What the work actually involves

You work in units of tasks, not shifts. A task typically means authoring one single-turn prompt in your area of practice — isotopic age dating, impurity and process signatures, portal monitor response, attribution methodology — assigning it a label across three tiers (benign, dual-use, adversarial), then evaluating the model response against a written policy standard and producing the reference answer: what a correct reply contains, what it withholds, and the technical reasoning for both. The hard part is rarely the physics. It is deciding whether a question about a detection threshold is a routine sensitivity question from a measurement professional or a request framed to establish what stays under the alarm, and then defending that call in prose.

This domain sits on the dual-use line more than most. The signature that attributes a sample is the signature someone would want to suppress. A model that refuses a textbook question about mass-spectrometric methodology is failing as clearly as one that helps with evasion, and the panel exists because generalist annotators cannot tell those two failures apart. Expect your labels to be contested in calibration review and expect to defend them with reference to the policy text, not intuition.

What the screen looks for

  • Hands-on characterisation of real nuclear material — casework, international round-robin exercises, or national laboratory research, pre- or post-detonation.
  • The ability to articulate why a given request falls where it does, in language a policy reviewer without a radiochemistry background can follow.
  • Written output under your own name: publications, technical reports, expert witness work. Have a sample or link ready; the screen asks for it.
  • Clean scoping around classified, export-controlled, NDA-bound or prepublication-review material. The work must not draw on any of it. Holding such obligations does not disqualify you — concealing them does. Declare them and the work gets scoped around them.

Logistics

Fully remote and asynchronous, with no fixed hours. Contributors generally commit a defined number of tasks per week rather than blocks of time; task rates observed in this band run $65–75, and are set by the platform, not guaranteed. The work involves sustained reading and writing about misuse scenarios in your field. Mercor briefs experts on this before onboarding, and you can pause or withdraw at any point without penalty.