We have a very fine line we want the agent to walk here. In these tests we want the agents to be willing to step around small issues that a human would deal with, like a UI element having a different name, and note it. However, if a test is blocked due to a failure we want it to stop and mark the test failed. This is pretty nuanced.