What the work involves

You will spend most of your time reading model output and deciding whether it would survive contact with an actual courtroom. That means checking whether a suppression argument correctly tracks the Fourth Amendment posture, whether a hypothetical plea agreement reflects real charging and sentencing dynamics, whether hearsay and confrontation issues are handled properly, and whether procedural sequencing — arraignment, discovery obligations, pretrial motions, notice requirements — is described accurately rather than plausibly. Alongside evaluation, you'll author reference material: fact patterns with worked analysis, drafted motions with explanations of the strategic choice behind each argument, and written rationales that make your reasoning legible to non-lawyer reviewers.

A recurring task type is comparative grading — you see two or more model responses to the same legal question and rank them, then justify the ranking in writing. Another is failure documentation: identifying where a response is confidently wrong, mislabels a standard of review, invents authority, or crosses an ethical line, and describing precisely why. Jurisdictional variation matters here; expect to flag when an answer is written as if one state's rule were universal.

What the screening looks for

micro1 runs an AI-led interview before any project assignment. It is checking for a verifiable active bar license, real criminal defense practice depth (not adjacent litigation experience), and the ability to hold up under follow-up questioning on doctrine you claim to know. Expect the interviewer to push a second and third layer on any answer — naming a standard is not enough, you'll be asked to apply it to a modified fact pattern. Writing quality is assessed directly, since much of the deliverable is prose explanation.

  • Active license in good standing, verifiable
  • Concrete case and motion experience you can describe without breaching confidentiality
  • Clear reasoning under time pressure, in writing
  • Judgment about what makes an answer wrong versus merely incomplete

Logistics

  • Fully remote, contractor engagement, no relocation
  • Asynchronous for most task work; occasional live calibration sessions with project leads and other reviewers
  • Typical commitments run 10–20 hours per week, though scope varies by project and some tracks are lighter
  • Project-based, so duration and volume can change; treat it as supplemental rather than replacement income
  • No prior AI or machine learning background expected — the training is on the annotation guidelines, not the models