrow keys.

A potential solution could be to wait for inactive items to become non-tabbable before hitting the Tab key during the test.