What the work actually is

You write single-turn prompts in your own domain and label each one benign, dual-use, or adversarial. Then you read the model's response and judge it against a written policy standard — did it answer a legitimate inspector's question fully, or did it hedge into uselessness? Did it decline a request that was fishing for concealment technique, or did it walk the requester through it? Finally, you write the reference answer: what a correct response looks like, and the technical reasoning behind it. A typical task bundles all three steps, and the rationale carries as much weight as the verdict.

The domain is unusually hard because the same knowledge sits on both sides. Material balance, MUF evaluation, measurement uncertainty, Cf-252 shuffler and HRGS interpretation, seal verification, enrichment cascade behaviour — every one of these is routine professional content and every one is adjacent to a concealment or diversion pathway. Your value to the panel is knowing which side a given phrasing falls on, and being able to explain that distinction to a policy reviewer who has never set foot in a facility.

What the screen looks for

  • Verified nuclear material, not read about it. Inspection planning and execution, facility declarations and ICRs, complementary access under the Additional Protocol, unattended monitoring, NDA measurement campaigns — the screen probes for operational specifics and follows up when answers stay abstract.
  • Dual-use judgment. You will be asked to reason aloud about where you would draw a line and why, including cases where reasonable practitioners disagree.
  • Written rationale under a non-specialist reader. Prior technical writing, published work, or expert witness experience is a strong signal; have a sample or link ready.
  • Clean information boundaries. The work must not touch classified or export-controlled material, NDA-covered content, or anything under prepublication review. Holding such obligations is not disqualifying — disclosing them up front is required so the work can be scoped.

Logistics

Fully remote and asynchronous, paid per completed task rather than per hour, with observed rates in the $65–75 band and no guaranteed volume. Most contributors work in blocks of a few hours and set their own schedule; sustained throughput matters more than any fixed weekly commitment. The role involves reading and writing about misuse scenarios in your field for extended periods — Mercor briefs experts on this in advance, and you may pause or step away at any point without penalty.