What the work involves

This is comparative benchmarking, not content writing. You receive a set of prompts covering Japanese corporate and commercial law — Companies Act mechanics, share transfer restrictions, tender offer and disclosure duties under FIEA, due diligence findings, share purchase and joint venture drafting questions — and run each one through two LLM platforms. You then place the outputs side by side and score them against a standardized rubric, with a short written rationale for each score. The written feedback is the substance of the deliverable: reviewers are expected to name the specific legal error, cite the provision or practice point the model missed, and explain why a Japanese practitioner would or would not rely on the answer.

The judgment being asked of you is practitioner judgment applied to text. A model answer can be doctrinally accurate and still useless in a deal context; it can also read fluently while quietly importing Delaware or English law assumptions into a Japanese structure. Distinguishing an outright misstatement from an incomplete-but-defensible answer, and both from an answer that is technically correct yet would embarrass you in front of a client, is the core skill.

What the platform screens for

  • Verifiable practice history — firm names, years, and the kind of transactions you actually staffed, not a description of the practice area in general terms.
  • Depth under follow-up — the AI screener will push on a specific area you claim, so nominated strengths should be ones you can discuss at the level of provisions and negotiated positions.
  • Rubric discipline — willingness to score what the rubric measures rather than substituting your own preferred standard, while flagging where the rubric fails to capture a real defect.
  • Written precision in Japanese and English — feedback must be compact and unambiguous.

Logistics

Fully remote and asynchronous. The initial engagement is around 10 hours, with stated potential for continuation depending on batch volume and reviewer quality. Observed pay for this listing is $140–150/hr; bands are as posted by the platform and are not guaranteed. All work sits under NDA, and prompts and model outputs may not be retained or reused.