What the work involves

You are the standard the model is measured against. On a typical assignment you will be handed a prompt and one or more model responses — say, a request to draft an earn-out provision, spot the problems in an indemnification cap, summarize the change-of-control triggers in a credit agreement, or advise a founder on a down-round protective provision — and asked to judge whether the output is something you would let leave your office. That means grading on correctness, but also on the things that separate a second-year's memo from a partner's: whether the response flags the deal risk that wasn't asked about, whether it hedges when the law is genuinely unsettled, whether the defined terms actually track. Other assignments run the other direction: you write the gold-standard answer or the adversarial prompt that a model is likely to fail.

Work is issued in batches with rubrics, and your written rationale matters as much as your score. A rating with no reasoning is unusable to the team training the model; a rating that says "the survival period conflicts with the escrow release schedule in §2.4, which makes the indemnity illusory" is exactly the signal they pay for.

What the platform screens for

  • Verifiable transactional depth. Firm names, deal types, deal sizes, and your actual role on them. Screens probe past the résumé line — expect follow-ups on specific negotiated points.
  • Willingness to be specific about a document. Vague answers about "M&A experience" score poorly against answers that name the provision and explain why it was fought over.
  • Evaluation judgment. Can you distinguish a wrong answer from a merely differently-drafted one? Over-penalizing stylistic variance and under-penalizing confident fabrication are both failure modes.
  • Calibration and honesty. Saying "this is jurisdiction-dependent and I'd only be confident on Delaware" is a strength, not a hedge.

Logistics

Fully remote and largely asynchronous, structured as contract work rather than employment. Most contributors take 5–20 hours a week around a practice or in-house role; a batch may carry a turnaround window of a few days, but you choose when inside it. There is no billable-hour target and no client contact. Expect an onboarding calibration exercise before paid work begins, and expect your early submissions to be spot-checked more heavily than later ones.