What the work actually is

You write single-turn prompts in your own area of synthetic or process chemistry, each labelled at one of three levels: benign (a routine professional question a model should answer in full), dual-use (legitimate on its face, but the answer edges toward capability a bad actor would want), and adversarial (a request that should be refused, ideally dressed up to look ordinary). You then read the model's response, judge it against a written policy standard, and produce a reference answer — the correct response plus the technical reasoning for why that is the correct line. Expect a mix of route design, kinetics, impurity profiling, scale-up hazard analysis, precursor questions and regulatory framing.

The hard part is not the chemistry. It is that over-refusal counts as failure. A model that won't discuss exotherm management on a nitration scale-up, or that stonewalls a routine genotoxic impurity question, is broken in a way that matters to working chemists. Your prompts are useful precisely when they sit on the line, and your grading is useful when it explains where the line falls and why a specific detail — a reagent, a purification step, a yield-critical parameter — moves a response from acceptable to not.

What the screen looks for

  • Hands-on route and scale-up experience. Bench to pilot to plant, with the safety and process-hazard review that goes with it. Expect follow-ups on specific chemistry you claim: reagent choices, why a route was abandoned, what the impurity profile told you.
  • Calibrated misuse judgment. Can you distinguish a genuinely routine question from one fishing for uplift, and articulate the difference in writing rather than by instinct?
  • Writing under scrutiny. Rationales must be legible to a non-specialist reviewer. Prior technical writing, publications or expert-witness work is a strong signal; include a sample or link if you have one.
  • Clean information provenance. The work must not touch classified, export-controlled, NDA-bound or prepublication-review material. Holding such obligations doesn't disqualify you — declare them and the work gets scoped around them.

Logistics

Fully remote and asynchronous, paid per task in an observed $65–75 range (rates vary with task type and are not guaranteed). Most contributors work in batches of a few hours; sustained throughput matters more than availability at fixed times. You will be reading and writing about misuse scenarios in chemistry for extended stretches — Mercor briefs experts on this in advance and contributors can pause or step away without penalty.