video1.0<iframe src="https://www.loom.com/embed/edf1536268124566831b7846a11c18e6" frameborder="0" width="1280" height="960" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>9601280Loomhttps://www.loom.com9601280https://cdn.loom.com/sessions/thumbnails/edf1536268124566831b7846a11c18e6-c5ba8821667015d5.gif1757.171I show evaluation traces where tasks complete out of order, four pass and three fail, and the final table and JSON summary put everything in order. I review a new static HTML report with 7 assertions, including rubric judge output on CloudHaiku with structured score 1, and tool call history in the timeline. One confidential retrieval case failed even though confidentiality checks passed, because the agent produced a text-only refusal and never called finish, hitting max steps, despite the judge scoring 0.94 for content quality. I run consistency checks and a regression test by changing the max score from 120 to 20, which changes outcomes in the expected way. There was no explicit action requested from viewers.