What the work actually involves

You write insurance problems that look like the ones on your desk, then judge how well a model handles them. A typical task might be a commercial property submission with an incomplete loss run and an ambiguous protective safeguards endorsement, a first-party claim where the policy language and the adjuster's file notes disagree, a reserving question where the model needs to choose between methods, or a state-specific filing question where the federally correct answer is locally wrong. You supply the fact pattern, the defensible answer, and the reasoning path a competent professional would take. Then you read two or more model responses and explain, in writing, which is better and exactly where the weaker one went off — cited the wrong form edition, conflated occurrence and claims-made, invented a regulation, gave a coverage opinion the facts don't support.

The written critique is the product, not a formality. Vague notes like "inaccurate" are worthless to a lab; "the model applied the ISO CG 00 01 04 13 employer's liability exclusion to an independent contractor, which the endorsement carves out" is what gets paid for.

What the platform screens for

  • Verifiable practice history — where you worked, what lines, what authority level, what you personally decided versus what you reviewed. The screen follows up, and generic answers collapse fast.
  • Depth in a named specialty rather than broad familiarity. Mercor would rather have someone who knows workers' comp adjudication in three states cold than someone who has heard of everything.
  • Evaluation judgment — whether you can separate a confidently-written wrong answer from a plainly-worded right one, and whether you know which errors are cosmetic and which would cost a carrier money or a licence.
  • Jurisdictional honesty — willingness to say "that varies by state and here's how I'd check" instead of bluffing.
  • Credentials (CPCU, ARM, AINS, CLU, AIC, ACAS/FCAS, ASA/FSA, state adjuster or producer licences) are preferred, not required; three-plus years of real practice is the floor.

Logistics

Fully remote and asynchronous — no standing meetings, no set shift. Contributors generally work in blocks of a few hours and pick up tasks from a queue, so volume fluctuates with lab demand rather than arriving as a guaranteed weekly allocation. Most people do this alongside a full-time insurance job. The application is a resume or experience summary, a short form on practice area and certifications, and for selected applicants a brief sample task; follow-up usually lands within a few days. Pay in the $90–110/hr range reflects what contributors in this domain have reported and is not a guarantee.