The test has multiple issues that make it hard to investigate the failures (e.g. see BF-30780):

• it runs multiple tests in a loop without logging the context for each iteration so it's not clear which iteration is failing
• it uses randomized dynamically generated datasets for its inputs without logging them (the purpose of randomizing the datasets isn't clear)
• it uses more complex pipeline than just $unionWith to enable validation on the dynamically created dataset

At the minimum the test should be updated to log the inputs/context, but I'd recommend rewriting it to simplify and to make the intent more explicit. Doing so might also allow to remove some of the tags.