The work

You will spend most of your time reading model outputs on tax questions and deciding whether they would survive contact with a real filing. That means checking whether a stated deduction limit reflects the current year's rules, whether an S-corp reasonable-compensation answer is defensible, whether a filing-deadline or extension answer accounts for state variation, and whether the model's chain of reasoning arrives at the right number for the wrong reason. Feedback is written, specific, and cites authority where possible — a score with no explanation is not a deliverable.

Alongside grading, you may prepare or review sample returns used as ground truth, draft prompts that probe known weak spots, and help refine the rubric itself when it fails to separate a competent answer from a confidently wrong one. Some tasks ask you to critique advisory language for clarity, not just correctness — whether a taxpayer could actually act on what the model said.

What the screen looks for

micro1 runs an AI-led interview, so expect follow-up questions that push on whatever you claim. The screen checks that your CPA license is active and verifiable, that your return-preparation experience genuinely spans both individual and pass-through entity work, and that you can explain a technical rule cleanly under time pressure. Vague answers about "handling small business clients" tend to stall; naming forms, thresholds, elections, and the states you actually practice in does not. Prior AI evaluation experience is a plus, not a gate.

Logistics

  • Contractor, part-time, remote — US only, with authorization to practice in the US
  • Asynchronous task queues; you set your own hours, though throughput expectations are real
  • Observed pay band of $80–110/hr, typically varying with credential depth and task type — stated as observed, not guaranteed
  • Volume fluctuates with client demand and can be seasonal against filing deadlines