What the work actually looks like

You work in discrete tasks rather than shifts. A task typically means authoring a prompt in your area of radiological practice — dose projection, shielding, source security, contamination control, protective action decision-making — and tagging it as benign, dual-use, or adversarial. You then read the model's response and judge it against a written policy standard: did it answer a legitimate operational question fully, hedge a routine one into uselessness, or hand over something it shouldn't have? Finally you write the reference answer: what a correct response contains, and the technical reasoning behind it.

The interesting part is the middle category. A shielding calculation that protects a radiographer is the same calculation that conceals a source in transit; a Gaussian plume estimate that drives an evacuation radius also describes an intended effect. Over-refusal is scored as a failure here, not as caution. That is why the panel is built from people who have held radiological response as a job responsibility — facility or state emergency planning, FRMAC or RAP field deployment, CBRN response, consequence modelling, incident command, or medical and public-health population monitoring — rather than from people who follow the subject with interest.

What the screen is looking for

  • Specific operational history. Which plan, which exercise, which deployment, what your role was, what decision you owned. Named frameworks and regulatory context help.
  • Calibration on the dual-use line. Can you articulate why a given question is routine and a near-identical one is fishing — and can you hold that distinction under follow-up rather than retreating to "refuse everything sensitive"?
  • Writing evidence. Technical reports, published research, expert witness work, drafted plans. Include a sample or link; this is writing-intensive work and the rationale is the deliverable.
  • Clean-source discipline. You must be able to work entirely from unclassified, non-export-controlled, non-NDA material. Existing obligations don't disqualify you — undisclosed ones do.

Logistics

Fully remote and asynchronous, paid per task in an observed band of $65–75 per task (rate varies by task complexity and is not guaranteed). Volume is project-driven rather than a fixed weekly commitment; experts who deliver clean rationales tend to get offered more batches. The work involves reading and writing about misuse scenarios in your own field for sustained periods. Mercor briefs experts on this before onboarding, and you can pause or step away at any point without penalty.