Our tests must sometimes sleep so that some condition is met for an ensuing assertion. Langchain4j uses Awaitility, for example:

awaitUntilAsserted(() -> assertThat(getAllEmbeddings()).hasSize(3));

// awaitUntilAsserted is:
Awaitility.await()
        .atMost(Duration.ofSeconds(60))
        .pollDelay(Duration.ofSeconds(0))
        .pollInterval(Duration.ofMillis(300))
        .untilAsserted(assertion);

We don't necessarily need to use Awaitility - it should be straightforward to implement a similar method that catches failed assertions for some period of time, and then fails. This should reduce test run times, and clarify what the tests are doing. This would mainly benefit our nondeterministic tests that have race conditions we can't get rid of.