What the work actually involves

You work in a task loop. First you write a single-turn prompt in your own domain — post-blast residue interpretation, storage magazine compliance, licensee records review, scene reconstruction — and label it benign, dual-use, or adversarial. Then a frontier model answers, and you judge that answer against a written policy standard: did it help a legitimate questioner fully, did it refuse something it should have refused, did it refuse something it had no business refusing? Finally you write the reference answer and the technical reasoning behind it, in prose a policy reviewer with no chemistry background can follow.

The hard part is not the chemistry. It is that a residue signature that closes a case reads in reverse as a formulation note, and that the diversion indicators an inspector is trained to spot are also a list of behaviours to avoid. An over-refusing model fails the forensic examiner; an over-helpful one fails everyone else. Calibrating that line, item by item, with a rationale attached, is the deliverable.

What the screen looks for

  • Verifiable casework under a national certification — scenes worked, samples identified, licensees inspected, testimony given. Expect follow-ups on instrumentation, chain of custody, and what you actually did versus what your unit did.
  • Evidence you can write. Published papers, lab reports, court-admissible findings, or standards documents are all strong; the platform asks for a sample or link and reads it.
  • Judgment under ambiguity — whether you can articulate why a request sits on one side of the line rather than gesturing at a gut feeling.
  • Clean disclosure of classified, export-controlled, NDA or prepublication obligations. Holding them is not disqualifying; failing to raise them is.

Logistics and conditions

Fully remote and asynchronous, paid per accepted task in an observed $65–75 range — rates are set by the platform per project and are not guaranteed. Volume fluctuates with batch availability; most contributors treat this as part-time alongside existing work. The content is sustained exposure to misuse scenarios in your own field, which Mercor briefs on in advance; you can pause or leave a batch without penalty.