For experts · async only
Judgment work, on your hours.
Bench pays practising lawyers to evaluate AI outputs against their jurisdiction's law — scoring, correcting, and writing the gold standards that AI teams pay to be measured against. Everything is asynchronous. No calls, no meetings, no public profile, ever.
The work
What an evaluation task looks like.
You receive a batch of AI-generated answers to legal questions in your jurisdiction and practice area. For each one you score five rubric dimensions, classify the failure, and — where the answer fails — write the correct one, grounded in primary authority. A typical task takes 20–40 minutes of focused legal judgment.
Compensation is per task, at rates reflecting the judgment involved (full evaluations pay more than verification passes). Payouts run weekly by bank transfer or Wise. We never pay in crypto.
The process
Vetting, without the theatre.
| 01 | Apply with credentials and availability — the form, nothing else |
| 02 | Paid test task ($100 flat): evaluate 5 sample outputs with our rubric, 72-hour window |
| 03 | Calibration round — your scores are compared with a second expert's; agreement is computed, not vibes |
| 04 | Admission decided by a human. Early tasks are fully QA-reviewed, then sampling drops as your track record builds |
AI assists our assessment scoring; admission and retention decisions are always made by a person. Reports show your credential class (e.g. “UK data protection · 6y PQE”) — never your name. See the methodology for how calibration works.