What the work involves

You write single-turn prompts in radiological source security and label each one benign, dual-use, or adversarial. Then you evaluate what the model returns against a defined policy standard — did it answer the legitimate dose or shielding question fully, and did it decline the request that was fishing for concealment or defeat of a physical protection layer? For each item you also author the reference answer: what a correct response should contain, and the technical reasoning behind it. Tasks are self-contained; most contributors work through them in batches rather than on a fixed schedule.

The domain sits on the dual-use line more than most. The isotope in a teletherapy head is the isotope in an industrial radiography camera. The shielding calculation that keeps a worker under limits is the calculation that hides a source from a portal monitor. A model that refuses a routine occupational dose question has failed as clearly as one that walks a user through attenuating a source in transit. Your prompts are meant to sit exactly on that seam, and your grading is meant to say — in writing — which side of it the response landed on.

What the platform screens for

  • Operational security exposure, not licensing paperwork. Screeners probe whether you have actually assessed how Category 1 and 2 sources get stolen, shielded, or moved: vulnerability assessments, detect-delay-assess-respond design, insider threat, source recovery and consolidation, 10 CFR 37 or national-equivalent security inspection, or transport security and chain of custody.
  • Calibration in both directions. Expect scenarios where the model over-refused, and you will be asked why that is a failure.
  • Writing. Rationales must be legible to a non-specialist reviewer. Prior technical writing, published research, or expert witness work is a strong signal — include a sample or link.
  • Clean-source discipline. The work must not involve or draw on classified or export-controlled material, NDA-covered content, or anything subject to prepublication review. Holding such obligations does not disqualify you; disclosing them lets the work be scoped.

Logistics

Fully remote and asynchronous. You will be reading and writing about misuse scenarios in your field for sustained stretches; Mercor briefs experts on this in advance and you can pause or step away at any point without penalty. Pay is stated as observed per task at $65–75, not guaranteed, and varies with task complexity and review outcomes.