What the work actually involves
You build transactional fact patterns and then judge how well a model handles them. A task might start with a term sheet for a mid-market carve-out and ask the model to flag the indemnity gaps, or hand it a data room summary and ask what escalates to the board. You write the scenario, write a reference answer that states the legal reasoning behind each drafting and diligence call, and then grade model output against it. The grading is where most of the value sits: labs care less about a score than about a written explanation of why a confident-sounding answer would have exposed a client — a survival period that doesn't match the rep, a materiality scrape applied where it wasn't bargained for, a consent that was never actually obtained.
What the platform screens for
- Verifiable practice history. JD, active US bar admission, and at least two years of corporate or M&A work you can describe at the deal-mechanics level.
- Hands-on document exposure. Purchase agreements, disclosure schedules, diligence memos, board consents and minutes — not adjacent work you supervised from a distance.
- Reasoning you can write down. Screeners probe whether you can articulate why a provision reads the way it does, not just that it does.
- Calibration. Whether you can distinguish an answer that is wrong from one that is defensible but not what you would have drafted. Over-penalising stylistic difference is as much a problem as missing a real risk.
Logistics
Fully remote and asynchronous — no standing calls, no client hours. You pick up tasks from a queue and turn them around on your own schedule, with a floor of roughly 10 hours per week. Volume flexes week to week, and contributors commonly scale down during closings and back up afterward. Pay is weekly via Stripe at a rate set from your experience; the $100–200/hr band is what has been observed, not a guarantee. Expect a written sample or graded trial task as part of onboarding.