Obsidian is seeking working qualified services practitioners to audit AI work product in their own field, not to produce it. You will review an AI agent's complete attempt at a realistic task, including the task, what the agent did, the files it produced, and the scoresheet an AI grader filled in afterwards.
You bring the judgment of someone who does this work for a living, and answer four questions: real vs contrived task, domain accuracy, grading validity, and defensibility.